Machine unlearning (MU) has emerged as a critical tool for removing sensitive or personal information from machine learning models, empowering individuals with the right to be forgotten. While MU has achieved success in classification and generative tasks, whether this technique can be effectively applied to segmentation foundation models remains uncertain. To address this issue, we propose an efficient method, Selective Concept Unlearning (SCU), to unlearn the segmentation capability of target concepts. SCU consists of several key aspects: (1) The Multi-level Forgetting Module, designed with a hierarchical three-level suppression strategy, including (i) distillation-level: Negative distillation steers model’s output distribution away from teacher’s correct outputs, erasing its learned concept recognition. (ii) attention-level: Attention suppression minimizes model’s attention to target regions. (iii) output-level: Directly erases predictions for the target by relabeling as background. (2) The Preservation Module ensures maintaining segmentation quality for non-target concepts. Additionally, we introduce a set of metrics to evaluate segmentation unlearning methods. Experiments demonstrate that SCU consistently outperforms existing baselines.

Recent studies have explored the capabilities of large language models (LLMs) in solving knowledge-intensive mathematical reasoning problems. However, existing benchmarks predominantly involve static theorems that LLMs have encountered during pretraining, failing to assess dynamic knowledge integration. In this work, we introduce TaxReasoning, a novel benchmark designed to evaluate LLMs’ abilities in real-world tax calculation scenarios. These tasks require not only mathematical reasoning and numerical computation, but also the extraction and application of complex, frequently updated tax regulations. Through extensive experiments with state-of-the-art LLMs using diverse prompting strategies and knowledge augmentation techniques, we uncover substantial limitations in their ability to handle dynamic, knowledge-intensive questions—primarily due to missing domain-specific knowledge and ineffective retrieval. Even the best-performing models fall significantly short of human-level performance. Our analysis points to key avenues for improvement, including enhancing LLMs' reasoning capabilities, developing more effective knowledge summarization techniques, and improving retrieval strategies. TaxReasoning offers a critical testbed for advancing LLMs in dynamic knowledge-intensive domains.

Recent advancements in large language models (LLMs) have demonstrated remarkable text generation capabilities. However, controlling specific attributes of generated text remains challenging without architectural modifications or extensive fine-tuning. Current methods typically toggle a single, basic attribute but struggle with precise multi-attribute control. In scenarios where attribute requirements conflict, existing methods lack coordination mechanisms, causing interference between desired attributes. Furthermore, these methods fail to incorporate iterative optimization processes in the controlled generation pipeline. To address these limitations, we propose Conflict-aware, Composite, and Collaborative Controlled Text Generation (C³TG), a two-phase framework for fine-grained, multi-dimensional text attribute control. During generation, C³TG selectively pairs the LLM with the required attribute classifiers from the 17 available dimensions and employs weighted KL-divergence to adjust token probabilities. The optimization phase then leverages an energy function combining classifier scores and penalty terms to resolve attribute conflicts through iterative feedback, enabling precise control over multiple dimensions simultaneously while preserving natural text flow. Experiments show that C³TG significantly outperforms baselines across multiple metrics including attribute accuracy, linguistic fluency, and output diversity, while simultaneously reducing toxicity. These results establish C³TG as an effective and flexible solution for multi-dimensional text attribute control that requires no costly model modifications.

Jiaqi Li, Zhijing Zhang, Jiahui Geng, Sheng Bi, Chuanyi Zhang, Fan Liu, Guilin Qi. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Table Retrieval (TR) has traditionally been formulated as an ad-hoc retrieval problem, where relevance is primarily determined by topical semantic similarity. With the growing adoption of LLM-based agentic systems, access to structured data is increasingly instruction-driven, where relevance is conditional on explicit content and schema constraints rather than topical similarity alone. We therefore formalize Instruction-Following Table Retrieval (IFTR), a new task that requires models to jointly satisfy topical relevance and fine-grained instruction constraints. We identify two core challenges in IFTR: (i) sensitivity to content scope, such as inclusion and exclusion constraints, and (ii) awareness of schema-grounded requirements, including column semantics and representation granularity--capabilities largely absent in existing retrievers. To support systematic evaluation, we introduce FollowTable, the first large-scale benchmark for IFTR, constructed via a taxonomy-driven annotation pipeline. We further propose a new metric, termed the Instruction Responsiveness Score, to evaluate whether retrieval rankings consistently adapt to user instructions relative to a topic-only baseline. Our results indicate that existing retrieval models struggle to follow fine-grained instructions over tabular data. In particular, they exhibit systematic biases toward surface-level semantic cues and remain limited in handling schema-grounded constraints, highlighting substantial room for future improvements.

Table Understanding (TU) has achieved promising advancements, but it faces the challenges of the scarcity of manually labeled tables and the presence of complex table structures. To address these challenges, we propose HeGTa, a heterogeneous graph (HG)-enhanced large language model (LLM) designed for few-shot TU tasks. This framework aligns structural table semantics with the LLM's parametric knowledge through soft prompts and instruction tuning. It also addresses complex tables with a multi-task pre-training scheme, incorporating three novel multi-granularity self-supervised HG pre-text tasks. We empirically demonstrate the effectiveness of HeGTa, showing that it outperforms the SOTA for few-shot complex TU on several benchmarks.

Nan Hu, Jiaoyan Chen, Yike Wu, Guilin Qi, Hongru Wang, Sheng Bi, Yongrui Chen, Tongtong Wu, Jeff Z. Pan. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

In e-commerce, effective product Attribute Mining (AM) is essential for enhancing product features and aiding consumer decisions.However, current AM methods often focus on extracting attributes from unimodal text, underutilizing multimodal data.In this paper, we propose a novel framework called Multimodal Self-Correction Instruction Tuning (MSIT) to mine new potential attributes from images and texts with Multimodal Large Language Models (MLLMs).The tuning process involves two datasets: Attribute Generation Tuning Data (AGTD) and Chain-of-Thought Tuning Data (CTTD).AGTD is constructed utilizing incontext learning with a small set of seed attributes, aiding the MLLMs in accurately extracting attribute-value pairs from multimodal information.To introduce explicit reasoning and improve the extraction accuracy, we construct CTTD, which incorporates a structured 5-step reasoning process for self-correction.Finally, we employ a 3-stage inference process to filter out redundant attributes and sequentially validate each generated attribute.Comprehensive experimental results on two datasets show that MSIT outperforms state-of-the-art methods.We will release our code and data in the near future.

Large language models (LLMs) have encoded vast amounts of knowledge in their parameters, but the acquired knowledge can sometimes be incorrect or outdated over time, necessitating rectification after pre-training.Traditional localized methods in knowledge-based model editing (KME) typically assume that knowledge is stored in particular intermediate layers.However, recent research suggests that these methods do not identify the optimal locations for parameter editing, as knowledge gradually accumulates across all layers in LLMs during the forward pass rather than being stored in specific layers.This paper, for the first time, introduces the concept of critical transmission paths into KME for parameter updating.Specifically, these paths capture the key information flows that significantly influence the model predictions for the editing process.To facilitate this process, we also design a parameter-aware contrastive rectifying algorithm that considers less important paths as contrastive examples.Experiments on two prominent datasets and three widely used LLMs demonstrate the superiority of our method in editing performance.

In a conversational system, dynamically generating follow-up questions based on context can help users explore information and provide a better user experience. Humans are usually able to ask questions that involve some general life knowledge and demonstrate higher order cognitive skills. However, the questions generated by existing methods are often limited to shallow contextual questions that are uninspiring and have a large gap to the human level. In this paper, we propose a three-stage external knowledge-enhanced follow-up question generation method, which generates questions by identifying contextual topics, constructing a knowledge graph (KG) online, and finally combining these with a large language model to generate the final question. The model generates information-rich and exploratory follow-up questions by introducing external common sense knowledge and performing a knowledge fusion operation. Experiments show that compared to baseline models, our method generates questions that are more informative and closer to human questioning levels while maintaining contextual relevance.

Taxonomy expansion is a primary method for enriching taxonomies, involving appending a large number of additional nodes (i.e., queries) to an existing taxonomy (i.e., seed), with the crucial step being the identification of the appropriate anchor (parent node) for each query by incorporating the structural information of the seed. Despite advancements, existing research still faces an inherent challenge of spurious query-anchor matching, often due to various interference factors (e.g., the consistency of sibling nodes), resulting in biased identifications. To address the bias in taxonomy expansion caused by unobserved factors, we introduce the Structural Causal Model (SCM), known for its bias elimination capabilities, to prevent these factors from confounding the task through backdoor paths. Specifically, we employ the Front-Door Criterion, which guides the decomposition of the expansion process into a parser module and a connector. This enables the proposed causal-aware Taxonomy Expansion model to isolate confounding effects and reveal the true causal relationship between the query and the anchor. Extensive experiments on three benchmarks validate the effectiveness of TEF, with a notable 6.1% accuracy improvement over the state-of-the-art on the SemEval16-Environment dataset.

Large Language Models (LLMs) have revolutionized various natural language processing tasks with their remarkable capabilities. However, challenges persist in effectively integrating new knowledge into LLMs without compromising their performance, particularly in the Large Language Models (LLMs) have revolutionized various natural language processing tasks with their remarkable capabilities. However, a challenge persists in effectively processing new information, particularly in the area of long-term knowledge updates without compromising model performance. To address this challenge, this paper introduces a novel memory augmentation framework that conceptualizes memory as a peripheral component (akin to physical RAM), with the LLM serving as the information processor (analogous to a CPU). Drawing inspiration from RAM architecture, we design memory as a sequence of memory banks, each modeled using Kolmogorov-Arnold Network (KAN) to ensure smooth state transitions. Memory read and write operations are dynamically controlled by query signals derived from the LLMs' internal states, closely mimicking the interaction between a CPU and RAM. Furthermore, a dedicated memory bank is used to generate a mask value that indicates the relevance of the retrieved data, inspired by the sign bit in binary coding schemes. The retrieved memory feature is then integrated as a prefix to enhance the model prediction. Extensive experiments on knowledge-based model editing validate the effectiveness and efficiency of our peripheral memory.

Virtual Knowledge Graphs (VKGs) provide an effective solution for data integration but typically require significant expertise for their construction. This process, involving ontology development, schema analysis, and mapping creation, is often hindered by naming ambiguities and matching issues, which traditional rule-based methods struggle to address. Large language models (LLMs), with their ability to process and generate contextually relevant text, offer a potential solution. In this work, we introduce LLM4VKG, a novel framework that leverages LLMs to automatize VKG construction. Experimental evaluation on the RODI benchmark demonstrates that LLM4VKG surpasses state-of-the-art methods, achieving an average F1-score improvement of +17% and a peak gain of +39%. Moreover, LLM4VKG proves robust against incomplete ontologies and can handle complex mappings where current methods fail.

Answer Set Programming (ASP) is a powerful paradigm for non-monotonic reasoning. Recently, large language models (LLMs) have demonstrated promising capabilities in logical reasoning. Despite this potential, current evaluations of LLM capabilities in ASP are often limited. Existing works normally employ overly simplified ASP programs, do not support negation, disjunction, or multiple answer sets. Furthermore, there is a lack of benchmarks that introduce tasks specifically designed for ASP solving. To bridge this gap, we introduce ASPBench, a comprehensive ASP benchmark, including three ASP specific tasks: ASP entailment, answer set verification, and answer set computation. Our extensive evaluations on ASPBench reveal that while 14 state-of-the-art LLMs, including deepseek-r1, o4-mini, and gemini-2.5-flash-thinking, perform relatively well on the first two simpler tasks, they struggle with answer set computation, which is the core of ASP solving. These findings offer insights into the current limitations of LLMs in ASP solving. This highlights the need for new approaches that integrate symbolic reasoning capabilities more effectively. The code and dataset are available at https://github.com/HomuraT/ASPBench.

Dehai Min, Zhiyang Xu, Guilin Qi, Lifu Huang, Chenyu You. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Continual Structured Knowledge Reasoning (CSKR) focuses on training models to handle sequential tasks, where each task involves translating natural language questions into structured queries grounded in structured knowledge. Existing general continual learning approaches face significant challenges when applied to this task, including poor generalization to heterogeneous structured knowledge and inefficient reasoning due to parameter growth as tasks increase. To address these limitations, we propose a novel CSKR framework, \textsc{K-DeCore}, which operates with a fixed number of tunable parameters. Unlike prior methods, \textsc{K-DeCore} introduces a knowledge decoupling mechanism that disentangles the reasoning process into task-specific and task-agnostic stages, effectively bridging the gaps across diverse tasks. Building on this foundation, \textsc{K-DeCore} integrates a dual-perspective memory consolidation mechanism for distinct stages and introduces a structure-guided pseudo-data synthesis strategy to further enhance the model's generalization capabilities. Extensive experiments on four benchmark datasets demonstrate the superiority of \textsc{K-DeCore} over existing continual learning methods across multiple metrics, leveraging various backbone large language models.

Uncertain knowledge graphs (UKGs) associate each triple with a confidence score to provide more precise knowledge representations. Recently, since real-world UKGs suffer from the incompleteness, uncertain knowledge graph (UKG) completion attracts more attention, aiming to complete missing triples and confidences. Current studies attempt to learn UKG embeddings to solve this problem, but they neglect the extremely imbalanced distributions of triple confidences. This causes that the learnt embeddings are insufficient to high-quality UKG completion. Thus, in this paper, to address the above issue, we propose a new semi-supervised Confidence Distribution Learning (ssCDL) method for UKG completion, where each triple confidence is transformed into a confidence distribution to introduce more supervision information of different confidences to reinforce the embedding learning process. ssCDL iteratively learns UKG embedding by relational learning on labeled data (i.e., existing triples with confidences) and unlabeled data with pseudo labels (i.e., unseen triples with the generated confidences), which are predicted by meta-learning to augment the training data and rebalance the distribution of triple confidences. Experiments on two UKG datasets demonstrate that ssCDL consistently outperforms the state-of-the-art baselines in different evaluation metrics.

Event causality identification (ECI) aims to identify causality between event pairs in a text, and is commonly approached as a supervised classification task using pre-trained language models (PLMs). However, limitations in implicit causality identification and insufficient event-knowledge interaction pose significant challenges to ECI. To address these issues, we propose a novel Knowledge Interaction Graph guided Prompt Tuning (KIGP), which leverages prompt tuning and knowledge interaction to fully exploit the potential of PLMs for ECI by integrating external knowledge. Specifically, to accurately capture implicit causality, we design the guidance mechanism and construct event-knowledge interaction graphs that enable external knowledge to enhance event representations through deep interaction between events and knowledge. Experimental results on two benchmark datasets demonstrate that our model outperforms existing approaches significantly.

When we integrate factual knowledge from knowledge graphs (KGs) into large language models (LLMs) to enhance their performance, the cost of injection through training increases with the scale of the models. Consequently, there is significant interest in developing prompt strategies that effectively incorporate KG information into LLMs. However, the community has not yet comprehensively understood how LLMs process and interpret KG information in different input formats and organizations within prompts, and researchers often rely on trial and error. To address this gap, we design extensive experiments to empirically study LLMs’ comprehension of different KG prompts. At the literal level, we reveal LLMs’ preferences for various input formats (from linearized triples to fluent natural language text). At the attention distribution level, we discuss the underlying mechanisms driving these preferences. We then investigate how the organization of structured knowledge impacts LLMs and evaluate LLMs’ robustness in processing and utilizing KG information in practical scenarios. Our experiments show that (1) linearized triples are more effective than fluent NL text in helping LLMs understand KG information and answer fact-intensive questions; (2) Different LLMs exhibit varying preferences for different organizational formats of triples; (3) LLMs with larger scales are more susceptible to noisy, incomplete subgraphs.

Open rules refer to the implication from premise atoms to hypothesis atoms, which captures various relationships between instances in the real world. Injecting open rule knowledge into the machine helps to improve the performance of downstream tasks such as dialogue and relation extraction. Existing approaches focus on single-hop open rule generation, ignoring scenarios involving multiple hops, leading to logical inconsistencies between premise and hypothesis atoms, as well as semantic duplication of generated rule atoms. To address these issues, we propose a progressive multi-stage open rule generation method called PRIMO. We introduce ontology information during the rule generation stage to reduce ambiguity and improve rule accuracy. PRIMO constructs a multi-stage structure consisting of generation, extraction, and rank modules to fully leverage the latent knowledge within the language model across multiple dimensions. Furthermore, we employ reinforcement learning from human feedback to further optimize model, enhancing the model’s understanding of commonsense knowledge. Experimental results demonstrate that compared to baseline models, PRIMO significantly enhances rule quality and diversity while reducing the repetition rate of rule atoms.

Recent studies have explored the use of Large Language Models (LLMs) with Retrieval Augmented Generation (RAG) for Knowledge Graph Question Answering (KGQA).They typically require rewriting retrieved subgraphs into natural language formats comprehensible to LLMs.However, when tackling complex questions, the knowledge rewritten by existing methods may include irrelevant information, omit crucial details, or fail to align with the question's semantics.To address them, we propose a novel rewriting method CoTKR, Chain-of-Thought Enhanced Knowledge Rewriting, for generating reasoning traces and corresponding knowledge in an interleaved manner, thereby mitigating the limitations of single-step knowledge rewriting.Additionally, to bridge the preference gap between the knowledge rewriter and the question answering (QA) model, we propose a training strategy PAQAF, Preference Alignment from Question Answering Feedback, for leveraging feedback from the QA model to further optimize the knowledge rewriter.We conduct experiments using various LLMs across several KGQA benchmarks.Experimental results demonstrate that, compared with previous knowledge rewriting methods, CoTKR generates the most beneficial knowledge representation for QA models, which significantly improves the performance of LLMs in KGQA 1 .* Equal contribution.†Corresponding author. 1 Our code is available at https://github.com/wuyike2000/CoTKR.Question: what is the unit of area that the measurement system that have an electric field strength units of volt per metre have?Answer: Square kilometer, Square meter KR Methods KR Results SummaryThe International System of Units includes various measurement systems for different physical quantities such as length, area, volume, conductance, resistance, current, force, and pressure.The measurement system with an electric field strength unit of Volt per meter does not directly relate to a specific area unit within the International System of Units.CoTKR Reason: I need to know the unit of area associated with the measurement system that has an electric field strength unit of Volt per meter.Summarize: The International System of Units includes various units for different physical quantities such as length, area, volume, etc. Reason: I need to find the specific unit of area in the International System of Units that corresponds to the electric field strength unit of Volt per meter.Summarize: The International System of Units includes units like Square kilometer, Square meter, etc. for measuring area.CoTKR+PA Reason: I need to know which measurement system includes the unit for electric field strength in Volt per meter.Knowledge: The International System of Units contains the unit for electric field strength in Volt per meter.Reason: I need to identify the unit of area within the International System of Units.Knowledge: The International System of Units includes the units Square kilometer and Square meter as area units.

Yongrui Chen, Haiyun Jiang, Xinting Huang, Shuming Shi, Guilin Qi. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Machine unlearning (MU) empowers individuals with the `right to be forgotten' by removing their private or sensitive information encoded in machine learning models. However, it remains uncertain whether MU can be effectively applied to Multimodal Large Language Models (MLLMs), particularly in scenarios of forgetting the leaked visual data of concepts. To overcome the challenge, we propose an efficient method, Single Image Unlearning (SIU), to unlearn the visual recognition of a concept by fine-tuning a single associated image for few steps. SIU consists of two key aspects: (i) Constructing Multifaceted fine-tuning data. We introduce four targets, based on which we construct fine-tuning data for the concepts to be forgotten; (ii) Joint training loss. To synchronously forget the visual recognition of concepts and preserve the utility of MLLMs, we fine-tune MLLMs through a novel Dual Masked KL-divergence Loss combined with Cross Entropy loss. Alongside our method, we establish MMUBench, a new benchmark for MU in MLLMs and introduce a collection of metrics for its evaluation. Experimental results on MMUBench show that SIU completely surpasses the performance of existing methods. Furthermore, we surprisingly find that SIU can avoid invasive membership inference attacks and jailbreak attacks. To the best of our knowledge, we are the first to explore MU in MLLMs. We will release the code and benchmark in the near future.

Digital publishing’s exponential growth has created vast scholarly collections. Guiding researchers to relevant resources is crucial, and knowledge graphs (KGs) are key tools for unlocking hidden knowledge. However, current methods focus on external links between concepts, ignoring the rich information within individual papers. Challenges like insufficient multi-relational data, name ambiguity, and cold-start issues further limit existing KG-based methods, failing to capture the intricate attributes of diverse entities. To solve these issues, we propose GLAMOR, a robust KG framework encompassing entities e.g., authors, papers, fields of study, and concepts, along with their semantic interconnections. GLAMOR uses a novel random walk-based KG text generation method and then fine-tunes the language model using the generated text. Subsequently, the acquired context-preserving embeddings facilitate superior top@k predictions. Evaluation results on two public benchmark datasets demonstrate our GLAMOR’s superiority against state-of-the-art methods especially in solving the cold-start problem.

Recently, uncertain knowledge graphs (UKGs), where each relation between entities is associated with a confidence score, have gained much attention. Compared with traditional knowledge graphs, UKGs possess the capability of uncertainty knowledge expression, which facilitates more reliable and precise knowledge graph reasoning by not only completing missing triples but also predicting triple confidences. In this paper, we release unKR, the first open-source python library for uncertain Knowledge graph (UKG) Reasoning by representation learning. We design a unified framework to implement two types of representation learning models for UKG reasoning, i.e., normal and few-shot ones. Besides, we standardize the evaluation tasks and metrics for UKG reasoning to ensure fair comparisons, and report the detailed results of each model under the consistent test setting. With unKR, it is effortless for users to reproduce existing models, as well as efficiently customize their own models. The library, documentation, demo, and re-implementing results are all publicly released at https://github.com/seucoin/unKR.

Conventional text-to-SQL studies are limited to a single task with a fixed-size training and test set. When confronted with a stream of tasks common in real-world applications, existing methods struggle with the problems of insufficient supervised data and high retraining costs. The former tends to cause overfitting on unseen databases for the new task, while the latter makes a full review of instances from past tasks impractical for the model, resulting in forgetting of learned SQL structures and database schemas. To address the problems, this paper proposes integrating semi-supervised learning (SSL) and continual learning (CL) in a stream of text-to-SQL tasks and offers two promising solutions in turn. The first solution Vanilla is to perform self-training, augmenting the supervised training data with predicted pseudo-labeled instances of the current task, while replacing the full volume retraining with episodic memory replay to balance the training efficiency with the performance of previous tasks. The improved solution SFNet takes advantage of the intrinsic connection between CL and SSL. It uses in-memory past information to help current SSL, while adding high-quality pseudo instances in memory to improve future replay. The experiments on two datasets shows that SFNet outperforms the widely-used SSL-only and CL-only baselines on multiple metrics.

Text-video based multimodal event extraction refers to identifying event information from the given text-video pairs.Existing methods predominantly utilize video appearance features (VAF) and text sequence features (TSF) as input information.Some of them employ contrastive learning to align VAF with the event types extracted from TSF.However, they disregard the motion representations in videos and the optimization of contrastive objective could be misguided by the background noise from RGB frames.We observe that the same event triggers correspond to similar motion trajectories, which are hardly affected by the background noise.motivated by this, we propose a Three Stream Multimodal Event Extraction framework (TSEE) that simultaneously utilizes the features of text sequence and video appearance, as well as the motion representations to enhance the event extraction capacity.Firstly, we extract the optical flow features (OFF) as motion representations from videos to incorporate with VAF and TSF.Then we introduce a Multi-level Event Contrastive Learning module to align the embedding space between OFF and event triggers, as well as between event triggers and types.Finally, a Dual Querying Text module is proposed to enhance the interaction between modalities.Experimental results show that TSEE outperforms the state-of-theart methods, which demonstrates its superiority.

Multimodal movie genre classification has always been regarded as a demanding multi-label classification task due to the diversity of multimodal data such as posters, plot summaries, trailers and metadata. Although existing works have made great progress in modeling and combining each modality, they still face three issues: 1) unutilized group relations in metadata, 2) unreliable attention allocation, and 3) indiscriminative fused features. Given that the knowledge graph has been proven to contain rich information, we present a novel framework that exploits the knowledge graph from various perspectives to address the above problems. As a preparation, the metadata is processed into a domain knowledge graph. A translate model for knowledge graph embedding is adopted to capture the relations between entities. Firstly we retrieve the relevant embedding from the knowledge graph by utilizing group relations in metadata and then integrate it with other modalities. Next, we introduce an Attention Teacher module for reliable attention allocation based on self-supervised learning. It learns the distribution of the knowledge graph and produces rational attention weights. Finally, a Genre-Centroid Anchored Contrastive Learning module is proposed to strengthen the discriminative ability of fused features. The embedding space of anchors is initialized from the genre entities in the knowledge graph. To verify the effectiveness of our framework, we collect a larger and more challenging dataset named MM-IMDb 2.0 compared with the MM-IMDb dataset. The experimental results on two datasets demonstrate that our model is superior to the state-of-the-art methods. Our code and dataset is available at https://github.com/aoluming/IDKG.git.

Continual table semantic parsing aims to train a parser on a sequence of tasks, where each task requires the parser to translate natural language into SQL based on task-specific tables but only offers limited training examples. Conventional methods tend to suffer from overfitting with limited supervision, as well as catastrophic forgetting due to parameter updates. Despite recent advancements that partially alleviate these issues through semi-supervised data augmentation and retention of a few past examples, the performance is still limited by the volume of unsupervised data and stored examples. To overcome these challenges, this paper introduces a novel method integrating parameter-efficient fine-tuning (PEFT) and in-context tuning (ICT) for training a continual table semantic parser. Initially, we present a task-adaptive PEFT framework capable of fully circumventing catastrophic forgetting, which is achieved by freezing the pre-trained model backbone and fine-tuning small-scale prompts. Building on this, we propose a teacher-student framework-based solution. The teacher addresses the few-shot problem using ICT, which procures contextual information by demonstrating a few training examples. In turn, the student leverages the proposed PEFT framework to learn from the teacher's output distribution, and subsequently compresses and saves the contextual information to the prompts, eliminating the need to store any training examples. Experimental evaluations on two benchmarks affirm the superiority of our method over prevalent few-shot and continual learning baselines across various metrics.

Query graph construction aims to construct the correct executable SPARQL on the KG to answer natural language questions. Although recent methods have achieved good results using neural network-based query graph ranking, they suffer from three new challenges when handling more complex questions: 1) complicated SPARQL syntax, 2) huge search space, and 3) locally ambiguous query graphs. In this paper, we provide a new solution. As a preparation, we extend the query graph by treating each SPARQL clause as a subgraph consisting of vertices and edges and define a unified graph grammar called AQG to describe the structure of query graphs. Based on these concepts, we propose a novel end-to-end model that performs hierarchical autoregressive decoding to generate query graphs. The high-level decoding generates an AQG as a constraint to prune the search space and reduce the locally ambiguous query graph. The bottom-level decoding accomplishes the query graph construction by selecting appropriate instances from the preprepared candidates to fill the slots in the AQG. The experimental results show that our method greatly improves the SOTA performance on complex KGQA benchmarks. Equipped with pre-trained models, the performance of our method is further improved, achieving SOTA for all three datasets used.

Abstract Event extraction is an important, but challenging task. Many existing techniques decompose it into event and argument detection/classification subtasks, which are complex structured prediction problems. Generation-based extraction techniques lessen the complexity of the problem formulation and are able to leverage the reasoning capabilities of large pretrained language models. However, they still suffer from poor zero-shot generalizability and are ineffective in handling long contexts such as documents. We propose a generative event extraction model, KC-GEE, that addresses these limitations. A key contribution of KC-GEE is a novel knowledge-based conditioning technique that injects the schema of candidate event types as the prefix into each layer of an encoder-decoder language model. This enables effective zero-shot learning and improves supervised learning. Our experiments on two benchmark datasets demonstrate the strong performance of our KC-GEE model. It achieves particularly strong results in the challenging document-level extraction task and in the zero-shot learning setting, outperforming state-of-the-art models by up to 5.4 absolute F1 points.

This paper studies event causality identification, which aims at predicting the causality relation for a pair of events in a sentence. Regarding event causality identification as a supervised classification task, most existing methods suffer from the problem of insufficient annotated data. In this paper, we propose a new derivative prompt joint learning model for event causality identification, which leverages potential causal knowledge in the pre-trained language model to tackle the data scarcity problem. Specifically, rather than external data or knowledge augmentation, we derive two relevant prompt tasks from event causality identification to enhance the model’s ability to identify explicit and implicit causality. We evaluate our model on two benchmark datasets and the results show that our model has great advantages over previous methods.

Relation extraction has focused on extracting semantic relationships between entities from the unstructured written textual data.However, with the vast and rapidly increasing amounts of spoken data, relation extraction from speech is an important but under-explored problem. In this paper, we propose a new information extraction task, speech relation extraction (SpeechRE).To facilitate further research, we construct the first synthetic training datasets, as well as the first human-spoken test set with native English speakers.We establish strong baseline performance for SpeechRE via two approaches.The pipeline approach connects a pretrained ASR module with a text-based relation extraction module.The end-to-end approach employs a cross-modal encoder-decoder architecture.Our comprehensive experiments reveal the relative strengths and weaknesses of these approaches, and shed light on important future directions in SpeechRE research.We share the source code and datasets on https://github.com/ wutong8023/SpeechRE.* denotes the equal contribution.

The few-shot problem is an urgent challenge for single-table text-to-SQL. Existing methods ignore the potential value of unlabeled data, and merely rely on a coarse-grained Meta-Learning (ML) algorithm that neglects the differences of column contributions to the optimization object. This paper proposes a Meta Self-Training text-to-SQL (MST-SQL) method to solve the problem. Specifically, MST-SQL is based on column-wise HydraNet and adopts self-training as an effective mechanism to learn from readily available unlabeled samples. During each epoch of training, it first predicts pseudo-labels for unlabeled samples and then leverages them to update the parameters. A fine-grained ML algorithm is used in updating, which weighs the contribution of columns by their specificity, in order to further improve the generalizability. Extensive experimental results on both open-domain and domain-specific benchmarks reveal that our MST-SQL has significant advantages in few-shot scenarios, and is also competitive in standard supervised settings.

Medication recommendation targets to provide a proper set of medicines according to patients’ diagnoses, which is a critical task in clinics. Currently, the recommendation is manually conducted by doctors. However, for complicated cases, like patients with multiple diseases at the same time, it’s difficult to propose a considerate recommendation even for experienced doctors. This urges the emergence of automatic medication recommendation which can help treat the diagnosed diseases without causing harmful drug-drug interactions. Due to the clinical value, medication recommendation has attracted growing research interests. Existing works mainly formulate medication recommendation as a multi-label classification task to predict the set of medicines. In this paper, we propose the Conditional Generation Net (COGNet) which introduces a novel copy-or-predict mechanism to generate the set of medicines. Given a patient, the proposed model first retrieves his or her historical diagnoses and medication recommendations and mines their relationship with current diagnoses. Then in predicting each medicine, the proposed model decides whether to copy a medicine from previous recommendations or to predict a new one. This process is quite similar to the decision process of human doctors. We validate the proposed model on the public MIMIC data set, and the experimental results show that the proposed model can outperform state-of-the-art approaches.

Single-table text-to-SQL aims to transform a natural language question into a SQL query according to one single table. Recent work has made promising progress on this task by pre-trained language models and a multi-submodule framework. However, zero-shot table, that is, the invisible table in the training set, is currently the most critical bottleneck restricting the application of existing approaches to real-world scenarios. Although some work has utilized auxiliary tasks to help handle zero-shot tables, expensive extra manual annotation limits their practicality. In this paper, we propose a new approach for the zero-shot text-to-SQL task which does not rely on any additional manual annotations. Our approach consists of two parts. First, we propose a new model that leverages the abundant information of table content to help establish the mapping between questions and zero-shot tables. Further, we propose a simple but efficient meta-learning strategy to train our model. The strategy utilizes the two-step gradient update to force the model to learn a generalization ability towards zero-shot tables. We conduct extensive experiments on a public open-domain text-to-SQL dataset WikiSQL and a domain-specific dataset ESQL. Compared to existing approaches using the same pre-trained model, our approach achieves significant improvements on both datasets. Compared to the larger pre-trained model and the tabular-specific pre-trained model, our approach is still competitive. More importantly, on the zero-shot subsets of both the datasets, our approach further increases the improvements.

Continual relation extraction is an important task that focuses on extracting new facts incrementally from unstructured text. Given the sequential arrival order of the relations, this task is prone to two serious challenges, namely catastrophic forgetting and order-sensitivity. We propose a novel curriculum-meta learning method to tackle the above two challenges in continual relation extraction. We combine meta learning and curriculum learning to quickly adapt model parameters to a new task and to reduce interference of previously seen tasks on the current task. We design a novel relation representation learning method through the distribution of domain and range types of relations. Such representations are utilized to quantify the difficulty of tasks for the construction of curricula. Moreover, we also present novel difficulty-based metrics to quantitatively measure the extent of order-sensitivity of a given model, suggesting new ways to evaluate model robustness. Our comprehensive experiments on three benchmark datasets show that our proposed method outperforms the state-of-the-art techniques. The code is available at the anonymous GitHub repository: https://github.com/wutong8023/AAAI_CML.

Visual modality recently has aroused extensive attention in the fields of knowledge graph and multimedia because a lot of real-world knowledge is multi-modal in nature. However, it is currently unclear to what extent the visual modality can improve the performance of knowledge graph tasks over unimodal models, and equally treating structural and visual features may encode too much irrelevant information from images. In this paper, we probe the utility of the auxiliary visual context from knowledge graph representation learning perspective by designing a Relation Sensitive Multi-modal Embedding model, RSME for short. RSME can automatically encourage or filter the influence of visual context during the representation learning. We also examine the effect of different visual feature encoders. Experimental results validate the superiority of our approach compared to the state-of-the-art methods. On the basis of in-depth analysis, we conclude that under appropriate circumstances models are capable of leveraging the visual input to generate better knowledge graph embeddings and vice versa.

The aim of visual relation detection is to provide a comprehensive understanding of an image by describing all the objects within the scene, and how they relate to each other, in < object-predicate-object > form; for example, < person-lean on-wall > . This ability is vital for image captioning, visual question answering, and many other applications. However, visual relationships have long-tailed distributions and, thus, the limited availability of training samples is hampering the practicability of conventional detection approaches. With this in mind, we designed a novel model for visual relation detection that works in one-shot settings. The embeddings of objects and predicates are extracted through a network that includes a feature-level attention mechanism. Attention alleviates some of the problems with feature sparsity, and the resulting representations capture more discriminative latent features. The core of our model is a dual graph neural network that passes and aggregates the context information of predicates and objects in an episodic training scheme to improve recognition of the one-shot predicates and then generate the triplets. To the best of our knowledge, we are the first to center on the viability of one-shot learning for visual relation detection. Extensive experiments on two newly-constructed datasets show that our model significantly improved the performance of two tasks PredCls and SGCls from 2.8% to 12.2% compared with state-of-the-art baselines.

Question generation over knowledge bases (KBQG) aims at generating natural-language questions about a subgraph, i.e. a set of triples.Two main challenges still face the current crop of encoder-decoder-based methods, especially on small subgraphs: (1) low diversity and poor fluency due to the limited information contained in the subgraphs, and (2) semantic drift due to the decoder's oblivion of the semantics of the answer entity.We propose an innovative knowledgeenriched, type-constrained and grammar-guided KBQG model, named KTG, to addresses the above challenges.In our model, the encoder is equipped with auxiliary information from the KB, and the decoder is constrained with word types during QG.Specifically, entity domain and description, as well as relation hierarchy information are considered to construct question contexts, while a conditional copy mechanism is incorporated to modulate question semantics according to current word types.Besides, a novel reward function featuring grammatical similarity is designed to improve both generative richness and syntactic correctness via reinforcement learning.Extensive experiments show that our proposed model outperforms existing methods by a significant margin on two widely-used benchmark datasets SimpleQuestion and PathQuestion.

Event extraction plays an important role in legal applications, including case push and auxiliary judgment. However, traditional event structure cannot express the connections between arguments, which are extremely important in legal events. Therefore, this paper defines a dynamic event structure for Chinese legal events. To distinguish between similar events, we design hierarchical event features for event detection. Moreover, to address the problem of long-distance semantic dependence and anaphora resolution in argument classification, we propose a novel pedal attention mechanism to extract the semantic relation between two words through their dependent adjacent words. We label a Chinese legal event dataset and evaluate our model on it. Experimental results demonstrate that our model can surpass other state-of-the-art models.

Knowledge graph representation learning, also called knowledge graph embedding, is the task of mapping entities and relations into a low-dimensional, continuous vector space, and, as a result, can support various machine learning models to perform knowledge completion tasks with good performance and robustness. However, most of existing embedding models focus on improving the link prediction accuracy while ignoring the time-efficiency in search-intensive applications over large-scale knowledge graphs. To tackle this problem, in this paper, we encode knowledge graph into Hamming space and introduce a novel HAsh Learning Framework (HALF) for search-oriented knowledge graph embedding. The proposed method can be applied to recent various knowledge graph embedding models for accelerating the computation of searching embeddings by utilizing the bitwise operations (XNOR and Bitcount). Experimental results on benchmark datasets demonstrate the effectiveness of our proposed method, which gets a bonus of speed-up in the searching embeddings while the accuracy and scalability of the original model are basically maintained.

Complex question-answering (CQA) involves answering complex natural-language questions on a knowledge base (KB). However, the conventional neural program induction (NPI) approach exhibits uneven performance when the questions have different types, harboring inherently different characteristics, e.g., difficulty level. This paper proposes a meta-reinforcement learning approach to program induction in CQA to tackle the potential distributional bias in questions. Our method quickly and effectively adapts the meta-learned programmer to new questions based on the most similar questions retrieved from the training data. The meta-learned policy is then used to learn a good programming policy, utilizing the trial trajectories and their rewards for similar questions in the support set. Our method achieves state-of-the-art performance on the CQA dataset (Saha et al., 2018) while using only five trial trajectories for the top-5 retrieved questions in each support set, and meta-training on tasks constructed from only 1% of the training set. We have released our code at https://github.com/DevinJake/MRL-CQA.

Formal query building is an important part of complex question answering over knowledge bases. It aims to build correct executable queries for questions. Recent methods try to rank candidate queries generated by a state-transition strategy. However, this candidate generation strategy ignores the structure of queries, resulting in a considerable number of noisy queries. In this paper, we propose a new formal query building approach that consists of two stages. In the first stage, we predict the query structure of the question and leverage the structure to constrain the generation of the candidate queries. We propose a novel graph generation framework to handle the structure prediction task and design an encoder-decoder model to predict the argument of the predetermined operation in each generative step. In the second stage, we follow the previous methods to rank the candidate queries. The experimental results show that our formal query building approach outperforms existing methods on complex questions while staying competitive on simple questions.

A compelling approach to complex question answering is to convert the question to a sequence of actions, which can then be executed on the knowledge base to yield the answer, aka the programmer-interpreter approach. Use similar training questions to the test question, meta-learning enables the programmer to adapt to unseen questions to tackle potential distributional biases quickly. However, this comes at the cost of manually labeling similar questions to learn a retrieval model, which is tedious and expensive. In this paper, we present a novel method that automatically learns a retrieval model alternately with the programmer from weak supervision, i.e., the system’s performance with respect to the produced answers. To the best of our knowledge, this is the first attempt to train the retrieval model with the programmer jointly. Our system leads to state-of-the-art performance on a large-scale task for complex question answering over knowledge bases. We have released our code at https://github.com/DevinJake/MARL.

Sentiment analysis in various languages has been a hot research topic with several applications. Most of the existing models have been reported to work well with widely used language. Were the lass directly applying these models to poor-quality corpora often leads to low results. Thus, to deal with these shortcoming we propose a cross-lingual sentiment topic model evolution over time (CLSTOT) which jointly models time with topic and sentiment. In CLSTOT, we consider the mapping between sentiment-aware topics under different cultures and analyze their evolution over time. The topic-specific sentiment is extracted using the entire data and not for each single document. As long as providing sentiment-topic, we can predict the timestamps for each test document by finding its most likely location over the timeline. This is achieved by using inference algorithm which is based on Gibbs Sampling. The experimental results on Chinese and English newsreader dataset; Chinese from SinaNews2, and English from Yahoo1, show that CLSTOT achieves significant improvement over the state-of-the-art.

Distant supervision has advantages of generating training data automatically for relation extraction by aligning triples in Knowledge Graphs with large-scale corpora. Some recent methods attempt to incorporate extra information to enhance the performance of relation extraction. However, there still exist two major limitations. Firstly, these methods are tailored for a specific type of information which is not enough to cover most of the cases. Secondly, the introduced extra information may contain noise. To address these issues, we propose a novel hybrid graph model, which can incorporate heterogeneous background information in a unified framework, such as entity types and human-constructed triples. These various kinds of knowledge can be integrated efficiently even with several missing cases. In addition, we further employ an attention mechanism to identify the most confident information which can alleviate the side effect of noise. Experimental results demonstrate that our model outperforms the state-of-the-art methods significantly in various evaluation metrics.

Knowledge graph (KG) embedding techniques represent entities and relations as low-dimensional, continuous vectors, and thus enables machine learning models to be easily adapted to KG completion and querying tasks. However, learned dense vectors are inefficient for large-scale similarity computations. Learning-to-hash is to learn compact binary codes from high-dimensional input data and provides a promising way to accelerate efficiency by measuring Hamming distance instead of Euclidean distance or dot-product. Unfortunately, most of learning-to-hash methods cannot be directly applied to KG structure encoding. In this paper, we introduce a novel framework for encoding incomplete KGs and graph queries in Hamming space. To preserve KG structure information from embeddings to hash codes and address the ill-posed gradient issue in optimization, we utilize a continuation method with convergence guarantees to jointly encode queries and KG entities with geometric operations. The hashed embedding of a query can be utilized to discover target answers from incomplete KGs whilst the efficiency has been greatly improved.We compared our model with state-of-the-art methods on real-world KGs. Experimental results show that our framework not only significantly speeds up the searching process, but also provides good results for unanswerable queries caused by incomplete information.

Intent determination is one of the main tasks of natural language understanding and dialogue systems, aiming to determine the intent of the user’s input. Incorporating contextual information for intent determination has shown much promise. Recently, memory networks have been used to encode context from the dialogue history at each turn. However, these methods lack general mechanism for encoding intent patterns. In this paper, we investigate the incorporation of intent patterns extracted from regular expression for the multi-turn intent determination task. We propose a novel neural network model with two memories. The first memory encodes context from dialogue history at each turn whereas the second memory contains the information obtained from the regular expression patterns. We evaluate the model on Frames and Key-Value Retrieval datasets, the experimental results demonstrate that encoding intent patterns with memory networks significantly improves the multi-turn intent determination.

Relation Extraction (RE) aims at extracting semantic relation of entities from text and it is a crucial task in natural language processing. Deep neural network (DNN) based models have achieved excellent performance in RE. However, they still have several problems remaining to be addressed: (1) humans can hardly take measures to amend the DNN-based RE systems because it is difficult to encode human intention to guide them to capture desired patterns. (2) DNN-based RE models may suffer from not having sufficient background information for making predictions. To handle these issues, we propose an RE framework based on reinforcement learning, which can enhance existing DNN-based RE models by incorporating human knowledge including soft rules and relation evidence. The introduction of soft rules enable human to impose an effect on the RE result and correct the RE system, while the relation evidence help supplement the background information without limiting its types and sources. The experimental results show that our approach can reinforce existing DNN-based RE models effectively and outperforms state-of-the-art RE methods.

Extracting knowledge from Wikipedia has attracted much attention in recent ten years. One of the most valuable kinds of knowledge is type information, which refers to the axioms stating that an instance is of a certain type. Current approaches for inferring the types of instances from Wikipedia mainly rely on some language-specific rules. Since these rules cannot catch the semantic associations between instances and classes (i.e. candidate types), it may lead to mistakes and omissions in the process of type inference. The authors propose a new approach leveraging attributes to perform language-independent type inference of the instances from Wikipedia. The proposed approach is applied to the whole English and Chinese Wikipedia, which results in the first version of MulType (Multilingual Type Information), a knowledge base describing the types of instances from multilingual Wikipedia. Experimental results show that not only the proposed approach outperforms the state-of-the-art comparison methods, but also MulType contains lots of new and high-quality type information.

This paper investigates a challenging problem,which is known as fine-grained image classification(FGIC). Different from conventional computer visionproblems, FGIC suffers from the large intraclassdiversities and subtle inter-class differences.Existing FGIC approaches are limited to exploreonly the visual information embedded in the images.In this paper, we present a novel approachwhich can use handy prior knowledge from eitherstructured knowledge bases or unstructured text tofacilitate FGIC. Specifically, we propose a visual-semanticembedding model which explores semanticembedding from knowledge bases and text, andfurther trains a novel end-to-end CNN frameworkto linearly map image features to a rich semanticembedding space. Experimental results on a challenginglarge-scale UCSD Bird-200-2011 datasetverify that our approach outperforms several state-of-the-art methods with significant advances.

With the development of the Semantic Web, more and more semantic data including many useful knowledge bases has been published on the Web. Such knowledge bases always lack expressive schema information, especially disjointness axioms and subclass axioms. This makes it difficult to perform many crit ical Semantic Web tasks like ontology reasoning, inconsistency handling and ontology mapping. To deal with this problem, a few approaches have been proposed to generate terminology axioms. However, they often adopt the closed world assumption which is opposite to the assumption adopted by the semantic data. This may lead to a lot of noisy negative examples so that existing learning approaches fail to perform well on such incomplete data. In this paper, a novel framework is proposed to automatically obtain disjointness axioms and subclass axioms from incomplete semantic data. This framework first obtains probabilistic type assertions by exploiting a type inference algorithm. Then a mining approach based on association rule mining is proposed to learn high-quality schema information. To address the incompleteness problem of semantic data, the mining model introduces novel definitions to compute the support and confidence for pruning false axioms. Our experimental evaluation shows promising results over several real-life incomplete knowledge bases like DBpedia and LUBM by comparing with existing relevant approaches.

The ontology language OWL 2 EL, is designed for knowledge modeling and has been widely used in real applications. However, modeling knowledge as OWL ontologies is an error-prone process, where logical errors or contradictions would be imported. Further, it is almost impossible to manually find erro rs occurring in large-scale ontologies. Finding justification, an important service for error pinpointing, has then attracted much attention of researchers and developers. However, current methods of finding justifications suffer from an issue: high-coupling of reasoners, i.e., there is a tight relation between reasoners and the task of finding justifications. This makes the performance of finding justifications be highly influenced by the utilized reasoner, and, it is also restricted to optimize the algorithms effectively. In order to tackle this problem, we consider giving a method such that the task of finding justifications is independent from the utilized reasoner. Specifically, we propose a kind of graph called Explanation Dependency Graph (EDG) which guides to compute justifications from the reasoning results directly. We further give several optimizing strategies and prove the correctness of our method. We implement our method and evaluate it on real ontologies, including SNOMED CT. The experimental results show that our method is practical and performs better than current methods.

Ontology mappings are regarded as the semantic bridges that link entities from different yet overlapping ontologies in order to support knowledge sharing and reuse on the Semantic Web. However, mappings can be wrong and result in logical conflicts among ontologies. Such kind of mappings are called incoherent mappings. As an important part of ontology matching, mapping validation aims at detecting the conflicts and restoring the coherence of mappings. In this paper, we propose a graph-based approach which is complete for detecting incoherent mappings among DL-Lite ontologies. The lightweight DL-Lite family of description logics stand out for tractable reasoning and efficient query answering capabilities. Our approach consists of a set of graph construction rules, a graph-based incoherence detection algorithm, and a graph-based incoherence repair algorithm. We propose and formalize three repair principles in an attempt to measure the wrong mappings, where the notion of common closures w.r.t. a mapping arc in the constructed graph is introduced. These principles feature a global removal strategy that is independent of individual ontology matchers. In order to relieve the loss of information among ontologies in the repair process, we further define a mapping revision operator so that common closures related to the removed mappings can be preserved in the graph. We implement the graph-based algorithms and evaluate their performance in a comparison with state-of-the-art systems on real-world ontologies. Experimental results show that our approach can remove more wrong mappings and achieve better repairing results in most of the cases.

Knowledge graph embedding, which aims to represent entities and relations in vector spaces, has shown outstanding performance on a few knowledge graph completion tasks. Most existing methods are based on the assumption that a knowledge graph is a set of separate triples, ignoring rich graph features, i.e., structural information in the graph. In this paper, we take advantages of structures in knowledge graphs, especially local structures around a triple, which we refer to as triple context. We then propose a Triple-Context-based knowledge Embedding model (TCE). For each triple, two kinds of structure information are considered as its context in the graph; one is the outgoing relations and neighboring entities of an entity and the other is relation paths between a pair of entities, both of which reflect various aspects of the triple. Triples along with their contexts are represented in a unified framework, in which way structural information in triple contexts can be embodied. The experimental results show that our model outperforms the state-of-the-art methods for link prediction.

Topic models have been widely applied in discovering topics that underly a collection of documents. Incorporating human knowledge can guide conventional topic models to produce topics which are easily interpreted and semantically coherent. Several knowledge-based topic models have been proposed, bu t these models just leverage lexical knowledge of words that are often not in accordance with topics. To solve the problem, we recognize entity mentions, besides words, in the documents and incorporate entity knowledge from external knowledge bases. In this paper, we study to utilize entity knowledge, concepts and categories in Wikipedia, as prior knowledge into topic models to discover more coherent topics. A novel knowledge-based topic model, WCM-LDA (Wikipedia-Category-concept-Mention Latent Dirichlet Allocation), is proposed, which not only models the relationship between words and topics, but also utilizes concept and category knowledge of entities to model the semantic relation of entities and topics. We compare WCM-LDA with the state-of-the-art knowledge-based topic models, on three datasets. Experimental results show that our approach outperforms the existing baseline methods on all three datasets. Moreover, our model can visualize topics with top words, concepts and categories such that topics are made easily to be interpreted and classified.

As more and more multilingual knowledge becomes available on the Web, knowledge sharing across languages has become an important task to benefit many applications. One of the most crucial kinds of knowledge on the Web is taxonomy, which is used to organize and classify the Web data. To facilitate knowledge sharing across languages, we need to deal with the problem of cross-lingual taxonomy alignment, which discovers the most relevant category in the target taxonomy of one language for each category in the source taxonomy of another language. Current approaches for aligning cross-lingual taxonomies strongly rely on domain-specific information and the features based on string similarities. In this paper, we present a new approach to deal with the problem of cross-lingual taxonomy alignment without using any domain-specific information. We first identify the candidate matched categories in the target taxonomy for each category in the source taxonomy using the cross-lingual string similarity. We then propose a novel bilingual topic model, called Bilingual Biterm Topic Model (BiBTM), to perform exact matching. BiBTM is trained by the textual contexts extracted from the Web. We conduct experiments on two kinds of real world datasets. The experimental results show that our approach significantly outperforms the designed state-of-the-art comparison methods.

Joint sentiment/topic models are widely applied in detecting sentiment-aware topics on the lengthy review data and they are achieved with Latent Dirichlet Allocation (LDA) based model. Nowadays plenty of user-generated posts, e.g., tweets and E-commerce short reviews, are published on the social media and the posts imply the public's sentiments (i.e., positive and negative) towards various topics. However, the existing sentiment/topic models are not applicable to detect sentiment-aware topics on the posts, i.e., short texts, because applying the models to the short texts directly will suffer from the context sparsity problem. In this paper, we propose a Time-User Sentiment/Topic Latent Dirichlet Allocation (TUS-LDA) which aggregates posts in the same timeslice or user as a pseudo-document to alleviate the context sparsity problem. Moreover, we design approaches for parameter inference and incorporating prior knowledge into TUS-LDA. Experiments on the Sentiment140 and tweets of electronic products from Twitter7 show that TUS-LDA outperforms previous models in the tasks of sentiment classification and sentiment-aware topic extraction. Finally, we visualize the sentiment-aware topics discovered by TUS-LDA.

Materialization is an important reasoning service for applications built on the Web Ontology Language (OWL). To make materialization efficient in practice, current research focuses on deciding tractability of an ontology language and designing parallel reasoning algorithms. However, some well-known large-scale ontologies, such as YAGO, have been shown to have good performance for parallel reasoning, but they are expressed in ontology languages that are not parallelly tractable, i.e., the reasoning is inherently sequential in the worst case. This motivates us to study the problem of parallel tractability of ontology materialization from a theoretical perspective. That is, we aim to identify the ontologies for which materialization is parallelly tractable, i.e., in NC complexity. In this work, we focus on datalog rewritable ontology languages. We identify several classes of datalog rewritable ontologies (called parallelly tractable classes) such that materialization over them is parallelly tractable. We further investigate the parallel tractability of materialization of a datalog rewritable OWL fragment DHL (Description Horn Logic) and an extension of DHL that allows complex role inclusion axioms. Based on the above results, we analyze real-world datasets and show that many ontologies expressed in DHL or its extension belong to the parallelly tractable classes.

Two essential tasks in managing description logic knowledge bases are eliminating problematic axioms and incorporating newly formed ones. Such elimination and incorporation are formalised as the operations of contraction and revision in belief change. In this paper, we deal with contraction and revision for the DL-Lite family through a model-theoretic approach. Standard description logic semantics yields an infinite number of models for DL-Lite knowledge bases, thus it is difficult to develop algorithms for contraction and revision that involve DL models. The key to our approach is the introduction of an alternative semantics called type semantics which can replace the standard semantics in characterising the standard inference tasks of DL-Lite. Type semantics has several advantages over the standard one. It is more succinct and importantly, with a finite signature, the semantics always yields a finite number of models. We then define model-based contraction and revision functions for DL-Lite knowledge bases under type semantics and provide representation theorems for them. Finally, the finiteness and succinctness of type semantics allow us to develop tractable algorithms for instantiating the functions.

Model-based approaches provide a semantically well justified way to revise ontologies. However, in general, model-based revision operators are limited due to lack of efficient algorithms and inexpressibility of the revision results. In this paper, we make both theoretical and practical contribution to efficient computation of model-based revisions in DL-Lite. Specifically, we show that maximal approximations of two well-known model-based revisions for DL-Lite_R can be computed using a syntactic algorithm. However, such a coincidence of model-based and syntactic approaches does not hold when role functionality axioms are allowed. As a result, we identify conditions that guarantee such a coincidence for DL-Lite_FR. Our result shows that both model-based and syntactic revisions can co-exist seamlessly and the advantages of both approaches can be taken in one revision operator. Based on our theoretical results, we develop a graph-based algorithm for the revision operat

The development and maintenance of large and complex ontologies are often time-consuming and error-prone. Thus, automated ontology learning and evolution have attracted intensive research interest. In data-centric applications where ontologies are designed from the data or automatically learnt from it, when new data instances are added that contradict the ontology, it is often desirable to incrementally revise the ontology according to the added data. In description logics, this problem can be intuitively formulated as the operation of TBox contraction, i.e., rational elimination of certain axioms from the logical consequences of a TBox, and it is w.r.t. an ABox. In this paper we introduce a model-theoretic approach to such a contraction problem by using an alternative semantic characterisation of DL-Lite TBoxes. We show that entailment checking (without necessarily first computing the contraction result) is in coNP, which does not shift the corresponding complexity in propositional logic, and the problem is tractable when the size of the new data is bounded.

OWL EL is a tractable fragment of OWL 2. Classification, which is the task of computing a subsumption hierarchy between concepts, plays an important role in many applications of OWL EL. The current classification methods are proposed based on the specific parallel computation platforms, thus can hardly adapt to other platforms and satisfy different needs of users or developers. In this paper, we propose a platform-independent approach to perform classification in OWL EL. We first give a novel and well defined graph formalism GEL for representing EL ontologies. Based on this formalism, we describe the classification using a platform-independent computation model, and discuss the lower bound of run-time. We further give a refined algorithm based on the proposed model, and show the correctness of the algorithm. We implement a prototype system, which can easily switch between multi-core and a distributed cluster. Finally, we conduct experiments on several real world ontologies. The experimental results show that our system outperforms two state-of-theart EL reasoning systems, and has a linear scalability on the extensions of GO ontologies.

Computing all ABox repairs is a key to cautious or brave reasoning over inconsistent description logic (DL) ontologies. However, the number of ABox repairs can be exponential in the number of assertions in the ABox even for very lightweight DLs. Hence we propose to compute a minimal representative set of ABox repairs. A set of ABox repairs is representative, if every assertion occurring in at least one ABox repair also occurs in at least one element of this set, while every assertion occurring in all ABox repairs occurs in all elements of this set. Cautious or brave reasoning then can be approximated by standard reasoning over a minimal representative set other than the complete set of ABox repairs. However, computing a minimal representative set of ABox repairs is still intractable in general. To guarantee the tractability in data complexity for computing a minimal representative set, we focus on a class of DL ontologies called the first-order rewritable class. We propose a tractable method for computing a minimal representative set of ABox repairs in an inconsistent first-order rewritable ontology. Experimental results demonstrate the high efficiency and scalability of the proposed method.

DL-Lite is an important family of description logics. Recently, there is an increasing interest in handling inconsistency in DL-Lite as the constraint imposed by a TBox can be easily violated by assertions in ABox in DL-Lite. In this paper, we present a distance-based paraconsistent semantics based on the notion of feature in DL-Lite, which provides a novel way to rationally draw meaningful conclusions even from an inconsistent knowledge base. Finally, we investigate several important logical properties of this entailment relation based on the new semantics and show its promising advantages in non-monotonic reasoning for DL-Lite.

Two essential tasks in managing Description Logic (DL) ontologies are eliminating problematic axioms and incorporating newly formed axioms. Such elimination and incorporation are formalised as the operations of contraction and revision in belief change.In this paper, we deal with contraction and revision for the DL-Lite family through a model-theoretic approach.Standard DL semantics yields infinite numbers of models for DL-Lite TBoxes, thus it is not practical to develop algorithms for contraction and revision that involve DL models. The key to our approach is the introduction of an alternative semantics called type semantics which is more succinct than DL semantics. More importantly, with a finite signature, type semantics always yields finite humber of models.We then define model-based contraction and revision for DL-Lite TBoxesunder type semantics and provide representation theorems for them.Finally, the succinctness of type semantics allows us to develop tractable algorithms for both operations.

Resolving incoherent terminologies is an important task in the maintenance of evolving OWL 2 DL ontologies. Existing approaches to this task are either semi-automatic or based on simple deletion of axioms. There is a need of fine-grained approaches to automatize this task. Since a fine-grained approach should consider multiple choices for modifying an axiom other than the deletion of axioms only, the primary challenges for developing such an approach lie in both the semantics of the repaired results and the efficiency in computing the repaired results. To tackle these challenges, we first introduce the notion of fine-grained repair based on modifying one axiom to zero or more axioms, then propose an efficient incremental method for computing all fine-grained repairs one by one. We also propose a modification function for axioms expressed in OWL 2 DL, which returns weaker axioms. Based on this modification function and the method for computing fine-grained repairs, we develop an automatic approach to resolving incoherent OWL 2 DL terminologies. Our extensive experimental results demonstrate that the proposed approach is efficient and practical.

Finding justifications for an entailment in an OWL ontology is an important reasoning service in the development and maintenance of OWL ontologies. In this paper, we consider the problem of finding all the justifications of an entailment in description logic EL+, which underpins OWL 2 EL. Our method for computing all the justifications utilizes the classification result of an EL+ ontology by an OWL EL reasoner. We first give an algorithm for finding all the justifications of an entailment by constructing an explanation dependency graph from the entailment and the classification result. We then propose some pruning strategies to optimize this algorithm. We implement a prototype system by using the multi-core technology. The experimental results show that in many cases, our optimized algorithm outperforms an existing algorithm for finding all the justifications for an entailment in EL+ that is optimized with the module extraction technology.

Possibilistic Description Logics (DLs) extend description logics with possibilistic semantics to reason with inconsistent and uncertain knowledge. In possibilistic DLs, a crucial reasoning task is to compute the inconsistency degree of a possibilistic DL knowledge base. In this work, we first point out a shortcoming of a previous tableaux algorithm for possibilistic DL ALC and propose a new tableaux algorithm. We then propose a tableaux algorithm for computing the inconsistency degree of a knowledge base in possibilistic DL ALCI(R+), which extends possibilistic DL ALC with inverse roles and transitive roles. A blocking condition is proposed to ensure the termination of the algorithm. Although the tableaux algorithm for possibilistic DL ALCI(R+) is easy to understand and to implement, it may need exponential space in the worst case. Therefore, we give another tableaux algorithm to improve it and show that the complexity of this algorithm is PSpace-complete. This shows that the flexibility in representing uncertain information is handled without extra computational costs.

In this paper, we propose a new operator for revising ABoxes in DL-Lite ontologies. We present a graph-based algorithm for ABox revision in DL-Lite, which implements the revision operator and we show it runs in polynomial time

In this work, we build a large scale reasoning engine under temporal RDFS semantics using MapReduce. We identify the major challenges of applying MapReduce framework to reason over temporal information, and present our solutions to tackle them.

In this paper, we investigate belief revision in possibilistic logic, which is a weighted logic proposed to deal with incomplete and uncertain information. Existing revision operators in possibilistic logic are restricted in the sense that the input information can only be a formula instead of a possibilistic knowledge base which is a set of weighted formulas. To break this restriction, we consider weighted prime implicants of a possibilistic knowledge base and use them to define novel revision operators in possibilistic logic. Intuitively, a weighted prime implicant of a possibilistic knowledge base is a logically weakest possibilistic term (i.e., a set of weighted literals) that can entail the knowledge base. We first show that the existing definition of a weighted prime implicant is problematic and need a modification. To define a revision operator using weighted prime implicants, we face two problems. The first problem is that we need to define the notion of a conflict set between two weighted prime implicants of two possibilistic knowledge bases to achieve minimal change. The second problem is that we need to define the disjunction of possibilistic terms. We solve these problems and define two conflict-based revision operators in possibilistic logic. We then adapt the well-known postulates for revision proposed by Katsuno and Mendelzon and show that our revision operators satisfy four of the basic adapted postulates and satisfy two others in some special cases.

In the last few years, RDF is becoming the dominating data model used in semantic web for knowledge representation and inference. In this paper, we revisit the problem of pattern matching query in RDF model, which is usually expensive in efficiency due to the huge cost on join operations. To alleviate the efficiency pain, view materialization techniques are usually deployed to accelerate the query processing. However, given an arbitrary view, it remains difficult to identify how to reuse the view for a particular query, because of the NP-hardness behind the algorithm matching patterns and views. To fully exploit the benefit of the materialized views, we propose a new paradigm to enhance the effectiveness of the materialized view. Instead of choosing materialized views in arbitrary form, our paradigm aims to select the views only if they are sortable. The property of sortability raises huge gains on the pattern-view matching, bringing down the cost to linear complexity in terms of the pattern size. On the other side, the costs on identifying sortable views and searching over the views using inverted index are affordable. Moreover, sortable views generally improve the overall performance of pattern matching, by means of a cost model used to optimize the query rewriting on the most appropriate views. Finally, we demonstrate extensive experimental results to verify the superiority of our proposal on both efficiency and effectiveness.

In order to directly reason over inconsistent OWL 2 DL ontologies, this paper considers linear order inference which comes from propositional logic. Consequences of this inference in an inconsistent ontology are defined as consequences in a certain consistent sub-ontology. This paper proposes a novel framework for compiling an OWL 2 DL ontology to a Horn propositional program so that the intended consistent sub-ontology for linear order inference can be approximated from the compiled result in polynomial time. A tractable method is proposed to realize this framework. It guarantees that the compiled result has a polynomial size. Experimental results show that the proposed method computes the exact intended sub-ontology for almost all test cases, while it is significantly more efficient and scalable than state-of-the-art exact methods.

ABox abduction is an important reasoning facility in Description Logics (DLs). It finds all minimal sets of ABox axioms, called abductive solutions, which should be added to a background ontology to enforce entailment of an observation which is a specified set of ABox axioms. However, ABox abduction is far from practical by now because there lack feasible methods working in finite time for expressive DLs. To pave a way to practical ABox abduction, this paper proposes a new problem for ABox abduction and a new method for computing abductive solutions accordingly. The proposed problem guarantees finite number of abductive solutions. The proposed method works in finite time for a very expressive DL, , which underpins the W3C standard language OWL 2, and guarantees soundness and conditional completeness of computed results. Experimental results on benchmark ontologies show that the method is feasible and can scale to large ABoxes.

Consistent query answering over description logic based ontologies is an important topic in ontology engineering as it can provide meaningful answers to queries posed over inconsistent ontologies. Current approaches for dealing with this problem usua

ABox abduction is an important aspect for abductive reasoning in Description Logics (DLs). It finds all minimal sets of ABox axioms that should be added to a background ontology to enforce entailment of a specified set of ABox axioms. As far as we know, by now there is only one ABox abduction method in expressive DLs computing abductive solutions with certain minimality. However, the method targets an ABox abduction problem that may have infinitely many abductive solutions and may not output an abductive solution in finite time. Hence, in this paper we propose a new ABox abduction problem which has only finitely many abductive solutions and also propose a novel method to solve it. The method reduces the original problem to an abduction problem in logic programming and solves it with Prolog engines. Experimental results show that the method is able to compute abductive solutions in benchmark OWL DL ontologies with large ABoxes.

Axiom pinpointing plays an important role in the development and maintenance of ontologies. It helps the user to comprehend an unwanted entailment of an ontology by presenting all minimal subsets of the ontology which are responsible for the entailment (called MinAs). In this paper, we consider the problem of axiom pinpointing in description logic EL+, which underpins OWL 2 EL, a profile of the latest version of Web Ontology Language (OWL). We propose a novel method to compute all MinAs that utilizes the hierarchy information obtained from the classification of an EL+ ontology. The advantage of our method over an existing labeled classification based method is that we do not attach labels to entailed subsumptions, which can be memory exhaustion for large scale ontologies. We further consider axiom pinpointing in EL+ when ontologies change. An incremental algorithm is given to compute all MinAs by reusing MinAs previously computed.

Finding all justifications of an OWL entailment is an important reasoning service for explaining logical inconsistencies. In this paper, we consider finding all justifications of an entailment in OWL pD* fragment, which is a fragment of OWL that makes possible decidable rule extensions of OWL. We first propose a novel approach to find all justifications of OWL pD* entailments using TMS and show the complexity of this approach. This approach is limited by the hardware capabilities of standalone systems. In order to improve its scalability to handle large scale semantic data, we optimize the proposed approach by exploiting the MapReduce technology. We implement our approach and the optimization, and do experiments on synthetic and real world data sets. Evaluation results show that our approach has the ability to scale to more than one billion triples.

Computing all diagnoses of an inconsistent ontology is important in ontology-based applications. However, the number of diagnoses can be very large. It is impractical to enumerate all diagnoses before identifying the target one to render the ontology consistent. Hence, we propose to represent all diagnoses by multiple sets of partial diagnoses, where the total number of partial diagnoses can be small and the target diagnosis can be directly retrieved from these partial diagnoses. We also propose methods for computing the new representation of all diagnoses in an OWL DL ontology. Experimental results show that computing the new representation of all diagnoses is much easier than directly computing all diagnoses.

In this paper, we propose fuzzy pD* semantics which generalizes pD* semantics to reason over fuzzy RDF data using OWL vocabulary. We first define the notions of fuzzy RDF graph and fuzzy pD* interpretation. We then present a set of fuzzy pD*entailment rules and define the Best Degree Bound (BDB) of a triple derived from a fuzzy RDF graph. We show the existence of the BDB of an arbitrary triple. After that, we generalize the partial and full pD* closures to obtain the BDBs of derived triples. We show that the partial fuzzy closure exists and can be computed within polynomial time. Finally, we prove soundness and completeness results for the entailment relation. We also prove that the consistency checking is in P, the entailment is NP-complete, and in P if the target fuzzy RDF graph is ground. Therefore, extending the pD* semantics with fuzzy semantics does not increase the computational complexity.

Possibilistic logic provides a convenient tool for dealing with uncertainty and handling inconsistency. In this paper, we propose possibilistic description logics as an extension of description logics, which are a family of well-known ontology languages. We first give the syntax and semantics of possibilistic description logics and define several inference services in possibilistic description logics. We show that these inference serviced can be reduced to the task of computing the inconsistency degree of a knowledge base in possibilistic description logics. Since possibilistic inference services suffer from the drowning problem, that is, axioms whose confidence degrees are less than or equal to the inconsistency are not used, we consider a drowning-free variant of possibilistic inference, called linear order inference. We propose an algorithm for computing the inconsistency degree of a possibilistic description logic knowledge base and an algorithm for the linear order inference. We consider the impact of our possibilistic description logics on ontology learning and ontology merging. Finally, we implement these algorithms and provide some interesting evaluation results. © 2011 Wiley Periodicals, Inc.

Measuring inconsistency in knowledge bases has been recognized as an important problem in several research areas. Many methods have been proposed to solve this problem and a main class of them is based on some kind of paraconsistent semantics. However, existing methods suffer from two limitations: (i) they are mostly restricted to propositional knowledge bases; (ii) very few of them discuss computational aspects of computing inconsistency measures. In this article, we try to solve these two limitations by exploring algorithms for computing an inconsistency measure of first-order knowledge bases. After introducing a four-valued semantics for first-order logic, we define an inconsistency measure of a first-order knowledge base, which is a sequence of inconsistency degrees. We then propose a precise algorithm to compute our inconsistency measure. We show that this algorithm reduces the computation of the inconsistency measure to classical satisfiability checking. This is done by introducing a new semantics, named S[n]-4 semantics, which can be calculated by invoking a classical SAT solver. Moreover, we show that this auxiliary semantics also gives a direct way to compute upper and lower bounds of inconsistency degrees. That is, it can be easily revised to compute approximating inconsistency measures. The approximating inconsistency measures converge to the precise values if enough resources are available. Finally, by some nice properties of the S[n]-4 semantics, we show that some upper and lower bounds can be computed in P-time, which says that the problem of computing these approximating inconsistency measures is tractable.

Performance and scalability are two issues that are becoming increasingly pressing as RDF data model is applied to real-world applications. Because neither vertical nor flat structures of RDF storage can handle frequent schema updates and meanwhile avoid possible long-chain joins, there is no clear winner between these two typical structures. In this paper, we propose an alternative storage schema called open user schema. The open user schema consists of flat tables automatically extracted from RDF query streams. A query is divided into two parts and conquered,respectively, on the flat tables in the open user schema and on the vertical table stored in a backend storage. At the core of this divide and conquer architecture with open user schema, an efficient isomorphism decision algorithm is given to guide a query to related flat tables in the open user schema. Our proposal in essence departs from existing methods in that it can accommodate schema updates without possible long-chain joins. We implement our approach and provide empirical evaluations to demonstrate both efficiency and effectiveness of our approach in evaluating complex RDF queries.

Falconer is a semantic Web search engine enhanced SIOC (Semantically-Interlinked Online Communities) application, which is designed to demonstrate the ability of accelerating the creation and reuse process of semantic Web data with easy-to-use user interfaces. In this process, semantic Web search engines feed existing semantic data into the SIOC framework, where new semantic data are composed by the community and indexed again by those search engines. Compared to existing social (semantic) Web applications, Falconer inherently conforms to SIOC specification. It provides semantic search engine based user registration suggestion, friends auto-discovery, and semantic annotation for forum post content. Another distinctive feature is that it enables users to subscribe any resource having a URI as the topic they are interested in. The relationships among users, topics, and posts are further visualized for analyzing the topic trends in the community. As all semantic data are formatted in RDF and RDFa, they can be queried with SPARQL query language.

Measuring Inconsistency in ontologies is an important topic in ontology engineering as it can provide extra information for dealing with inconsistency. Many approaches have been proposed to deal with this issue. However, the main drawback of these algorithms is their high computational complexity. One of the main sources of the high complexity is the intractability of the underlying Description Logics (DLs). In this paper, we focus on an important tractable DL family, \emph{DL-Lite}. We define an inconsistency degree of a \emph{DL-Lite} ontology based on a three-valued semantics. We also present an algorithm to compute this inconsistency degree and show that its time-complexity is PTime in the size of ABox and TBox.

International audience

Knowledge representation is very important in intelligent systems -e.g. for knowledge discovery, data mining, and machine learning. The human brain, a significant intelligent system, works with a huge number of spiking neurons. Based on spiking neuron models a new generation of spiking neural networks (SNNs) has been developed for artificial intelligence systems. SNNs are computationally more powerful than conventional artificial neural networks. In this paper, the spiking neuron model is applied to represent logic rules and fuzzy rules. Based on the STDP (Spike Timing Dependent Plasticity) principle, a new SNN model is proposed for pattern recognition. An efficient learning rule derived from the STDP is applied for self-organizing the input training set efficiently. An example, Animal-Growth-Record, is used to explain the principle of the SNN model. Benchmark data sets are applied to compare the proposed approach with other approaches. As there are very efficient learning rules in the SNN model, the model can be applied not only for fusion of multi-sensory data, but also for data mining in large databases with large numbers of attributes.