Large Vision-Language Models (LVLMs) have demonstrated remarkable capabilities in multimodal understanding and generation by integrating visual and textual data. However, these models frequently exhibit object hallucination problems: generating outputs that are inconsistent with the input image. Existing improved methods for mitigating hallucinations still suffer from two key limitations: dynamic approaches based on logits or attention mechanisms risk suppressing valuable linguistic priors, whereas static methods that employ fixed intervention vectors lack the flexibility to adapt to diverse images and questions. To address these issues, we propose RFI (Rectified Flow Intervention), a novel approach that harnesses the linear trajectory design of rectified flow for input-specific adaptation and employs gradient correction to ensure coherent generation, effectively combining the adaptability of dynamic methods with the stability of static ones. RFI dynamically predicts latent-space intervention vectors while requiring only a single forward pass in LVLMs per question, achieving computational efficiency (1.09x latency overhead for 100 new tokens). Extensive experiments show RFI significantly reduces hallucinations, achieving superior performance compared to existing advanced methods, highlighting its effectiveness as a lightweight plug-and-play method for reducing LVLM's hallucination in practical applications.

Inspired by theories in language psychology, it is natural to consider more constraints, such as intentions, logic, knowledge, etc., when a complex or multi-hop question is generated.As the subtask of Multi-Hop Question Generation (MHQG), the task of Multi-Constraint Multi-Hop Question Generation (MCHQG) is more aligned with human question theories.However, it is hard to determine how to bring various high-dimensional semantic constraints, and how to integrate each constraint across all hops when a multi-hop question is being generating.To address these challenges, we introduce an effective framework which includes constraint dimensionality reduction and divide-andconquer-based dynamic projection; we call it RUBY.The proposed RUBY contains a module of high-dimensional semantic constraint dimension reduction and a module of sub-question answer pairs-based multi-hop question generation.Meanwhile, a Reasoning Dynamic Projection strategy is tailored to effectively incorporate the constraints into every hop of the multi-hop question.The experimental results demonstrate that RUBY consistently outperforms baseline models, which suggest that RUBY is able to effectively capture and integrate semantic constraints, leading to more accurate and humanlike multi-hop question generation.We release the code and data to public 1 .

Lyrics generation has garnered increasing attention within the artificial intelligence community.Our task focuses on generating harmonious Cantonese lyrics.Unlike other languages, Cantonese has a unique system of nine contours and six tones, making it essential to satisfy the harmony rules that ensure the alignment between the melody and the tonal contours of the lyrics when composing lyrics.Current research has not yet addressed the challenge of generating lyrics that adhere to Cantonese harmony rules.To tackle this issue, we propose ToneCraft, a novel framework for generating Cantonese lyrics that ensures tonal and melodic harmony.It enables LLMs to generate lyrics with a fixed character count while aligning with tonal and melodic structures.We present an algorithm that combines characterlevel control, melodic guidance, and a taskspecific loss to achieve tonal harmony without compromising generation flexibility and quality.By incorporating domain-specific expertise, we leverage pure lyric datasets to train our model, eliminating the need for aligned data.Both objective evaluations and subjective assessments show that our generated lyrics align with melodic contours significantly better than existing methods.All code and data are available at: https://github.com/purepasser- by/ToneCraft.

Citation classification, which identifies the intention behind academic citations, is pivotal for scholarly analysis. Previous works suggest fine-tuning pretrained language models (PLMs) on citation classification datasets, reaping the reward of the linguistic knowledge they gained during pretraining. However, directly fine-tuning for citation classification is challenging due to labeled data scarcity, contextual noise, and spurious keyphrase correlations. In this paper, we present a novel framework, Citss, that adapts the PLMs to overcome these challenges. Citss introduces self-supervised contrastive learning to alleviate data scarcity, and is equipped with two specialized strategies to obtain the contrastive pairs: sentence-level cropping, which enhances focus on target citations within long contexts, and keyphrase perturbation, which mitigates reliance on specific keyphrases. Compared with previous works that are only designed for encoder-based PLMs, Citss is carefully developed to be compatible with both encoder-based PLMs and decoder-based LLMs, to embrace the benefits of enlarged pretraining. Experiments with three benchmark datasets with both encoder-based PLMs and decoder-based LLMs demonstrate our superiority compared to the previous state of the art. Our code is available at: github.com/LITONG99/Citss

Single-cell RNA sequencing (scRNA-seq) technology has emerged as a powerful tool for studying complex biological systems and diseases with high resolution. However, generating sufficient high quality scRNA-seq samples for accurate downstream analyses and ensuring reproducibility remains a challenge. Previous methods often produce low-quality samples or a limited number of useful specific cell subpopulations. To address this issue, we propose a novel Efficient Single-Cell RNA-seq Flow Diffusion (named ESCFD) method based on the Diffusion Model, capable of synthesizing large-scale, high-quality scRNA-seq samples and generating either ''holistic'' or specific cellular subpopulations within a single framework. A pre-guidance mechanism is designed for synthesizing specific cellular subpopulations of high quality, and a post-guidance mechanism was designed to accelerate the sampling of scRNA-seq samples. The ESCFD can efficiently synthesize large-scale and high-quality scRNA-seq samples for various downstream tasks, surpassing previous work with faster sampling speed. Experimental results demonstrate that the ESCFD model achieves state-of-the-art performance in cell classification and cell distribution similarity when evaluated on three scRNA-seq benchmarks, and verified the effectiveness of accelerated sampling. Visualization experiments also reveal the ESCFD's ability to synthesize specific cellular subpopulations. This innovative approach has the potential to significantly advance our understanding of complex biological systems and accelerate the development of novel therapeutic interventions and personalized medicine strategies.

Recently, continual learning (CL) has attracted much attention due to its widespread applications in the real world. Given a set of data sets sequentially, continual learning aims to achieve good performance on the new data sets while avoiding deterioration in performance on the old data sets. Despite the success, most CL models follow the supervised setting, which limits their potential in data scarcity cases. Thus, some pioneering works study unsupervised CL (UCL) to discuss what CL tricks suit the unsupervised setting. However, their advancements lack in-depth analysis of the characteristics of UCL, especially the lack of attention to the use of old data. We identify that using old data sets is essential for improving the UCL model performance while existing works ignore them. Unfortunately, given a limited data storage budget, it is a nontrivial task to select representative data and effectively replay them without label assistance. To further improve the UCL performance, we present a new method in this paper, named Effective Data Selection and Replay (EDSR) for UCL. Specifi-cally, we analyze that entropy can be an effective data selection metric, where representative data usually exhibit the highest entropy in the representation space. Then, to balance the model stability for old data and the plasticity for new data, we adopt a strategy of replaying those stored representative data with a noise-enhanced knowledge distillation process. The empirical study demonstrates the outstanding performance of EDSR on benchmark computer vision data sets. Especially, EDSR shows strong resistance to forgetting old data knowledge while maintaining high accuracy. The implementation is publicly available at https://github.com/LeeJarvis996/edsr_project/tree/main/EDSR.

In the search process, it is essential to strike a balance between effectiveness and efficiency to improve search experience. Thus, ranking list truncation has become increasingly crucial. Especially in the legal domain, irrelevant cases can severely increase search costs and even compromise the pursuit of legal justice. However, there are truncation challenges that mainly arise from the distinctive structure of legal case documents, where the elements such as fact, reasoning, and judgement in a case serve as different but multi-view texts, which could result in a bad performance if the multi-view texts cannot be well-modeled. Existing approaches are limited due to their inability to handle multi-view elements information and their neglect of semantic interconnections between cases in the ranking list. In this paper, we propose a multi-view truncation framework for legal case retrieval, named MileCut. MileCut employs a case elements extraction module to fully exploit the multi-view information of cases in the ranking list. Then, MileCut applies a multi-view truncation module to select the most informative view and make a more comprehensive cut-off decision, similar to how legal experts look over retrieval results. As a practical evaluation, MileCut is assessed across three datasets, including criminal and civil case retrieval scenarios, and the results show that MileCut outperforms other methods on F1, DCG, and OIE metrics.

The enhancement of unsupervised learning of sentence representations has been significantly achieved by the utility of contrastive learning. This approach clusters the augmented positive instance with the anchor instance to create a desired embedding space. However, relying solely on the contrastive objective can result in sub-optimal outcomes due to its inability to differentiate subtle semantic variations between positive pairs. Specifically, common data augmentation techniques frequently introduce semantic distortion, leading to a semantic margin between the positive pair. While the InfoNCE loss function overlooks the semantic margin and prioritizes similarity maximization between positive pairs during training, leading to the insensitive semantic comprehension ability of the trained model. In this paper, we introduce a novel Identical and Fraternal Twins of Contrastive Learning (named IFTCL) framework, capable of simultaneously adapting to various positive pairs generated by different augmentation techniques. We propose a Twins Loss to preserve the innate margin during training and promote the potential of data enhancement in order to overcome the sub-optimal issue. We also present proof-of-concept experiments combined with the contrastive objective to prove the validity of the proposed Twins Loss. Furthermore, we propose a hippocampus queue mechanism to restore and reuse the negative instances without additional calculation, which further enhances the efficiency and performance of the IFCL. We verify the IFCL framework on nine semantic textual similarity tasks with both English and Chinese datasets, and the experimental results show that IFCL outperforms state-of-the-art methods.

Millions of smart contracts have been deployed onto the Ethereum platform, posing potential attack subjects. Therefore, analyzing contract binaries is vital since their sources are unavailable, involving identification comprising function entry identification and detecting its boundaries. Such boundaries are critical to many smart contract applications, e.g. reverse engineering and profiling. Unfortunately, it is challenging to identify functions from these stripped contract binaries due to the lack of internal function call statements and the compiler-inducing instruction reshuffling. Recently, several existing works excessively relied on a set of handcrafted heuristic rules which impose several faults. To address this issue, we propose a novel neural network-based framework for EVM bytecode Function Entries and Boundaries Identification (neural-FEBI) that does not rely on a fixed set of handcrafted rules. Instead, it used a two-level bi-Long Short-Term Memory network and a Conditional Random Field network to locate the function entries. The suggested framework also devises a control flow traversal algorithm to determine the code segments reachable from the function entry as its boundary. Several experiments on 38,996 publicly available smart contracts collected as binary demonstrate that neural-FEBI confirms the lowest and highest F1-scores for the function entries identification task across different datasets of 88.3 to 99.7, respectively. Its performance on the function boundary identification task is also increased from 79.4% to 97.1% compared with state-of-the-art. We further demonstrate that the identified function information can be used to construct more accurate intra-procedural CFGs and call graphs. The experimental results confirm that the proposed framework significantly outperforms state-of-the-art, often based on handcrafted heuristic rules.

The Unmanned Aerial Vehicles (UAVs) delivery service is being increasingly used in logistics. However, it is challenging for a UAV to precisely identify the position for parcel delivering if it is only aided by the GPS, especially in some complex environments with weak signals and high interference. For this issue, we present a knowledge distillation empowered edge intelligence architecture, KeepEdge, to achieve visual information-assisted positioning for the UAV delivery services. Specifically, we integrate deep neural networks (DNN) into an edge computing framework to enable edge intelligence which empowers the UAVs to autonomously identify the expected delivery position. Deploying the DNN model and conducting model inference on UAVs however, requires high computing performance. To manage the trade-off between the limited resources onboard the UAVs and high-performance requirements, we employ knowledge distillation to produce a lightweight model with high accuracy based on the full model trained in the cloud. The lightweight model with significantly lower complexity and less inference latency is used onboard of the UAVs for accurate positioning. Comprehensive experiments show that the proposed architecture achieves satisfactory performance for assisted positioning. A real-world case study is presented to demonstrate the effectiveness of the proposed edge intelligence solution for UAV delivery services.

Patents are legal rights issued to inventors to protect their inventions for a certain period and play an important role in today's artificial innovation. With the ever-increasing number of patents each year, an effective and efficient patent management and search system is indispensable for determining how different an invention is from prior works from the vast amount of patent data. However, the chnologists are using now is still based on the strategy of traditional keyword-based Boolean, which requires complex bool expressions. This type of strategy leads to poor performance and costs too much labor power to filter in post-processing. To address these issues, we proposed CoPatE: a novel Contrastive Learning Framework for Patent Embeddings to capture the high-level semantics of the large-scale patents, where a patent semantic compression module learns the informative claims to reduce the computational complexity, and a tags auxiliary learning module is to enhance the semantics of a patent from the structure to learn the high-quality patent embeddings. The CoPatE is trained with the patents from USPTO from 2013 to 2020 and tested by the patents from 2021 with the CPC scheme. The experimental results demonstrate that our model achieves a 17.7% increase at [email protected] compared to the second-best method on the patent retrieval task and achieves 64.5% at Micro-F1 in the patent classification task.

EOSIO smart contracts are programs that can be collectively executed by a network of mutually untrusted nodes. As EOSIO smart contracts manage valuable assets, they become high-value targets and are subjected to more and more attacks. Tools for protecting EOSIO smart contracts are imperative. This paper proposes EOSIOAnalyzer, an effective static secu-rity analysis framework for EOSIO smart contracts to counter the three most common attacks. The framework consists of three components, the control flow graph builder, the static analyzer and the vulnerability detector. This paper implements an approach to transforming low-level Wasm bytecode into a high-level intermediate representation (Register Transfer Language). Besides, this paper also implements vulnerability detection speci-fications for three popular EOSIO smart contracts vulnerabilities, including Fake EOS Transfer, Forged Transfer Notification and Block Information Dependency. As a proof of concept, this paper conducts experiments to evaluate the effectiveness and efficiency of the EOSIOAnalyzer. The experiment results show that the detection accuracy of the three vulnerabilities is 100 %, 98.8 % and 100%, respectively.

The financial property of Ethereum makes smart contract attacks frequently bring about tremendous economic loss. Method for effective detection of vulnerabilities in contracts imperative. Existing efforts for contract security analysis heavily rely on rigid rules defined by experts, which are labor-intensive and non-scalable. There is still a lack of effort that considers combining expert-defined security patterns with deep learning. This paper proposes EtherGIS, a vulnerability detection framework that utilizes graph neural networks (GNN) and expert knowledge to extract the graph feature from smart contract control flow graphs (CFG). To gain multi-dimensional contract information and reinforce the attention of vulnerability-related graph features, sensitive EVM instruction corpora are constructed by analyzing EVM underlying logic and diverse vulnerability triggering mechanisms. The characteristic of nodes and edges in a CFG is initially confirmed according to the corpora, generating the corresponding attribute graph. GNN is adopted to aggregate the whole graph's attribute and structure information, bridging the semantic gap between low-level graph features and high-level contract features. The feature representation of the graph is finally input into the graph classification model for vulnerability detection. Furthermore, automated machine learning (AutoML) is adopted to automate the entire deep learning process. Data for this research was collected from Ethereum to build up a dataset of six vulnerabilities for evaluation. Experimental results demonstrate that EtherGIS can productively detect vulnerabilities in Ethereum smart contracts in terms of accuracy, precision, recall, and F1-score. All aspects outperform the existing work.

Sentence representation approaches have been widely used and proven to be effective in many text modeling tasks and downstream applications. Many recent proposals are available on learning sentence representations based on deep neural frameworks. However, these methods are pre-trained in open domains and depend on the availability of large-scale data for model fitting. As a result, they may fail in some special scenarios, where data are sparse and embedding interpretations are required, such as legal, medical, or technical fields. In this paper, we present an unsupervised learning method to exploit representations of sentences for some closed domains via topic modeling. We reformulate the inference process of the sentences with the corresponding contextual sentences and the associated words, and propose an effective context-enhanced process called the bi-Directional Context-enhanced Sentence Representation Learning (bi-DCSR). This method takes advantage of the semantic distributions of the nearby contextual sentences and the associated words to form a context-enhanced sentence representation. To support the bi-DCSR, we develop a novel Bayesian topic model to embed sentences and words into the same latent interpretable topic space called the Hybrid Priors Topic Model (HPTM). Based on the defined topic space by the HPTM, the bi-DCSR method learns the embedding of a sentence by the two-directional contextual sentences and the words in it, which allows us to efficiently learn high-quality sentence representations in such closed domains. In addition to an open-domain dataset from Wikipedia, our method is validated using three closed-domain datasets from legal cases, electronic medical records, and technical reports. Our experiments indicate that the HPTM significantly outperforms on language modeling and topic coherence, compared with the existing topic models. Meanwhile, the bi-DCSR method does not only outperform the state-of-the-art unsupervised learning methods on closed domain sentence classification tasks, but also yields competitive performance compared to these established approaches on the open domain. Additionally, the visualizations of the semantics of sentences and words demonstrate the interpretable capacity of our model.

BACKGROUND: In medicine, chromosome karyotyping analysis plays a crucial role in prenatal diagnosis for diagnosing whether a fetus has severe defects or genetic diseases. However, chromosome instance segmentation is the most critical obstacle to automatic chromosome karyotyping analysis due to the complicated morphological characteristics of chromosome clusters, restricting chromosome karyotyping analysis to highly depend on skilled clinical analysts. METHOD: In this paper, we build a clinical dataset and propose multiple segmentation baselines to tackle the chromosome instance segmentation problem of various overlapping and touching chromosome clusters. First, we construct a clinical dataset for deep learning-based chromosome instance segmentation models by collecting and annotating 1,655 privacy-removal chromosome clusters. After that, we design a chromosome instance labeled dataset augmentation (CILA) algorithm for the clinical dataset to improve the generalization performance of deep learning-based models. Last, we propose a chromosome instance segmentation framework and implement multiple baselines for the proposed framework based on various instance segmentation models. RESULTS AND CONCLUSIONS: segmentation precision, and 95.38% accuracy, which exceeds results reported in current chromosome instance segmentation methods. The quantitative evaluation results demonstrate the effectiveness and advancement of the proposed method for the chromosome instance segmentation problem. The experimental code and privacy-removal clinical dataset can be found at Github.

Abstract Web service discovery is a fundamental task in service-oriented architectures which searches for suitable web services based on users’ goals and preferences. In this paper, we present a novel service discovery approach that can support user queries with various-size-grained text elements. Compared with existing approaches that only support semantics matchmaking in single texture granularity (either word level or paragraph level), our approach enables the requester to search for services with any type of query content with high performance, including word, phrase, sentence, or paragraph. Specifically, we present an unsupervised Bayesian probabilistic model, bi-Directional Sentence-Word Topic Model (bi-SWTM), to achieve semantic matchmaking between possible textual types of queries (word, phrase, sentence, paragraph) and the texts in web service descriptions, by mapping words and sentences in the same semantic space. The bi-SWTM captures textual semantics of the words and sentences in a probabilistic simplex, which provides a flexible method to build the semantic links from user queries to service descriptions. The novel approach is validated using a collection of comprehensive experiments on ProgrammableWeb data. The results demonstrate that the bi-SWTM outperforms state-of-the-art methods on service discovery and classification. The visualization of the nearest-neighbored queries and descriptions shows the capability of our model on capturing the latent semantics of web services.

In medicine, any abnormalities in the number of chromosomes or the structure of chromosomes may cause the newborn baby to suffer from genetic diseases, such as Edward syndrome and so on. Chromosome karyotype analysis is the most important and common method for prenatal diagnosis to determine whether a newborn baby has chromosome defects refers to segment chromosome instances from stained cell images and arrange chromosome instances according to their categories. However, due to the non-rigid nature of chromosomes, chromosome instances may overlap and adhere to each other, which makes the task of segmenting chromosome instances time-consuming and error-prone. This paper proposes a novel chromosome instance segmentation method that includes three stages. First, we segment a given stained cell image into several segments using geometric connectivity. Second, a machine learning method is proposed to distinguish chromosome of individual instances and clusters. Finally, a deep learning-based method is applied to separate chromosome instances from clusters. It shows that the proposed method achieves 97.61% instance segmentation accuracy in a hold-out clinical dataset with 162 cell images consisting of 7,452 chromosome instances, which is a promising result in clinical application. The innovation of this work is to combine geometry and deep learning to handle tasks for different stages of chromosome instance segmentation issue. The benefit of this innovation is that it can obtain a much better performance than existing geometric-based methods with a small number of training samples. Meanwhile, the segmentation performance of the proposed method is superior to existing methods fully based on deep learning.

Knowledge representation learning has received a lot of attention in the past few years. The success of existing methods heavily relies on the quality of knowledge graphs. The entities with few triplets tend to be learned with less expressive power. Fortunately, there are many knowledge graphs constructed from various sources, the representations of which could contain much information. We propose an adversarial embedding transfer network ATransN, which transfers knowledge from one or more teacher knowledge graphs to a target one through an aligned entity set without explicit data leakage. Specifically, we add soft constraints on aligned entity pairs and neighbours to the existing knowledge representation learning methods. To handle the problem of possible distribution differences between teacher and target knowledge graphs, we introduce an adversarial adaption module. The discriminator of this module evaluates the degree of consistency between the embeddings of an aligned entity pair. The consistency score is then used as the weights of soft constraints. It is not necessary to acquire the relations and triplets in teacher knowledge graphs because we only utilize the entity representations. Knowledge graph completion results show that ATransN achieves better performance against baselines without transfer on three datasets, CN3l, WK3l, and DWY100k. The ablation study demonstrates that ATransN can bring steady and consistent improvement in different settings. The extension of combining other knowledge graph embedding algorithms and the extension with three teacher graphs display the promising generalization of the adversarial transfer network.

Because of its efficiency, word embedding has been widely used in many natural language processing and text modeling tasks. It aims to represent each word by a vector so such that the geometry between these vectors can capture the semantic correlations between words. An ambiguous word can often have diverse meanings in different contexts, a quality which is called polysemy. The bulk of studies aimed to generate only one single embedding for each word, whereas a few studies have made a small number of embeddings to present different meanings of each word. However, it is hard to determine the exact number of senses for each word, as meanings depend on contexts. To address this problem, this paper proposes a novel adaptive cross-contextual word embedding (ACWE) method for capturing the word polysemy in different contexts based on topic modeling, in which the word polysemy is defined over a latent interpretable semantic space. The proposed ACWE consists of two main parts, in the first of which an unsupervised cross-contextual probabilistic word embedding model is designed to obtain the global word embeddings, and each word is represented by an embedding in the unified latent semantic space. Based on the global word embeddings, an adaptive cross-contextual word embedding process is then devised in the second part to learn the local embeddings for each polysemous word in different contexts. In fact, a word embedding is adaptively adjusted and updated with respect to different contexts to generate different word embeddings tailored to the corresponding contexts. The proposed ACWE is validated on two datasets collected from Wikipedia and IMDb on different tasks including word similarity, polysemy induction, semantic interpretability, and text classification. Experimental results indicate that ACWE does not only outperform the established word embedding methods, which consider word polysemy on six popular benchmark datasets, but it also yields competitive performance compared with state-of-the-art deep learning-based approaches without considering polysemy. Moreover, the proposed ACWE significantly improves the performances of text classification both in precision and F1, and the visualizations of the semantics of words demonstrate the feasibility and advantage of the proposed ACWE model on polysemy.

Analyzing textual semantics in matching user query and service description is critical for Web service discovery. Existing works mostly extract the features of the description and query independently, downgrading them into word-level calculation, which can not jointly extract the accurate semantics. For this issue, this work explores a way to enable the semantic matching for the contents (including words, phrases, or sentences) in query and the sentences in service description, by mapping words and sentences into the same semantic space. Specifically, we propose an unsupervised Bayesian probabilistic model, bi-Directional Hybrid Priors Topic Model (bi-HPTM), to capture the textual semantics of the words and sentences in a probabilistic simplex, which provides a flexible operation to build the semantic links from the queries to service descriptions. Meanwhile, the textual semantics generated by bi-HPTM is highly interpretable that help to understand the user requirements. The proposed model is examined by ProgrammableWeb. Experimental results demonstrate that bi-HPTM outperforms state-of-the-art methods for semantic service discovery on service classification and retrieval. The visualizations of the nearest-neighbored queries and descriptions show the insights of our model on capturing the latent semantics of Web services.

Word embeddings have been widely used and proven to be effective in many natural language processing and text modeling tasks. It is obvious that one ambiguous word could have very different semantics in various contexts, which is called polysemy. Most existing works aim at generating only one single embedding for each word while a few works build a limited number of embeddings to present different meanings for each word. However, it is hard to determine the exact number of senses for each word as the word meaning is dependent on contexts. To address this problem, we propose a novel Adaptive Probabilistic Word Embedding (APWE) model, where the word polysemy is defined over a latent interpretable semantic space. Specifically, at first each word is represented by an embedding in the latent semantic space and then based on the proposed APWE model, the word embedding can be adaptively adjusted and updated based on different contexts to obtain the tailored word embedding. Empirical comparisons with state-of-the-art models demonstrate the superiority of the proposed APWE model.

In a document, the topic distribution of a sentence depends on both the topics of its neighbored sentences and its own content, and it is usually affected by the topics of the neighbored sentences with different weights. The neighbored sentences of a sentence include the preceding sentences and the subsequent sentences. Meanwhile, it is natural that a document can be treated as a sequence of sentences. Most existing works for Bayesian document modeling do not take these points into consideration. To fill this gap, we propose a bi-Directional Recurrent Attentional Topic Model (bi-RATM) for document embedding. The bi-RATM not only takes advantage of the sequential orders among sentences but also uses the attention mechanism to model the relations among successive sentences. To support to the bi-RATM, we propose a bi-Directional Recurrent Attentional Bayesian Process (bi-RABP) to handle the sequences. Based on the bi-RABP, bi-RATM fully utilizes the bi-directional sequential information of the sentences in a document. Online bi-RATM is proposed to handle large-scale corpus. Experiments on two corpora show that the proposed model outperforms state-of-the-art methods on document modeling and classification.

Heng Wang, Shuangyin Li, Rong Pan, Mingzhi Mao. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

It is difficult to train a personalized task-oriented dialogue system because the data collected from each individual is often insufficient. Personalized dialogue systems trained on a small dataset is likely to overfit and make it difficult to adapt to different user needs. One way to solve this problem is to consider a collection of multiple users as a source domain and an individual user as a target domain, and to perform transfer learning from the source domain to the target domain. By following this idea, we propose a PErsonalized Task-oriented diALogue (PETAL) system, a transfer reinforcement learning framework based on POMDP, to construct a personalized dialogue system. The PETAL system first learns common dialogue knowledge from the source domain and then adapts this knowledge to the target domain. The proposed PETAL system can avoid the negative transfer problem by considering differences between the source and target users in a personalized Q-function. Experimental results on a real-world coffee-shopping data and simulation data show that the proposed PETAL system can learn optimal policies for different users, and thus effectively improve the dialogue quality under the personalized setting.

Knowledge representation learning aims at modeling knowledge graph by encoding entities and relations into a low dimensional space. Most of the traditional works for knowledge embedding need negative sampling to minimize a margin-based ranking loss. However, those works construct negative samples through a random mode, by which the samples are often too trivial to fit the model efficiently. In this paper, we propose a novel knowledge representation learning framework based on Generative Adversarial Networks (GAN). In this GAN-based framework, we take advantage of a generator to obtain high-quality negative samples. Meanwhile, the discriminator in GAN learns the embeddings of the entities and relations in knowledge graph. Thus, we can incorporate the proposed GAN-based framework into various traditional models to improve the ability of knowledge representation learning. Experimental results show that our proposed GAN-based framework outperforms baselines on triplets classification and link prediction tasks.

In a document, the topic distribution of a sentence depends on both the topics of preceding sentences and its own content, and it is usually affected by the topics of the preceding sentences with different weights. It is natural that a document can be treated as a sequence of sentences. Most existing works for Bayesian document modeling do not take these points into consideration. To fill this gap, we propose a Recurrent Attentional Topic Model (RATM) for document embedding. The RATM not only takes advantage of the sequential orders among sentence but also use the attention mechanism to model the relations among successive sentences. In RATM, we propose a Recurrent Attentional Bayesian Process (RABP) to handle the sequences. Based on the RABP, RATM fully utilizes the sequential information of the sentences in a document. Experiments on two copora show that our model outperforms state-of-the-art methods on document modeling and classification.

In recent years, recurrent neural networks have been widely used for various text classification tasks. However, most of the recurrent architectures will not assign a class label to a text until they read the last word, while human beings are able to determine the text class before reading the whole text. In this paper, we propose a Length Adaptive Recurrent Model (LARM) which can automatically determine the minimum text length that is necessary to perform the classification. With three parts includingReader, Predictor andAgent, our model is designed to read a text word by word, and terminate the process when the adequate information has been caught for the text classification task. The experimental results show that our model has comparable or even better performance compared to the vanilla LSTM when both are fed with partial text input. Besides, we can speed up text classification by truncating the text when sufficient evidence is found for classification. Furthermore, we also visualize our model and show that our model works like human beings, who can gradually come up with the general idea of a text while reading texts sequentially.

In the last decade, there has been a huge amount of documents with different types of rich metadata information, which belongs to the Semi-Structured Documents (SSDs), appearing in many real applications. It is an interesting research work to model this type of text data following the way how humans understand text with informative metadata. In the paper, we introduce a Self-paced Compensatory Deep Boltzmann Machine (SCDBM) architecture that learns a deep neural network by using metadata information to learn deep structure layer-wisely for Semi-Structured Documents (SSDs) embedding in a self-paced way. Inspired by the way how humans understand text, the model defines a deep process of document vector extraction beyond the space of words by jointing the metadata where each layer selects different types of metadata. We present efficient learning and inference algorithms for the SCDBM model and empirically demonstrate that using the representation discovered by this model has better performance on semi-structured document classification and retrieval, and tag prediction comparing with state-of-the-art baselines.

In the past two decades, there has been a huge amount of document data with rich tag information during the evolution of the Internet, which can be called semi-structured data. These semi-structured data contain both unstructured features (e.g., plain text) and metadata, such as tags in html files or author and venue information in research articles. It's of great interest to model such kind of data. Most previous works focused on modeling the unstructured data. Some other methods have been proposed to model the unstructured data with specific tags. To build a general model for semi-structured documents remains an important problem in terms of both model fitness and efficiency. In this paper, we propose a novel method to model the tagged documents by a so-called Tag-Weighted Dirichlet Allocation (TWDA). TWDA is a framework that leverages both the tags and words in each document to infer the topic components for the documents. This allows not only to learn the document-topic and topic-word distributions, but also to infer the tag-topic distributions for text mining (e.g., classification, clustering, and recommendations). Moreover, TWDA can automatically infer the probabilistic weights of tags for each document, that can be used to predict the tags in one document. We present an efficient variational inference method with an EM algorithm for estimating the model parameters. The experimental results show the effectiveness, efficiency and robustness of our TWDA approach by comparing it with the state-of-the-art methods on four corpora in document modeling, tags prediction and text classification.