Deep Research (DR) requires LLM agents to autonomously perform multi-step information seeking, processing, and reasoning to generate comprehensive reports. In contrast to existing studies that mainly focus on unstructured web content, a more challenging DR task should additionally utilize structured knowledge to provide a solid data foundation, facilitate quantitative computation, and lead to in-depth analyses. In this paper, we refer to this novel task as Knowledgeable Deep Research (KDR), which requires DR agents to generate reports with both structured and unstructured knowledge. Furthermore, we propose the Hybrid Knowledge Analysis framework (HKA), a multi-agent architecture that reasons over both kinds of knowledge and integrates the texts, figures, and tables into coherent multimodal reports. The key design is the Structured Knowledge Analyzer, which utilizes both coding and vision-language models to produce figures, tables, and corresponding insights. To support systematic evaluation, we construct KDR-Bench, which covers 9 domains, includes 41 expert-level questions, and incorporates a large number of structured knowledge resources (e.g., 1,252 tables). We further annotate the main conclusions and key points for each question and propose three categories of evaluation metrics including general-purpose, knowledge-centric, and vision-enhanced ones. Experimental results demonstrate that HKA consistently outperforms most existing DR agents on general-purpose and knowledge-centric metrics, and even surpasses the Gemini DR agent on vision-enhanced metrics, highlighting its effectiveness in deep, structure-aware knowledge analysis. Finally, we hope this work can serve as a new foundation for structured knowledge analysis in DR agents and facilitate future multimodal DR studies.

Inductive knowledge graph completion (KGC) aims to predict missing triples with unseen entities. Recent works focus on modeling reasoning paths between the head and tail entity as direct supporting evidence. However, these methods depend heavily on the existence and quality of reasoning paths, which limits their general applicability in different scenarios. In addition, we observe that latent type constraints and neighboring facts inherent in KGs are also vital in inferring missing triples. To effectively utilize all useful information in KGs, we introduce CATS, a novel context-aware inductive KGC solution. With sufficient guidance from proper prompts and supervised fine-tuning, CATS activates the strong semantic understanding and reasoning capabilities of large language models to assess the existence of query triples, which consist of two modules. First, the type-aware reasoning module evaluates whether the candidate entity matches the latent entity type as required by the query relation. Then, the subgraph reasoning module selects relevant reasoning paths and neighboring facts, and evaluates their correlation to the query triple. Experiment results on three widely used datasets demonstrate that CATS significantly outperforms state-of-the-art methods in 16 out of 18 transductive, inductive, and few-shot settings with an average absolute MRR improvement of 7.2%.

Retrieval-augmented generation (RAG) has improved large language models (LLMs) by using knowledge retrieval to overcome knowledge deficiencies. However, current RAG methods often fall short of ensuring the depth and completeness of retrieved information, which is necessary for complex reasoning tasks. In this work, we introduce Think-on-Graph 2.0 (ToG-2), a hybrid RAG framework that iteratively retrieves information from both unstructured and structured knowledge sources in a tight-coupling manner. Specifically, ToG-2 leverages knowledge graphs (KGs) to link documents via entities, facilitating deep and knowledge-guided context retrieval. Simultaneously, it utilizes documents as entity contexts to achieve precise and efficient graph retrieval. ToG-2 alternates between graph retrieval and context retrieval to search for in-depth clues relevant to the question, enabling LLMs to generate answers. We conduct a series of well-designed experiments to highlight the following advantages of ToG-2: 1) ToG-2 tightly couples the processes of context retrieval and graph retrieval, deepening context retrieval via the KG while enabling reliable graph retrieval based on contexts; 2) it achieves deep and faithful reasoning in LLMs through an iterative knowledge retrieval process of collaboration between contexts and the KG; and 3) ToG-2 is training-free and plug-and-play compatible with various LLMs. Extensive experiments demonstrate that ToG-2 achieves overall state-of-the-art (SOTA) performance on 6 out of 7 knowledge-intensive datasets with GPT-3.5, and can elevate the performance of smaller models (e.g., LLAMA-2-13B) to the level of GPT-3.5’s direct reasoning. The source code is available on https://anonymous.4open.science/r/ToG2.

Muzhi Li, Cehao Yang, Chengjin Xu, Xuhui Jiang, Yiyan Qi, Jian Guo, Ho-fung Leung, Irwin King. 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.

Natural Language to SQL (NL2SQL) enables intuitive interactions with databases by transforming natural language queries into structured SQL statements. Despite recent advancements in enhancing human-computer interaction within database applications, significant challenges persist, particularly regarding the inference performance in complex scenarios involving multi-table joins and nested queries. Current methodologies primarily utilize supervised fine-tuning (SFT) to train the NL2SQL model, which may limit adaptability and interpretability in new environments (e.g., finance and healthcare). In order to enhance the reasoning performance of the NL2SQL model in the above complex situations, we introduce SQL-R1, a novel NL2SQL reasoning model trained by the reinforcement learning (RL) algorithms. We design a specialized RL-based reward function tailored for NL2SQL tasks and discussed the impact of cold start and synthetic data on the effectiveness of intensive training. In addition, we achieve competitive accuracy using only a tiny amount of synthetic NL2SQL data for augmented training and further explore data engineering for RL. In existing experiments, SQL-R1 achieves execution accuracy of 88.6\% and 67.1\% on the benchmark Spider and BIRD, respectively. The code is available at https://github.com/IDEA-FinAI/SQL-R1.

ABSTRACT Dams, as crucial water conservancy engineering facilities, play a role in safe guarding people's livelihoods and providing economic benefits. However, due to the impact of natural factors and human activities, dams may develop cracks and other potential safety hazards during operation. Crack detection can identify these potential issues in a timely manner, allowing for appropriate measures to be taken for repair and reinforcement, thereby preventing catastrophic consequences such as dam breaches under extreme weather or geological conditions. In the process of dam crack detection, this paper presents a method, YOLOv9‐LAE, which may solve missed or false detections. Firstly, the large separable kernel attention (LSKA) module is introduced, which emphasises positional information while focusing on channel features. Secondly, the SPPFELAN in YOLOV9 is replaced by the AIFI module, as capturing the key information needed in the image will enable the following modules to accurately detect the crack information. Finally, the EIOU to calculate the loss, accelerating training convergence and improving the accuracy of crack detection. The research results indicate that YOLOV9‐LAE achieves a precision of 90.7%, the recall rate is 75.1%, with at 81.5% and at 60.6%. Compared to YOLOv9, the precision has improved by 9.9%, the recall has increased by 2%, has risen by 1.5% and has been enhanced by 1.5%.

Xuhui Jiang, Yinghan Shen, Zhichao Shi, Chengjin Xu, Wei Li, Zixuan Li, Jian Guo, Huawei Shen, Yuanzhuo Wang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Stance detection aims to identify the expressed attitude towards a target from the text, which is significant for learning public cognition from social media. The short and implicit nature of social media users’ expressions potentially results in the stance understanding bias of the model. To address this problem, introducing external background information is helpful to mitigate these biases and enhance explainability. The core view, reflecting the motivations and reasons behind an individual’s stance toward the target, can be summarized and extracted from collective tweets, which can serve as a reference for stance detection. In this study, we propose the Stance Detection via Core View Discovery (SD-CVM), where the core views are used for background information modeling. Specifically, we construct a joint classifier combining the semantic understanding of tweets and their relevant core views from the public. We utilize the Large Language Model (LLM) to extract core views with stances from tweets and use these core views as background references for tweets. To further optimize the tweet understanding, we develop the contrastive and rebalancing mechanism by incorporating stance supervision signals for training. Experiments on two representative datasets demonstrate the excellent performance of our method.

The flourishing of knowledge graph (KG) applications has driven the need for entity alignment (EA) across KGs. However, the heterogeneity of practical KGs, characterized by differing scales, structures, and limited overlapping entities, greatly surpasses that of existing EA datasets. This discrepancy highlights an oversimplified heterogeneity in current EA datasets, which obstructs the exploration of the EA application. In this paper, we study the performance of EA methods on the alignment of highly heterogeneous KGs (HHKGs). Firstly, we address the oversimplified heterogeneity settings of current datasets and propose two new HHKG datasets that closely mimic practical EA scenarios. Then, based on these datasets, we conduct extensive experiments to evaluate previous representative EA methods. Our findings reveal that, in aligning HHKGs, valuable structure information can hardly be exploited, which leads to inferior performance of existing EA methods, especially those based on GNNs. These findings shed light on the potential problems associated with the conventional application of GNN-based methods as a panacea for all EA datasets. Consequently, to elucidate what EA methodology is genuinely beneficial in practical scenarios, we undertake an in-depth analysis by implementing a simple but effective approach: Simple-HHEA. Our experiment results conclude that the key to the future EA model design in practice lies in their adaptability and efficiency to varying information quality conditions, as well as their capability to capture patterns across HHKGs. The datasets and source code are available at https://github.com/IDEA-FinAI/Simple-HHEA.

Session search is a widely adopted technique in search engines that seeks to leverage the complete interaction history of a search session to better understand the information needs of users and provide more relevant ranking results. The vast majority of existing methods model a search session as a sequence of queries and previously clicked documents. However, if we simply represent a search session as a sequence we will lose the topological information in the original search session. It is non-trivial to model the intra-session interactions and complicated structural patterns among the previously issued queries, clicked documents, as well as the terms or entities that appeared in them. To solve this problem, in this paper, we propose a novel Session Search with Graph Classification Model (SSGC), which regards session search as a graph classification task on a heterogeneous graph that represents the search history in each session. To improve the performance of the graph classification, we design a specific pre-training strategy for our proposed GNN-based classification model. Extensive experiments on two public session search datasets demonstrate the effectiveness of our model in the session search task.

Complex question generation over knowledge bases (KB) aims to generate natural language questions involving multiple KB relations or functional constraints. Existing methods train one encoder-decoder-based model to fit all questions. However, such a one-size-fits-all strategy may not perform well since complex questions exhibit an uneven distribution in many dimensions, such as question types, involved KB relations, and query structures, resulting in insufficient learning for long-tailed samples under different dimensions. To address this problem, we propose a meta-learning framework for complex question generation. The meta-trained generator can acquire universal and transferable meta-knowledge and quickly adapt to long-tailed samples through a few most related training samples. To retrieve similar samples for each input query, we design a self-supervised graph retriever to learn distributed representations for samples, and contrastive learning is leveraged to improve the learned representations. We conduct experiments on both WebQuestionsSP and ComplexWebQuestion, and results on long-tailed samples of different dimensions have been significantly improved, which demonstrates the effectiveness of the proposed framework.