Zhenhao Zhou, Zhuochen Huang, Yike He, Chong Wang, Jiajun Wang, Yijian Wu, Xin Peng, Yiling Lou. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recently, Automated Vulnerability Localization (AVL) has attracted growing attention, aiming to facilitate diagnosis by pinpointing the specific lines of code responsible for vulnerabilities. Large Language Models (LLMs) have shown potential in various domains, yet their effectiveness in line-level vulnerability localization remains underexplored.In this work, we present the first comprehensive empirical evaluation of LLMs for AVL. Our study examines 19 leading LLMs suitable for code analysis, including ChatGPT and multiple open-source models, spanning encoder-only, encoder-decoder, and decoder-only architectures, with model sizes from 60M to 70B parameters. We evaluate three paradigms—few-shot prompting, discriminative fine-tuning, and generative fine-tuning—with and without Low-Rank Adaptation (LoRA), on both a BigVul-derived dataset for C/C++ and a smart contract vulnerability dataset.Our results show that discriminative fine-tuning achieves substantial performance gains over existing learning-based AVL methods when sufficient training data is available. In low-data settings, prompting advanced LLMs such as ChatGPT proves more effective. We also identify challenges related to input length and unidirectional context during fine-tuning, and propose two remedial strategies: a sliding window approach and right-forward embedding, both of which yield significant improvements. Moreover, we provide the first assessment of LLM generalizability in AVL, showing that certain models can transfer effectively across Common Weakness Enumerations (CWEs) and projects. However, performance degrades notably for newly discovered vulnerabilities containing unfamiliar lexical or structural patterns, underscoring the need for continual adaptation. These findings offer practical guidance for deploying LLM-based AVL systems in realistic software security workflows.

Yihao Huang, Chong Wang, Xiaojun Jia, Qing Guo, Felix Juefei-Xu, Jian Zhang, Yang Liu, Geguang Pu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Alperen Yildiz, Sin G Teo, Yiling Lou, Yebo Feng, Chong Wang, Dinil Mon Divakaran. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Python's dynamic typing system offers flexibility and expressiveness but can lead to type-related errors, prompting the need for automated type inference to enhance type hinting. While existing learning-based approaches show promising inference accuracy, they struggle with practical challenges in comprehensively handling various types, including complex parameterized types and (unseen) user-defined types. In this paper, we introduce TIGER, a two-stage generating-then-ranking (GTR) framework, designed to effectively handle Python's diverse type categories. TIGER leverages fine-tuned pre-trained code models to train a generative model with a span masking objective and a similarity model with a contrastive training objective. This approach allows TIGER to generate a wide range of type candidates, including complex parameterized types in the generating stage, and accurately rank them with user-defined types in the ranking stage. Our evaluation on the ManyTypes4Py dataset shows TIGER's advantage over existing methods in various type categories, notably improving accuracy in inferring user-defined and unseen types by 11.2% and 20.1% respectively in Top-5 Exact Match. Moreover, the experimental results not only demonstrate TIGER's superior performance and efficiency, but also underscore the significance of its generating and ranking stages in enhancing automated type inference.

Automated library APIs testing is difficult as it requires exploring a vast space of parameter inputs that may involve objects with complex data types. Existing search based approaches, with limited knowledge of relations between object states and program branches, often suffer from the low efficiency issue, i.e., tending to generate invalid inputs. Symbolic execution based approaches can effectively identify such relations, but fail to scale to large programs. In this work, we present an LLM-based input space partitioning testing approach, LISP, for library APIs. The approach lever-ages LLMs to understand the code of a library API under test and perform input space partitioning based on its understanding and rich common knowledge. Specifically, we provide the signature and code of the API under test to LLMs, with the expectation of obtaining a text description of each input space partition of the API under test. Then, we generate inputs through employing the generated text description to sample inputs from each partition, ultimately resulting in test suites that systematically explore the program behavior of the API. We evaluate LISP on more than 2,205 library API meth-ods taken from 10 popular open-source Java libraries (e.g.,$a$p$a$che/commons-lang with 2.6k stars, guava with 48.8k stars on GitHub). Our experiment results show that LISP is effective in library API testing. It significantly outperforms state-of-the-art tool EvoSuite in terms of edge coverage. On average, LISP achieves 67.82 % branch coverage, surpassing EvoSuite by 1.21 times. In total, LISP triggers 404 exceptions or errors in the experiments, and discovers 13 previously unknown vulnerabilities during evaluation, which have been assigned CVE IDs.

Code Large Language Models (LLMs) have demonstrated remarkable capabilities in generating, understanding, and manipulating programming code. However, their training process inadvertently leads to the memorization of sensitive information, posing severe privacy risks. Existing studies on memorization in LLMs primarily rely on prompt engineering techniques, which suffer from limitations such as widespread hallucination and inefficient extraction of the target sensitive information. In this paper, we present a novel approach to characterize real and fake secrets generated by Code LLMs based on token probabilities. We identify four key characteristics that differentiate genuine secrets from hallucinated ones, providing insights into distinguishing real and fake secrets. To overcome the limitations of existing works, we propose DeSec,a two-stage method that leverages token-level features derived from the identified characteristics to guide the token decoding process. DeSec consists of constructing an offline token scoring model using a proxy Code LLM and employing the scoring model to guide the decoding process by reassigning token likelihoods. Through extensive experiments on four state-of-the-art Code LLMs using a diverse dataset, we demonstrate the superior performance of DeSec in achieving a higher plausible rate and extracting more real secrets compared to existing baselines. Our findings highlight the effectiveness of our token-level approach in enabling an extensive assessment of the privacy leakage risks associated with Code LLMs.

Neural code models (NCMs) have demonstrated extraordinary capabilities in code intelligence tasks. Meanwhile, the security of NCMs and NCMs-based systems has garnered increasing attention. In particular, NCMs are often trained on large-scale data from potentially untrustworthy sources, providing attackers with the opportunity to manipulate them by inserting crafted samples into the data. This type of attack is called a code poisoning attack (also known as a backdoor attack). It allows attackers to implant backdoors in NCMs and thus control model behavior, which poses a significant security threat. However, there is still a lack of effective techniques for detecting various complex code poisoning attacks. In this paper, we propose an innovative and lightweight technique for code poisoning detection named KillbadCode. KillbadCode is designed based on our insight that code poisoning disrupts the naturalness of code. Specifically, KillBADCODE first builds a code language model (CodeLM) on a lightweight$n$-gram language model. Then, given poisoned data, KillbadCode utilizes CodeLM to identify those tokens in (poisoned) code snippets that will make the code snippets more natural after being deleted as trigger tokens. Considering that the removal of some normal tokens in a single sample might also enhance code naturalness, leading to a high false positive rate (FPR), we aggregate the cumulative improvement of each token across all samples. Finally, KillbadCode purifies the poisoned data by removing all poisoned samples containing the identified trigger tokens. We conduct extensive experiments to evaluate the effectiveness and efficiency of KillbadCode, involving two types of advanced code poisoning attacks (a total of five poisoning strategies) and datasets from four representative code intelligence tasks. The experimental results demonstrate that across 20 code poisoning detection scenarios, KillbadCode achieves an average FPR of 8.30 % and an average Recall of 100 %, significantly outperforming four baselines. More importantly, KillBadCode is very efficient, with a minimum time consumption of only 5 minutes, and is 25 times faster than the best baseline on average.

Large language models (LLMs), pre-trained or fine-tuned on large code corpora, have shown effectiveness in generating code completions. However, in LLM-based code completion, LLMs may struggle to use correct and up-to-date Application Programming Interfaces (APIs) due to the rapid and continuous evolution of libraries. While existing studies have highlighted issues with predicting incorrect APIs, the specific problem of deprecated API usage in LLM-based code completion has not been thoroughly investigated. To address this gap, we conducted the first evaluation study on deprecated API usage in LLM-based code completion. This study involved seven advanced LLMs, 145 API mappings from eight popular Python libraries, and$\mathbf{2 8, 1 2 5}$completion prompts. The study results reveal the status quo (i.e., API usage plausibility and deprecated usage rate) of deprecated API and replacing API usage in LLM-based code completion from the perspectives of model, prompt, and library, and indicate the root causes behind. Based on these findings, we propose two lightweight fixing approaches, Replaceapi and InsertPrompt, which can serve as baseline approaches for future research on mitigating deprecated API usage in LLM-based completion. Additionally, we provide implications for future research on integrating library evolution with LLMdriven software development.

Resource leaks, caused by resources not being released after acquisition, often lead to performance issues and system crashes. Existing static detection techniques rely on mechanical matching of predefined resource acquisition/release APIs and null-checking conditions to find unreleased resources, suffering from both (1) false negatives caused by the incompleteness of predefined resource acquisition/release APIs and (2) false positives caused by the incompleteness of resource reachability validation identification. To overcome these challenges, we propose InferROI, a novel approach that leverages the exceptional code comprehension capability of large language models (LLMs) to directly infer resource-oriented intentions (acquisition, release, and reachability validation) in code. InferROI first prompts the LLM to infer involved intentions for a given code snippet, and then incorporates a two-stage static analysis approach to check control-flow paths for resource leak detection based on the inferred intentions. We evaluate the effectiveness of InferROI in both resource-oriented intention inference and resource leak detection. Experimental results on the DroidLeaks and JLeaks datasets demonstrate InferROI achieves promising bug detection rate (59.3% and 62.5%) and false alarm rate (18.6% and 19.5%). Compared to three industrial static detectors, InferROI detects 14~45 and 149~485 more bugs in DroidLeaks and JLeaks, respectively. When applied to real-world open-source projects, InferROI identifies 29 unknown resource leak bugs (verified by authors), with 7 of them being confirmed by developers. In addition, the results of an ablation study underscores the importance of combining LLM-based inference with static analysis. Finally, manual annotation indicated that InferROI achieved a precision of 74.6% and a recall of 81.8% in intention inference, covering more than 60% resource types involved in the datasets.

Large language models (LLMs) have demonstrated impressive capabilities in code generation, by leveraging retrieval-augmented generation (RAG) methods. However, the computational costs associated with LLM inference, particularly in terms of latency and energy consumption, have received limited attention in the security context. This paper introduces DrainCode, the first adversarial attack targeting the computational efficiency of RAG-based code generation systems. By strategically poisoning retrieval contexts through mutation-based approach, DrainCode forces LLMs to produce significantly longer outputs, thereby increasing GPU latency and energy consumption. We evaluate the effectiveness of DrainCode across multiple models. Our experiments show that DrainCode achieves up to a 85% increase in latency, a 49% increase in energy consumption, and more than a 3× increase in output length compared to the baseline. Furthermore, we demonstrate the generalizability of the attack across different prompting strategies and its effectiveness compared to different defenses. The results highlight DrainCode as a potential method for increasing the computational overhead of LLMs, making it useful for evaluating LLM security in resource-constrained environments. We provide code and data at https://github.com/DeepSoftwareAnalytics/DrainCode.

Abstract Researchers have recently achieved significant advances in deep learning techniques, which in turn has substantially advanced other research disciplines, such as natural language processing, image processing, speech recognition, and software engineering. Various deep learning techniques have been successfully employed to facilitate software engineering tasks, including code generation, software refactoring, and fault localization. Many studies have also been presented in top conferences and journals, demonstrating the applications of deep learning techniques in resolving various software engineering tasks. However, although several surveys have provided overall pictures of the application of deep learning techniques in software engineering, they focus more on learning techniques, that is, what kind of deep learning techniques are employed and how deep models are trained or fine-tuned for software engineering tasks. We still lack surveys explaining the advances of subareas in software engineering driven by deep learning techniques, as well as challenges and opportunities in each subarea. To this end, in this study, we present the first task-oriented survey on deep learning-based software engineering. It covers twelve major software engineering subareas significantly impacted by deep learning techniques. Such subareas spread out through the whole lifecycle of software development and maintenance, including requirements engineering, software development, testing, maintenance, and developer collaboration. As we believe that deep learning may provide an opportunity to revolutionize the whole discipline of software engineering, providing one survey covering as many subareas as possible in software engineering can help future research push forward the frontier of deep learning-based software engineering more systematically. For each of the selected subareas, we highlight the major advances achieved by applying deep learning techniques with pointers to the available datasets in such a subarea. We also discuss the challenges and opportunities concerning each of the surveyed software engineering subareas.

Recent code large language models (LLMs) have shown promising performance in generating standalone functions. However, they face limitations in repository-level code generation due to their lack of awareness of repository-level dependencies ( e.g., user-defined attributes), resulting in dependency errors such as undefined-variable and no-member errors. In this work, we introduce ToolGen , an approach that integrates autocompletion tools into the code LLM generation process to address these dependencies. ToolGen comprises two main phases: Trigger Insertion and Model Fine-tuning (Offline), and Tool-integrated Code Generation (Online). During the offline phase, ToolGen augments functions within a given code corpus with a special mark token, indicating positions to trigger autocompletion tools. These augmented functions, along with their corresponding descriptions, are then used to fine-tune a selected code LLM. In the online phase, ToolGen iteratively generates functions by predicting tokens step-by-step using the fine-tuned LLM. Whenever a mark token is encountered, ToolGen invokes the autocompletion tool to suggest code completions and selects the most appropriate one through constrained greedy search. We conduct comprehensive experiments to evaluate ToolGen ’s effectiveness in repository-level code generation across three distinct code LLMs: CodeGPT, CodeT5, and CodeLlama. To facilitate this evaluation, we create a benchmark comprising 671 real-world code repositories and introduce two new dependency-based metrics: Dependency Coverage and Static Validity Rate . The results demonstrate that ToolGen significantly improves Dependency Coverage by 31.4% to 39.1% and Static Validity Rate by 44.9% to 57.7% across the three LLMs, while maintaining competitive or improved performance in widely recognized similarity metrics such as BLEU-4, CodeBLEU, Edit Similarity, and Exact Match. On the CoderEval dataset, ToolGen achieves improvements of 40.0% and 25.0% in test pass rate (Pass@1) for CodeT5 and CodeLlama, respectively, while maintaining the same pass rate for CodeGPT. ToolGen also demonstrates high efficiency in repository-level code generation, with latency ranging from 0.63 to 2.34 seconds for generating each function. Furthermore, our generalizability evaluation confirms ToolGen ’s consistent performance when applied to diverse code LLMs, encompassing various model architectures and scales.

Open source software (OSS) supply chain enlarges the attack surface of a software system, which makes package registries attractive targets for attacks. Recently, multiple package registries have received intensified attacks with malicious packages. Of those package registries, NPM and PyPI are two of the most severe victims. Existing malicious package detectors are developed with features from a list of packages of the same ecosystem and deployed within the same ecosystem exclusively, which is infeasible to utilize the knowledge of a new malicious NPM package detected recently to detect the new malicious package in PyPI. Moreover, existing detectors lack support to model malicious behavior of OSS packages in a sequential way. To address the two limitations, we propose a single detection model using malicious behavior sequence, named Cerebro , to detect malicious packages in NPM and PyPI. We curate a feature set based on a high-level abstraction of malicious behavior to enable multi-lingual knowledge fusing. We organize extracted features into a behavior sequence to model sequential malicious behavior. We fine-tune the pre-trained language model to understand the semantics of malicious behavior. Extensive evaluation has demonstrated the effectiveness of Cerebro over the state-of-the-art as well as the practically acceptable efficiency. Cerebro has detected 683 and 799 new malicious packages in PyPI and NPM, and received 707 thank letters from the official PyPI and NPM teams.

Software vulnerabilities pose serious threats to the security of modern software systems. Deep Learning-based Automated Vulnerability Repair (AVR) has gained attention as a potential solution to accelerate the remediation of vulnerabilities. However, recent studies indicate that existing AVR approaches often only generate patches, which may not align with developers' current repair practices or expectations. In this paper, we introduce VulAdvisor, an automated approach that generates natural language suggestions to guide developers or AVR tools in repairing vulnerabilities. VulAdvisor comprises two main components: oracle extraction and suggestion learning. To address the challenge of limited historical data, we propose an oracle extraction method facilitating ChatGPT to construct a comprehensive and high-quality dataset. For suggestion learning, we take the supervised fine-tuning CodeT5 model as the basis, integrating local context into Multi-Head Attention and introducing a repair action loss, to improve the relevance and meaningfulness of the generated suggestions. Extensive experiments on a large-scale dataset from real-world C/C++ projects demonstrate the effectiveness of VulAdvisor, surpassing several alternatives in terms of both lexical and semantic metrics. Moreover, we show that the generated suggestions enhance the patch generation capabilities of existing AVR tools. Human evaluations further validate the quality and utility of VulAdvisor's suggestions, confirming their potential to improve software vulnerability repair practices.

As basic elements in program, variables convey essential information that is critical for program comprehension and maintenance. However, understanding the meanings of variables in program is not always easy for developers, since poor-quality variable names are prevalent while such variable are less informative for program comprehension. Therefore, in this paper, we target at generating concise natural language explanations for variables to facilitate program comprehension. In particular, there are two challenges in variable explanation generation, including the lack of training data and the association with complex code contexts around the variable. To address these issues, we propose a novel approach ZeroVar,which leverages code pre-trained models and zero-shot prompt learning to generate explanations for the variable based on its code context. ZeroVarcontains two stages: (i) a pre-training stage that continually pre-trains a base model (i.e., CodeT5) to recover the randomly-masked parameter descriptions in method docstrings; and (ii) a zero-shot prompt learning stage that leverages the pre-trained model to generate explanations for a given variable via the prompt constructed with the variable and its belonging method context. We then extensively evaluate the quality and usefulness of the variable explanations generated by ZeroVar.We construct an evaluation dataset of 773 variables and their reference explanations. Our results show that ZeroVarcan generate higher-quality explanations than baselines, not only on automated metrics such as BLEU and ROUGE, but also on human metrics such as correctness, completeness, and conciseness. Moreover, we further assess the usefulness of ZeroVAR-generated explanations on two downstream tasks related to variable naming quality, i.e., abbreviation expansion and spelling correction. For abbreviation expansion, the generated variable explanations can help improve the present rate (+13.1%), precision (+3.6%), and recall (+10.0%) of the state-of-the-art abbreviation explanation approach. For spelling correction, by using the generated explanations we can achieve higher hit@1 (+162.9(%) and hit@3 (+49.6%) than the recent variable representation learning approach.

Resource leaks, which are caused by acquired resources not being released, often result in performance degradation and system crashes. Resource leak detection relies on two essential components: identifying potential Resource Acquisition and Release (RAR) API pairs, and subsequently analyze code to uncover instances where the corresponding release API call is absent after an acquisition API call. Yet, existing techniques confine themselves to an incomplete pair pool, either pre-defined manually or mined from project-specific code corpus, thus limiting coverage across libraries/APIs and po- tentially overlooking latent resource leaks.

When searching code, developers may express additional constraints (e.g., functional constraints and nonfunctional constraints) on the implementations of desired functionalities in the queries. Existing code search tools treat the queries as a whole and ignore the different implications of different parts of the queries. Moreover, these tools usually return a ranked list of candidate code snippets without any explanations. Therefore, the developers often find it hard to choose the desired results and build confidence on them. In this article, we conduct a developer survey to better understand and address these issues and induct some insights from the survey results. Based on the insights, we propose XCoS, an explainable code search approach based on query scoping and knowledge graph. XCoS extracts a background knowledge graph from general knowledge bases like Wikidata and Wikipedia. Given a code search query, XCoS identifies different parts (i.e., functionalities, functional constraints, nonfunctional constraints) from it and use the expressions of functionalities and functional constraints to search the codebase. It then links both the query and the candidate code snippets to the concepts in the background knowledge graph and generates explanations based on the association paths between these two parts of concepts together with relevant descriptions. XCoS uses an interactive user interface that allows the user to better understand the associations between candidate code snippets and the query from different aspects and choose the desired results. Our evaluation shows that the quality of the extracted background knowledge and the concept linkings in codebase is generally high. Furthermore, the generated explanations are considered complete, concise, and readable, and the approach can help developers find the desired code snippets more accurately and confidently.

When reusing or modifying code, developers need to understand the implicit knowledge behind a piece of code in addition to the literal meaning of code. Such implicit knowledge involves related concepts and their explanations. Uncovering and understanding the implicit knowledge in code are challenging due to the extensive use of abbreviations, scattered expressions of concepts, and ambiguity of concept mentions. In this paper, we propose an automatic approach (called CoLiCo) that can uncover implicit concepts in code and link the uncovered concepts to Wikipedia. Based on a trained identifier embedding model, CoLiCo identifies Wikipedia concepts mentioned in a given code snippet and excerpts a paragraph-level explanation from Wikipedia for each concept. During the process, CoLiCo resolves identifier abbreviation (i.e., concepts mentioned in the form of abbreviations) and identifier aggregation (i.e., concepts mentioned by an aggregation of multiple identifiers) based on identifier embedding and mining of identifier abbreviation/aggregation relations. Experimental study shows that CoLiCo outperforms a general entity linking approach by 38.7% in the correctness of concept linking and identifies 96.7% more correct concept linkings on a dataset with 629 code snippets. The concept linking is significant for program understanding in 54% code snippets. Our user study shows that CoLiCo can significantly shorten the time and improve the correctness in code comprehension tasks that intensively involve implicit knowledge.

Annotations have been widely used in Java programs to support additional compile-time, deployment-time, and runtime processing. Developers use annotations to delegate repetitive logics such as object initialization and request forwarding to compilers and runtime frameworks. Therefore, these annotations are important for the correct execution of programs. In practice, however, developers often find it hard to correctly use annotations and the misuse of annotations has led to real bugs in Java programs. In this paper, we conduct an empirical study on Stack Overflow questions to investigate the major development frameworks that are involved in questions about Java annotations and the main problems encountered by developers in the use of Java annotations. Based on the findings of the study, we propose DeepAnna, a deep learning based Java annotation recommendation and misuse detection approach. Based on a corpus of Java programs with intensive use of annotations, DeepAnna trains a deep learning based multi-label classification model by considering both the structural and textual contexts of source code. DeepAnna can recommend annotations at both class level and method level. Our evaluation with a large corpus of open-source Java projects shows that DeepAnna outperforms state-of-the-art text multi-label classification approaches in annotation recommendation and can effectively detect annotation misuses. Based on our analysis, we submit 85 bug-fixing pull requests for annotation misuses in open-source projects and 20 of them have been accepted and merged.

Assigning bugs to the right components is the prerequisite to get the bugs analyzed and fixed. Classification-based techniques have been used in practice for assisting bug component assignments, for example, the BugBug tool developed by Mozilla. However, our study on 124,477 bugs in Mozilla products reveals that erroneous bug component assignments occur frequently and widely. Most errors are repeated errors and some errors are even misled by the BugBug tool. Our study reveals that complex component designs and misleading component names and bug report keywords confuse bug component assignment not only for bug reporters but also developers and even bug triaging tools. In this work, we propose a learning to rank framework that learns to assign components to bugs from correct, erroneous and irrelevant bug-component assignments in the history. To inform the learning, we construct a bug tossing knowledge graph which incorporates not only goal-oriented component tossing relationships but also rich information about component tossing community, component descriptions, and historical closed and tossed bugs, from which three categories and seven types of features for bug, component and bug-component relation can be derived. We evaluate our approach on a dataset of 98,587 closed bugs (including 29,100 tossed bugs) of 186 components in six Mozilla products. Our results show that our approach significantly improves bug component assignments for both tossed and non-tossed bugs over the BugBug tool and the BugBug tool enhanced with component tossing relationships, with >20% Top-k accuracies and >30% NDCG@k (k=1,3,5,10).

This report introduces the approach that we have designed and implemented for the DeClutter challenge of Doc-Gen2, which detects non-informative code comments. The approach combines both comment based text classification and code context based prediction. Based on the approach, our "fduse" team achieved the best F1 score (0.847) in the competition.

A domain glossary that organizes domain-specific concepts and their aliases and relations is essential for knowledge acquisition and software development. Existing approaches use linguistic heuristics or term-frequency-based statistics to identify domain specific terms from software documentation, and thus the accuracy is often low. In this paper, we propose a learning-based approach for automatic construction of domain glossary from source code and software documentation. The approach uses a set of high-quality seed terms identified from code identifiers and natural language concept definitions to train a domain-specific prediction model to recognize glossary terms based on the lexical and semantic context of the sentences mentioning domain-specific concepts. It then merges the aliases of the same concepts to their canonical names, selects a set of explanation sentences for each concept, and identifies "is a", "has a", and "related to" relations between the concepts. We apply our approach to deep learning domain and Hadoop domain and harvest 5,382 and 2,069 concepts together with 16,962 and 6,815 relations respectively. Our evaluation validates the accuracy of the extracted domain glossary and its usefulness for the fusion and acquisition of knowledge from different documents of different projects.