XZXianju Zheng

Xianju Zheng

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Xianju Zheng has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Semantic Spectrum Fault Localization via Method Behavioral Divergence

    by Tu Peng, Xianju Zheng, Yazhi Yang, Shuliang Wang, Ling Xiong, Yunqi Ye

    Context: Automated fault localization techniques, ranging from standard Spectrum-Based Fault Localization (SBFL) to deep learning and large language model (LLM) based approaches, typically rely on extracting structural execution profiles. While these approaches capture the program's control structure, they suffer from a fundamental weakness: they merely record which path the program took (control flow) while ignoring what values the program produced (data semantics). Consequently, because they rely on incomplete characteristic information, these techniques often fail to distinguish defective behaviors that manifest as incorrect values rather than incorrect paths. This focus on structural features leads to frequent "critical ties" (where multiple methods share identical execution profiles) and inherently limited localization precision.Objective: To address these limitations, we propose a novel approach, Semantic Spectrum-based Fault Localization (SSFL), which incorporates method output semantics into the fault localization process.Method: We introduce a unified tokenization scheme that transforms heterogeneous return types into a common token vocabulary, and model runtime behaviors as output frequency matrices. By quantifying the behavioral divergence using Bidirectional Normalized Euclidean Distance and Weighted Frequency Difference, our technique captures fine-grained semantic differences that structural data ignore. Consequently, this approach not only resolves critical ties by differentiating coverage-identical methods but, more importantly, significantly enhances localization precision by leveraging the magnitude of behavioral semantics.Results: Empirical evaluation on five projects from the Defects4J benchmark demonstrates the effectiveness of our proposal. SSFL reduces MFR by 60%-93% compared to the best-performing SBFL formulae, and localizes 269 bugs within Top-5 across 334 evaluated versions---surpassing state-of-the-art learning-based and LLM-based techniques that require project-specific training data or large language model reasoning.Conclusion: These findings confirm that output semantics provide a powerful and previously untapped diagnostic signal that breaks the performance ceiling of traditional structure-based localization.

  • Open Access

    Correlation-enhanced feature learning for multi-label classification

    by Zhengjuan Zhou, Xianju Zheng, Yue Yu, Shaolong Li, Shiwen Li, Lingli Zhou

    Label correlations take important roles in solving multi-label classification problems. However, many approaches extract label correlations from the corresponding label sets of data instances, which cannot provide satisfactory per-formance, because incompletely tagged label sets may exist in reality. To avoid such adverse effects, we directly extract label correlations from data instances and propose a multi-label approach called correlation-enhanced feature learning, named CeFL. First, a new revised label matrix is obtained by multiplying the incomplete label matrix with the label correlations, in which the label correlations are obtained from the data matrix rather than the incomplete label matrix. Then, the revised label information is used to expand the feature of the original data, and a neural network is used to learn the correlation-enhanced feature which contains relationships between data features, between labels and data features, and between labels. Compared with established approaches, ours is more intuitive and easier to understand, and label correlations obtained from the data can implicitly reflect the different importance of labels observed for identical data representation. Experiments on several databases demonstrate the effectiveness of our approach.