Thananchai Khamket

Bug severity classification supports software maintenance by helping developers prioritize defect resolution. Most existing approaches combine bug report titles and descriptions into a single representation, often assuming that both textual components contribute equally to severity prediction. However, concise titles and detailed descriptions may provide different types of severity-related information. This study investigates their relative contribution and proposes an Adaptive Title–Description Fusion (ATDF) framework that separately encodes title and description representations and adaptively estimates their contribution during feature fusion. Experiments were conducted on Mozilla Bugzilla repositories using a multiclass classification setting with five severity categories. The proposed framework was compared with title-only, description-only, direct concatenation, and standard gated fusion baselines. ATDF achieved the highest overall performance, with a Macro-F1 score of 0.771, compared with 0.756 for standard gated fusion and 0.739 for direct concatenation. Additional analyses showed that title and description contributions varied across severity categories and repositories under the current experimental setting. Overall, the findings suggest that considering the relative contribution of title and description information can provide a practical approach to improving bug severity classification while offering additional insight into title–description fusion behavior in software repositories.

Public opinion on online platforms now plays an important role in agricultural markets, which have always been unpredictable. Although sentiment analysis has been widely applied to agricultural texts, most existing studies typically focus only on classification accuracy without connecting results to actual market intelligence systems, especially in multilingual contexts. This paper introduces a reliability-aware transformer-based framework for analyzing sentiment in agricultural market intelligence across multiple languages. The framework leverages weakly supervised multilingual transformers to extract sentiment signals from large-scale unlabeled Thai and English texts about major agricultural commodities found online. To enhance robustness under weak supervision, the framework incorporates reliability-aware mechanisms, including confidence-based pseudo-label filtering, cross-source consistency refinement, and expert-guided calibration to reduce noise and account for bias between different data sources. Sentiment predictions are further aligned with market intelligence objectives through reliability-weighted aggregation, yielding interpretable sentiment indices that enable cross-lingual and cross-source comparability. We tested the framework extensively using a multilingual agricultural corpus derived from social media and news coverage of agriculture. The results show consistent improvements over both classical machine learning approaches and standard multilingual transformer baselines. Additional ablation studies and sensitivity analyses confirmed that reliability-aware mechanisms, particularly confidence thresholding, play a crucial role in getting the right balance between label quality and data coverage. Overall, the results indicate that reliability-aware multilingual sentiment analytics provide robust and actionable insights for agricultural market monitoring and policy analysis.

Cross-domain sentiment analysis remains challenging due to distributional shifts and heterogeneous sentiment expressions across platforms. Existing domain adaptation approaches primarily focus on enforcing domain-invariant representations. However, such symmetry-preserving strategies often overlook directional and expression-level asymmetries. These asymmetries naturally arise in real-world sentiment data, particularly for context-inferred sentiment expressions. In this work, we propose a novel symmetry- and asymmetry-aware domain adaptation framework for cross-domain sentiment classification. The framework models symmetry through explicit multi-source distribution alignment, which captures transferable sentiment knowledge across domains. Additionally, aspect-level structural supervision organizes representations according to shared linguistic aspects. To address asymmetry, a directional divergence regularization is introduced. This component models expression-level and directional discrepancies between source and target domains. Importantly, the framework operates without requiring target-domain annotations. Experiments are conducted under a multi-source unsupervised domain adaptation setting using sentence-level hotel review datasets collected from multiple online platforms. Empirical results demonstrate strong performance for the proposed framework. It achieves an average Accuracy of 82.0% and Macro-F1 of 80.6%. The framework consistently and statistically significantly outperforms source-only, multi-source, and transformer-based adversarial adaptation baselines across all evaluated target domains (p < 0.05). Additional analyses confirm improved robustness to implicit sentiment expressions and platform-induced asymmetries. These findings highlight the importance of jointly modeling symmetry and asymmetry for robust cross-domain sentiment adaptation and provide a unified and deployable solution for sentiment analysis under realistic platform shifts.