Short Texts Keyword Extraction

Implementation of Transfer Learning for Automatic Summarization in Research Article Synthesis

The findings indicate that BERT-based extractive summarization can support preliminary literature screening, but further improvement is needed through stronger baseline comparison, human evaluation, and redundancy-aware optimization.

Evaluation of the BERT model for text semantic similarity

This study validates the effectiveness of the BERT model in semantic similarity calculation, providing more accurate technical support for related application scenarios, and laying the foundation for subsequent model optimization and lightweighting research.

Sequran: Composite Scoring and Reranking Techniques for Refining Quran Search Engine Results

A Hybrid Framework for Automated News Classification for the Low-Resource Assamese Language

Assamese is a low resource language that presents significant challenges for automated news classification due to the scarcity of curated datasets. To address these challenges, this paper introduces a dedicated corpus of 6,582 Assamese news articles and proposes a hybrid framework that merges a transformer-based subword tokenizer with a linear Support Vector Machine through a TF-IDF vectorization pipeline. Empirical evaluations demonstrate that the proposed architecture balances accuracy and computational efficiency, outperforming sixteen alternative model configurations and establishing itself as an optimal candidate for real-world deployment. This study further led to the development of a web application named ANC v1.0 using Flutter. The application features a responsive interface that provides real-time top-3 news-category predictions with probabilistic confidence scores. By bridging the gap between high-dimensional machine learning and practical web deployment, this work provides a scalable solution for the digital management of Assamese news articles.

Development and Evaluation of an Indonesian Thesis-Title Similarity Detection System Using Hybrid TF-IDF–SBERT Retrieval and LightGBM Reranking

This study developed and evaluated an Indonesian thesis-title similarity detection system to support the early screening of potentially similar research titles. The reference corpus consisted of 28,117 unique thesis titles collected from a publicly accessible institutional repository. The system integrates TF-IDF-based lexical retrieval, SBERT-based semantic retrieval, Union Top-K candidate selection, and machine learning reranking. Logistic Regression, XGBoost, and LightGBM were trained using threshold-derived pseudo-labels, while the final model performance was evaluated on an independent human-validated dataset constructed from a title-disjoint evaluation partition. A total of 900 candidate title pairs were independently assessed by two academic validators, achieving a Cohen’s kappa of 0.877. LightGBM achieved the best performance, with an accuracy of 94.67% and a macro F1-score of 0.9467. ISO/IEC 25010 evaluation showed a 100% functional pass rate, an 84.17% usability score, performance scores of 92.3% on mobile and 96.8% on desktop, while the automated security assessment categorized the staging deployment as Fairly Secure with two medium-risk findings. The system can support academic title-similarity screening; however, its generalizability remains limited by the use of a single-institution corpus and requires further multi-institutional validation.

Research on Text Information Extraction and Imbalanced Classification Methods for Enterprise Profiling

This research focuses on enterprise profiling in scenarios where large volumes of diverse texts—such as registration records, annual reports, news articles, and bidding notices—are continuously generated. Instead of relying solely on a single data representation or classification model, we developed a comprehensive natural language processing (NLP) pipeline for extracting key information and identifying industries. The pipeline consists of several steps. First, we use a BERT-BiLSTM-CRF model to identify core e/nterprise entities. Then, we combine TF-IDF with BERT embeddings to create a hybrid feature scheme that captures both lexical cues and contextual semantics. To address the challenge of imbalanced industry labels, we apply SMOTE in the dense semantic space and pair it with Focal Loss to enhance learning for minority classes. Additionally, we introduce a Stacking strategy to integrate outputs from different models, making predictions more stable. Tests on a self-compiled dataset covering ten national economic sectors and about 50,000 enterprises show that our method achieves a macro-F1 score of 95.4%. It outperforms traditional machine learning baselines and single deep learning models, offering more reliable recognition for minority classes. These results suggest that our framework is well-suited for applications such as supply chain partner discovery, industrial mapping, and targeted investment promotion.