Scholarlink

<p>Mining information from massive texts and extracting key terminology is crucial for enhancing information processing. Addressing the challenges of complex term recognition and the limitations of single-feature methods in the field of information technology terminology extraction, this study proposes a simple yet effective approach that integrates morphological features for identifying IT terms. The method extracts textual features through root, stem, and affix vectors and dynamically incorporates lexicon-based features. Multiple morphological features are then fed into a BiLSTM-CRF model to improve recognition performance. Experimental results demonstrate that this method significantly outperforms other comparative experiments in recognizing Kazakh terminology in the IT domain, effectively enhancing term identification for agglutinative languages in specialized fields. On the test set, it achieved a maximum F1-score of 91.9%, surpassing the best baseline by 13% points. Ablation studies further validate the effectiveness of fused features in boosting recognition performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on deep learning and Morphological Features enhanced method in Kazakh IT term

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Muheyat Niyazbek,
  • Kunsaule Talp

摘要

Mining information from massive texts and extracting key terminology is crucial for enhancing information processing. Addressing the challenges of complex term recognition and the limitations of single-feature methods in the field of information technology terminology extraction, this study proposes a simple yet effective approach that integrates morphological features for identifying IT terms. The method extracts textual features through root, stem, and affix vectors and dynamically incorporates lexicon-based features. Multiple morphological features are then fed into a BiLSTM-CRF model to improve recognition performance. Experimental results demonstrate that this method significantly outperforms other comparative experiments in recognizing Kazakh terminology in the IT domain, effectively enhancing term identification for agglutinative languages in specialized fields. On the test set, it achieved a maximum F1-score of 91.9%, surpassing the best baseline by 13% points. Ablation studies further validate the effectiveness of fused features in boosting recognition performance.