СРАВНИТЕЛЬНЫЙ АНАЛИЗ ЭФФЕКТИВНОСТИ МЕТОДОВ ПОИСКА И ГЕНЕРАЦИИ ОТВЕТОВ В ДИАЛОГОВЫХ СИСТЕМАХ

Аннотация

This study presents a comparative analysis of three response-generation methods in dialogue systems: TF-IDF, BM25, and a generative approach. The results demonstrate the superiority of BM25 in terms of response relevance (4.40/5.0) with optimal execution time (0.156 ms), while the generative approach achieved the highest processing speed (0.010 ms) at the cost of reduced answer quality. The automatic BLEU and ROUGE metrics showed limited effectiveness in evaluating the quality of dialogue systems [1, 2]. Keywords: dialogue systems, information retrieval, BM25, TF-IDF, generative models, quality evaluation, relevance, response time.