Query expansion based on context-dependent sentiment analysis in databases with domain-specific filtering

The object of this study is an improved query expansion method based on context-dependent text sentiment analysis in information retrieval systems for working with databases. Using natural language processing (NLP) methods, in particular contextual embeddings and transformer architectures, this paper focuses on adaptively determining user intent within the submitted query. The study involves analyzing and improving text processing mechanisms using subject-specific filtering to increase the accuracy and relevance of search results. The proposed method demonstrates an increase in the accuracy of context-sensitive models by 6 % compared to baseline approaches. The aggregate F1-measure indicator, which combines precision, completeness, and accuracy, reflects the relevance of the constructed models, showing an increase of 6–8 %. The difference between the least and most effective methods is 16 % in accuracy and 17 % in relevance. The proposed approach overcomes the limitations of static traditional synonym and statistical methods by dynamically interpreting the relationship between tone, context, and domain specificity of content. Improved semantic understanding allows for more accurate matching of extended queries with user goals. This method could be effectively applied in practice in settings where information retrieval systems operate within domain-specific databases. This applies to scenarios in which user queries contain complex, emotionally colored language constructs that require a deeper understanding of context and tone. However, its implementation requires training on high-quality domain-specific datasets with contextual labels that provide accurate adaptation