Optimal Keyword Selection by Hybrid Optimization with Itemset Mining for Text Summarization in Biomedical Sector

Abstract

Biomedical literature is growing exponentially, creating challenges for researchers and clinicians to access relevant information efficiently. Automatic biomedical text summarization is a promising solution to this issue, allowing the extraction of essential information while reducing redundancy. Traditional summarization techniques often rely on shallow linguistic features or simple term frequencies, which fail to capture the complex relationships in biomedical texts. This paper explores advanced methods for biomedical text summarization, including graph-based approaches, frequent item set mining, and deep learning models such as BERT. The proposed framework introduces a hybrid optimization technique combining Colliding Bodies Optimization (CBO) and Cuckoo Search Optimization (CSO) for optimal keyword selection, alongside a Recurrent Neural Network (RNN) for sentence categorization. Extensive experimentation using UMLS-based concept extraction, keyword selection, and Apriori-based itemset mining demonstrates that the proposed method significantly outperforms existing models in both the in formativeness and coherence of generated summaries. The results reveal that combining deep language models with domain-specific knowledge enhances summarization quality and can be effectively applied to diverse types of biomedical text.