Graph-based biomedical text summarization: An itemset mining and sentence clustering approach.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Data Mining

Background:

• The rapid growth of biomedical literature necessitates efficient summarization tools.
• Existing methods may not fully capture domain-specific nuances or sentence correlations.

Purpose of the Study:

• To develop a novel graph-based text summarization method for the biomedical domain.
• To enhance the informativeness and relevance of summaries by incorporating domain knowledge and frequent itemset mining.

Main Methods:

• Utilized the Unified Medical Language System (UMLS) for concept-based document modeling.
• Employed frequent itemset mining to identify correlations among medical concepts.
• Constructed a similarity function and a represented graph.
• Applied a minimum spanning tree clustering algorithm to identify document subthemes.
• Selected sentences from subthemes to generate the final summary.

Main Results:

• The proposed summarization system demonstrated superior performance compared to baselines and benchmarks.
• Automatic evaluation using Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metrics confirmed effectiveness.

Conclusions:

• Integrating domain-specific knowledge and frequent itemset mining significantly improves sentence informativeness measurement.
• Graph-based clustering enables efficient targeting of diverse subthemes within biomedical documents.
• The approach offers a substantial performance improvement for biomedical text summarization.