Frequent Itemsets as Meaningful Events in Graphs for Summarizing Biomedical Texts

In this paper, we introduce a method using graph modeling for summarizing biomedical texts. We address the challenges of identifying meaningful topics of the input document, modeling the relations between the sentences, and selecting the most relevant sentences. Our summarizer utilizes an itemset mining method to discover the topics from the concepts extracted from the input document. It uses a meaningfulness measure to identify the meaningful events, i.e. the essential topics of the input document. Theses meaningful topics are used to construct a small-world network that models the relations between the sentences within the text. Those sentences identified as highly-informative and relevant are selected to generate the final summary. Conducting a set of evaluations, we assess the efficiency of the graph-based approach for summarization of biomedical documents. The evaluations demonstrate that our method can achieve a better performance in comparison with a number of available methods with respect to some widely-used metrics. The summaries produced by the graph-based summarizer contain more informative content compared to the summaries generated by the other methods. This shows that the combination of itemset mining on the concepts, the meaningfulness measure, and modeling the text as a small-world network can perform well in biomedical text summarization.