MOO-CMDS+NER: Named Entity Recognition-Based Extractive Comment-Oriented Multi-document Summarization
Résumé
In this work, we propose an unsupervised extractive summarization framework for generating good quality summaries which are supplemented by the comments posted by the end-users. Using the evolutionary multi-objective optimization concept, different objective functions for assessing the quality of a summary, like diversity and the relevance of sentences in relation to comments, are optimized simultaneously. In the literature, named entity recognition (NER) has been shown to be useful in the summarization process. The current work is the first of its kind where we have introduced a new objective function that utilizes the concept of NER in news documents and user comments to score the news sentences. To test how well the new objective function works, different combinations of the NER-based objective function with already existing objective functions were tested on the English and French datasets using ROUGE 1, 2, and SU4 F1-scores. We have also investigated the abstractive and compressive summarization approaches for our comparative analysis. The code of the proposed work is available at the github repository https://github.com/vishalsinghroha/Unsupervised-Comment-based-Multi-document-Extractive-Summarization .