SubModRank: Monotone Submodularity for Opinionated Key-phrase Extraction
Akash Sheoran, G.Sudhagar Rahul Namdeo Jadhav, Amit Sarkar · 2022
With the rise of World Wide Web (WWW), a large volume of structured and unstructured data in textual form, has become available. This has led to various data and information processing tasks, one of which is key-phrase extraction. It is concerned with identifying important words and phrases which can comprise a meaningful summary of the input text. Current unsupervised methods for key-phrase extraction do not explicitly account for the opinion expressing side of the text. This is important for certain type of data, for example, product reviews, since they contain user opinion about the product and its aspects. Therefore, the extracted key-phrases should reflect both factual information as well as the opinionated part of the text. In this paper, we present SubModRank, an unsupervised approach for automated key-phrase extraction, based on submodular function optimization, for which a partial enumeration based greedy algorithm has been used. The resultant key-phrases are able to capture both objective as well as subjective side of the text, outperforming state of the art unsupervised key-phrase extraction methods in terms of precision, recall and F1-score.