Crowdsourced Social Media Reaction Analysis for Recommendation

A pre-analysis is always important for crucial decision making in many events where reviews, feedback, and comments posted by different stakeholders play an important role. Summaries generated by humans are mostly based on abstractive summarization. It sometimes changes the meaning of the text. This paper proposes a customized extractive summarization approach to generate a summary of large text extracted from social media viz. Twitter, YouTube review, feedback, comments, etc. for a movie. The proposed approach where PageRank with k-means clustering was used to select representative sentences from a large number of reviews and feedback. Cluster heads were selected based on the customization of PageRank. The proposed approach shows improved results over the graph-based TextRank approach with and without synonyms. It can be applied to predict trends for items other than movies through the social media platform.