Sentiment Polarity Analysis for Generating Search Result Snippets based on Paragraph Vector
Although sentiment polarity is important for under- standing public reputations, conventional web search snippets do not include sentiment polarity. Therefore, we propose a system to generate search result snippets that considers sentiment polarity. We also propose a method for extracting reputation sentences from search result snippets based on a given search query. To extract such sentences, we use a Paragraph Vector (PV) to calculate the similarity of words and sentences. Overall, our method is based on this PV and logistic regression for analyzing sentiment polarities. We evaluated the accuracy of our method via the Rakuten dataset. In our evaluation, we focused on methods pertaining to what supports sentiment polarity analysis in our method. We found that considering both a search query and reputation is important in generating our modified snippets. Furthermore, we identified the need for a method to remove meaningless sentences, which will be part of our future studies.