Opinion Summarization for Hotel Reviews.

This paper presents a new approach for finding the best ngrams that efficiently summarize a large set of reviews. The proposed unsupervised method uses a readability score and a representativeness score to select those n-grams that best convey the main opinions contained in the processed reviews. In order to further refine the selected n-grams, we use sentiment analysis and part of speech (POS) tagging to impose certain requirements that the n-grams that we are looking for should meet. Furthermore, the best n-grams were classified into several topics, which allowed a better prevention of redundancy among the summarizing n-grams. Therefore we offer an unsupervised, mostly non-aspect based, unstructured opinion summarization algorithm that can be easily implemented for any web platform that accepts reviews, due to its genericity. In order to assess the results of our algorithm, we summarized hotel reviews extracted for the TripAdvisor 1 website. The algorithm produces readable results that convey relevant opinions about the hotels that we used for testing.