Category-oriented Sentiment Polarity Dictionary for Rating Prediction of Japanese Hotels

摘要

Hotel booking sites provide us hotel evaluations, each of which includes a textual review and numeric ratings for multiple categories such as Service, Location, Rooms, etc., submitted by a hotel guest. However, some evaluations have no numeric rating. Also, textual reviews and numeric ratings of some evaluations are inconsistent. For example, a good textual review is submitted with low ratings. Such evaluations may confuse site users. To resolve such problems, we propose a high accuracy method to predict a numeric rating for each category from a textual review. Our new idea is to use Category-oriented Sentiment Polarity Dictionaries (CSPD), each of which is automatically compiled for each category using a Rakuten Travel review database in advance. The CSPD gives the sentiment polarity value (i.e., the positivity/negativity value) for each sentiment word such as "bad", "light", and "delicious" for each category. In our experiments, CSPD showed higher precision in rating prediction than the existing dictionaries when we use only sentiment polarity values for rating prediction. We also combine the vectorization method of words and the CSPD to obtain an expected rating value from a textual review. Our experimental results show that our combined method attains higher accuracy than the previously published sentence vectorization method.