Time-Series Transformation of Review Scores Based on the Each User’s Self-Information

Abstract

This study aims to quantify the evaluations of items and categories on online review sites as time-series data, and transform the review scores using probability distributions of scores for each user. Generally, the meaning of each score may vary between users with lenient or strict evaluations. Therefore, by utilizing the self-information of each user's probability distribution of scores estimated from their review history, it is inferred to be possible to objectively quantify the review scores. However, since the probability distributions of scores may change over time, we constructed a model that considers the time-series changes of these distributions. In the experimental results, we generated the time-series data of estimated evaluation values.