Mining User Reviews for Online Game on Handheld Devices

• Main Authors: Re-Jiau Sung, 宋瑞蛟
• Other Authors: Chaochang Chiu
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/z2zk49

博士 === 元智大學 === 資訊管理學系 === 102 === This research investigates the review analytics of Google Play games using a proposed text analytics approach to extract user sentiments about games expressed in Chinese. Based on a corpus study, document- and feature-level sentiment classification are examined to discover about reviews and which attributes of games users felt good or bad. A supervised GA/k-means for classification approach is proposed for document-level sentiment classification. For feature-level sentiment classification, an approach mixed with supervised semantic orientation algorithms and heuristic n-phrase rule is developed to find out user opinions on game attributes including overall impression of a game and different aspects of gameplay, aesthetic, musicality, stability, and developer. The experimental results show that the proposed approaches can help construct effective classification models with acceptable performance. The user reviews are further analyzed various characteristics such as reviewer’s gender, games categories, star ratings, game attributes, and high sentiment games. The results of content analysis accompanied with correspondence analysis show many interesting facts that can provide in-depth insight into users’ concern as well as best practices for developers or managers. Additionally, the star rating is the most common mechanism for users’ confidence by crowd-sourced app ratings. Three GA-based hybrid time series models, GA/ARIMA, GA/SVR, and GA/ARIMA_SVR, are developed for daily star rating forecasting. The evidences show that GA/ARIMA, linear model with optimized parameters, outperforms other models for star rating forecasting.