A text mining-based framework to discover the important factors in text reviews for predicting the views of live streaming

Abstract

Live streaming has become one of the leisure activities of most people due to the rich and various contents. For young generation, to watch other people playing games on the live streaming platform is becoming very popular. Related researches mainly focused on predicting the number of viewers, finding popular streamer, studying the gift giving behaviors, and so on. Relatively few studies focused on how viewers’ comments affect users’ viewing behaviors, since the power of text comments in social media have been confirmed. In addition, published studies usually employed questionnaire survey methods which are prone to experimental effects. And online text comments will be more objective and less sampling bias than data collected by questionnaires. Consequently, this study focuses live streaming of games and uses viewers’ text comments for experimental analysis. A text mining-based framework which includes Least Absolute Shrinkage and Selection Operator (LASSO), Support Vector Machine-Recursive Feature Elimination (SVM-RFE), Chi-square test will be proposed to determine the important keywords of predicting the number of views in live streaming. Support Vector Machine (SVM) will be utilized to evaluate the performances of candidate feature subsets. Then, K-means and Latent Semantic Analysis (LSA) using Singular Value Decomposition (SVD) have been used to organize the selected keywords into understandable concepts. Real cases of game live streaming cases will be collected from Twitch.tv for our experiments. Results can be used as a reference for live streaming platforms and live channels, and help them to increase the number of viewers for further income enhancement.