Bigram Collocation Extraction of Mobile Game Comments
Download as PDF
DOI: 10.25236/icemit.2019.025
Author(s)
Jing Zhang, Haiqi Chen, Shoubin Dong
Corresponding Author
Jing Zhang
Abstract
Mining user opinions about game attributes from comment data becomes key to both the player and the game developers. As it’s impractical to take the user score as category information for the training set, a solution based on Co-training and Pu-training is proposed to gain the training corpus for the classifier. A concept of 2-dimensional attribute lexicon and a cluster-based construction solution are proposed. A dependency tree-based comment collocation template extraction method is proposed with the training corpus and 2D attribute lexicon as knowledge base. Experiment shows comment collocations as “picture–fabulous” and “teammate–really lame”, which leads to satisfactory accuracy and recall.
Keywords
Mobile game comments, 2-dimensional attribute lexicon, comment collocation, sentiment tendency analysis