An Efficient Tag Recommendation Method using Topic Modeling Approaches

Abstract

Software information sites such as Stack Overflow, Super User, and Ask Ubuntu allow users to post software-related questions, answer the questions asked by other users, and add tags to their questions. Tagging is a popular system across web communities because allowing users to classify their contents is less costly than employing an expert to categorize them. However, tagging systems suffer from the problem of the tag explosion and the tag synonym. To solve these problems, we propose a tag recommendation method using topic modeling approaches. Topic models have advantages of dimensionality reduction and document similarity. We also emphasize highest topics in calculating document similarity to retrieve more relevant documents. Our tag recommendation method considers the document similarity and the historical tag occurrence to calculate tag scores. Experiment results show that emphasizing highest topic distributions increases overall performance of tag recommendation.