Video topic detection on Micro-Blog using Relational Topic Model

Abstract

To enhance the performance of the personalized tag recommendation method, a microblog video topic discovery algorithm based on common knowledge atlas analysis was proposed. Firstly, graphical form was used for expressing latent local expression in the microblog video topic and top-k similar user discovery of users represented by user topic distribution. Then, the frequency of all the tags occurring in these users was calculated and the tags most relevant with users were recommended. Next, to mine potential topic information, enhanced cosine similarity RTM model with penalty term was used for naming the tag of topic for microblog video, which enhanced the impact of united modeling on tag generation for potential topic and might discover the relationship between global tab and topic. At last, the real experiment result showed that the recommended method was superior to typical tag recommendation algorithms of several selected microblog video topics. Meanwhile, the validity of the algorithm was verified.