Hot Topic Discovery in Online Community using Topic Labels and Hot Features

Abstract

With huge volumes of information on Internet, how to extract user-concerned hot topics quickly and effectively has become a fundamental task for information processing on Internet. Generally, hot topic detection includes two tasks, the first one is topic discovery and the other is its hotness evaluation. In this paper, we propose a hot topic detection method. For topic discovery, topics are identified by clustering based on extracted topic labels. For hotness evaluation, the proposed model has fully considered the internal and external dual features and combined them together. The experimental results over TianYa BBS demonstrate the efficiency of the proposed method: compared with topic discovery based on latent semantic indexing, the improved vector space model based on topic labels gets better results and the identified topics are more accurate. Moreover, the proposed hotness features could reflect the popularity of a topic, and hence have obtained better hot topic results finally.

Keywords

• external features
• improved vector space model
• internal features
• topic hotness
• topic label