A Fuzzy, Incremental and Semantic Trending Topic Detection in Social Feeds

Nowadays, a huge number of people participating in social networks is triggering a fast and wide spectrum of topics. Such trending topics are usually derived from the most frequent searches, the published posts and the daily news. The automated analysis for such data requires topics detection and tracking methods. Many challenges are being faced. It is difficult to discover the semantic relatedness when the same event is presented by different titles and to handle merging semantically identical topics from different channels (aggregation). Other hardships are the vagueness regarding the vast web collection, the scalability to analyze them, and the fact that it is a time consuming task. The framework introduced in this paper aims to solve these issues. Because a web document often consists of several topics, the suggested model employs a fuzzy C-Means (FCM) clustering based trending topics detection. It applies a semantic document similarity algorithm to resolve such ambiguity issues caused by the usage of synonyms, homonyms or different abstraction levels. This algorithm is also used to summarize the long documents. Furthermore, an incremental clustering technique is utilized to preserve high cohesiveness up-to-date top trending topics. The experimental results finally illustrate the effectiveness and the superiority of this model, compared with other trending topics detection algorithms, in terms of entropy and F-score measures.