GPU accelerated hot term extraction from user generated content

Abstract

This thesis aims at developing and investigating an efficient approach to hot term extraction. In the Web 2.0, the user generated content (UGC) is increased dramatically in different Consumer Generated Media (CGM) such as forums and blogs. People easily search their knowledge and opinions in CGM as well as generate Word Of Mouth (WOM) in different online channels. Facing the huge amount of data, it is not easy to find the useful information even using a search engine. Having a good hot term extraction algorithm can reveal hidden information to users and also provide an indicator in the search results, so that users can easily know which terms are popular in the search results. In this thesis, a GPU based hot term extraction algorithm is presented. Graphics Processing Units (GPUs) is designed for data-parallel computations. Comparing to running a single program with multiple data in CPU, GPU can have faster execution. The hot term is defined as a word that appears frequently in the search result. We assume that the greater the frequency of appearance of a term, the more the relevancy of the term to the users. As there are lots of terms in the searched results, processing them is time-consuming. The proposed GPU based hot term extraction algorithm can achieve a fast performance and works well in real-time applications.