Fusing and mining opinions for reputation generation

The Internet provides a convenient platform for people to freely share their opinions on any entities. The opinions expressed in natural languages carry the subjective attitudes and preferences of humans. They represent the public perspectives on any entity, thus impact user decisions and behaviors in some way. Therefore, opinions have been recognized as useful and valuable pieces of information for reputation generation. Fusing and mining opinions offer a promising approach to extract reputation information and track public perspectives. However, the literature lacks studies on this topic. In this paper, we propose a novel reputation generation approach based on opinion fusion and mining. In our approach, opinions are filtered to eliminate unrelated ones, and then grouped into a number of fused principal opinion sets that contain opinions with a similar or the same attitude or preference. By aggregating the ratings attached to the fused opinions, we normalize the reputation of an entity. Meanwhile, various types of recommendations can be generated based on relationships among opinions. To offer sufficient reputation information to users, we also propose a new way of reputation visualization. It shows the details of opinion fusing and mining results, such as the normalized reputation value, principal opinions with popularity and other statistics. Experimental results coming from an analysis of big real-world data collected from several popular commercial websites in both English and Chinese demonstrate the generality and accuracy of the proposed approach, especially the effectiveness of opinion filtering for reputation generation. A small-scale real-world user study further quantifies the user acceptance of the developed reputation visualization method. In the sequel, this implies that the proposed approach can be applied in practice to generate reputation.