Mining entity latent semantic relationships by two-phase clustering

Web Integration System (WIS) provides abundant structured information about entities in a domain. Inter-relationships for entities in a WIS are valuable for further analysis and decision-making. This paper proposes a method to mine the latent semantic relationships between two entities in WIS and labels more entity pairs in WIS to the relationship predefined in WIS. We first extract related entities and corresponding contexts from web texts for each entity in WIS, and then cluster the entities related to it into different sets to finding the latent semantic relationships, and cluster all sets to find similar latent relationships. The number of labeled entity pairs in our method is much more than compared method. This paper evaluate the method by using a real-world dataset generated by search engine, and the results show that the proposed approach could analyze the similarity of latent semantic relationships between entities efficiently. Compared to CO-algorithm [4], the method gets higher recall and F-measure. Copyright © 2015 Binary Information Press.