Semantic similarity between complex named entities: An approach using multiple web resources

In this paper, we suggest a novel method for a new problem: measuring the similarity between Complex Named Entities (CNEs) by a hybrid model using multiple Web resources. CNEs like book names or names of commodities appear everywhere in our daily life. Different from previous works, we focus on fine-grained similarities within a CNE class. We measure the similarity of two CNEs based on their co-occurrences on the Web as well as similarities of two URL sets associated with them through a Web search. Additionally, we extract hierarchical information from Web sites such as Baidu Tieba and Taobao to distinguish between pure relatedness and strong similarity. The use of Web information ensures that we can handle new CNEs that have not been listed in any ontology or corpus. Evaluation results show that our approach achieves satisfactory results. The suggested hybrid approach will be beneficial for a lot of real world semantic applications. © 2011 ISSN.