An Empirical Comparison of Statistical Term Association Graphs with DBpedia and ConceptNet for Query Expansion

Term graphs constructed from document collections as well as external resources, such as encyclopedias (DBpedia) and knowledge bases (ConceptNet), can be used as sources of semantically related terms for query expansion. Although these resources individually have been shown to be effective for IR, it is not known how their retrieval effectiveness compares with each other. In this work, we use standard TREC collections to perform systematic evaluation and empirical comparison of retrieval effectiveness of both types of term graphs for all and difficult queries. Our results indicate that of the term association graphs constructed automatically from document collection using information theoretic measures are more effective for Web collections, while the term graphs derived from DBpedia and ConceptNet are more effective for newswire collections.