Vector-Based Semantic Expansion Approach: An Application to Patent Retrieval

Patent collection is increasing incrementally. Most of the new technological information is from patent documents, and retrieving specific patents in such a large pool of documents has become a challenging issue for both patent examiners and normal/inexperienced users. Nowadays, most of the retrieval systems (patent retrieval systems) search for the exact match of the query string inside the collection to return a result. However, based on natural language processing, most words have different meanings and each concept can be expressed by more than one word. Since patent documents include lots of new technical terms, searching among them by exact match needs lots of experiences and it is an exhausting task. In this study we aim to investigate this problem by using expansion techniques. We propose a new semantic expansion approach, namely, vector-based semantic expansion (VSE). To do the expansion, we use external knowledge sources, WordNet and Wikipedia. We also examine the affects of using their combination on the expansion results. We do word (VSWE), query (VSQE), and document (VSDE) expansion in this thesis to demonstrate the performance of our approach. We show that our approach with combination of these two knowledge sources can be effective to find similarity between two units of language which might be word, sentence or text. We apply our technique on Miller and Charles dataset to find word-word similarity. Also, our experiments which are based on clef-ip 2011 shows that our technique increases the recall-rate in query expansion, especially for shorter queries.Computer science, Information Architecture (IA) track, Web information systems (WIS) groupSoftware and Computer TechnologyElectrical Engineering, Mathematics and Computer Scienc