Discovering Latent Concepts and Exploiting Ontological Features for Semantic Text Search

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 571–579, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Discovering Latent Concepts and Exploiting Ontological Features for Semantic Text Search Vuong M. Ngo1,2,3 and Tru H. Cao1,2 1HCMC University of Technology and 2John von Neumann Institute, VNU-HCM, and 3VNG Corporation {vuong,tru}@cse.hcmut.edu.vn Abstract Named entities and WordNet words are im- portant in defining the content of a text in which they occur. Named entities have onto- logical features, namely, their aliases, classes, and identifiers. WordNet words also have on- tological features, namely, their synonyms, hypernyms, hyponyms, and senses. Those fea- tures of concepts may be hidden from their textual appearance. Besides, there are related concepts that do not appear in a query, but can bring out the meaning of the query if they are added. The traditional constrained spreading activation algorithms use all relations of a node in the network that will add unsuitable information into the query. Meanwhile, we on- ly use relations represented in the query. We propose an ontology-based generalized Vector Space Model to semantic text search. It dis- covers relevant latent concepts in a query by relation constrained spreading activation. Be- sides, to represent a word having more than one possible direct sense, it combines the most specific common hypernym of the remaining undisambiguated multi-senses with the form of the word. Experiments on a benchmark da- taset in terms of the MAP measure for the re- trieval performance show that our model is 41.9% and 29.3% better than the purely key- word-based model and the traditional con- strained spreading activation model, respec- tively. 1. Introduction With rapid development of the World Wide Web and e-societies, Information Retrieval (IR) has many challenges in discovering and exploiting those rich and huge information resources. Se- mantic search improves search precision and re- call by understanding user's intent and the con- textual meaning of concepts in documents and queries (Huston and Croft, 2010; Losada, et al, 2010; Egozi, et al, 2011). Concepts are named entities or WordNet words (unnamed entities). Named entities are those that are referred to by names such as people, organizations, and locations (Sekine, 2004) and could be described in ontologies. Each fully recognized named entity (NE) has three features, namely, name, class, and identifier. WordNet words are words in a lexical database (e.g. WordNet database). Each fully recognized WordNet word (WW) has three features, namely, form, direct hypernym, and sense. Lexical search is not adequate to represent the semantics of queries referring to NEs or WWs. Some examples of NE-based queries are: (1) Search for documents about “football clubs”; (2) Search for documents about “Barcelona”; (3) Search for documents about “Paris City”; (4) Search for documents about “Paris City, Texas, USA”. In fact, the first query searches for docu- ments containing NEs of the class Football Club, e.g. Chelsea or Barcelona, rather than those con- taining the keywords “football club”. For the second query, target documents may mention Football Club Barcelona under other …

Content truncated to 500 words