| In recent years, with the popularization of network and the development of information society, a large number of information has increased greatly. Therefore, computer text processing has become a more and more important field. Evaluating the semantic similarity between words is an important branch. Semantic similarity can be understood as a degree of taxonomic likeness between terms. It plays an important role in many fields, such as word-sense disambiguation, document classification or documents clustering, information retrieval or extraction.The Information Content(i.e., IC) provides an estimation of concept’s generality/concreteness which has been applied in the evaluation of semantic similarity between concepts successfully. we proposed two algorithms based on ontology. Recently, computing IC of a concept by using the taxonomic features extracting from ontology has obtained better results. In the taxonomic structure of the ontology, the height of a concept can also affect concept’s IC. In the dissertation, we consider the superconcepts, leaves, height of a concept, and propose a new IC computation method. Some general methods like path-based methods, feature-based methods and vector-based methods have some problems about computational complexity, the limitations of parameter adjustment and it’s hard to find taxonomic features in ontology. To overcome these problems, we propose a new IC-based method according to information theory, based on the full consideration of the ontology knowledge.In order to verify the validity of the method, we use the medical ontology SNOMED CT and the general ontology Word Net ontology as input ontologies, have carried on the experiment on three datasets. In the experiment, we compute the correlation between our values and human subjective judgment. Also we compare our method with other related works. Results show that the proposed method got a better precision, improved the Batet and Sánchez’s algorithm. It is verified that the IC computation method of the concept and IC-based semantic similarity method are valid. |
