Research And Application On Multidimensional Conceptual Association Model For Text Similarity Calculation

Text similarity calculation technology plays a basis and important role in the field of natural language processing technology. In order to adapt to the knowledge development in the new era, as well as to meet the needs of information processing technology, this thesis puts forward multidimensional concept model on which the sentence similarity calculation method is based. And the model can be applied to practical teaching.The thesis mainly aims at the limits of the semantic network and HowNet, and proposes the concept of multidimensional model. Compared with the knowledge representation method of semantic network, definition of the mentioned concept in the present model is an entity class rather than just a semantic unit. With the widespread usage of network or HowNet, relationship concept designed by the model contains not only words correlation but also the context semantic relations. Multidimensional concept model realizes the mapping from the nouns to verbs, making characteristics relationship more manifest, and come up with a relationship describing concepts’ relationship - an analogy analysis of relationship, which promotes the relationship between the concepts to be clearer, and the calculation to be more accurate when calculating the sentence similarity. When compared with the existing method of vector space, the merge of word sets based on concepts increases of coarse search space, however can get accurate and precise expectations in the second retrieval after downsizing.Then the thesis establishes a link between multidimensional concept model and statements of the architecture, and advances the model based on multidimensional concept to calculate sentence similarity.The idea of sentence similarity calculation consists of following steps:firstly, to extract the text of the statement to be processed architecture. And then establishes multidimensional concept model from sentences, by comparing core words and local multidimensional concept model. Finally we must calculate their similarity. The thesis integrates similarity computation of semantic structure and a sentence and lexical semantic, and achieves a multidimensional concept correlation model based on the text similarity computing system ultimately. The method can be applied to computer assisting instruction in automatic grading of subjective topics, which proves its validity.