A Dynamic Knowledge Extraction Method Based on Sentence-Clustering Recognition

In view of the clustering phenomenon of natural language as well as the demand for dynamic alternation of knowledge architecture, a dynamic knowledge extraction method based on sentence clustering recognition is put forward in this paper. First of all, the paper gives a research framework of the proposed method, which describes the transformation process from natural language texts to object oriented knowledge architecture. Some problems related to sentence vectorization are investigated, several fundamental definitions as well as one judgement theorem are given, and the postpositional processing on attribute vectors of sentence cells is discussed. A sentence clustering recognition approach is constructed using ART2 neural network and the concept of prior belief degree is adopted to measure the outcome of sentence recognition. A simulation procedure of ART2 neural network is compiled by Matlab, the effects of sentence recognition by the procedure are given, and the corresponding analysis is made. A width first code generating method is proposed so as to make knowledge transformation for the clustered sentences according to the outcome of sentence recognition by ART2 neural network, and the posterior belief degree is defined as a final evaluation index to sentence recognition and semantic model construction. The implementation steps of the above method are further introduced for a specific sentence pattern. Finally, the derived relation is generated by utilizing the new approach of structural modeling proposed by the authors, such that the knowledge extraction process from natural language texts to object oriented knowledge architecture can be accomplished. The effectiveness of the proposed method is verified by a practical example of mechanical CAD, which is adopted to penetrate the whole paper so as to demonstrate the complete implementation details.