Research On Intelligent Extraction Of Knowledge Element Based On Feature Item Weight And Sentence Similarity

As the knowledge age comes, human cognitive granularity has changed. Rather than staying in the literature level, people's access to knowledge has been deeped into the knowledge element level and people need to get a solution to a specific problem. People need to locate and acquire the knowledge contained in the literature quickly.The knowledge-organizing on knowledge element level has solved the problem of schizophrenia and realized knowledge cooperation and knowledge sharing. The knowledge element link theory can generate a new form of knowledge management — Knowledge Grid and then establish a complete knowledge service system. It can also speed up the process of knowledge creation, realize the value-added of knowledge, promote the transition from information service to knowledge service, improve the effect of knowledge service, and provide the method for human knowledge learning and knowledge innovation. It has practical significance to research on the knowledge extraction technology.This thesis studies the knowledge element definition and structure model for the existing problems of knowledge element extraction and defines five-tuple structure model of knowledge element based on the existing knowledge element structure model. The knowledge element is extracted by improved MF-S-TFIDF algorithm and the sentence weight algorithm. The whole process is also verified by experiments. The main work of this thesis includes five parts as following:(1) Study knowledge element theory and technology related to the extraction of knowledge element and analyse the development of knowledge element extraction technology at home and abroad. Summarize the definition of knowledge element and define five-tuple structure model of knowledge element based on the existing knowledge structure model.(2) Study the feature weighting algorithm and focus on the TF-IDF algorithm to make up for the deficiency of the traditional TF-IDF algorithm in the semantic level of words.Similarity calculation is carried out based on the CiLin and word classification is finished .An improved S-TFIDF algorithm is presented based on semantic similarity and TF-IDF algorithm. The improved MF-S-TFIDF algorithm which is fusioned of multi-character factors in words is applied to the process of feature item weight calculation.(3) Improve the sentence similarity algorithm by analyzing the shortcomings of sentence similarity algorithm based on word matching and edit distance. The semantic information of words is considered in sentence similarity algorithm based on word matching.The exchange operation of nonadjacent blocks is added to the sentence similarity algorithm based on edit distance and also consider the influence of word semantic information and word internal factors on the editing distance. The optimized algorithm improves the accuracy of sentence similarity calculation and is applied to the process of sentence weight calculation.(4) Study the sentence weight calculation algorithm and incorporate the sentence multi-feature factors to carry out the sentence weight calculation.(5) Integrate knowledge element intelligence extraction algorithm and extract the attributes of knowledge element to gain structured knowledge elements. Make the experimental comparision and finish the algorithm evaluation.Finally, the content of this thesis is summarized and the outlook of the work is put forward.