Research On Sentence Similarity Calculation Method Of Fusion Knowledge

The calculation of similarity of sentence semantic is an important task in the processing of NLP,which has been widely used in the fields of NLP processing including information retrieval,information extraction,text classification,semantic cancellation,and instance-based machine translation.At present,with the deep application of deep learning technology in the field of NLP,semantic similarity calculation completes the text similarity calculation by mainly using the automatic learning vector representation of the context information of words in large-scale text to,which achieves better results than the traditional methods do.Due to its lack of interpretability,the development of deep learning encounters an unbreakable bottleneck.Some people believe that the future scientific breakthrough of AI may be determined by building an AI system based on knowledge and data at the same time,Therefore,on the task of semantic similarity calculation at sentence level,two methods of sentence similarity calculation on knowledge fusion are proposed in this paper.The research methods are as follows:1.A method of sentence similarity calculation based on SBA model.The SBA(Semem-Bilstm-Attention)model consists of five parts: input layer,vector layer that integrates sememe information,Bi LSTM network layer,Attention layer and output layer.First,the input words are transformed into vector representations by the SAT model incorporating sememe information,and then input into the Bi LSTM network layer Then,the weight distribution of the words in the sentence is calculated through the attention layer,and the similarity of the two sentences will be obtained.It is an improvement by 6.5% in accuracy compared to the baseline used in this article.2.A method of sentence similarity calculation based on SBA-LRSF model.The semantic similarity calculation based on How Net is combined with that based deep learning of sentence similarity calculation.The context is encoded through Bi LSTM network,knowledge branches are added from Bi LSTM network,and sememe inference is carried out for the words in the input sentences.By comparing the predictive sememe sequences of two input sentences,the sentence similarity of fusion knowledge branches is obtained,and then it will be weighted with the sentence similarity of SBA model.The accuracy is improved by1.4% compared with that of SBA model.