Research On Named Entity Recognition Relation Extraction And Recommendation Algorithm In Chinese Tourism

In recent years,people's demand for information services has changed from simple information acquisition to knowledge acquisition,and the rapid development of Knowledge Graph has provided many conveniences for this purpose.However,the General Knowledge Graph is not specialized in the application of knowledge services in specific fields,and the tourism field is no exception.The realization of tourism services requires the professional Knowledge Graph of tourism field as the basis.To construct domain Knowledge Graph,the basic task is to extract domain knowledge.To construct domain Knowledge Graph,the most important thing is domain knowledge extraction.Because most of the knowledge entities in the field of tourism exist in the form of unstructured text,the acquisition process becomes very difficult,and how to construct the domain Knowledge Graph effectively Interpretable recommendation combined with Knowledge Graph is also a hot.This paper focuses on the following problems:Knowledge extraction of unstructured text in tourism field and interpreta-bility recommendation based on Knowledge Graph.In the task of identifying named entities in Chinese tourism knowledge extraction,the current mainstream model can not solve the ambiguity and nesting of words effectively,and because of the restriction of training data set,it can not identify entities such as time and things in the tourism field.In the task of Entity Relation Extraction(Classification)in Chinese tourism field knowledge extraction,attention mechanism is added at present.The convolution neural network model of(Attention)can not effectively consider the local information of the entity to the location and the context of the statement,and is also subject to the lack of the Chinese entity relation extraction corpus,which leads to the inaccurate extractionof the trained model entity relation.The training time cost of the KPRN model which currently performs the best in the Knowledge Graph interpretable Top-N recommen-dation task is high.In view of the above problems,the contents of this paper are:(1)Using Google pre-training model Bert to replace word2vec for word embedding in named entity recognition and constructing Bert-BiLSTM-CRF combi-nation model to realize domain text knowledge entity extraction.(2)Propose a Chinese tourism relationship extraction model Transformer-PCNN,in which Transformer's self-attention mechanism deepens the network model,and segmented pooled PCNN can effectively extract the local features of the sentence,which can effectively solve the above problems;Construct a Chinese tourism relationship extraction corpus to Used for model training;(3)Using GRU instead of the LSTM layer in the model to improve the model training efficiency by using the feature of parameter reduction brought by one less gating unit than LSTM.The experimental results of multiple comparison models on multiple data sets show that the accuracy of the Bert-BiLSTM-CRF named entity recognition method has been improved in time and place,the Transformer-PCNN based entity relationship extraction method has made some progress compared with the existing methods,and the training efficiency of the model based on the improved KPRN model has also been improved.