Research On Named Entity Recognition For Tourism

With the rapid development of information technology,China's tourism industry is also developing rapidly in the direction of information technology.For tourists who travel outside,it is a very convenient way to use intelligent applications to solve problems encountered during travel,such as intelligent route recommendations,intelligent question and answer systems for scenic spots,etc.The tourism text contains a lot of key information with application value,such as scenic spot names,place names,special snacks,etc.How to extract these information and provide more intelligent services on this basis is particularly important.Therefore,the task of named entity recognition in tourism field is also getting more and more attention.Named entity recognition is the basic research of various tasks in natural language processing,and named entity recognition in the tourism field is an important part of the process of building tourism knowledge graph.Compared with the entities in the general field,the entities in the tourism text are longer in length and have more diversified definitions.These features make the task of named entity recognition on the tourism domain more challenging.Currently,for a field named entity recognition task,the introduction of the dictionary is to solve the above problems the most common way,but consider the introduction of a dictionary in the past mostly based on convolutional neural network or recurrent neural network,this fusion dictionary because of wrong word and text inherent word order caused a Series of ambiguity problems,so the dictionary information cannot be used well.In addition,there are few common datasets in the domain,which makes it difficult to conduct research.In order to solve such problems,this article introduces the following aspects:(1)To address the problem of the lack of named entity recognition datasets in the tourism domain,this paper first collected the relevant text information field of tourism in Xinjiang,with Xinjiang regional characteristics on the basis of previous studies,marked and build a small tourist area of NER datasets.(2)In terms of fusion lexicons,a named entity recognition method based on directed graph model is proposed.The main idea of this method is to use graph neural network to eliminate ambiguities caused by word segmentation errors,sentence inherent order and word order.The model first uses external lexicons information to construct a directed graph of sentences,generates corresponding adjacency matrices to match word information,and then uses convolution neural network to further obtain local features of each character.Finally,word vectors and sentence adjacency matrices containing local features are fed into a gated graph neural network.Word information is dynamically learned and conditional random field is introduced to obtain the optimal entity tag sequence.Experiments show that this method can improve the performance of named entity recognition.(3)A named entity recognition method of graph neural network combined with attention mechanism is proposed.Although the method based on directed graph can fuse lexical information well,it does not highlight the core words by giving the same degree of attention to each word in the sentence during the learning process.In this paper,we further improve the performance of named entity recognition in tourism domain by fusing attention approach.