Fine-Grained Entity Type Classification And Short Text Entity Linking Research

The text data in the Internet is growing explosively in the current age of big data.How to find valuable information accurately from the massive Internet text has become a hot research issue at present.The generation of knowledge graph makes it possible to make full use of massive Internet data.Fine-grained entity type classification is an important part in information extraction,which can enhance the performance of entity linking and the downstream application of knowledge graph.Entity liking is a key technology in the construction and application of knowledge graph.The main research content of this paper is fine-grained entity type classification and short text entity linking.The research results are as follows:(1)A fine-grained entity type classification method based on BERT is proposed.Firstly,Samples are sent to BERT layer for feature extraction,and then take the corresponding vector of [CLS] in BERT output as the sentence vector of the input text.According to the begin and end positions of entity mention in the input text,the corresponding vectors in BERT output are taken and concatenated as the entity mention eigenvector.The sentence feature vector and entity mention eigenvector are concatenated to for a joint vector,and the joint feature vector in sent to the classification layer to classify the entity mention.Experiments show that the fine-grained entity type classification method proposed in this paper can effectively solve the problem of fine-grained entity type classification in Chinese.(2)A short text entity linking method based on multi-task learning is proposed,and the multi-task learning method is introduced into the short text entity linking process.Constructing a multi-task learning model,short text entity linking is the main task,and entity classification task is introduced as auxiliary task.Auxiliary tasks can alleviate the problem of information failure in the process of short text entity linking,so a more general text vector representation can be obtained and the performance of short text entity linking can be enhanced.Experiments on the dataset of CCKS2020(National Knowledge Mapping and Semantic Computing Conference)task 2 show that the method proposed in this paper has a good performance in entity linking and can effectively solve the problem of information failure in the process of short text entity linking.