Research On Entity Relation Extract Based On LSTM

Internet technology is developing rapidly nowadays,and the scale of Internet users shows obvious growth.Massive text data appears on the network.Automatically,quickly and effectively extract the useful knowledge has already become our urgent demand,such as the Internet search,automatic navigation,automatic question answering,machine translation and speech recognition applications,which are inseparable from the knowledge base,and entity relation extraction is one of the key technologies of building knowledge base.The purpose is to extract from natural language text of semantic relations between entities.Entity relationship extraction has become a research hotspot in data mining,machine learning,artificial intelligence and natural language processing.At the same time,it has great application value and broad application prospect.In fact,relation extraction is relation classification.Among the current multiple deep models,the recurrent neural network(RNN)model based on long-term memory(LSTM)is considered to be particularly suitable for the processing of text sequence data because of its ability to effectively utilize the long-distance information dependence in sequence data.Therefore,this paper proposes a deep learning model based on LSTM to solve the problem of relationship classification.On the basis of LSTM,this paper firstly studies word embedding technology.Word embedding can map words to a low-dimensional real number vector through neural network.It can avoid the shortcoming of lacking semantic information of traditional word vector.Secondly,the text feature extraction is studied.In this stage,four text feature schemes are proposed to feed the feature vector into the bi-directional LSTM(BLSTM)model.After that,the state of each time is given different weights in combination with the Attention mechanism,so that the model can solve the problem of information redundancy to the greatest extent and optimize the text feature vector on the basis of retaining effective information.Finally,using Gradient Boosting Decision Tree(GBDT)as a classifier.It can solve the weak generalization ability caused by traditional Softmax as a classifier.To some extent,it can further improve the accuracy of relation classification.In addition,in the process of implementation,this paper also experimented with the visualization of entity relationship results,so that we can view the whole or part situation in the graph database,allowing users to view the association between entities through simple operations.