Research On Named Entity Recognition And Knowledge Representation In Knowledge Graph Construction

The advent of the artificial intelligence has provided new opportunities and challenges for the effective organization and management of massive,heterogeneous,and complex data information that is flooded into people's lives,and the mining and utilization of hidden knowledge,which have become industry and academic hot research issues in the world.Among them,knowledge is the cornerstone of artificial intelligence,so knowledge graph technology has emerged and developed rapidly in recent years.The knowledge graph organizes various kinds of data information to form a huge knowledge base,provides rich structured data for the machine,and becomes an important resource in artificial intelligence technology.Although there have been many research results of knowledge graph and the technology is gradually mature,there are still some shortcomings in named entity recognition and knowledge representation.These shortcomings directly affect the quality of knowledge map construction.Therefore,the research on named entity recognition and knowledge representation learning has become a hot issue in the field of knowledge graph.This paper makes in-depth research on named entity recognition and knowledge representation in the construction of knowledge graph,and aims to improve the accuracy of named entity recognition in knowledge graph,and the performance of knowledge representation learning.The main research contents are as follows:(1)In order to solve the problem that traditional named entity recognition heavily depend on a large number of artificial features and cause insufficient text feature representation,this paper proposed a named entity recognition method based on Seq2 Seq model(BERT-Seq2Seq-Attention).Firstly,the method first uses the BRET pre-training model to dynamically generate semantic vectors for words.Secondly,encode the word vector through the encoder in the Seq2 Seq model,and introduce the attention mechanism to assign weights of the words to obtain local and global features of the text;Finally,the obtained features are input into the decoder,and through the softmax layer to predict the sequence labels.The experimental results show that the method has improved in accuracy,recall and F1,and has better applicability.(2)In view of the fact that the traditional knowledge representation model only considered direct relationships between entities are modeled,and not fully considered the diversity of relationships and the indirect relationship problems in multi-step relationshippaths,this paper proposed a knowledge representation learning model based on dynamic relationship mapping and path building(PDRM).Firstly,the model uses the TransD model to model the direct relationship,and sets a dynamic weight matrix to optimize the scoring function.Secondly,the model uses path modeling to obtain the indirect relationships between entities,which enriches the entity's semantic information.In addition,K-means clustering is used during model training to reduce the probability of error triads in constructs negative triad samples.Finally,the experimental results verify the effectiveness of the method.(3)Aiming at the problem of data sparseness and lack of semantic information in the information recommendation algorithm,a information recommendation algorithm based on knowledge graph representation learning(SAEKG-CF)is proposed.In this paper,firstly,the rating matrix is input as a stack autoencoder,and obtained the implicit feature representation of the project to calculate the feature similarity matrix;secondly,using the knowledge graph representation learning algorithm(PDRM)to map the entities in the project to the low-dimensional vector space,and calculating semantic similarity matrix between entities;then,on this basis,the feature similarity matrix is merged with the semantic similarity matrix to obtain a fusion similarity matrix,finally,according to the optimal fusion similarity matrix generate a top-k recommendation list.(4)Based on the service statistics data information,a smart service platform system is constructed,and the aforementioned research results were successfully applied to the knowledge question and answer and service information recommendation modules in the system.