Research On Construction Technology Of Knowledge Graph Of Network Film

Online movies is a concept proposed by qiyi in 2014 for movies played on the Internet.With the progress and development of the Internet,the number and types of online movies on online platforms have increased accordingly,which has led to the independence and siloing of online movies,making it difficult for users to find the relationships among online movies.Because of its powerful relationship processing and reasoning ability,and its ability to accurately show the inner connection of knowledge,this paper constructs a knowledge graph of online movies and designs a recommendation system based on it,which makes full use of the semantic information of the knowledge graph to recommend movies for users in a personalized way.In this paper,we study and construct the knowledge graph of online movies,and the main work is as follows.(1)Realizing multi-source data acquisition and processing.For the lack of professional data sets of online movies,this paper designs a scrapy-based web crawler program to crawl unstructured and semi-structured data from major news websites and video websites as the data sources of online movie knowledge graph.(2)A named entity recognition algorithm based on phonetic information is proposed.Since the current entity extraction models lack the ability to deal with the phenomenon of "multiple meanings of words" in Chinese,this paper proposes a Transformer-based entity extraction algorithm,which first vectorizes the character and word information through Lattice structure,vectorizes the pinyin information through convolutional neural network,and then vectors the information through Transformer.The algorithm then fuses the word and pinyin information through Transformer,and finally decodes them using conditional random fields to obtain the label sequences of characters,which has good performance on datasets such as Weibo and Onto Notes4.(3)Building a knowledge graph of online movies.For the crawler data and the data extracted using the named entity recognition algorithm go through the operations of deduplication and pre-processing,and again go through the steps of grouping,transforming entities and relationships into values,and finally store the data into the Neo4 j graph database and construct the web movie knowledge graph.(4)Design a recommendation system based on the web movie knowledge graph.The algorithm uses the Trans H model to embed the knowledge graph into the low-dimensional space,and then takes into account the long-and short-term recommendations by modeling short-term preferences with attention mechanisms and residual networks,long-term preferences with metric learning,and finally the results of both are weighted and fused.Both have good performance on public datasets.By constructing the knowledge graph of online movies,users can clearly and intuitively understand online movies and the connections between them,and can explore deeper information of online movies,which provides technical support for users to recommend their favorite movies.