| At present,the research in the field of knowledge graph focused on two aspects of knowledge graph's construction technology and application based on knowledge graph.In recent years,great progress has been made in the research of knowledge graph construction technology.Knowledge reasoning has also been greatly developed as an important link and key technology in knowledge graph construction.Through knowledge reasoning,new knowledge can be obtained from existing knowledge.Thereby further improving the knowledge graph.In terms of knowledge graph application,the use of knowledge graph to realize automatic text generation in recent years is also an important aspect of artificial intelligence research.We look forward to the day when computers can write high-quality natural language text likes humans write.This paper proposes a knowledge-reasoning algorithm based on random walk and an automatic text generation technology based on knowledge graph.The following are the main work and contributions of this paper:(1)A goal-oriented random walk algorithm is proposed,which uses a specific inferred target as the direction of each step in the random walk process.Specifically,in order to achieve such a goal-oriented mechanism,at each step of random walk,the algorithm dynamically estimates the potential of each adjacent point by using the final target,and assigns greater probability to the adjacent points with higher potential.Therefore,the algorithm is more inclined to visit adjacent points that are beneficial to infer the target and avoid shifting to the noisy structure.Compared with the basic random walk,the method in this paper directly random walks through inference targets,which improves the efficiency of mining useful formulas and has a strong anti-noise ability.Compared with the heuristic method,the method in this paper can automatically learn the random walk strategy and dynamically adjust the strategy of different inference goals,while the heuristic method needs to manually write heuristic rules and always follow the same rules.The experiment of link prediction task proves that the method in this paper has high efficiency in mining formulas,and has good performance on WN18 and FB15 K data sets.(2)Constructed the movie knowledge graph,first determined the initial data needed to obtain from Douban,then analyzed the website's webpage structure,and chose to use Python's requests library and bs4 library and then used regular expressions to grab the original data from Douban.Then process all the one-to-many relationships in the movie data,all decompose into one-to-one relationships and store them in the corresponding database tables,and then export the corresponding CSV files.Then,all CSV files are directly imported into the Neo4 j graph database and we have reached the knowledge graph we need.Finally,the target-guided random walk algorithm is applied to the movie knowledge graph to make the constructed knowledge graph perfect.(3)Using the constructed movie knowledge graph,an automatic text generation system is realized.The text generation is an automatic question-and-answer type text generation,which is an extension of the automatic question and answer system.Based on the knowledge graph,this paper decomposes the text to be generated into several subquestions,queries the answer in the knowledge graph,and finally uses the query answer to generate the final text.In the system implementation process,the Naive Bayes algorithm and TF-IDF feature extraction are used to classify the problems,so that various problems can find their corresponding Cypher query templates.The combination of the Word2 Vec word-vector language model,BiLSTM and CRF is used to realize the function of named entity recognition,so that the Cypher query template can be filled to obtain a complete Cypher query statement.In the query stage,the fuzzy query function is also set for the direct query failure caused by the possible situations such as user input error,named entity recognition error and so on. |
PDF Full Text Request