Research And Application Of Recommendation Algorithm Based On Knowledge Graph Technology

In the Internet era,the explosive growth of online content and services has provided users with overwhelming choices,The research of personalized recommendation system is an important breakthrough in realizing information filtering,usually based on the historical interactive information of users to mine potential interests and preferences to push projects that satisfy users.However,the recommendation algorithm based on collaborative filtering only constructs user-item scoring matrix based on the analysis of user project interaction information to complete interest recommendation of follow-up algorithm,which often has the problem of data sparsity and cold start.In order to overcome these limitations,researchers propose to incorporate auxiliary information into the recommendation algorithm.From the semantic aspect,this paper proposes to introduce rich semantic information of knowledge atlas and use knowledge representation learning method to generate recommendation results by embedding data into low-dimensional vector space to calculate semantic similarity on the basis of retaining the semantic information of users or projects themselves.This algorithm enriches the semantic information of users or items in traditional recommendation algorithms,and effectively enhances the recommendation performance.The contents of this paper are as follows:1.In this paper,Bernoulli sampling strategy is introduced to improve the negative sampling algorithm in the training process of translation model in knowledge representation.Different substitution probabilities are set according to different relationship types between head and tail entities to reduce the probability of constructing false negative triples.2.This thesis uses the improved translation model to vectorize the user and project and measure the semantic similarity,effectively integrate the information from different sources and solve the data sparse problem.Based on the translation model to learn the semantic information of the user and the project accurately,the idea of the maximum distance is used to separate the positive and negative samples in the vector space,which means that similar entities are gathered in the same regional space,and the link prediction is used to verify the model.Training efficiency,choose the Euclidean distance to measure the semantic similarity between entities,the closer the distance,the higher the similarity,the lower the opposite.3.This paper proposes a recommendation algorithm based on knowledge mapping technology.The main idea is: according to the analysis of user and projectinteraction data,obtain the user-item scoring matrix,extract knowledge from different data sources of the Internet,build knowledge map,and find knowledge for the items in the matrix.The matching entities in the map use the improved TransH translation model to learn and vectorize the semantic information of the entities in the knowledge map and the relationship information with the connections between different entities to generate a semantic similarity matrix of the project-item.By constructing the user interest model,the user interest can be dynamically reflected with time,and the user-item score matrix can be used to calculate the similarity of the project similarity,and the final project similarity matrix can be calculated by combining the weights to predict the user's score on the project.Sort a certain number of items in descending order by rating as a list of recommended items and push them to the user.4.In this paper,three data sets of film field,Book Field and e-commerce field are used to verify the effectiveness of the proposed algorithm compared with the traditional recommendation algorithm in terms of average absolute error and F value.