Research On Video Recommendation Based On Knowledge Reasoning Of Knowledge Graph

With the rapid development of the Internet,various kinds of software and hardware technology such as video codec and streaming media are becoming more and more matured and improved,which leads to the explosion of various video websites and video apps and the video data size in the network is increasing day by day.Efficient retrieval of information in large-scale video data has become increasingly difficult,and owing to this circumstance the personalized recommendation appears.At present,the collaborative filtering recommendation algorithm is widely used in video recommendation because it is simple and easy to implement,has wide application range and good recommendation effect,and becomes a more successful method.However,problems such as data sparseness,cold start,and scalability still exist in practical applications.Collaborative filtering only relies on the interactively scoring data among user projects,ignoring the attribute characteristics of those projects themselves and the impact of implicit feature data on the recommendation results.At the same time,the impact of popularity is not considered when calculating project similarity.All the factors talked above limit the efficiency of recommendation.In order to solve those problems,the main work of this paper is as follows:Firstly,in order to improve the impact of popularity on video similarity calculation,this paper improves the item-based collaborative filtering,by adding popularity weighting factor when calculating video similarity and proposes a popularity?CF algorithm combined with video popularity.Secondly,considering the influence of video attribute feature information and implicit data features on recommendation accuracy,we propose a video recommendation algorithm called PTrans E?CF,based on knowledge reasoning of knowledge graph and combining with video attribute data.For item-based collaborative filtering,it has the problems that it ignores video attribute information and doesn't extract the hidden feature.In order to solve these problems,our algorithm uses the following four steps.Firstly,we transform explicit attribute feature information into a unified triple form using knowledge mapping technology.Secondly,all entities and relationships are extracted and embedded into the low-dimensional space using PTrans E knowledge reasoning technology.Thirdly,we use the PRA path algorithm to mine and discover implicit relationships between entities in the low-dimensional space.Finally,we use the information obtained in the third step to calculate semantic similarity between videos and integrate it into collaborative filtering.The advantages of our algorithm are as follows.Firstly,adding the semantic similarity alleviates the existing data sparseness problem to a certain extent when calculating the similarity.Secondly,the interpretation of the recommendation result is better by finding the connection among entities through path discovery.Thirdly,it improves the efficiency of recommendation by adding semantic similarity into the collaborative filtering recommendation.Finally,in experiment section,we use movie Lens100 K datasets and determine the proportion of semantic data in collaborative filtering by adjusting the fusion parameters.Then we calculate the accuracy,recall and F1 value of the popularity?CF algorithm combined with the popularity weighting factor and the PTrans E?CF video recommendation algorithm based on knowledge reasoning of knowledge graph.The experimental results show that our algorithm has better effects in alleviating data sparseness,and remarkably improves the accuracy of recommendation.Besides,we also test the scalability of PTrans E?CF,and the experimental results show that the algorithm has scalability.