| Nowadays,with the rapid development of mobile internet and the continuous rise of social media,the way users obtain information has undergone unprecedented changes.The way video content is disseminated has replaced text dissemination as the mainstream way for users to obtain information.However,as the number of users has increased,excessive video information has gradually caused information overload problems,seriously affecting users’ information acquisition experience.Therefore,to solve this problem,this article focuses on the research of mass diffusion recommendation algorithms based on graph networks,and considers user negative feedback behavior to implement a video recommendation system,to solve the problem of information overload and provide personalized recommendation services to users.Firstly,in order to obtain an initial energy value that can reflect the importance difference of tags in the user interaction graph and to solve the problem of unreasonable initial resource allocation of traditional mass diffusion algorithms,this article proposes a tag energy initialization method based on random walks on a hypergraph.This method treats tags as hypernodes,and users or associated videos that are the source of tags as hyperedges to construct a hypergraph.By considering different point-edge weighting strategies based on activity factors for the hypergraph,a basis for generating the transition probability matrix for random walks is provided,thereby ultimately calculating reasonable tag initial energy values.Secondly,in order to address the problem of sparse user interaction matrices and user interaction with video projects,this article applies bidirectional diffusion algorithms to obtain user interest vectors for tags and tag relevance vectors for videos.Based on the mediating role of tags,the normalized interest vector and relevance vector are multiplied to obtain a predicted rating for videos that aren’t interacting.By sorting the rating values,a video recommendation list is obtained.In the algorithm performance comparison experiment,the bidirectional mass diffusion algorithm incorporating hypergraph random walks has certain advantages in precision,recall,and F1 value compared to other similar algorithms.Next,this article analyzes the complementary effect of negative feedback on recommendation algorithms and proposes a negative feedback definition method based on video playback duration and playback duration ratio to address the problem of difficult collection of negative feedback behavior.Subsequently,using the improved TF-IDF algorithm,the user’s dislike for each tag is dynamically calculated within a sliding time window,and the user’s negative feedback tags are obtained based on the maximum antipathy value.This is used to screen candidate recommendation objects containing these negative feedback tags and to negatively optimize the bidirectional mass diffusion algorithm incorporating hypergraph random walks.Experiments show that the negative feedback tag determination algorithm proposed in this article can effectively combine positive recommendation methods and filter out videos that users are not interested in,improving the quality of recommendations compared to other baseline algorithms on the Kuai Rec dataset.Finally,this article carries out software development work for the video recommendation system,tests the relevant functional modules,and demonstrates its actual running effects.The test results show that the system has achieved the expected design,can help users better explore video content of interest,enhance user stickiness and loyalty,bring certain economic benefits,and have practical guiding significance for the development and operation of video platforms. |
