A Novel Recommendation Model Based On GNN And Overlapping Community Detection

In recent years,network literature has flourished and greatly enriched the amateur life of netizens.However,it takes a lot of time for users to find their favorite novels,and recommendation technology is an effective solution to this problem.At present,many small and medium-sized novel reading websites do not have satisfactory recommendation ability,so the novel recommendation research for small and medium-sized novel reading websites is of certain significance.Traditional recommendation technology needs to deal with the problems of sparse data and cold start,but the combination of community discovery and recommendation technology has excellent effect on reducing the impact of the above problems.In this paper,a novel recommendation model is proposed by combining NOCD model and rating prediction recommendation technique.The main contents of this paper are as follows:(1)User similarity network is constructed through user data,and NOCD,the model of overlapping community detection,is used to divide the community of user network.(2)On the basis of overlapping community detection,select communities with high degree of belonging to target users to establish candidate user sets.This process fills candidate user score matrix by data smoothing technology to reduce the impact of data sparsity on the recommendation process.And further in the search for the target user’s nearest neighbor user set.(3)The method based on matrix decomposition is used to score and predict target users,and the final recommendation is made according to the prediction results.When new users join the system,update the user community to reduce the impact of cold starts.(4)Through comparative experiments,it is proved that the recommendation model proposed in this paper can achieve more impressive recommendation effect under the same experimental conditions,and has good performance in the face of data sparse problems and cold start problems.