Program Recommendation Algorithm Based On LDA Topic Model And ALS Collaborative Filtering

In recent years,with the rapid development of radio and television industry and Internet technology,intelligent TV and Internet audio-visual terminals are becoming more and more popular,program source and content are becoming more and more rich,making the audience quickly enter the era of content surplus from the era of program scarcity.For viewers and audio-visual operators,how to recommend the programs has become a more and more important topic.With the large-scale upgrading of the two-way set-top box,the audience’s viewing behavior has the technical conditions of return.At present,the audience rating data generated by national audio-visual operators every day can reach TB level.In this big data environment,audience rating behavior and interest can be accurately depicted.Therefore,the research of program recommendation algorithm has the premise of technology and business.At present,researchers have proposed collaborative filtering recommendation,content recommendation,similarity recommendation,association rule recommendation and other recommendation algorithms.However,the collaborative filtering recommendation algorithm is the most widespread in various fields.In the big data environment,when the collaborative filtering recommendation algorithm is applied to the program recommendation scenario,there are the following problems:1)when the program rating matrix with low sparsity is used as input,the problem of data sparsity,cold start and low recommendation accuracy will occur in the program recommendation algorithm;2)if the recommendation algorithm iterates for a long time,there will be many new problems,such as single result type,unstable recommendation accuracy,etc.This dissertation designs a program recommendation algorithm based on the theme model of implicit Dirichlet allocation(LDA)and the collaborative filtering of alternative least squares(ALS).It not only improves the accuracy of program recommendation results,but also increases the diversity of recommendation results under the premise of stable recommendation,which meets the needs of users’ program viewing.Algorithm working mechanism:1)using LDA theme model to obtain program characteristics and program similarity matrix,then designing an algorithm to optimize the sparse degree of rating matrix,taking the program similarity matrix as the weight factor,finally filling in the blank value of the rating matrix,so that the problems caused by sparse rating matrix can be alleviated;2)taking ALS collaborative filtering algorithm as the basic algorithm,reducing The sparse program rating matrix is used as input,combining with the credibility of collaborative filtering recommendation between programs and users,a model of dynamic adjustment of weight is designed to improve ALS collaborative filtering algorithm,so as to achieve the effect of "stable recommendation" and"diversified recommendation";3)when facing the big data environment,spark technology is used to realize the complex matrix iteration process of program recommendation algorithm,so as to achieve the improvement section In order to solve the problem of low efficiency,parallel algorithm is recommended.The simulation experiment is carried out by building spark big data experiment platform.The experimental data shows that the mean squared error(MAE)value of the program recommendation algorithm designed in this dissertation is stable at about 0.78,which is about 15%higher than the traditional ALS collaborative filtering recommendation algorithm,and the effectiveness of the program recommendation algorithm is verified.