An Optimized Video Recommender System Based On Implicit Feedback

Now we are in an era of information explosion.With the promotion of network technology and the increase in the number of applications,the amount of data in the Internet is also rising rapidly.Due to the impact of the Internet,the video site also has the problem of information overload.For this situation,the current consensus is to use the recommended system to solve.Internet film and television sites in recent years has developed rapidly,there are many users explicit feedback data.Most of the recommended systems are based on the explicit feedback data to recommend,but still have more huge implicit data has not yet been used.Compared to explicit feedback data,implicit feedback data exists in a wider range,record a large number of user selection behaviors,are a very valuable recommendation resources.Based on this,this paper presents a new implicit feedback recommendation technique.The proposed technique is inspired by the matrix factorization model.That is,the model by mapping the user / item data set to low-dimensional space.Based on this,the feature vector of the user / item is estimated,and the result set of the user selection behavior is estimated.Based on the matrix factorization model,this paper successfully transforms the recommendation problem into the optimization of the probability of user preference estimation by dimension reduction technique.This type of conversion does not require the introduction of negative cases,but only concerned about the user behavior itself.So the algorithm does not have the drawbacks which are due to the increase in noise caused by the introduction of other implicit recommendation algorithms.Then,through the stochastic gradient descent algorithm SGD,the user selection behavior is optimized step by step,and finally the user selection tendency probability estimation matrix is obtained.The algorithm is similar to the matrix factorization model,has a very good scalability,and the complexity of the model and the user's choice of goods only the number of linear correlation.At the same time,in order to deal with the huge implicit data and further improve the running speed of the model,this paper puts forward the scheme of binning the user-item data set.The scheme divides the data into multiple disjoint subsets,enabling parallelization of the SGD.Finally,we designed a detailed experiment to verify our system.First,we introduce the way to get the data set and give an example;Second,several evaluation methods of the recommendation system are discussed,and the evaluation scheme is selected for the implicit feedback recommendation system;Finally,the results were analyzed: Firstly,compared with other algorithms,the advantages of the algorithm are verified under the same data set.Secondly,the parallelization of the algorithm is realized by multi-process technology,which proves the efficiency of the parallel optimization algorithm.