Matrix Completion And Recommendation Algorithm Based On Word2vec

With the increasing popularity of the Internet to people's life,people's life has been closely related to the Internet,inseparable.Nowadays,the main way for people to get information is through the Internet,which leads to the problem of information overload.Therefore,the recommendation system emerges as The Times require.As a tool to filter information based on users' personalized information,the recommendation system can recommend the products most needed by users through their previous historical behaviors and the similarity between products,which saves time and energy for users to browse invalid information.Therefore,the emergence of recommendation system solves the problem of information explosion to a certain extent.Content-based recommendation algorithm and collaborative filtering algorithm are the main components of today's recommendation system.The problem of content-based recommendation algorithm is that when calculating the similarity between user preference characteristics and items,the number of the same tag is mainly taken into account,and the internal connection between different tags is ignored,which affects the recommendation accuracy.The inaccuracy of recommendation in collaborative filtering algorithm is mainly divided into two aspects.On the one hand,there is a project cold start problem,that is,when there is a new project,the scoring information of the new project is blank,and it is impossible to find the graded project similar to the new project through the user score,so that the score of the new project cannot be predicted.On the other hand,the collaborative filtering algorithm relies heavily on the user rating data and ignores the internal similarity between projects,resulting in the recommendation relying too much on the user's subjective rating data.To solve these problems,this paper proposes a recommendation algorithm based on word2 vec and a matrix completion collaborative filtering algorithm based on word2 vec.The above algorithms can solve the following problems:First of all,in order to solve the problem of inaccurate recommendation caused by ignoring the potential connection between words in content-based recommendation algorithm.In this paper,the definition of item attribute value similarity and a new method of item similarity calculation are proposed.Word2 vec is used to train the word vector of the main attribute values of the project and calculate the similarity between the two attribute values.Through the training of multiple linear regression model based on the historical score data,the weight of each attribute was calculated,and the similarity between items was calculated based on the similarity between attribute values,and the recommendation was made according to the similarity ranking.Secondly,the collaborative filtering algorithm is highly dependent on user ratings while ignoring the similarity between projects.In this paper,word vectors are trained by word2 vec and word vectors are used to calculate the word similarity between item attributes.In addition,the proposed new item similarity calculation method is used to complete the sparse scoring matrix.Based on the original Pearson correlation coefficient,the item similarity parameters proposed in this paper are integrated.Finally,the improved algorithm is applied to MovieLens data set to design experiments.The matrix completion algorithm and recommendation algorithm based on word2 vec are compared with collaborative filtering algorithm and content-based recommendation algorithm respectively,which have obvious advantages in recommendation accuracy and rating prediction error rate,proving that the algorithm proposed in this paper is effective and feasible.