Algorithm For Rating Prediction In Recommendation

With the rapid development of Internet and e-commerce in the last decade,overload of information has become a major problem for online users.Recom-mender systems,which are able to provide a list of interested items to a spe-cific user by collaborative filtering,can settle this problem.Recommendation aims to recommend items of interest to a given user by applying user' profiles,items' attributes or a history of feedbacks on items by a set of users.Rating is one kind of explicit feedbacks from recommender system,and it directly shows how much a user likes an item.As a result,rating has attracted a lot of attention in the past decades.Rating prediction is an important issue in recommenda-tion,because it can help recommender systems discovering user' attention and understanding users' preferences.The traditional algorithms for rating prediction either use the regression model or the Bayesian model to predict ratings according to metadata of items,or using the collaborative filtering,in which recommender systems make pre-diction for target user based on similarly users 'feedbacks.Although they have greatly promoted the development of recommendation,they still have their lim-itations.In recent years,representation learning which is represented by deep learning has been widely concerned in the field of speech recognition,image analysis and natural language processing.In this paper,we mainly study repre-sentation learning methods which can be used to solve rating prediction prob-lem,so as to effectively discover the semantic relation between users and items in a low dimensional space,alleviate sparseness of data,and help recommender system understanding user's preference and combining heterogeneous informa-tion.The contributions of our work are as follows:1.We aim to propose a novel model,called Aspect-based Latent Factor Model(ALFM)to integrate ratings and review texts via latent factor model,in which by integrating rating matrix,user-review matrix and item-attribute ma-trix,the user latent factors and item latent factors with word latent factors can be derived.Our proposed model aggregates all review texts of the same user on the respective items and builds a user-review matrix by word frequencies.Similarly,an item's review is considered as all review texts of the same item collected from respective users.According to different information abstracted from review texts,we introduce two different kinds of item-attribute matrix to integrate the item-word frequencies and polarity scores of corresponding words.Experimental results on real-world data sets illustrate that our model can not only perform better than traditional models and art-of-state models on rating prediction task,but also accomplish cross-domain task through transfer-ring word embedding.2.We bring a new perspective of traditional recommendation task by re-defining it as link prediction problem in a multi-relational bipartite graph,where valid values of ratings correspond to relation types.We propose a novel users'Feedback Embedding Model(FEM)to represent each rating by a translation vector and two embedding matrices,which can capture the rating-dependent aspects of users and items for enhancing the matching level of their embed-dings.FEM represents each rating as a translation between users and items in the rating-dependent subspace.Then various types of initializations are incor-porated into the proposed model to enhance the learned embeddings of users,items and ratings for better optimization.Empirical results on real world data sets show that our proposed model outperforms state-of-the-art methods for rat-ing prediction task.The effectiveness of auxiliary information is also demon-strated in experiments.3.We propose a multi semantic path based Latent Social Embedding Model(LSEM).LSEM considers rating prediction a weighted heterogeneous informa-tion network.It proposes a similarity based on the weighted meta path,which can help to construct the potential social network of users and items.Then LSEM learns representations of users and items in their potential social net-work through graph learning method respectively.Finally,it makes prediction via link embeddings of users and items through matrices which reflects user'cluster-level rating patterns.In this paper,the experiment is carried out in the real data set.The experimental results show that the potential social information of users and commodities can improve the accuracy of rating prediction.