Study On Several Key Problems Of The Ranking-based Recommendation System

In the study of recommendation systems,many methods based on predicting ratings have been put forward.However,the rating-predicting methods have some shortages.It pays too much attention to predicting,instead of the nature of recommendation.In this article,we use a pairwise-based learning algorithm to learn the model to make it more sensible.Our study including the following aspects:1.Research on the fitness of the Learning to Rank algorithm on recommendation problems.The LTR algorithm can be used in three ways:pointwise,pairwise and list-wise.We focus on how to use the Learning to Rank algorithm to solve the recommendation problem.We use a pairwise-based learning algorithm to learn our model and take the zero-sampling method to improve our model.2.Text modeling.A common used method,collaborative filtering,focuses only on the users' ratings.And it factorizes the rating matrix to get the user vector and item vector in the recommendation task.Thus,in order to get more information,we focus on how to model the reviews of the users.3.Combination of the rating model and the text model.In our work,we separate the modeling process of the ratings and the reviews.And that lead us to focus on the third study point which is how to combine these two parts.