Cold Start Recommendation Based on Attribute-Fused Singular Value Decomposition

Abstract

Collaborative filtering plays an important role in promoting the service recommendation ecosystem, and the matrix decomposition technology has been proven to be one of the most effective recommendation methods. However, the traditional collaborative filtering algorithm has great shortcomings in the recommendation of cold start items, especially the emergence of new items will be largely ignored. This not only has a very bad impact on the development of the item, but also greatly reduces the diversity of the recommendation system. The rise of mobile devices has also brought a large number of mobile applications, and these emerging applications need to be promoted in order to maintain the robustness of the application system. In order to solve this problem, we propose a method of combining the attribute information of the item with the historical rating matrix to predict the potential preferences of the user. It combines the attribute and time information into a matrix decomposition model. By testing our method on the movielens and the climbed JD dataset, the experimental results show that, compared with the baseline method, the proposed method achieves a significant improvement in recommendation accuracy. Therefore, this method is an effective way to solve the cold start problem of new items.

Evolving Hierarchical and Tag Information via the Deeply Enhanced Weighted Non-Negative Matrix Factorization of Rating Predictions

“…. The rating predictions were made by combining Equations (9) and (12) and utilizing the following objective function for the minimization task:…”

“…Memory-and model-based techniques are commonly used to elucidate CF recommendations [3,[11][12][13][14][15]. Past studies have demonstrated the benefits of memory-based CF, wherein rating predictions are computed from the preferences of similar users via a rating matrix [12,[16][17][18][19]. Conversely, the model-based CF technique leverages a user-item rating matrix to initially build a predictive model using deep learning methods and then source the rating predictions from it [3,20].…”

Factorization Meets Neural Networks: A Scalable and Efficient Recommender for Solving the New User Problem

“…However, this form of representation is prone to produce considerable invalid information, especially when the dimensions are very high [27]. Although the truncated SVD (singular value decomposition) [8], [11] technique for low-rank factorization can extract the metadata from high-dimensional data, the correlation between various types of data has not been fully considered. Moreover, this category of principal component extraction technique cannot flexibly combine extracted metadata to improve performance.…”

“…The key to solving this problem is how to make full use of user profiles to predict their behavior. A user's profile can be the user's own attribute [6]- [8] or other information besides the behavior. We collectively refer to this available user information as user profiles.…”

“…Second, in addition to the user context, old users contain much more information than new users. In particular, the rich information associated with old users can include not only user contexts, but also item attributes, and user-item interactions [8], [13]. To clearly express different types of user information, we uniformly define the information of both new users and old users from the perspective of context-aware recommender systems (CARS) [12], [14] as follows.…”