Between Dec 19, 2024 and Jan 2, 2025, datasets can be submitted to DRUM but will not be processed until after the break. Staff will not be available to answer email during this period, and will not be able to provide DOIs until after Jan 2. If you are in need of a DOI during this period, consider Dryad or OpenICPSR. Submission responses to the UDC may also be delayed during this time.

Browsing by Subject "Supervised Clustering"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    A Personalized Recommender System with Correlation Estimation
    (2018-05) Yang, Fan
    Recommender systems aim to predict users’ ratings on items and suggest certain items to users that they are most likely to be interested in. Recent years there has been a lot of interest in developing recommender systems, especially personalized recommender systems to efficiently provide personalized services and increase conversion rates in commerce. Personalized recommender systems identify every individual’s preferences through analyzing users’ behavior, and sometimes also analyzing user and item feature information. Existing recommender system methods typically ignore the correlations between ratings given by a user. However, based on our observation the correlations can be strong. We propose a new personalized recommender system method that takes into account the correlation structure of ratings by a user. General precision matrices are estimated for the ratings of each user and clustered among users by supervised clustering. Moreover, in the proposed model we utilize user and item feature information, such as the demographic information of users and genres of movies. Individual preferences are estimated and grouped over users and items to find similar individuals that are close in nature. Computationally, we designed an algorithm applying the difference of convex method and the alternating direction method of multipliers to deal with the nonconvexity of the loss function and the fusion type penalty respectively. Theoretical rate of convergence is investigated for our new method. We also show theoretically that incorporating the correlation structure gives higher asymptotic efficiency of the estimators compared to ignoring it. Both simulation studies and Movielens data indicate that our method outperforms existing competitive recommender system methods.