A survey on using side information in recommendation systems

Repository

A survey on using side information in recommendation systems

Show full record

Title: A survey on using side information in recommendation systems
Author: Gunasekar, Suriya
Abstract: This report presents a survey of the state-of-the-art methods for building recommendation systems. The report mainly concentrates on systems that use the available side information in addition to a fraction of known affinity values such as ratings. Such data is referred to as Dyadic Data with Covariates (DyadC). The sources of side information being considered includes user/item entity attributes, temporal information and social network attributes. Further, two new models for recommendation systems that make use of the available side information within the collaborative filtering (CF) framework, are proposed. Review Quality Aware Collaborative Filtering, uses external side information, especially review text to evaluate the quality of available ratings. These quality scores are then incorporated into probabilistic matrix factorization (PMF) to develop a weighted PMF model for recommendation. The second model, Mixed Membership Bayesian Affinity Estimation (MMBAE), is based on the paradigm of Simultaneous Decomposition and Prediction (SDaP). This model simultaneously learns mixed membership cluster assignments for users and items along with a predictive model for rating prediction within each co-cluster. Experimental evaluation on benchmark datasets are provided for these two models.
Department: Electrical and Computer Engineering
Subject: Side information Recommendation systems SDaP MMBAE Survey Affinity estimation
URI: http://hdl.handle.net/2152/ETD-UT-2012-05-5740
Date: 2012-05

Files in this work

Size: 670.4Kb
Format: application/pdf

This work appears in the following Collection(s)

Show full record


Advanced Search

Browse

My Account

Statistics

Information