Collaborative Filtering Techniques
Pallavi Mishra, Sachi Nandan Mohanty · 2021
The perceptible transformation from an era of scarcity to the era of abundance has made availability of complex and diverse information (Big Data) all over the internet. As such, it becomes difficult for online users to locate and access useful information about the précised item of their interest for making accurate decisions. The extremely large amount of data available requires the creation of mechanisms for efficient information filtering. This led to the development of Personalised Recommendation Search Engines. Collaborative Filtering is one of the approaches of Recommendation Systems for dealing with this problem which encompasses memory-based techniques for matching people with similar interests and making recommendation on the basis of the reactions of similar groups. The key challenges of Collaborative Filtering are that the Collaborative Filtering algorithms should have the ability to deal with highly sparse data, to scale with the increasing numbers of users and items to make satisfactory recommendations in real time. The Netflix Prize Challenge, held in 2006, has fuelled much recent progress in Model-based Collaborative Filtering System adopting Machine Learning Techniques (such as Singular Value Decomposition (SVD) and Probabilistic Matrix Factorization (PMF)) to predict user ratings of unrated items with a lower prediction error. These developments have been successfully implemented in Personalised Recommendation Systems. This study focuses solely on recent advances in the field of Collaborative Filtering Approaches using Matrix Factorization Models and Nearest Neighborhood Models. Various techniques involved in the Matrix Factorization Models, such as SVD and PMF, will be projected with recently evolved models, incorporating a kind of implicit user feedback using SVD++ methods and further extended to a Factor Model using time SVD++ methods, which aim to account for temporal effects. The recent advances in the Nearest Neighborhood Model will be presented with extension to the Global Neighborhood Model and the Factorized Neighborhood Model in addition to temporal dynamics to bring enrichment in the predictive accuracy. This chapter will summarize the following broad areas: Overview of Collaborative Filtering Model, Baseline Predictor Estimation Techniques (Temporal Effects), Matrix Factorization Models with Advanced Dimensionality Reduction Techniques, Evolution of Nearest Neighborhood Models with Temporal effects.