Exploiting Hierarchy for Ranking-based Recommendation

TL;DR

This thesis introduces a novel hierarchical structure-based algorithmic framework, HIR, for collaborative filtering that leverages NCD properties to enhance recommendation quality and address data sparsity.

Contribution

It develops the HIR framework, exploiting hierarchical item structures and NCD properties to improve collaborative filtering recommendations.

Findings

HIR improves recommendation accuracy over baseline methods.

Utilizes hierarchical item structures to better model item relations.

Addresses data sparsity issues in collaborative filtering.

Abstract

The purpose of this master's thesis is to study and develop a new algorithmic framework for collaborative filtering (CF) to generate recommendations. The method we propose is based on the exploitation of the hierarchical structure of the item space and intuitively "stands" on the property of Near Complete Decomposability (NCD) which is inherent in the structure of the majority of hierarchical systems. Building on the intuition behind the NCDawareRank algorithm and its related concept of NCD proximity, we model our system in a way that illuminates its endemic characteristics and we propose a new algorithmic framework for recommendations, called HIR. We focus on combining the direct with the NCD "neighborhoods" of items to achieve better characterization of the inter-item relations, in order to improve the quality of recommendations and alleviate sparsity related problems.