Similar but Different: Exploiting Users' Congruity for Recommendation Systems

TL;DR

This paper explores how incorporating users' congruity, or varying degrees of agreement among similar users, can enhance social media recommendation systems, showing that embedding this information improves performance.

Contribution

It introduces the concept of users' congruity into recommendation systems and demonstrates its effectiveness through experimental validation.

Findings

Embedding congruity improves recommendation accuracy

Users' interests vary despite social relations

Exogenous data helps discern user agreement levels

Abstract

The pervasive use of social media provides massive data about individuals' online social activities and their social relations. The building block of most existing recommendation systems is the similarity between users with social relations, i.e., friends. While friendship ensures some homophily, the similarity of a user with her friends can vary as the number of friends increases. Research from sociology suggests that friends are more similar than strangers, but friends can have different interests. Exogenous information such as comments and ratings may help discern different degrees of agreement (i.e., congruity) among similar users. In this paper, we investigate if users' congruity can be incorporated into recommendation systems to improve it's performance. Experimental results demonstrate the effectiveness of embedding congruity related information into recommendation systems.