Improved user similarity computation for finding friends in your location

Georgios Tsakalakis,Polychronis Koutsakis

doi:10.1186/s13673-018-0160-7

Abstract

Recommender systems are most often used to predict possible ratings that a user would assign to items, in order to find and propose items of possible interest to each user. In our work, we are interested in a system that will analyze user preferences in order to find and connect people with common interests that happen to be in the same geographical area, i.e., a “friend” recommendation system. We present and propose an algorithm, Egosimilar+, which is shown to achieve superior performance against a number of well-known similarity computation methods from the literature. The algorithm adapts ideas and techniques from the recommender systems literature and the skyline queries literature and combines them with our own ideas on the importance and utilization of item popularity.

Highlights

The diversity of social networks makes the problem of correctly estimating user preferences essential for personalized applications [1]
We are interested in a system that will analyze user preferences in order to find and connect people with common interests that happen to be in the same geographical area, i.e., a “friend” recommendation system
This paper continues our recent work [2], where we presented the architectural design, the functional requirements and the user interface of eMatch [3], an Android application which was inspired by the idea of finding people with common interests in the same geographical area

Summary

Introduction

The diversity of social networks makes the problem of correctly estimating user preferences essential for personalized applications [1]. To compare the results we ran the K-means clustering algorithm, each time with a different similarity computation measure (EgoSimilar and five other measures that are presented later ).

Results

Conclusion