Contact recommendation in social networks: algorithmic models, diversity and network evolution
Title (trans.)
Recomendación de contactos en redes sociales: modelos algorítmicos, diversidad y evolución de la redAuthor
Sanz-Cruzado Puig, JavierAdvisor
Castells Azpilicueta, PabloEntity
UAM. Departamento de Ingeniería InformáticaDate
2021-05-25Subjects
Recomendación - Redes sociales - Evaluación - Diversidad - Novedad - Tesis doctorales; Bandidos multi-brazo - Análisis - Axiomático - Difusión de información - Enlace débil - Tesis doctorales; InformáticaNote
Tesis doctoral inédita leída en la Universidad Autónoma de Madrid, Escuela Politécnica Superior, Departamento de Ingeniería informática. Fecha de lectura: 25-05-2021Esta obra está bajo una licencia de Creative Commons Reconocimiento-NoComercial-SinObraDerivada 4.0 Internacional.
Abstract
Online social networks like Twitter, Facebook and LinkedIn are used daily by hundreds of millions of people
to connect with other users around the world and share information with then. The massive scale of these
platforms has led to the development of automated tools to prevent users from being overwhelmed by the vast
amount of information they have access to on these sites. Recommender systems are a family of such tools,
designed to make individual suggestions of items or people that users might be interested in according to their
past personal preferences.
This thesis focuses on the study of a particular problem at the con uence of online social networks and
recommender systems: the problem of nding users in the network with whom other people wants to connect
– the problem known as contact recommendation. We explore this problem from three di erent perspectives.
We rst aim to identify the factors leading to the development of e ective contact recommendation approaches,
targeting the density of the network. For this, we explore the relation between contact recommendation
in social networks and text information retrieval. Considering a collaborative ltering perspective, we
explore the utility of adapting search-based models for their use in three di erent aspects of contact recommendation:
as standalone recommendation algorithms, as similarity measures, and as samplers and features
in learning to rank. Thorough experiments over Twitter and Facebook samples show the e ectiveness of the
adapted models in the three roles when compared to the best state of the art approaches.
We also explore the potential of contact recommendation algorithms to drive the evolution of social networks
towards desirable properties of the network as a whole – beyond aggregating the isolated gains of each
user. We investigate the de nition of novel diversity metrics that quantify the e ects of people recommendation
over the structure of the network, with a particular focus on notions of structural diversity and weak ties.
Over samples from Twitter, we nd that recommending weak ties leads to increased novelty and diversity in
the information that reaches the users in the network, with potential implications on the mitigation of lter
bubbles.
Finally, following up on the evolution perspectives, we address the recommendation task as an interactive
cyclic process, where information is constantly exchanged between the users and the system.We develop a simple
stochastic approach, based on the classical user-based k nearest neighbors collaborative ltering algorithm,
that deals with the uncertainty of the available data for selecting the optimal neighbors of the user we want to
generate recommendations for. We explore the utility of this method in dealing with cold start situations over
di erent datasets from di erent recommendation domains – including contact recommendation as a particularly
compelling one.
Files in this item
Google Scholar:Sanz-Cruzado Puig, Javier
This item appears in the following Collection(s)
Related items
Showing items related by title, author, creator and subject.