Date

2017

Author

Tarakçı, Hilal

Metadata

Show full item record

Item Usage Stats

298
views

211
downloads

Cite This

In this thesis, we present a hypergraph based user modeling framework to aggregate partial profiles of the individual and obtain a complete, semantically enriched, multi-domain user model. We also show that the constructed user model can be used to support different personalization services including recommendation. We evaluated the user model against datasets consisting of user's social accounts including Facebook, Twitter, LinkedIn and Stack Overflow. The evaluation results confirmed that the proposed user model improves the quality of the constructed user model in every case. The results also showed that the improvement is higher for generic domain datasets than datasets representing the user in terms of one domain. We propose a recommender system which exploits the proposed framework as case study. The presented system is capable of displaying semantic user model, making domain based, cross domain and general recommendations, discovery of similar users, discovery of users that might be interested in a given item and computation of a user's interest on a given item. We also show that the proposed framework is extendible by extending the framework by adding context information. We also present another user modeling approach based on hypernetworks. The methodology is based on modelling the individual as hypernetwork with a multi-level approach. Initially, lower level terms are represented with hyperedges. Afterwards, higher level terms are modeled by reusing lower level hyperedges. Hypernetwork is clustered to obtain a dynamically tailored user profile. Basically, tailoring a user profile is achieved by filtering the clusters which we want to focus on. Other clusters are eliminated. Q-Analysis technique is used to cluster the hypernetwork. The technique clusters the hypernetwork at level q by listing hyperedges which share q vertices. Eccentricity is a metric which indicates the amount of new and unshared vertices introduced by a hyperedge. We optimize clustering algorithm by using eccentricity of clusters. We define an eccentricity threshold by trial and error. When there exist clusters which have eccentricity at least equal to this threshold, clustering iterations are terminated. The methodology is evaluated against one month long Yandex search logs which contain over 167 million records and slightly improved Yandex's non-personalized ranking which is already a well performing baseline. 

Subject Keywords

Information filtering systems., Expert systems (Computer science)., Recommender systems (Information filtering).

URI

http://etd.lib.metu.edu.tr/upload/12621075/index.pdf
https://hdl.handle.net/11511/26506

Collections

Graduate School of Natural and Applied Sciences, Thesis