Recommender systems aim to capture interests of users to provide tailored recommendations. User interests are however often unique and depend on many unobservable factors including a user's mood and the local weather. We take a contextual session-based approach and propose a sequential framework using factored Markov decision processes (fMDPs) to detect the user's goal (the topic) of a session. We show that an independence assumption on the attributes of items leads to a set of independent models that can be optimised efficiently. Our approach results in interpretable topics that can be effectively turned into recommendations. Empirical results on a real world click log from a large e-commerce company exhibit highly accurate topic prediction rates of about 90%. Translating our approach into a topic-driven recommender system outperforms several baseline competitors.
Description
Factored MDPs for detecting topics of user sessions
%0 Conference Paper
%1 tavakol2014factored
%A Tavakol, Maryam
%A Brefeld, Ulf
%B Proceedings of the 8th ACM Conference on Recommender Systems
%C New York, NY, USA
%D 2014
%I ACM
%K dataset:Zalando detection factored mdp session topic user
%P 33--40
%R 10.1145/2645710.2645739
%T Factored MDPs for Detecting Topics of User Sessions
%U http://doi.acm.org/10.1145/2645710.2645739
%X Recommender systems aim to capture interests of users to provide tailored recommendations. User interests are however often unique and depend on many unobservable factors including a user's mood and the local weather. We take a contextual session-based approach and propose a sequential framework using factored Markov decision processes (fMDPs) to detect the user's goal (the topic) of a session. We show that an independence assumption on the attributes of items leads to a set of independent models that can be optimised efficiently. Our approach results in interpretable topics that can be effectively turned into recommendations. Empirical results on a real world click log from a large e-commerce company exhibit highly accurate topic prediction rates of about 90%. Translating our approach into a topic-driven recommender system outperforms several baseline competitors.
%@ 978-1-4503-2668-1
@inproceedings{tavakol2014factored,
abstract = {Recommender systems aim to capture interests of users to provide tailored recommendations. User interests are however often unique and depend on many unobservable factors including a user's mood and the local weather. We take a contextual session-based approach and propose a sequential framework using factored Markov decision processes (fMDPs) to detect the user's goal (the topic) of a session. We show that an independence assumption on the attributes of items leads to a set of independent models that can be optimised efficiently. Our approach results in interpretable topics that can be effectively turned into recommendations. Empirical results on a real world click log from a large e-commerce company exhibit highly accurate topic prediction rates of about 90%. Translating our approach into a topic-driven recommender system outperforms several baseline competitors.},
acmid = {2645739},
added-at = {2014-11-16T21:54:17.000+0100},
address = {New York, NY, USA},
author = {Tavakol, Maryam and Brefeld, Ulf},
biburl = {https://www.bibsonomy.org/bibtex/2d4e510f7cc1c93b1fbc6f6ad52178c08/nosebrain},
booktitle = {Proceedings of the 8th ACM Conference on Recommender Systems},
description = {Factored MDPs for detecting topics of user sessions},
doi = {10.1145/2645710.2645739},
interhash = {4349cb8328e893f388785c407e70dc97},
intrahash = {d4e510f7cc1c93b1fbc6f6ad52178c08},
isbn = {978-1-4503-2668-1},
keywords = {dataset:Zalando detection factored mdp session topic user},
location = {Foster City, Silicon Valley, California, USA},
numpages = {8},
pages = {33--40},
publisher = {ACM},
series = {RecSys '14},
timestamp = {2014-11-28T15:29:52.000+0100},
title = {Factored MDPs for Detecting Topics of User Sessions},
url = {http://doi.acm.org/10.1145/2645710.2645739},
year = 2014
}