Paper Abstract and Keywords |
Presentation |
2012-03-12 10:50
Detecting Latent Structural Changes via Latent Dirichlet Allocation Masashi Ueda, Ryota Tomioka, Kenji Yamanishi (The Univ. of Tokyo), Katsuhiko Ishiguro, Hiroshi Sawada, Naonori Ueda (NTT) IBISML2011-89 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Detecting changes in consumers' latent preference is a fundamental challenge for improving recommendation systems as well as market analysis. In this paper, we propose a method for detecting changes in consumers' preference from purchase logs in an online setting. The proposed method employs a Latent Dirichlet Allocation (LDA) model that shares common hyperparameters before and after an assumed change point. The change in latent preference is measured as the distance between the two LDA distributions corresponding to before and after the assumed change point. Furthermore, a dynamic threshold optimization technique is used to raise alarms at change points based on the above distance. We create data sets with artificial change points by concatenating purchase logs from different consumers. We demonstrate on these data sets that the proposed method performs significantly better than a simple multinomial model that does not have latent variables. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Latent Dirichlet Allocation / Change point detection / user preference model / purchase logs / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 111, no. 480, IBISML2011-89, pp. 15-20, March 2012. |
Paper # |
IBISML2011-89 |
Date of Issue |
2012-03-05 (IBISML) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
IBISML2011-89 |
|