Challenges in recommender systems : scalability, privacy, and structured recommendations
Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2015.
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | eng |
| Published: |
Massachusetts Institute of Technology
2015
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/1721.1/99785 |
| _version_ | 1826198697740861440 |
|---|---|
| author | Xin, Yu, Ph. D. Massachusetts Institute of Technology |
| author2 | Tommi Jaakkola. |
| author_facet | Tommi Jaakkola. Xin, Yu, Ph. D. Massachusetts Institute of Technology |
| author_sort | Xin, Yu, Ph. D. Massachusetts Institute of Technology |
| collection | MIT |
| description | Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2015. |
| first_indexed | 2024-09-23T11:08:43Z |
| format | Thesis |
| id | mit-1721.1/99785 |
| institution | Massachusetts Institute of Technology |
| language | eng |
| last_indexed | 2024-09-23T11:08:43Z |
| publishDate | 2015 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/997852019-04-11T12:34:37Z Challenges in recommender systems : scalability, privacy, and structured recommendations Xin, Yu, Ph. D. Massachusetts Institute of Technology Tommi Jaakkola. Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science. Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science. Electrical Engineering and Computer Science. Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2015. This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. Cataloged from student-submitted PDF version of thesis. Includes bibliographical references (pages 121-128). In this thesis, we tackle three challenges in recommender systems (RS): scalability, privacy and structured recommendations. We first develop a scalable primal dual algorithm for matrix completion based on trace norm regularization. The regularization problem is solved via a constraint generation method that explicitly maintains a sparse dual and the corresponding low rank primal solution. We provide a new dual block coordinate descent algorithm for solving the dual problem with a few spectral constraints. Empirical results illustrate the effectiveness of our method in comparison to recently proposed alternatives. In addition, we extend the method to non-negative matrix factorization (NMF) and dictionary learning for sparse coding. Privacy is another important issue in RS. Indeed, there is an inherent trade-off between accuracy of recommendations and the extent to which users are willing to release information about their preferences. We explore a two-tiered notion of privacy where there is a small set of public users who are willing to share their preferences openly, and a large set of private users who require privacy guarantees. We show theoretically, and demonstrate empirically, that a moderate number of public users with no access to private user information already suffices for reasonable accuracy. Moreover, we introduce a new privacy concept for gleaning relational information from private users while maintaining a first order deniability. We demonstrate gains from controlled access to private user preferences. We further extend matrix completion to high-order tensors. We illustrate the problem of recommending a set of items to users as a tensor completion problem. We develop methods for directly controlling tensor factorizations in terms of the degree of nonlinearity (the number of non-uniform modes in rank-1 components) as well as the overall number of rank-1 components. Finally, we develop a tensor factorization for dependency parsing. Instead of manually selecting features, we use tensors to map high-dimensional sparse features into low dimensional (dense) features. Our parser achieves state of the art results across multiple languages. by Yu Xin. Ph. D. 2015-11-09T19:13:01Z 2015-11-09T19:13:01Z 2015 2015 Thesis http://hdl.handle.net/1721.1/99785 927438195 eng M.I.T. theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. See provided URL for inquiries about permission. http://dspace.mit.edu/handle/1721.1/7582 128 pages application/pdf Massachusetts Institute of Technology |
| spellingShingle | Electrical Engineering and Computer Science. Xin, Yu, Ph. D. Massachusetts Institute of Technology Challenges in recommender systems : scalability, privacy, and structured recommendations |
| title | Challenges in recommender systems : scalability, privacy, and structured recommendations |
| title_full | Challenges in recommender systems : scalability, privacy, and structured recommendations |
| title_fullStr | Challenges in recommender systems : scalability, privacy, and structured recommendations |
| title_full_unstemmed | Challenges in recommender systems : scalability, privacy, and structured recommendations |
| title_short | Challenges in recommender systems : scalability, privacy, and structured recommendations |
| title_sort | challenges in recommender systems scalability privacy and structured recommendations |
| topic | Electrical Engineering and Computer Science. |
| url | http://hdl.handle.net/1721.1/99785 |
| work_keys_str_mv | AT xinyuphdmassachusettsinstituteoftechnology challengesinrecommendersystemsscalabilityprivacyandstructuredrecommendations |