Distributed user profiling via spectral methods
User profiling is a useful primitive for constructing personalised services, such as content recommendation. In the present paper we investigate the feasibility of user profiling in a distributed setting, with no central authority and only local information exchanges between users. We compute a prof...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Institute for Operations Research and the Management Sciences (INFORMS)
2014-09-01
|
| Series: | Stochastic Systems |
| Subjects: | |
| Online Access: | http://www.i-journals.org/ssy/viewarticle.php?id=36&layout=abstract |
| _version_ | 1811239695445655552 |
|---|---|
| author | Dan-Cristian Tomozei Laurent Massoulié |
| author_facet | Dan-Cristian Tomozei Laurent Massoulié |
| author_sort | Dan-Cristian Tomozei |
| collection | DOAJ |
| description | User profiling is a useful primitive for constructing personalised services, such as content recommendation. In the present paper we investigate the feasibility of user profiling in a distributed setting, with no central authority and only local information exchanges between users. We compute a profile vector for each user (i.e., a low-dimensional vector that characterises her taste) via spectral transformation of observed user-produced ratings for items. Our two main contributions follow:<br/>
(i) We consider a low-rank probabilistic model of user taste. More specifically, we consider that users and items are partitioned in a constant number of classes, such that users and items within the same class are statistically identical. We prove that without prior knowledge of the compositions of the classes, based solely on few random observed ratings (namely O(N log N) such ratings for N users), we can predict user preference with high probability for unrated items by running a local vote among users with similar profile vectors. In addition, we provide empirical evaluations characterising the way in which spectral profiling performance depends on the dimension of the profile space. Such evaluations are performed on a data set of real user ratings provided by Netflix.<br/>
(ii) We develop distributed algorithms which provably achieve an embedding of users into a low-dimensional space, based on spectral transformation. These involve simple message passing among users, and provably converge to the desired embedding. Our method essentially relies on a novel combination of gossiping and the algorithm proposed by Oja and Karhunen. |
| first_indexed | 2024-04-12T13:05:37Z |
| format | Article |
| id | doaj.art-2ac3c4c511d9401b9bd7fd00056eddcf |
| institution | Directory Open Access Journal |
| issn | 1946-5238 1946-5238 |
| language | English |
| last_indexed | 2024-04-12T13:05:37Z |
| publishDate | 2014-09-01 |
| publisher | Institute for Operations Research and the Management Sciences (INFORMS) |
| record_format | Article |
| series | Stochastic Systems |
| spelling | doaj.art-2ac3c4c511d9401b9bd7fd00056eddcf2022-12-22T03:32:03ZengInstitute for Operations Research and the Management Sciences (INFORMS)Stochastic Systems1946-52381946-52382014-09-014114310.1214/11-SSY036Distributed user profiling via spectral methodsDan-Cristian Tomozei0Laurent Massoulié1EPFLINRIAUser profiling is a useful primitive for constructing personalised services, such as content recommendation. In the present paper we investigate the feasibility of user profiling in a distributed setting, with no central authority and only local information exchanges between users. We compute a profile vector for each user (i.e., a low-dimensional vector that characterises her taste) via spectral transformation of observed user-produced ratings for items. Our two main contributions follow:<br/> (i) We consider a low-rank probabilistic model of user taste. More specifically, we consider that users and items are partitioned in a constant number of classes, such that users and items within the same class are statistically identical. We prove that without prior knowledge of the compositions of the classes, based solely on few random observed ratings (namely O(N log N) such ratings for N users), we can predict user preference with high probability for unrated items by running a local vote among users with similar profile vectors. In addition, we provide empirical evaluations characterising the way in which spectral profiling performance depends on the dimension of the profile space. Such evaluations are performed on a data set of real user ratings provided by Netflix.<br/> (ii) We develop distributed algorithms which provably achieve an embedding of users into a low-dimensional space, based on spectral transformation. These involve simple message passing among users, and provably converge to the desired embedding. Our method essentially relies on a novel combination of gossiping and the algorithm proposed by Oja and Karhunen.http://www.i-journals.org/ssy/viewarticle.php?id=36&layout=abstractrandom matrixmessage passingdis- tributed spectral embeddingdistributed recommendation systemdistributed spectral embeddingSpectral decompositiondistributed recommendation system |
| spellingShingle | Dan-Cristian Tomozei Laurent Massoulié Distributed user profiling via spectral methods Stochastic Systems random matrix message passing dis- tributed spectral embedding distributed recommendation system distributed spectral embedding Spectral decomposition distributed recommendation system |
| title | Distributed user profiling via spectral methods |
| title_full | Distributed user profiling via spectral methods |
| title_fullStr | Distributed user profiling via spectral methods |
| title_full_unstemmed | Distributed user profiling via spectral methods |
| title_short | Distributed user profiling via spectral methods |
| title_sort | distributed user profiling via spectral methods |
| topic | random matrix message passing dis- tributed spectral embedding distributed recommendation system distributed spectral embedding Spectral decomposition distributed recommendation system |
| url | http://www.i-journals.org/ssy/viewarticle.php?id=36&layout=abstract |
| work_keys_str_mv | AT dancristiantomozei distributeduserprofilingviaspectralmethods AT laurentmassoulie distributeduserprofilingviaspectralmethods |