## Collaborate With Strangers To Find Own Preferences (2005)

### Cached

### Download Links

Venue: | In Proc. 17th ACM Symp. on Parallelism in Algorithms and Architectures |

Citations: | 13 - 6 self |

### BibTeX

@INPROCEEDINGS{Awerbuch05collaboratewith,

author = {Baruch Awerbuch and Yossi Azar and Zvi Lotker and Boaz Patt-shamir and Mark R. Tuttle},

title = {Collaborate With Strangers To Find Own Preferences},

booktitle = {In Proc. 17th ACM Symp. on Parallelism in Algorithms and Architectures},

year = {2005},

pages = {263--269}

}

### OpenURL

### Abstract

We consider a model with n players and m objects. Each player has a “preference vector ” of length m, that models his grades for all objects. The grades are initially unknown to the players. A player can learn his grade for an object by probing that object, but performing a probe incurs cost. The goal of a player is to learn his preference vector with minimal cost, by adopting the results of probes performed by other players. To facilitate communication, we assume that players collaborate by posting their grades for objects on a shared billboard: reading from the billboard is free. We consider players whose preference vectors are popular, i.e., players whose preferences are common to many other players. We present a sequential and a parallel algorithm to solve the problem with logarithmic cost overhead.

### Citations

1872 | Randomized Algorithms
- Motwani, Raghavan
- 1995
(Show Context)
Citation Context ...ly the value v received more than θi of the qi qualified votes. In this case it is straightforward to bound the probability that v j i �= v using the Chernoff bound as follows (we use the versions of =-=[9]-=-). The voters are chosen at random (Step 4a), and therefore, by assumption on the popularity of the type, the probability that a random vote is v j i is at least α. Hence � P � v j i �= v � ≤ P v j i ... |

1116 | Grouplens: an open architecture for collaborative filtering of netnews
- Resnick, Iacovou, et al.
- 1994
(Show Context)
Citation Context ...which objects to probe, and process their feedback to achieve the desired result. Let us first review passive algorithms. Passive algorithm usually work by heuristically identifying clusters of users =-=[9]-=- (or products [10]) in the data set, and using past grades by users in a cluster to predict future grades by other users in the same cluster. Singular Value Decomposition (SVD) was also shown to be an... |

706 | Item-based collaborative filtering recommendation algorithms
- Sarwar, Karypis, et al.
- 2001
(Show Context)
Citation Context ...probe, and process their feedback to achieve the desired result. Let us first review passive algorithms. Passive algorithm usually work by heuristically identifying clusters of users [9] (or products =-=[10]-=-) in the data set, and using past grades by users in a cluster to predict future grades by other users in the same cluster. Singular Value Decomposition (SVD) was also shown to be an effective algebra... |

354 | Analysis of Recommendation Algorithms for ECommerce
- Sarwar, Karypis, et al.
- 2000
(Show Context)
Citation Context ...er to predict future grades by other users in the same cluster. Singular Value Decomposition (SVD) was also shown to be an effective algebraic technique for the off-line single recommendation problem =-=[11]-=-. Some of these systems enjoy industrial success, but they are known to perform poorly when prior data is less than plentiful [12], and they are quite vulnerable even to mild attacks [7, 8]. Theoretic... |

165 | Methods and metrics for cold-start recommendations
- Schein, Popescul, et al.
- 2002
(Show Context)
Citation Context ...ve algebraic technique for the off-line single recommendation problem [11]. Some of these systems enjoy industrial success, but they are known to perform poorly when prior data is less than plentiful =-=[12]-=-, and they are quite vulnerable even to mild attacks [7, 8]. Theoretical studies of recommendation systems usually take the latent variable model approach: a stochastic process is assumed to generate ... |

150 | Spectral analysis of data
- Azar, Fiat, et al.
- 2001
(Show Context)
Citation Context ...n a cluster is governed by an arbitrary probability distribution, and also consider the mixture model, in which each cluster is a probability distribution over all products. Azar et al. introduced in =-=[2]-=- the idea of using SVD to reconstruct the unknown preference vectors. The attractive feature of the SVD method is its ability to deal with some noise. However, SVD is inherently incapable of handling ... |

123 | Collaborative Filtering with Privacy
- Canny
- 2002
(Show Context)
Citation Context ...Some of these systems enjoy industrial success, but they are known to perform poorly when prior data is less than plentiful [14], and they are extremely vulnerable even to mild attacks [8, 10]. Canny =-=[3]-=- gives a distributed secure and private SVD computation for the off-line single recommendation version of the problem. Algorithmic results. Theoretical studies of recommendation systems usually take t... |

108 |
Shilling recommender systems for fun and profit
- Lam, Riedl
- 2004
(Show Context)
Citation Context ...tion problem [11]. Some of these systems enjoy industrial success, but they are known to perform poorly when prior data is less than plentiful [12], and they are quite vulnerable even to mild attacks =-=[7, 8]-=-. Theoretical studies of recommendation systems usually take the latent variable model approach: a stochastic process is assumed to generate noisy observations, and the goal of an algorithm is to appr... |

102 | Chernoff-Hoeffding Bounds for Applications with Limited Independence
- Schmidt, Siegel, et al.
- 1994
(Show Context)
Citation Context ...received more than `i of the qi qualified votes. In this case it is straightforward to bound the probability that vji 6= v using the tail bound for hypergeometric distributions as follows (see, e.g., =-=[13]-=-). The voters are chosen at random (Step 4a), and therefore, by assumption on the popularity of the type, the probability that a random vote is vji is at least ff. Hence the expected number of votes f... |

53 | Recommendation Systems: A Probabilistic Analysis
- Kumar, Raghavan, et al.
- 2001
(Show Context)
Citation Context ...e the latent variable model approach: a stochastic process is assumed to generate noisy observations, and the goal of an algorithm is to approximate some unknown parameters of the model. Kumar et al. =-=[6]-=- study passive algorithms for a model where preferences are identified with past choices (purchases). In this model there are clusters of products. Each user has a probability distribution over cluste... |

50 |
Competitive Recommendation Systems
- Drineas, Kerenidis, et al.
- 2002
(Show Context)
Citation Context ... she.” 2ssolutions to the problem (see Sec. 5.1). However, such algorithms, while possibly suitable for centralized settings, seem unrealistic in the distributed model, for the following reasons (see =-=[3, 1]-=- for more discussion of the issue): • Lack of fairness: the number of objects may be huge, and the tolerance of players for doing a lot of work for others may be in short supply. • Distrust of committ... |

34 | Learning binary relations and total orders
- Goldman, Rivest, et al.
- 1989
(Show Context)
Citation Context ...stand any number of dishonest users (the performance depends only logarithmically on the total number of users). 12sLearning relations. Another related problem is the learning model of Goldman et al. =-=[4]-=-, where the algorithm works for all inputs. The setting in [4] is that a centralized algorithm needs to learn a binary relation: the relation value for all (i, j) pairs must be output. In each step, t... |

33 |
Improved Recommendation Systems
- Awerbuch, Patt-Shamir, et al.
(Show Context)
Citation Context ... she.” 2ssolutions to the problem (see Sec. 5.1). However, such algorithms, while possibly suitable for centralized settings, seem unrealistic in the distributed model, for the following reasons (see =-=[3, 1]-=- for more discussion of the issue): • Lack of fairness: the number of objects may be huge, and the tolerance of players for doing a lot of work for others may be in short supply. • Distrust of committ... |

23 | Convergent algorithms for collaborative filtering
- Kleinberg, Sandler
- 2003
(Show Context)
Citation Context ...ses a cluster by his distribution, and then chooses a product uniformly at random from that cluster. The goal is to recommended a product from the user’s most preferred cluster. Kleinberg and Sandler =-=[5]-=- generalized this model to the case where the choice within a cluster is governed by an arbitrary probability distribution, and also consider the mixture model, in which each cluster is a probability ... |

23 |
Chernoff-Hoeffding bounds for applications with limited independence
- Schmidt, Siegel, et al.
- 1995
(Show Context)
Citation Context ...ceived more than θi of the qi qualified votes. In this case it is straightforward to bound the probability that v j i �= v using the tail bound for hypergeometric distributions as follows (see, e.g., =-=[13]-=-). The voters are chosen at random (Step 4a), and therefore, by assumption on the popularity of the type, the probability that a random vote is v j i is at least α. Hence the expected number of votes ... |

20 | Collaboration of Untrusting Peers with Changing Interests - Awerbuch, Patt-Shamir, et al. - 2004 |

11 |
Utility-based neighbourhood formation for efficient and robust collaborative filtering
- OMahony, Hurley, et al.
- 2004
(Show Context)
Citation Context ...tion problem [11]. Some of these systems enjoy industrial success, but they are known to perform poorly when prior data is less than plentiful [12], and they are quite vulnerable even to mild attacks =-=[7, 8]-=-. Theoretical studies of recommendation systems usually take the latent variable model approach: a stochastic process is assumed to generate noisy observations, and the goal of an algorithm is to appr... |