## A maximum entropy approach to collaborative filtering in dynamic, sparse, high-dimensional domains (2002)

### Cached

### Download Links

- [dpennock.com]
- [labs.yahoo.com]
- [books.nips.cc]
- DBLP

### Other Repositories/Bibliography

Citations: | 31 - 6 self |

### BibTeX

@TECHREPORT{Pavlov02amaximum,

author = {Dmitry Y. Pavlov and David M. Pennock},

title = {A maximum entropy approach to collaborative filtering in dynamic, sparse, high-dimensional domains},

institution = {},

year = {2002}

}

### Years of Citing Articles

### OpenURL

### Abstract

We develop a maximum entropy (maxent) approach to generating recommendations in the context of a user’s current navigation stream, suitable for environments where data is sparse, highdimensional, and dynamic—conditions typical of many recommendation applications. We address sparsity and dimensionality reduction by first clustering items based on user access patterns so as to attempt to minimize the apriori probability that recommendations will cross cluster boundaries and then recommending only within clusters. We address the inherent dynamic nature of the problem by explicitly modeling the data as a time series; we show how this representational expressivity fits naturally into a maxent framework. We conduct experiments on data from ResearchIndex, a popular online repository of over 470,000 computer science documents. We show that our maxent formulation outperforms several competing algorithms in offline tests simulating the recommendation of documents to ResearchIndex users. 1

### Citations

1115 | Grouplens: An Open Architecture for Collaborative Filtering of Netnews
- Resnick, Iacovou, et al.
- 1994
(Show Context)
Citation Context ...owing documents written by the same author(s), or textually similar documents to D). These methods have been shown to be good predictors [3]. Another possibility is to perform collaborative filtering =-=[13]-=- by assessing the similarities between the documents requested by the current user and the users who interacted with ResearchIndex in the past. Once the users with browsing histories similar to that o... |

1019 | C.M.: Empirical analysis of predictive algorithms for collaborative filtering
- Breese, Heckerman, et al.
- 1998
(Show Context)
Citation Context ...ll be similar as well, and the prediction is made accordingly. Common measures of similarity between users include Pearson correlation coefficient [13], mean squared error [16], and vector similarity =-=[1]-=-. More recent work includes application of statistical machine learning techniques, such as Bayesian networks [1], dependency networks [6], singular value decomposition [14] and latent class models [7... |

881 | Social information filtering: algorithms for automating 'Word of Mouth
- Shardanand, Maes
- 1995
(Show Context)
Citation Context ... future browsing patterns will be similar as well, and the prediction is made accordingly. Common measures of similarity between users include Pearson correlation coefficient [13], mean squared error =-=[16]-=-, and vector similarity [1]. More recent work includes application of statistical machine learning techniques, such as Bayesian networks [1], dependency networks [6], singular value decomposition [14]... |

553 | Inducing features of random fields
- Pietra, Pietra, et al.
- 1997
(Show Context)
Citation Context ...HS is the actual frequency (up to the same normalization factor) of this feature in the training data. There exist efficient algorithms for finding the parameters {λ} (e.g. improved iterative scaling =-=[11]-=-) that are known to converge if the constraints imposed on P are consistent. Under fairly general assumptions, the maxent model can also be shown to be a maximum likelihood model [11]. Employing a Gau... |

354 | Analysis of Recommendation Algorithms for ECommerce
- Sarwar, Karypis, et al.
- 2000
(Show Context)
Citation Context ...dingly sparse and thus challenging to model. In this paper, we work with the ResearchIndex data, since it is an interesting application domain, and is typical of many recommendation application areas =-=[14]-=-. There are two conceptually different ways of making recommendations. A content filtering approach is to recommend solely based on the features of a document D (e.g., showing documents written by the... |

268 | Digital libraries and autonomous citation indexing
- Lawrence, Giles, et al.
- 1998
(Show Context)
Citation Context ...utomatically locates computer science papers found on the Web, indexes their full text, allows browsing via the literature citation graph, and isolates the text around citations, among other services =-=[8]-=-. The archive contains over 470,000 documents including the full text of each document, citation links between documents, and a wealth of user access data. With so many documents, and only seven acces... |

229 | A Gaussian prior for smoothing maximum entropy models
- Chen, Rosenfeld
- 1999
(Show Context)
Citation Context ...a zero mean on parameters λ yields a maximum aposteriori solution that has been shown to be more accurate than the related maximum likelihood solution and other smoothing techniques for maxent models =-=[2]-=-. We use Gaussian smoothing in our experiments with a maxent model. HTable 2: Average number of hits ¯ h and height ¯ H of predictions across the clusters for different ranges of heights and using va... |

158 | Latent class models for collaborative filtering
- Hofmann, Puzicha
- 1999
(Show Context)
Citation Context ...1]. More recent work includes application of statistical machine learning techniques, such as Bayesian networks [1], dependency networks [6], singular value decomposition [14] and latent class models =-=[7, 12]-=-. Most of these recommendation algorithms are context and order independent: that is, the rank of recommendations does not depend on the context of the user’s current navigation or on recency effects ... |

132 | Probabilistic Models for Unified Collaborative and Content-Based Recommendation in Sparse-Data Environments
- Popescul, Ungar, et al.
- 2001
(Show Context)
Citation Context ...1]. More recent work includes application of statistical machine learning techniques, such as Bayesian networks [1], dependency networks [6], singular value decomposition [14] and latent class models =-=[7, 12]-=-. Most of these recommendation algorithms are context and order independent: that is, the rank of recommendations does not depend on the context of the user’s current navigation or on recency effects ... |

47 | An MDP-based recommender system
- Shani, Brafman, et al.
- 2002
(Show Context)
Citation Context ... knowledge, this is the first application of maxent for collaborative filtering, and one of the few published formulations that makes accurate recommendations in the context of a dynamic user session =-=[3, 15]-=-. We perform offline empirical tests of our recommender and compare it to competing models. The comparison shows our method is quite accurate, outperforming several other less-expressive models.The r... |

35 | REFEREE: An open framework for practical testing of recommender systems using researchindex
- Cosley, Lawrence, et al.
- 2002
(Show Context)
Citation Context ...ecommend solely based on the features of a document D (e.g., showing documents written by the same author(s), or textually similar documents to D). These methods have been shown to be good predictors =-=[3]-=-. Another possibility is to perform collaborative filtering [13] by assessing the similarities between the documents requested by the current user and the users who interacted with ResearchIndex in th... |

29 | Classes for fast maximum entropy training
- Goodman
- 2001
(Show Context)
Citation Context ...hnique with the end goal of recommendation, our approach appears to do a good job at maintaining high recall (sensitivity). Similar ideas in the context of maxent were proposed recently by Goodman in =-=[5]-=-. We explicitly model time: each user is associated with a set of sessions, and each session is modeled as a time sequence of document accesses. We present a maxent model that effectively estimates th... |

29 |
Dependency networks for density estimation, collaborative filtering, and data visualization
- Heckerman, Chickering, et al.
(Show Context)
Citation Context ...fficient [13], mean squared error [16], and vector similarity [1]. More recent work includes application of statistical machine learning techniques, such as Bayesian networks [1], dependency networks =-=[6]-=-, singular value decomposition [14] and latent class models [7, 12]. Most of these recommendation algorithms are context and order independent: that is, the rank of recommendations does not depend on ... |

23 | Inferring hierarchical descriptions
- Glover, Pennock, et al.
(Show Context)
Citation Context ...o a series of prediction problems for each cluster. We studied the clusters by trying to find out if the documents within a cluster are topically related. We ran code previously developed at NEC Labs =-=[4]-=- that uses information gain to find the top features that distinguish each cluster from the rest. Table 1 shows the top features for some of the created clusters. The top features are quite consistent... |

14 | Mixtures of conditional maximum entropy models
- Pavlov, Popescul, et al.
- 2003
(Show Context)
Citation Context ...ral important directions to extend the work described in this paper. First, we plan to perform “live” testing of the clustering approach and various models in ResearchIndex. Secondly, our recent work =-=[10]-=- suggests that for difficult prediction problems improvement beyond the plain maxent models can be soughtby employing the mixtures of maxent models. We also plan to look at different clustering metho... |