Probabilistic Models for Unified Collaborative and Content-Based Recommendation in Sparse-Data Environments

By Alexandrin Popescul, Lyle Ungar, David M. Pennock, Steve Lawrence. Conference on Uncertainty in Artificial Intelligence, 2001

534 citations recorded by the source corpus.

Recommender systems leverage product and community information to target products to consumers. Researchers have developed collaborative recommenders, content-based recommenders, and a few hybrid systems. We propose a unified probabilistic framework for merging collaborative and content-based recommendations. We extend Hofmarm's (1999) aspect model to incorporate three-way co-occurrence data among users, items, and item content. The relative influence of collaboration data versus content data is not imposed as an exogenous parameter, but rather emerges naturally from the given data sources. However, global probabilistic models coupled with standard EM learning algorithms tend to drastically overfit in the sparsedata situations typical of recommendation applications. We show that secondary content information can often be used to overcome sparsity. Experiments on data from the Researchlndex library of Computer Science publications show that appropriate mixture models incorporating secondary data produce significantly better quality recommenders than k-nearest neighbors (k-NN). Global probabilistic models also allow more general inferences than local methods like k-NN.

Subfields: Artificial Intelligence, Computer Networks and Communications, Information Systems

View Paper: https://www.semanticscholar.org/paper/d2914c1cfc5cc68afba9cf4f016725abf742d0c1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Networks and Communications

In Information Systems

Citations in the corpus

Review of "Mining the Web: discovering knowledge from hypertext data" by Soumen Chakrabati. Morgan Kaufman 2003. (2002, unknown): “The dyadic framework has been extended to handle contentbased and collaborative analysis by Popescul and others [174].”
Addressing cold-start problem in recommendation systems (2008, methodology): “…users into classes [3], simultaneously hard-clustering users and items [16], soft-clustering users and items [8, 11], singular value decomposition [13], inferring item-item similarities [14], probabilistic modeling [3, 11, 12, 20], machine learning [2], and list ranking [4, 10] have been employed.”
Content-boosted collaborative filtering for improved recommendations (2002, methodology): “Popescul et al. (2001) extended Hofmann's aspect model to incorporate three-way co-occurrence data among users, items, and item content.”
LCARS: a location-content-aware recommender system (2013, background): “As a result, a certain amount of research has focused on combining the advantages of both collaborative filtering and content-based methods [18, 4, 12].”
Facing the cold start problem in recommender systems (2014, methodology): “Additionally, Hybrid methods have been proposed (Popescul et al., 2001; Schein et al., 2002) in order to cover the disadvantages of CF and CB models.”
Personalized recommendation on dynamic content using predictive bilinear models (2009, unknown): “Hybrid methods are especially useful when data is sparse, for example in cold-start situations [35], but to our best knowledge none of previous work has been integrated with continuous online attributes, such as popularity or freshness.”
Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions (2005, methodology): “Popescul et al. [80] and Schein et al. [94] propose a unified probabilistic method for combining collaborative and content-based recommendations, which is based on the probabilistic latent semantic analysis [46].”
Large-Scale Parallel Collaborative Filtering for the Netflix Prize (2008, methodology): “Efforts to combine both content-based approach and collaborative filtering include the Fab system [3] and unified probabilistic framework [18].”
Collaborative Filtering Recommender Systems (2007, background): “There are many ways to combine them, and no consensus exists among researchers [5, 12, 13, 21, 49].”
Alleviating the Sparsity Problem of Collaborative Filtering Using Trust Inferences (2005, background): “Item-based [12, 13] in addition to Content-boosted CF [13, 14] approaches require additional information regarding items as well as a metric to compute meaningful similarities among them [25].”
Collaborative Filtering Recommender Systems (2011, background): “There is continued work, however, on various methods for estimating and propagating trust through social networks [12, 47, 48, 114, 155].”
Stating Preference for the Ethereal but Choosing the Concrete: How the Tangibility of Attributes Affects Attribute Weighting in Value Elicitation and Choice (2004, unknown): “It is not our goal to declare one type of agent as superior; after all, several authors have been working on combining individual and collaborative approaches into a single agent (e.g., Chung & Rao, 2002; Good et al., 1999; Popescul, Ungar, Pennock, & Lawrence, 2001; Schein et al., 2002).”
A Survey of Collaborative Filtering Techniques (2009, background): “However, hybrid recommenders rely on external information that is usually not available, and they generally have increased complexity of implementation [110, 115, 126].”
A Survey on Context Learning (2017, background): “[73] proposes a generative probabilistic recommender system which combines collaborative and content-based recommendations.”