@article{Burke2002,
  added-at = {2009-04-20T12:11:37.000+0200},
  address = {Hingham, MA, USA},
  author = {Burke, Robin},
  biburl = {http://www.bibsonomy.org/bibtex/2460b623792e13b4ec0e990563e57f26c/maida},
  description = {BibSonomy :: bibtex :: Hybrid Recommender Systems: Survey and Experiments},
  doi = {http://dx.doi.org/10.1023/A:1021240730564},
  file = {:D\:\\jorge\\PhD\\Papers\\fulltext.pdf:PDF;Review:D\:\\jorge\\PhD\\Papers\\Resumos\\HybridRecommenderSystems.pdf:PDF},
  interhash = {f40020400b8bc08adca29a987caf25d8},
  intrahash = {460b623792e13b4ec0e990563e57f26c},
  issn = {0924-1868},
  journal = {User Modeling and User-Adapted Interaction},
  keywords = {jabref:noKeywordAssigned},
  number = 4,
  pages = {331--370},
  publisher = {Kluwer Academic Publishers},
  review = {In [Burke, 2002], Robin Burke makes a survey of existing recommender techniques and systems and he also presents and evaluates a hybrid recommender system.   Burke defines a recommender system as "any system that produces individualized recommendations as output or has the effect of guiding the user in a personalized way to interesting or useful objects in a large space of possible options." Put differently, a recommender system assists users in finding relevant information (for the specific user) from a large information space.   There are several classifications possible for recommender systems, taking into account different perspectives. Burke describes a classification based on the "data that supports the recommendation and the algorithms that operate on that data."   Recommender systems can be classified as: collaborative, demographic, content-based, utility-based and knowledge-based.   Collaborative recommender systems use ratings from a community of users to discover commonalities between a given user and other users and recommend items that similar users have rated highly.  Demographic recommender systems categorize the user, usually by requiring the user to explicitly answer a set of questions, and generate recommendations based on demographic classes.  In content-based recommender systems, objects have a set of features or properties that can be used to compare them and find similar objects. These system build a user profile with features the user finds interesting (usually, the profile is built over time, as the user interacts with the system) and is then able to search for objects with similar features to the ones in the user profile.  Utility- and knowledge-based recommender systems require that knowledge about how a particular object satisfies the user needs and can use this knowledge to search the information space for objects relevant for the user in a particular situation. These systems "do not attempt to build long-term generalizations about their users". Utility-based systems can be considered a special case of knowledge-based systems where the a utility function is defined for each user (automatically or manually, by the user himself).   These types of recommender systems suffer from different weaknesses (and have different strengths) so, usually, they are combined in order increase performance. These hybrid recommender systems can be built by combining any two (or more) pure techniques in a number of different ways:  "A weighted hybrid recommender is one in which the score of a recommended item is computed from the results of all of the available recommendation techniques present in the system." As the name implies, each technique may have different weights in the final calculation.  A switching hybrid uses one technique or another, depending on some criterion. A system may try one technique and, if the confidence of the result is not satisfying, it may switch to another technique.  A mixed hybrid approach uses more than one technique at the same time and presents the results from all techniques to the user.  In a feature combination hybrid, collaborative data is treated as a feature and a content-based approach is used on this data.  The "cascade hybrid involves a staged process. In this technique, one recommendation technique is employed first to produce a coarse ranking of candidates and a second technique refines the recommendation from among the candidate set."  Feature augmentation is similar to the cascade one, but in this case the resulting information (ranking or classification) from the first technique is used by the second as an added feature.  In the meta-level approach, two recommendation techniques are combined by using the model produce by one as input for the other.   Burke analyzes all the possible combinations for producing hybrid systems and presents and evaluates a new hybrid: a knowledge-based/collaborative cascade hybrid as an evolution of an existing knowledge-based recommender system for restaurants. The results indicate that this hybrid performs better than the pure recommender system.},
  timestamp = {2009-04-20T12:11:37.000+0200},
  title = {Hybrid Recommender Systems: Survey and Experiments},
  volume = 12,
  year = 2002
}