Collaborative filtering

from Wikipedia, the free encyclopedia

When collaborative filtering ( collaborative filtering ) behavior patterns are groups of users to connect to the interests of individuals. This is a form of data mining that makes explicit user input superfluous.

target

Collaborative filtering is mostly used for very large amounts of data. Collaborative filtering is used in a wide variety of areas such as B. in the financial services sector for the integration of financial sources or in applications in eCommerce and Web 2.0 . This article deals with collaborative filtering for user data, although some methods and approaches can be transferred to other areas.

The aim of the method is an automatic prediction (filtering) of user interests. For this purpose, information about the behavior and preferences of as many users as possible is collected. The underlying assumption of collaborative filtering is that if two people have the same preferences for similar products, they will also agree on other products. Hence the term collaboration : If you want to know what opinion user A has about an article, you have to look at what other users think about this article. Whereby one only considers users whose opinion agrees with the opinion of user A for as many articles as possible. The other users work together to solve the question of what opinion user A is.

Through collaborative filtering, e.g. B. for a television program a prediction can be made which television program a viewer might like. One looks at the viewing habits of the viewer and compares them with the habits of other viewers. The viewers whose habits are most similar to those of the viewer in question are now used to recommend new programs. The output would be a list of possible favorite TV shows. It should be noted that this prediction is made individually for each individual viewer. The data basis for the prediction is collected from the entirety of the users. This is where collaborative filtering differs from simpler methods in which a non-specific mean value is calculated.

A specific problem of collaborative filters is their latency: A new user enters the system with an empty user profile. Since his interests are not yet known, he cannot get any meaningful recommendations at the beginning. The same applies to new elements entering the system (e.g. products in an online shop). They have no quantifiable similarity to other elements and can therefore not be recommended. In other words, collaborative filters are learning systems and thus a form of artificial intelligence.

methodology

Collaborative filtering usually takes place in two steps.

  1. Look for users who have the same behavior pattern as the active user. (= the user for whom the prediction is made)
  2. Use of the behavioral patterns to make a prediction for the active user.

Alternatively, there is article-based collaborative filtering, made famous by Amazon.com (“You might be interested in that, too”) and first introduced by Vucetic and Obradovic in 2000.

  1. Create a similarity matrix to determine relationships between articles.
  2. The preferences of the active user are derived from the matrix.

Other forms of collaborative filtering can be based on implicit observation of user behavior. These forms of filtering compare the behavior of the individual user with the behavior of all other users (what music did they listen to? What products did they buy?). This data is used to predict the future behavior of the user. It does not make sense to offer a certain piece of music to a user if he has made it clear through his behavior that he already owns it. Likewise, it does not make sense to offer a user additional Paris travel guides if they already have a travel guide for this city.

In today's information age, these and similar technologies are proving to be extremely helpful for product selection, especially when certain product groups (e.g. music, films, books, news, websites) have become so large that individual people cannot see the entire range .

application

In commercial systems

Commercial websites that use collaborative filtering:

In non-commercial systems

literature

  • Andreas Meier, Henrik Stormer: eBusiness & eCommerce: Management of the digital value chain. Springer, Berlin 2009, ISBN 978-3-540-85016-8 .
  • Robert Buchberger: When it gets personal ... - Web personalization. ( Memento of February 5, 2009 in the Internet Archive ) on: contentmanager.de , 6/2001, accessed on April 14, 2010
  • David Goldberg, David Nichols, Brain M. Oki, Douglas Terry: Using collaborative filtering to weave an information tapestry. In: Communications of the ACM. 35 (12), 1992, pp. 61-70.
  • Torben Brodt: Collaborative Filtering: for automatic recommendations . VDM Verlag, Saarbrücken 2010, ISBN 978-3-639-25509-6 .

swell

  1. ↑ Abhinandan S. Das, Mayur Datar, Ashutosh Garg, Shyam Rajaram: Google news personalization: scalable online collaborative filtering. In: WWW '07 Proceedings of the 16th international conference on World Wide Web. ACM, New York 2007, ISBN 978-1-59593-654-7 . doi: 10.1145 / 1242572.1242610 online
  2. ↑ Image from Slideshare "BigDataEurope" presentation "How Apache Drives Music Recommendations At Spotify". September 29, 2015, accessed January 11, 2016 .