Personalization (information technology)
In information technology, personalization refers to the nominal assignment of features to a user and the adaptation of programs, services or information to the personal preferences, needs and abilities of a user .
A distinction must be made between formal and content personalization. A distinction must also be made between personalizations for individuals and for groups of users. All features of an individual personalization are summarized in a feature vector and these feature vectors are stored with an identification feature.
Formally, for example, the background color or the number of columns in an Internet portal can be adapted to the personal preferences of an individual user. In terms of content, the information displayed can be tailored to the needs and wishes of the user.
From personalization, the user hopes for a more convenient use of a program or offer. Through personalization, companies attempt to offer services or goods that correspond to actual or perceived preferences of users. One use case is recommendation services in many online shops ("Customers who bought x also bought y"). Another example is the personalized delivery of advertising to reduce wastage .
requirements
A prerequisite for personalization that can be used repeatedly is the ability to distinguish or identify individual users, for example by name or ID number and by assigned feature vectors. These feature vectors of users are stored in so-called user profiles and must be able to be clearly assigned to them again at a later point in time. In a computer system , even in a network , the login can be used, for example, to determine the user identity and thus the profile assignment.
The recommendation systems of some personal video recorders , which are used alternately by several household members without being able to differentiate between them, represent an example of the violation of this elementary requirement . If these systems suggest new television content on the basis of an Electronic Program Guide , they will only inadequately differentiate the needs of the individual household members from a stored list of features of the last settings and selections.
Determination of user preferences
The preferences stored in the user profiles can be recorded in two ways, which can be combined with one another:
- through explicit input by the user himself (explicit personalization)
- through the observation of his behavior ( usually unnoticed by the user) (implicit personalization) as a continuous survey, possibly with a ring memory for the stored features.
Personalization techniques
A distinction can be made between three techniques for adapting the content or form of offers to the collected user profiles:
- Rule-based personalization
- Collaborative filtering
- Content-based personalization
Rule-based personalization
Rule-based personalization, a relatively simple technique, adapts content to the user profiles initially available on the basis of a given, relatively rigid set of rules.
For example, in some direct mail campaigns, not only the name of the recipient and his address, but also other recipient-specific information is inserted. Depending on the printing process, only the cover letter or all other printed pieces can be personalized (see digital printing ). With modern solutions for what is known as database publishing , rule-based approaches make it possible to set complex documents (catalogs, reports, etc.) fully automatically, taking into account extensive layout specifications. There are various providers and solutions such as Adobe Inc. or DocScape .
Collaborative filtering
With collaborative filtering (English community based personalization or collaborative filtering ) behavioral patterns of user groups are evaluated in order to infer the interests of individuals. This is a form of data mining that makes explicit user input superfluous. Based on the implicitly observed user behavior,
- a similarity matrix can be formed between the users of an offer. This allows users to be presented with those elements that have been used and / or rated positively by their statistical neighbors .
- a similarity matrix can be formed between the elements of an offer. It defines content that is often used and / or rated positively by the same people as similar.
A specific problem of collaborative filters is their latency: A new user enters the system with an empty user profile and therefore cannot receive 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 therefore cannot be recommended. In other words, collaborative filters are learning systems and thus a form of artificial intelligence .
Content-based personalization
Content-based personalization techniques define the similarity of individual elements on the basis of metadata that describes the content of the respective elements. This requires the elements to be indexed , which can be done either manually (e.g. by tagging ) or automatically (e.g. with the help of search engines ). The personalization system suggests to users those elements that match their preferences in terms of content.
A typical example of a content-based personalization process is the recommendation system from the Internet radio provider Pandora : Each piece of music is described manually using hundreds of meta-information, from which a similarity matrix between the music titles is formed.
See also
- Customized mass production (mass customization)
- Product configurator
- Recommendation engine
literature
- Sarabjot Singh Anand, Bamshad Mobasher (Ed.): Intelligent Techniques for Web Personalization. Berlin 2005, ISBN 3-540-29846-0 .
- Grob, Heinz Lothar, Holling, Heinz, Bensberg, Frank: Personalization of EUS for decision-making processes by experts , work report computer-assisted controlling, Münster 2008.