Case-based closing

from Wikipedia, the free encyclopedia

The case-based Close (Engl. Case-based reasoning, briefly CBR , French. Raisonnement par cas , span. Razonamiento basado en casos ) is a machine learning method to solve problems by analogy . The central element in a CBR system is a so-called case base (case database, case memory ) in which problems that have already been solved are stored as a case. Such a case consists at least of a problem description and an associated problem solution. The aim is to use the solution of a similar and previously solved problem to solve a given problem. This mimics a human behavior: When faced with a new problem, people often remember a comparable situation that they experienced in the past and try to master the current task in a similar way.

Occasionally one speaks of memory-based reasoning .

Some of the earliest realizations were by Roger Schank and his students in the early 1980s (his Dynamic Memory model), such as Janet Kolodner in CYRUS and Michael Lebowitz in IPP. Another pioneer was David Waltz in the 1980s (Memory Based Reasoning) on ​​the massively parallel computers of Danny Hillis' Thinking Machine Corporation . In 1995 there was a first international conference on CBR.

method

Illustration of the CBR cycle

Probably the best-known model goes back to the scientists Agnar Aamodt and Enric Plaza , who described the basic principle of case-based reasoning as a process with four phases, the so-called CBR cycle (source: see below).

  1. Retrieve . Starting from a given description of the problem, it is important to identify a problem that is as similar as possible in the case base. The challenge in this phase is to determine the similarity of the problem descriptions.
  2. Trap . The solution of the case that is most similar to the one given is adopted as a first proposed solution. This gives you a starting point for solving the new problem.
  3. Revise . The current problem cannot always be solved in the same way as the previous one. In the revise phase, the initial solution obtained beforehand is checked and, if necessary, adjusted to the specific conditions.
  4. Retain . The revised case is finally saved in the case base and is therefore available for future inquiries. In this way, the system learns with every further problem that is solved and thus improves its performance.

application

Case-Based Reasoning has proven itself particularly in application systems for customer service , so-called help desk systems. B. uses for diagnosis and therapy of customer problems. More recently it has been used increasingly in (product) advisory systems, for example in e-commerce , and for the classification of texts.

It is considered advantageous that CBR can also be used for poorly structured and incompletely described problems. In contrast to neighboring concepts (see below), a comparatively small collection of reference cases is sufficient initially, which gradually grows as a result of working with the CBR system. CBR is also suitable in application domains whose precise interactions are not fully known.

As always, when arguing with analogies, care must be taken that the proposed solutions generated by the system are adequate for the problem at hand, i.e. whether, for example, the prerequisites on which the historical solution was based are still fulfilled, etc. (obsolescence of knowledge ).

classification

Case-Based Reasoning is a sub-area of Artificial Intelligence and can be counted among the machine learning processes. The learning process is based on analogy , in contrast to learning by induction and deduction . Due to the numerous possible applications in companies (see above), CBR is not only used in (core) computer science , but also in business informatics .

literature

  • Janet Kolodner: Case-Based Reasoning . Morgan Kaufmann Series in Representation & Reasoning. Morgan Kaufmann Publishers In, 1993, ISBN 978-1-55860-237-3 .
  • Agnar Aamodt, Enric Plaza: Case-based reasoning; Foundational issues, methodological variations, and system approaches . In: AI COMMUNICATIONS . 7, 1994, pp. 39-59. , CiteSeerX: 10.1.1.15.9093
  • Michael M. Richter: Case-based reasoning. In: Görz, Günther; Rollinger, Claus-Rainer; Schneeberger, Josef (Hrsg.): Handbook of artificial intelligence. 4th edition, Munich / Vienna 2003, pp. 407-430. ISBN 3486272128
  • Ralph Bergmann: Experience Management: Foundations, Development Methodology, and Internet-Based Applications . Lecture Notes in Artificial Intelligence 2432. Springer, Berlin 2002, ISBN 978-3-540-44191-5 .
  • Ralph Bergmann, Althoff, KD, Breen, S., Göker, M., Manago, M., Traphöner, R. & Wess, S .: Developing industrial case-based reasoning applications: The INRECA methodology. 2nd revised edition. Lecture Notes in Artificial Intelligence 1612, Springer Verlag, Berlin 2003
  • E. Hüllermeier: Case-Based Approximate Reasoning. Springer Verlag, Berlin, 2007.

Web links

Commons : case-based inference  - collection of images, videos and audio files