Case-based Reasoning Applied to Information Retrieval

ABSTRACT The main idea under Case-Based Reasoning (CBR) is to store experience, problem-solving processes, as cases, in case libraries; when a new problem is encountered, the system uses memory of relevant past cases to interpret or solve the problem. Adaptation of cases plays an important role in this process. Thus, the more cases the system stores the better it performs. CBR allows exceptional cases to be easily stored and used in problem solving, makes learning from cases possible, and provides more convincing explanations, based upon the stored cases. Moreover, cases can be used to establish and analyse general rules. Due to this characteristics, CBR is specially useful in areas that (a) are particularly difficult to formalise (b) planning and explanations are required (c) persuasion is important, or (d) hypothetical scenarios are needed. Many approaches have been developed to deal with problems on information retrieval (IR). In traditional methods, the process of records of files from database, to find similarities between the records and queries, is very common, basically using boolean models. Some other methods are based on probabilistic or statistical analysis of occurrences of words, some use phrases and structured queries, others work with vector space models, semantic models and so on. Knowledge-Based approaches are not uncommon: Expert intermediary systems for formulating and evaluating queries, analysis of queries using natural-language-processing techniques, knowledge representations used for searching, etc. However, there is little experimental evidence to demonstrate the effectiveness of knowledge-based techniques, thus it remains a significant challenge for research.[1] Learning and adaptation have long been viewed as crucial part of an IR system[2]; CBR might prove to be a valuable approach to this area. In this paper, a CBR model is suggested as an alternative approach to to IR. In terms of knowledge representation, the model is integrated by three main components: a case base, a users' profile and an auxiliary case- specific context base. Cases are obtained from descriptions of problems and stored in particular structures, developed for this application. Auxiliary context is obtained from the same problem description. CBR methodologies are to be used to process the features in the description, to retrieve from the case library, to adapt analogous cases and to learn by adding and modifying cases, assisted by the case-context base and by the user profile.