Application of Markov chains in an interactive information retrieval system

This paper explores the application of Markov chains in an interactive information retrieval system. Building on the ideas of end-user's state of uncertainty, the system's model of language use, application of the results of information seeking to a work situation, and the user's feedback to the retrieval set through a graphic display of related nodes, the paper details how a Markov chain forms the basis of the IR process and whose transition probabilities are used in an interactive graphic user display to suggest preferred search paths. The transition probabilities between concepts related to the user's query reflect the aggregate behavior of a limited group of information seekers; the transition-matrix of probabilities suggest to the user which concepts are related, and the impact his/her individual feedback will have on the next state. Furthermore, a graphic user display of interactive nodes visualizes the relationships of concepts and edges between nodes suggest the most likely path to follow in order to achieve a successful outcome. A Java application of the system was constructed and tested in a small group of computer help-desk staff in a university budget office. The conclusion suggests that Markov models that integrate the group preferences may make a more useful retrieval model for domain-specific work groups. Introduction The model of most information retrieval systems (IR) is based on matching an information seeker's expression of need (the query) to a document collection's representation of intellectual content and then presenting the retrieval set as a ranked hierarchical list (Baeza-Yates & Ribierto-Neto, 1999, 28). This model has generated useful retrieval systems, but in the end is lacking essentially because it isolates information seeking behavior of the individual from how the resources are evaluated and applied within a social group, and clouds the user's interpretation of potential resources. While there is a rich literature of browsing, searching, and interface design, this paper proposes a Markov model of information retrieval (IR) that uses transition probabilities to guide information seeking behavior and group information seeking history to weight those probabilities. The whole was implemented in a Java application whose user interface reflects the underlying transition matrix. This paper details a Markov chain driven IR model. The motivation for this approach is based on potential weaknesses in traditional IR from the perspective of group awareness, query chains, and interactive visualization. a) Group awareness. Information seeking is not performed in a vacuum: the individual person looking for …