Do the right thing: a component theory for indexing stories as social advice
Eric A. Domeshek · 1992
Within the Artificial Intelligence paradigm and Case-Based Reasoning (CBR), an important problem is specifying how appropriate old cases are to be retrieved so as to assist in coping with new situations. This has been called the indexing problem. This dissertation describes an indexing system supporting retrieval of past cases as advice about everyday social problems; the system has been implemented in the Abby lovelorn advising program. Abby currently contains over 500 indices, each giving access to a story of some social problem. One major result of indexing such a large case-base, was the design and validation of a comprehensive vocabulary for describing social situations. Much of this dissertation can be read as a reference work detailing representational components that have proved useful in this context, and that are likely to serve well in others as well. In addition, this work helps to clarify and systematize a methodology for constructing component theories in support of indexing. Two other points are emphasized throughout the discussion of Abby's indexing system: (1) indices are descriptions of problems and their causes, couched in a vocabulary centered on intentional causality, and (2) indices fit a fixed format that allows reification of identity and thematic relationships as features, and thus, parallel associative retrieval sensitive to important aspects of input situations. Abby answers several of the central questions that any indexing system must address, and offers some advantages over less restrictive systems.