Qualitative Object Description: Initial Reports of the Exploration of the Frontier

The increased ability to access repositories of representations of complex objects, such as biological molecules or financial time series, has not been matched by the availability of tools that permit locating them, visualizing their characteristics, and describe them in terms that are close to the language of the intended users of those data collections. The representation methods and organization schemes employed in the majority of these repositories are based, rather, on considerations related to computational simplicity and efficiency. The observation that humans typically resort to qualitative descriptions of complex objects to describe interesting features of these entities suggests that the automated generation of those descriptions might substantially improve the accessibility and usefulness of repositories of complex objects. We present progress in a program of research devoted to the automated identification of significant qualitative features of complex objects, the discovery and representation of interesting relations between those features, the generation of structured indexes and textual annotations describing features and their relations, and the discovery of knowledge (or data mining) by analysis of collections of qualitative descriptions. ∗On leave at Departmento de Ciencias de la Computación e Inteligencia Artificial, Universidad de Granada, Spain. Currently visiting the Artificial Intelligence Center, SRI International, Menlo Park, California We focus primarily on methods for the succinct description of interesting features lying in the effective frontier—or the set of all Pareto-optimal solutions— of a multiobjective problem. The multiobjective, generalized clustering, problem being considered is that of extracting features deemed to be interesting from the viewpoint of domain experts. Since, typically, it is easier to derive good explanations of smaller portions of a data object (e.g., a short-lived uptrend) than those of extended portions, any optimization-based formulation of the qualitative feature-identification problem must contend with two conflicting objectives: quality and extent of the description. We present a Pareto genetic algorithm for the solution of the multiobjective optimization problem. The formulation of this problem is noteworthy because of its treatment of clustering as the isolation of individual clusters (i.e., as opposed to the determination of optimal partitions), its lack of assumptions about the number of clusters, and its ability to deal simultaneously with multiple types of interesting structures. We introduce also an approach based on elimination, simplification, and clustering techniques to summarize and organize the Pareto-optimal solutions lying along the effective frontier. We present experimental results of the application of our methodology to the qualitative description of financial time series.