Avoiding misconstruals in database systems: a default logic approach

A.S. Hemerly, M.A. Casanova, A.L. Furtado
1993 IEEE Transactions on Knowledge and Data Engineering  
guanine \ ' 0 Fig. 2. Portion of DNA molecule and discovered substructure. instance to nodes inside the instance now connect to the new node. Edges internal to the instance are removed. The program is then run a second time, with heavier weight given to substructures which utilize the previously discovered substructure. The increased weight reflects increased attention to this substructure. Fig. 2 shows the results after each pass. Note that on the third pass, SUBDUE linked together the
more » ... s of the substructure in the second pass to find the chains of the double helix. Results indicate that SUBDUE can discover pertinent substructures and find a hierarchical description of the input data by replacing previously discovered substructures on successive passes. V I . CONCLUSIONS Automated knowledge discovery is essential for extracting information from databases [2]. Extiacting knowledge from structural databases requires the identification of repetitive substructures in the data. The previous examples show how SUBDUE'S heuristic search and inexact graph match can discover interesting and repetitive substructures in real structural domains. Applying SUBDUE to scene analysis assists in compression of the image and identification of similar objects in the scene. Application to chemical analysis assists the discovery of previously unknown molecules and cognitive compression of the compound by abstracting over newly discovered molecules. Further experimentation is underway in both artificial domains and other real domains in order to determine the effects of parameters and reduce the computational requirements of SUBDUE'S substructure discovery algorithm. REFERENCES H. Bunke and G. Allerman, "Inexact graph matching for structural pattem recognition," Pattern Recognition Lerr., vol. 1 , no. 4, pp. 245-253, 1983. Abstract-This paper describes a cooperative interface that, using suitable user models, alters the processing of the user's queries to include additional information that will block faulty inferences. In a sense, the interface actively teaches the user facts about the database that he did not explicitly asked for. User interaction with the database then becomes a learning and diseovery process guided by the queries he poses to the interface. The paper also introduces a semantics for user models that captures, with the help of default logic, the nonmonotonic behavior users normally exhibit. Finally, the paper contains results showing that the cooperative interface generates enough additional information to block all faulty inferences. Index Terms-Cooperative user interfaces, database systems, default logic, user modeling.
doi:10.1109/69.250086 fatcat:qtfqadiwjfczpkixo4ltlbtq34