BINGO2

This project concerns the design of new methods and tools for supporting knowledge discovery from databases. We strongly believe that basic research in data mining has to consider real-life scenarios and we decided to address a few concrete KDD processes for molecular biology, e.g., gene synexpression group discovery or transcription binding site discovery. The partners of the Bingo2 consortium have a solid background in machine learning and data mining. Furthermore, they already cooperate on the development of the promising inductive database framework within the almost completed ACI Masse de Données project Bingo. The idea is that KDD processes might be considered as querying processes, i.e., sequences of queries, that exploit either the data or the patterns and models which hold in the data. Queries which have to return patterns or models are called inductive queries. Developing the IDB framework is challenging: for a so-called pattern domain, we have to identify which primitives are to be used for building queries and how such query can be evaluated by means of solvers (i.e., constraintbased mining algorithms). A key observation about the state-of-the-art is that most of the research on constraint-based mining does not really address the question of the origins of constraints. Once an analyst knows he/she can specify constraints over patterns or models, is it possible to support his/her querying process by guiding the de nition of relevant constraints? In Bingo2 , not only we will continue to design new data mining or machine learning tools but also we will consider how (partial) domain knowledge (possibly discovered by means of inductive querying) can be exploited to support knowledge discovery and avoid the pattern ooding follows data ooding scheme that is so typical in exploratory KDD processes. Technically, new methods and tools will be developed for grammatical inference and sequential pattern mining, inductive logic programming, pattern discovery in textual and 0/1 data.