Managing Data and Knowledge in the Industrial Unilever Environment

In data-and knowledge-intensive areas of the industrial Unilever environment, usually many database systems and knowledge-based systems exist and are being built. However, most systems do not interact which results in various problems. As a solution, one could strive for one huge, multipurpose data-knowledge-based system having the advantages of both the database research and the knowledge representation research. However, due to both practical and technical problems, this option is not viable. Instead, the vision is to come to an intelligent software system that is build on top of the available database and knowledge-based systems. This software system should be able to manage all underlying systems and to present itself to users as one overall, intelligent system. 1 CURRENT SITUATION In the Unilever business and research environment, both data and knowledge play a crucial role in decision making in general , and in research, innovation and business processing in particular. Huge amounts of data and knowledge of both empirical and theoretical nature are available. Within Unilever, more and more attention is being paid to an information culture, where data and knowledge in electronic format are shared and reused between various groups and disciplines. In this information culture, empirical data are considered a treasure. This data is usually stored in relational databases (Oracle is the Unilever standard). In addition, Unilever possesses a large amount of empirical knowledge in various disciplines, including processing experiences in the foods and detergents business. This knowledge is used in various tasks, including process assessment, monitoring , diagnosis, and design. For these tasks, a progressively increasing number of knowledge-based systems is being designed and implemented in object-oriented programming environments (G2 of Gensym is one of the Unilever standards). Finally, knowledge about processes that are well understood is usually captured in physical models and implemented in mathematical tools like Matlab. Using these process simu-lators, all kinds of prediction tasks can be performed. At the moment, database systems and knowledge-based systems (including mathematical models in the remainder of this paper) exist next to each other, without interaction. Various problems arise due to this situation, including the following ones: 1. redundant information is contained in these systems, which is diicult to maintain and may lead to inconsistencies ; 2. user frustrations arise if data that is included in one system need to be re-entered in another system; 3. terminology diiers in diierent systems focused on the same eld, leading to confusion; 4. …