Efficiently Storing and Discovering Knowledge in Databases via Inductive Logic Programming Implemented Directly in Databases

Abstract

Inductive Logic Programming (ILP) uses inductive, statistical techniques to generate hypotheses which incorporate the given background knowledge to induce concepts that cover most of the positive examples and few of the negative examples. ILP uses techniques from both logic programming and machine learning. Research has been evolving from several years in this field and many systems are developed to solve ILP problems and most of these systems are developed in Prolog and take the input in the form of text files or other similar formats. This thesis proposes to use a relational database to store background knowledge, positive and negative examples in the form of database entities. This information is then manipulated directly uses ILP techniques efficiently in the process of generating hypotheses. The database does the heavy lifting by efficiently handling and storing a very large number of intermediate rules which are generated in the process of finding the required hypotheses. The proposed system will be helpful to generate hypotheses from relational databases. The system also provides a mechanism to store the given data into a database which exists in text files. Sequential covering algorithm is used to find the hypotheses which cover all positive examples and few or none of the negative examples. The proposed system is tested on real world datasets, Mutagenesis and Chess Endgame, and the generated hypotheses and its accuracy are similar to the results of existing systems which were tested on the same datasets. The results are promising and this encourages researchers to use the system in future to discover the knowledge for other datasets or in relational databases.