Techniques to enhance efficiency and effectiveness of inductive logic programming systems: the TWEETY approaches
Chongbing Liu, Enrico Pontelli · Journal of Experimental & Theoretical Artificial Intelligence · 2013
This paper presents four novel approaches to enhance efficiency and effectiveness of Inductive Logic Programming (ILP) systems, along with their implementation in a new ILP system, called TWEETY. The proposed approaches include (1) a new declaration mechanism, called connection declarations, for bottom clause construction, which is simpler but more expressive than the commonly used mode declarations; (2) a new covering technique, called super_covering, which reduces the examples in such a way that recursion can be learned, independently from the ordering of the examples; (3) a new search heuristics, called neg_coverage heuristics, which guides the search using only the number of negative examples covered by each hypothesis and (4) a new search algorithm, called doubly_guided_search, which searches for best clauses by alternating the use of two search heuristics, i.e. the traditional coverage search heuristics and the new neg_coverage search heuristics. The TWEETY system is shown to be more effective and efficient than the state-of-the-art ILP system ALEPH; the proposed techniques can be used to enhance efficiency and effectiveness of ALEPH and other systems based on the same ILP principles.