Predictive Sequence Miner in ILP Learning (Extended Abstract)

Abstract

In this work we present an optimized version of XMuSer, an ILP based framework suitable to explore temporal patterns available in multi-relational databases. The main idea behind XMuSer consists of exploiting frequent sequence mining, an efficient and direct method to learn temporal patterns in the form of sequences. The efficiency of XMuSer comes from a new coding methodology and on the use of a predictive sequential miner, which finds discriminative frequent patterns. After finding the discriminative sequences, we map the most interesting ones into a new table that encodes the multi-relational temporal information. The original database is enlarged with a new table that encodes the temporal information in the form of sequences. The last step of our framework consists of applying an ILP algorithm to learn a theory on the enlarged relational database. We evaluate our framework by addressing three classification multi-relational problems.