Pattern-oriented associative rule-based patent classification

This paper proposes an innovative pattern-oriented associative rule-based approach to construct automatic TRIZ-based patent classification system. Derived from associative rule-based text categorization, the new approach does not only discover the semantic relationship among features in a document by their co-occurrence, but also captures the syntactic information by manually generalized patterns. We choose 7 classes which address 20 of the 40 TRIZ Principles and perform experiments upon the binary set for each class. Compared with three currently popular classification algorithms (SVM, C4.5 and NB), the new approach shows some improvement. More importantly, this new approach has its own advantages, which were discussed in this paper as well.