Learning WordNet-Based Classification Rules

A new classification paradigm, which automatically acquires WordNet-Based rules from a corpus, is presented. The approach is applied to developing an autonomous software agent that can recognize emotions which are expressed in natural language during an interactive human-computer environment. Such an agent could adapt to a user’s emotional state and dynamically adjust its interaction etiquette. Hierarchical concepts of WordNet’s noun and verb hypernymy are the basic building blocks of the classification rules. A greedy learning algorithm automatically determines which hierarchical concepts are best suited for each rule. A corpus of 5000 emotional sentences has been compiled from 502 test subjects and serves as input to the system.