Extracting Emotive Patterns for Languages with Rich Morphology

摘要

This paper describes a method of acquiring emotive patterns for morphosyntactically rich languages. The goal is to maximize the recall of automatically generated sentiment lexicons in a resource lean fashion. The algorithm requires a small corpus with mor-phosyntactic annotations to acquire candidates for emotive patterns and evalute the vocabulary, and web as corpus for lexical expansion. The approach, which involves rule mining and contrast sets discovery, is demonstrated and evaluated for Polish.