A STEPWISE DETECTION OF CONJUNCTIVE STRUCTURES IN QUESTIONS USING MAXIMUM ENTROPY MODEL

摘要

This paper presents a maximum entropy model approach to identifying conjuncts of conjunctive structures in questions of financial domain from on-line discussion groups.To avoid phrasal ambiguity, only features in lexical and shallow syntactic level are used.The conjunct detection problem is converted into a stepwise boundary identification task, reducing the search space of a n-word sentence from O(n2) to O(n), The best performance on the test set achieves 85.88% recall and 96% rejection.This approach itself is domain-independent and can be used for conjunct identification in questions universally.