Structure Analysis and Computation-Based Chinese Question Classification

Question classification is the key element of question answering system (QA). It governs answer extraction range and method, and further affects entire system performance. Most question classification methods are based on pattern sets or knowledge databases. Their disadvantages are that the set or database scale will become larger more and more, and need much human work. By comparing question classification with document classification, and analyzing Chinese question sentences´ characters, this paper proposes a new question classification computation model, which is based on Bayes theory and takes the relations between question structures and question types into account. Experiments show that in this model, feature vector extension can reduce the negative influences of shorter question sentences and smaller corpora; and interrogative-based 2-gram can make use of question structure features to efficiently improve question classification precision.