PGM-WV: A context-aware hybrid model for heuristic and semantic question classification in question-answering system

In the field of information retrieval, with the rapid growth of the amount of questions and answers, automatic question-answering system comes up to be a hot research direction, which consists of three procedures: question classification, information retrieval and answer extraction. Question classification is the first and most important part of the whole task. Currently, two kinds of algorithms are employed, rule-based algorithms and statistical-model-based algorithms. Rule-based algorithms have good performance in accuracy and pertinence with the shortcoming of relying on professional knowledge and poor scalability. Statistical-model-based algorithms get classification models from training dataset, these methods extract syntax features heuristically and provide better scalability and thus most question classification algorithms are based on statistical-model. However, semantic features have largely been overlooked in existing statistical-model-based question classification algorithms. In this paper, we propose a context-aware hybrid model based on a statistical-model PGM and a semantic language model word2vec. The experimental evaluations demonstrate the capability of the proposed model.