Question Classification Using Graph Convolutional Network

Question classification represents a significant undertaking in the field of Natural Language Processing (NLP), wherein the objective is to categorize questions into predefined classes. The function of question classification can be utilized in a multitude of ways, including Question Answering Systems (QAS), Information Retrieval (IR), or in the field of education. As the number of functions associated with question classification continues to expand, the construction of classification systems becomes more important. This paper proposes a novel approach to classification that makes use of advanced techniques in machine learning, with a particular focus on Graph Convolution Networks (GCN). GCN provides a robust and reliable approach to question classification due to its capacity to address structural and semantic intricacies inherent in natural language data. In this case we transform a classification problem from its question form into a graph. The objective was to ascertain whether this method could produce optimal question classification. To test the model created, a dataset comprising questions submitted by users was used. This dataset consisted of 4,959 questions taken from Reddit and Yahoo! Answers and was classified into 10 cognitive categories. These categories are as follows: extent, cause, disjunction, concept, comparison, consequence, verification, procedural, judgmental, example. After a series of experiments and evaluations, the technique attained an accuracy rate of 0.6362, and F1-Score is 0.5868. These findings show that the GCN approach has tremendous potential for overcoming the complexity and subtleties in natural language data in order to provide precise question classification results.