Question Classification Based on Weak Supervision and Interrogative Pronouns Attention Mechanism

Question classification (QC) plays a vital role in the question answering system, particularly in understanding the user intent. The major challenges for QC are related to annotating the question corpora and extracting the valuable features in the short question text. Intuitively, a lack of annotation corpora has appeared in the QC task, especially in the low-resource languages (LRLs) scenario. This study first tries to augment the corpus with a weakly supervised approach. After that, we also introduce Kazakh QC models based on deep learning networks and interrogative pronouns attention. Namely, we construct the original QC corpora and augment it with synonym dictionaries. The interrogative pronouns matrix is constructed through the question dictionary to strengthen the meaning features in the question. The experiment results demonstrate that the proposed QC model significantly improves the performance of the QC task. Moreover, the corpora presented in this paper are extremely useful for further studies.