Text-based neural networks for question intent recognition

An effective organization of their knowledge bases is pivotal in keeping social networks vibrant, namely in designing successful personalization and contextualization strategies. This way, enhancing dedicated displays and encouraging the production of better content. Particularly for question answering communities, splitting archived material according to their intent is essential to reuse their knowledge and social capital. Recently, deep neural networks have shown breakthrough capabilities on multiple tasks related to language understanding. Thus the main contribution of this work is a thorough comparison of assorted architectures applied to the detection of question intents (i.e., informational and conversational). Evaluated on two collections, DEBERTA and RoBERTa demonstrated to be the best options by finishing with an accuracy of 71.19% and 74.10%, respectively. As for conventional neural networks, RCNNs proven to be the most effective technique. Overall, best models signal the usefulness of both question titles and bodies, and that fusing diverse learning strategies hold promise since they focus on learning different discriminative patterns.