Abstract:
Currently, the processing of natural language holds diverse applications with distinct limitations. A common issue in specialized tasks is the limited availability of data, necessitating the search for appropriate models and datasets that align with these constraints. For event platform service providers, a prevalent challenge is the abundance of unorganized questions in the database. These questions often exhibit repetition and lack proper categorization. This research presents the development of a deep learning model for question categorization within event-related content using a CNN-BiLSTM hybrid neural network. Experimental results demonstrate that the presented model consistently outperforms other existing models, exhibiting significant improvements in performance. Furthermore, a method is proposed to identify potential issues within the training dataset by utilizing interpretability through artificial intelligence. This approach facilitates the explanation of the model's prediction outcomes, aiding researchers in better understanding the model's behavior. This, in turn, enables the researchers to analyze and address the model's performance more effectively. As the dataset quality improves, it enhances the model's predictive capabilities, resulting in better prediction outcomes.