Convolutional Neural Network Model With Hybrid Feature Learning For Sematic Text Mining

In this article, we present a generic inference hybrid structure for Convolutional Recurrent Neural Network (conv-RNN) text semantic modelling, combining the strengths of convolutional and recurrent neural network structures in extracting distinct linguistic features. In addition to the conv-RNN-based sentence matching as well as classification models, we also offer a new sentence classification model as well as an attention-based response selection model with improved performance. Two difficult answer selection (AS) tasks & five standard datasets for sentence classification (SC) are used to verify the suggested models. As far as we are aware, these are the most comprehensive comparison findings in AS and SC to date. We demonstrate the superiority of conv-RNN over other state-of-the-art methods experimentally on a variety of hard tasks and benchmark datasets. Deep training-based architectures outperform ensemble learning methods and supervised training approaches for the sentiment analysis issue in educational data mining. Overall classification accuracy was 95.80% for long-term short-term memory networks using representations built upon the GloVe word-embedding method, making them the most efficient prediction arrangement.