A Two-channel Intent Recognition Algorithm Based on Convolutional Neural Networks

Intention recognition is a crucial process in semantic analysis and can be viewed as a text classification task. Traditional convolutional neural networks, such as TextCNN, effectively capture local features. However, they struggle to capture contextual information and fully understand the semantic content of the text. In contrast, text convolutional recurrent neural networks (TextrCNNs) can capture long-term dependencies. This paper proposes a two-channel intention recognition algorithm named GTT. The GTT algorithm combines the strengths of TextCNN and TextRCNN, allowing it to fuse local and global features. As a result, it effectively extracts semantic information from the text, leading to a better understanding of the intent behind user queries. In the THUNews public dataset, the GTT model achieved an accuracy of 88.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. It also recorded an accuracy rate of 90.2% and a recall rate of 88.3%. The F1 score was 87.9%. Overall, the model outperformed GloVe-CNN, GloVe-BiLSTM, and ELMo-CNN-BiGRU. These results demonstrate that the GTT model excels in extracting semantic information and understanding user intentions.