Construction and optimization of deep learning intent classification model for smart home dialogue

In response to the lack of high-quality annotated corpora in the current smart home field and the insufficient accuracy and interpretability of existing intent classification models in complex home scenarios, this study proposes an intent classification model system that integrates unsupervised topic discovery and deep learning classification. Firstly, a Chinese smart home corpus containing 120,000 real conversations with a dual-layer annotation structure is constructed, and an unsupervised topic recognition model combining latent Dirichlet allocation and K-Means clustering algorithm is designed to mine semantic structures from a data-driven perspective. Finally, a deep learning classification model is designed that integrates a bidirectional transformer model with a text convolutional neural network to achieve accurate intent classification. The unsupervised topic model was superior to the single model in terms of normalized mutual information and adjusted Rand index reaching 0.752 and 0.698 respectively, which verified the rationality of the semantic structure. In terms of supervised learning, the accuracy of the bidirectional transformer-text convolutional neural network model was 96.7%, and the macro F1 score was 96.4%, both of which were better than the baseline model, especially in tasks such as "light control" (F1 score = 98.2%) and "temperature query" (F1 score = 97.6%). Intent distribution analysis further indicated that "device control" (31.2%) and "status query" (24.2%) were the most frequent user needs. Overall, the research provides high-performance intent classification solutions, and also reveals user behavior characteristics at the data level, providing important basis for optimizing the functionality and enhancing the interactive experience of smart home systems.