Abstract: With the continuous popularization of smart homes, voice interaction has gradually become an important interactive entry for users to control home appliances. Due to the diversity of types of household appliances and the richness of functions of a single device, it is difficult for ordinary users to control the household appliances through accurate voice control commands due to their lack of understanding of product function definitions in the early stages of use. To address this problem, proposing a semantic-guided reply technique that combines text classification, entity extraction, and knowledge graph schemes. Firstly, the text classification method is used to recall user requests containing valid intentions, and then semantic analysis is carried out by combining entity extraction and knowledge graph technology. Using 3 months of user voice interaction data to test and verify the technology, the correct rate of guidance response reached 87%, which can effectively guide users to use voice commands to control home appliances.

Key words: Voice interaction, Text classification, Entity extraction, Knowledge graph, Smart home