Comparative analysis of platforms for the development of intelligent chatbots in construction automation systems

Abstract

The article presents a comparative analysis of Rasa, Dialogflow, Botpress and OpenAI ChatGPT API platforms for creating intelligent chatbots in the construction industry using the example of processing 200 real Russian-language requests for building materials, technologies and equipment. The object of the research is software environments for developing chatbots with support for natural language processing in the context of construction production, the subject is their accuracy in recognizing intentions and entities(F1 metrics), flexibility of configuration, ease of integration into existing construction project management systems and cost-effectiveness. To assess the quality of NLU, F1 metrics were calculated for a specialized construction building, including requests for technical characteristics of materials, equipment specifications, regulatory requirements, and technological processes. Multi-criteria benchmarking was applied with an expert assessment of the key parameters of integration into building information systems. The results showed that the ChatGPT API demonstrated the highest average F1(0.92) when processing complex technical requests and the ability to generate detailed technical recommendations, but requires significant costs when scaling; Rasa provided a high level of customization for specific construction tasks and the ability to deploy offline on corporate servers. Dialogflow benefits from its intuitive interface and deep integration with Google Cloud Platform for building data management; Botpress combines visual dialog construction with minimal costs, which is critical for small construction enterprises. The practical significance of the work consists in the development of methodological recommendations on the choice of a platform for intelligent assistants in construction, depending on the scale of the enterprise, the type of construction facilities and budget constraints.