Conversation dialogue structure modeling is a task in conversational bots powered by artificial intelligence. A useful automated bot requires a properly designed dialogue model. Moreover, the dialogue model generated from the set of real-life conversations can be used for analytical purposes. Effective data mining in conversational intelligence systems is also a principal subject. The companies with large call-center systems demand automated AI tools for data mining. In this paper, we describe our solution for fully automated dialogue structure modeling based on deep neural networks. The proposed solution is based on a transformer neural network fine-tuned for keyword generation and multi-language sentence embedding. We combined these methods with our algorithm for dialogue structure modeling. We evaluate the proposed algorithm on the conversation dataset DiaBiz which is available to the public. The evaluation was done manually by a team of human annotators. Our proposed solution has \(89.04\%\) overall accuracy, indicating the fraction of generated graphs having an acceptable topology similar to the conversation flow based on the scenario script. The performed evaluation showed that is possible to build production-ready software for automated conversation graph analysis.