Social robots are devoted to interacting and communicating with humans. Traditionally, the interaction capabilities of social robots are limited because the dialogues they can maintain are perceived as predictable, repetitive, and unnatural. This can lead the user to lose interest in the robot. If we want to bet on a successful and long coexistence of humans and robots, it is necessary to provide robots with more varied speeches that can be easily adapted to the users’ needs. In this contribution, we propose a methodology that uses social media mining techniques to find topics that might interest a user. Then, using machine learning techniques, we create the robot’s verbal communication. This methodology, implemented in our social robot Mini, uses three types of deep learning models for natural language processing: a summarization model, a long-form query-answer model, and a generative model. We rely on pre-trained models that have been integrated into Mini, allowing our robot to maintain conversations about different topics that change dynamically.