In this paper, we present a proposal for emotion recognition in a game-based therapy for patients with cognitive impairment. Cognitive impairment is a problem of global scope, where therapies have been sought to mitigate their progress, a way to do it is by using computer-assisted therapy because it performs a cognitive stimulation through software with exercises (designed by specialists) to be solved by the patient. Within the computer system, we seek to analyze specific characteristics to help psychologists to define a series of exercises appropriate to the patient using computational intelligence. Additionally, we aim to include other parameters, in particular, the emotional state of the user, as it has been found that they play an important role when interacting with the software. In this research, we are using novel technologies designed by IBM, such as the TJBot and IBM Watson platform. With the help of these tools, the proposal for the detection of user emotions in human-agent interactions is being considered and preliminary experiments were performed where good results were obtained.