A significant part of the global population suffers from psychiatric diseases, which impede their capacity to reason properly and behave normally. Despite the difficulty of diagnosing these difficulties in their early stages, it is vital that individuals obtain treatment before their condition worsens. Observing how people present themselves on social media, including the language they use, the way they write, and even the emotions they express, is one way to accomplish this. This study investigates two computer models designed to approximate the temporal evolution of people's emotional states as they interact with social media. We employ two publicly available data sets for anorexia and depression, both of which are common mental diseases. The findings show that it is possible to call attention to important details regarding melancholy or anorexia on social media by using suggested representations of both constant and variable emotions. Incorporating both representations may improve performance, providing results comparable to the best technique for managing anorexia and depression (with a small 1% difference). Furthermore, these formats make it easier to incorporate interpretability into created outputs.