This study by Cao et al. aims to create a continuous suicide risk assessment machine learning model using data from social media. A dataset was constructed by first identifying a group of suicidal social media users, then labelling their severity risk levels with the Columbia Suicide Severity Rating Scale. The model was built with short term, long term, and global memory in order to account for changes in suicidal behavior over time. It then looked for signals indicative of suicidal behavior from users such as posting frequency, phrases associated with suicide, or negative emotions. With the given dataset, the model was able to achieve 96% accuracy in distinguishing suicidal from non-suicidal users. Furthermore, since this model is able to support continuous risk assessment rather than mere static classification, it is able to update its predictions over time as well as give personalized reports on users. The authors do note that the model may not perform as well with other samples as the conducted study only accounted for a fixed user base. Additionally, due to the complexity of human mental illness, solely relying on social media and machine learning to evaluate suicide risk severity is not enough to make an accurate judgement. As such, models such as this one should play a supporting role along with other comprehensive mental health assessment strategies.