A hybrid approach for the detection and monitoring of people having personality disorders on social networks

Abstract

Research in the medical field does not stop evolving. This evolution obliges doctors to be up-to-date in order to well manage every situation that may occur with their patients. However, the medical field is very sensitive and requires a great deal of precision, all of that poses a major problem. Consequently, there is a recourse to computer science, to resolve all of these issues. In this context, we propose in this paper an architecture, taking advantage of artificial intelligence (AI) and text mining techniques to: (i) identify individuals with personality disorder from their textual production on social networks by classifying their set of tweets into distinct classes representing respectively the presence, the category and the type of the disease and (ii) guarantee personalized monitoring by filtering inappropriate tweets according to patient’s circumstance. The first phase was achieved by taking advantage of a deep neuronal approach that benefits of: (i) CNN layers for features extraction from the textual part, (ii) two LSTM layers to preserve long-term dependencies between different lexical units, (iii) SVM classifier to detect the sick person using the dependency links found from the previous layer. The second phase was accomplished by applying a hybrid approach that combined linguistic and statistical techniques in order to filter inappropriate tweets according to the state of each patient. Following the evaluation of our approach, we acquire an F-measure rate equivalent to 84% for the detection of personality disorder, 64% for the detection of the type of disease and 70% for the task of filtering inappropriate content. The obtained results are motivating and may encourage researchers to improve them in view of the interest and the importance of this research axis.

Problem Analysis and Application Research in Legal Judgment Prediction Based on Data Mining

“…Although the machine learning method is effective, its model does not have a clear reflection in the interpretation of prediction results. On the one hand, such as deep neural networks or complex models, this type of method usually has high accuracy, but the internal principles and mechanisms of these methods and models are difficult to understand, and the influence of features on the model prediction results cannot be obtained [9][10][11]. On the other hand, simple models like linear regression and decision trees usually have better interpretability, but their predictive power is usually limited and their accuracy is lower.…”