Hyper-graph-based attention curriculum learning using a lexical algorithm for mental health

Highlights

• Extract depressive symptoms from writings written by patients.

• Enables the identification and visualization of depressive symptoms using a deep attention based method.

• The framework acts as a decision support system as it incorporates entropy based labeling of data.

• Our online interactive tool (ICT) helps provide context as well as visualization of detected depressive symptoms.

• Evaluated different architecture design for the attention mechanism combined with fuzzy rules.

Abstract

In this paper, we propose a structure hypergraph and an emotional lexicon for word representation. Our method can solve problems related to vocabulary size, grammatical representation of words, and the lack of an emotional lexicon. Natural Language Processing (NLP) and attention-based curriculum learning are then used in the developed model. The goal is to achieve semantic word representations using a graph model. Later, embedding is used to label the text using clinical procedures. The experimental results show the emotional word representation with the structure hypergraph. The bidirectional Long Short Term Memory (LSTM) architecture with an attention mechanism achieved a Receiver Operating Characteristic (ROC) value of 0.96. The learning method can help psychiatrists in note taking and contributes to the detection rate of depression symptoms.