從社群媒體使用者貼文探討憂鬱偵測：深度學習的方法

Abstract

重性抑鬱疾患（Major Depressive Disorder，簡稱MDD，也可簡稱為憂鬱症）為近年來非常重要的一個議題；患者不只會感到低落、負面情緒，甚至會有自殘與自殺的念頭。全球重性抑鬱疾患者的數量愈來愈多，不僅造成患者情緒上的負擔，更造成了社會生產力與經濟的壓力。及早發現並且給予適當治療可幫助減緩情緒與經濟上的負擔。本研究旨在以深度學習方法，利用社群媒體的資料來偵測有憂鬱傾向之使用者。本研究以使用者為層級（user-level）進行分析，採用長短期記憶神經網路（Long Short-Term Memory network，簡稱LSTM），透過將資料以時序區隔建立預測模型，檢驗時序因子對於社群媒體資料判別憂鬱傾向使用者是否有效用的同時，期待能夠透過該方法找出潛在憂鬱症使用者。

Major Depressive Disorder (MDD, also known as depression) has become a severe issue in recent years. People who suffer from depression feel not only depressed but also negative emotions. At worst, MDD patients would have suicidal ideation or conduct suicide. The number of MDD patients worldwide is increasing gradually. Apart from causing emotional burdens on patients, MDD also results in the productivity and economic stress of society. Early detection and appropriate treatment of depression can reduce the emotional and financial hardship of MDD patients. This study aims at detecting depressive users on the basis of social media data using a deep learning method, which analyzes data at a user-level. Through conducting Long Short-Term Memory networks (LSTM) with time-series segmentation to examine whether the time factor affects the detection of depressive users on social media data, as well as hoping to find out potential depressive users.