Scholarlink

The detection of mental health problems, such as depression, through user-generated content on social networks has become a key research topic. This study proposes a machine learning-based model to identify signs of depression in social network posts by comparing several algorithms: Random Forest (RF), Logistic Regression (LR), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Long Short-Term Memory networks (LSTM), and Multilayer Perceptrons (MLP) through three feature extraction techniques: Bag of Words (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), and Word2Vec (W2V). Initial results obtained before hyperparameter tuning indicated that LR with BoW obtained the highest accuracy (95.86%) and highest F1 score (95.68%), making it ideal for scenarios in which accuracy is prioritized. After applying careful hyperparameter optimization, performance improved significantly: SVM with BoW obtained the highest F1 score (96.09%), which was closely followed by LSTM and logistic regression with TF-IDF, both above 95.73%. These results underscore the substantial impact of hyperparameter tuning and highlight the potential of carefully aligning machine learning models with BoW and TF-IDF feature extraction techniques to advance automated mental health monitoring in social network contexts.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Proposal of a Model to Detect Depression in Social Media User Posts Using Machine Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Fernando Rodriguez,
  • Junior Fabian,
  • Wilfredo Ticona

摘要

The detection of mental health problems, such as depression, through user-generated content on social networks has become a key research topic. This study proposes a machine learning-based model to identify signs of depression in social network posts by comparing several algorithms: Random Forest (RF), Logistic Regression (LR), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Long Short-Term Memory networks (LSTM), and Multilayer Perceptrons (MLP) through three feature extraction techniques: Bag of Words (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), and Word2Vec (W2V). Initial results obtained before hyperparameter tuning indicated that LR with BoW obtained the highest accuracy (95.86%) and highest F1 score (95.68%), making it ideal for scenarios in which accuracy is prioritized. After applying careful hyperparameter optimization, performance improved significantly: SVM with BoW obtained the highest F1 score (96.09%), which was closely followed by LSTM and logistic regression with TF-IDF, both above 95.73%. These results underscore the substantial impact of hyperparameter tuning and highlight the potential of carefully aligning machine learning models with BoW and TF-IDF feature extraction techniques to advance automated mental health monitoring in social network contexts.