Deep Learning-Based Frameworks for Automated Identifying Depression Through Social Media
- Publication Type:
- Thesis
- Issue Date:
- 2023
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
The data generated by users on Twitter is extremely valuable for healthcare technology as it can reveal important patterns that can greatly benefit the field in multiple ways. Notably, the existing research on online depression detection is limited, with main challenges. First, previous research on identifying depressed users on social media primarily focused on analyzing user behavior and language patterns, including their social interactions. However, these methods have a drawback as they tend to be trained on irrelevant content that may not be essential for detecting depression, which can negatively impact the model's efficiency and effectiveness. Second, limited research has been conducted to identify the changes and variations in depression levels at a more specific level, such as state or neighborhood level, during the COVID-19 pandemic. Third, the ability to explain a model's predictions is crucial for gaining trust, as it offers an understanding of how the model came to a certain conclusion. Unfortunately, many machine learning techniques lack explainability, which is a concern. For example, in the task of automatically predicting depression, most machine learning models produce predictions that are not easily understandable to humans. Fourth, creating new tasks aimed specifically towards modeling narrative elements in social media and to what extent using social media posts makes it possible to extract such features for a narrative explanation of a series of events, which could be significant if we compare people with a mental disorder. Therefore, this thesis aims to develop approaches to identifying depression using online social media, particularly Twitter, and build prediction models that can identify users who are likely to be experiencing mental problems or displaying symptoms that might soon lead to mental disorders. This thesis is organized into five main themes: (1) A depression classification model for understanding how the COVID-19 pandemic has affected people's depression; (2) Depression detection at the User level and its impact during the pandemic; (3) A new, scalable hybrid model that utilizes a combination of deep learning techniques to identify depressed individuals on social media platforms like Twitter through the use of multiple features; (4) An explicable deep learning-based system for depression detection; (5) Modeling narrative elements to identify depression. This thesis's empirical results and findings show the advantages of the proposed approaches in that achieve outstanding performance and provide sufficient evidence to justify the predictions, and demonstrate the narrative elements of a depressed user.
Please use this identifier to cite or link to this item: