| Autor: |
Vajrobol, Vajratiya, Aggarwal, Nitisha, Shukla, Unmesh, Saxena, Geetika Jain, Singh, Sanjeev, Pundir, Amit |
| Zdroj: |
International Journal of Information Technology; 20230101, Issue: Preprints p1-16, 16p |
| Abstrakt: |
Depression is a significant global mental health challenge, and its early detection is crucial for effective treatment. Social media platforms are intricately linked with users' emotions, and thus, on many levels, reflect the users' personal lives through written content. Researchers have access to English data regarding depression detection on social media. However, identifying Depression in low-resource languages can be challenging due to the limited availability of annotated data and language models, especially in the Thai language. This study introduces an approach to tackle the scarcity of resources in low-resource languages. It proposes knowledge transfer from English to Thai as a viable strategy. This approach takes a specific focus on the domain of depression detection, a growing global concern. Additionally, the study delves into the topic analysis of depression within the Thai context. Furthermore, an attempt has been made to determine the most effective architectures used for cross-lingual Thai-English applications for depression detection. Results show that RoBERTa achieved the highest accuracy with 77.97%, recall 77.81%, precision 77.97%, and F1-score 77.86%. Explainable NLP also indicated that RoBERTa has the highest prediction probabilities to capture the context in Depression and non-depression classes. |
| Databáze: |
Supplemental Index |
| Externí odkaz: |
|
Full text from SpringerLink