The proliferation of social media-based research on mental health offers exciting possibilities to complement traditional methods in mental health care. As ascertained by psychology experts, the online platform should get priority over offline as it offers considerably reliable diagnosis than granted in person. Early detection does not only alleviate the effects of depression on the patient but also benefits the whole community. In this thesis, we explore computational methods in tackling some of the research challenges in depression analysis and make four contributions to the body of knowledge. First, we develop a binary classification model for classifying depression-indicative text from social media. We propose three feature engineering strategies and assess the effectiveness of supervised model to enhance the classification performance in predicting posts indicate depression. To tackle the short and sparse social media data, we particularly integrate the coherent sentiment-topic extracted from the topic model. Additionally, we propose strategies to investigate the effectiveness of affective lexicon in the task of depression classification. Second, we propose a computational method for analysing potential causes of depression from text. With this study, we demonstrate the ability to employ the topic model to discover the potential factors that might lead to depression. We show the most prominent causes and how it evolved over time. Furthermore, we highlight some differences in causes triggered between two different groups, i.e. high-risk of depression and low-risk. Hence, this study significantly expands the ability to discover the potential factors that trigger depression, making it possible to increase the efficiency of depression treatment. Third, we develop a computational method for monitoring the psychotherapy outcome from the individual psychotherapy counselling. Third, we develop a computational method for monitoring the psychotherapy outcome from the individual psychotherapy counselling. By doing this, we show the possibilities of utilising the topic model to track the treatment progress of each patient by assessing the sentiment and topic discussed throughout the course of psychotherapy treatment. Fourth, we propose an unsupervised method called split over-training for identifying user's intention expressed in social media text. We develop a binary classification model for classifying intentions in texts. With this study, we want to show the possibility of applying the intention analysis in mental health domain. Overall, we demonstrate how computational analysis can be fully utilised to benefit clinical settings in mental health analysis. We suggest that more future work could be further explored to complement the traditional settings in mental health care.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:760034 |
Date | January 2018 |
Creators | Abd Yusof, Noor Fazilla |
Contributors | Lin, Chenghua ; Guerin, Frank |
Publisher | University of Aberdeen |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://digitool.abdn.ac.uk:80/webclient/DeliveryManager?pid=238393 |
Page generated in 0.0021 seconds