Back
Texte intégral
Doctoral thesis

Text mining for online mental health state and personality assessment

    09.09.2021

168 p

Thèse de doctorat: Università della Svizzera italiana, 2021

Information retrieval
Natural language processing
Social media analysis
Text mining
Data stream mining
Data analysis
Data visualisation
Data science
Healthcare informatics
Machine learning
Capsule neural networks
Applied computing
Depression assessment
Personality assessment
English Advances on psycho-linguistics have evidenced that the ways in which people use words could act as a reliable source to assess a wide array of behaviours. Language use acts as an indicator of the individuals’ current mental state, personality and even personal values. In this thesis, we focus on language analysis to study two closely related processes which encompass integral components of persons’ psychological profiles: mental health state and personality. Mental health state assessment by analysing online user-generated content is a field that has recently attracted considerable attention. We start this dissertation by analysing the online digital traces left by individuals in order to ascertain their mental state condition at a particular point in time. To this aim, we exploit the latent semantic structure of social media users posts to spot early traces of depression. Next, present a weak-supervision framework to derive large quantities of data for the study of depression on online settings. Moreover, we conduct a series of analytical studies aimed at gaining insights and extending the current knowledge on how mental disorders are manifested through language and online behaviour in order to be able to detect the early onset of such disorders. While the mental state condition of individuals may fluctuate over their lives, there is a core set of patterns concerning thought, affect and behaviour which is consistent across time and context, constituting the basis of what is commonly referred to as personality. In the second part of this dissertation, we focus on the computational assessment of personality from language cues. We present a novel approach to personality recognition in conversations based on capsule neural networks and exploit its inherent interpretability potential to gain insights from its inner functioning. Moreover, we propose a novel open-vocabulary approach based on multiword expressions which aims at discovering distinctive linguistic patterns of a personality trait. Such technologies will open new avenues to building more empathetic and naturalistic conversational systems.
Language
  • English
Classification
Computer science and technology
License
License undefined
Identifiers
Persistent URL
https://n2t.net/ark:/12658/srd1319225
Statistics
Document views: 204 File downloads:
  • 2021INFO009.pdf: 158