Janna Lipenkova, Anacode
Janna Lipenkova is an entrepreneur with a background in linguistics, Machine Learning and software development. After a PhD in Computational Linguistics, she founded Anacode, a company focussing on international market intelligence using NLP. Her work focuses on the interface between data, tech and business purpose, developing applications that distill business-relevant insights from Big Data. Janna is an active member of the research community in Data Mining and NLP, regularly contributing to conferences and publications in these areas.
The Linguistic Expression of Mental Health
Mental diseases such as depression are a major health issue in modern society. Social media contains a range of cues about the mental state of a user, and Natural Language Processing can be used to identify anomalies and thus facilitate intervention and treatment.
Linguistic behaviour provides important cues about one’s emotional situation. Our language systematically correlates with our mental state – for instance, as expected, the use of terms denoting negative emotions such as sadness and anger signals a negative mindset. Beyond obvious lexical indicators, implicit patterns on the syntactic and discourse level can be even more revealing since they are more difficult to control: thus, while intensive use of first-person singular pronouns signals a greater self-focus and social disengagement, a lack of coherence and readability might indicate general mental confusion.
In this talk, I will show how the lexical, syntactic and discourse signals of depression in English social data can be analysed using a range of NLP algorithms from the Deep Learning family. These unfiltered descriptions of our own mental states are probably the most genuine emotional expression that is publicly available. The methods applied in this area can be useful not only for detecting and addressing critical cases of depression, but also for the transfer to more commercial settings, thus allowing for the analysis of emotions on a more subtle linguistic level.