Study highlights ability to predict depression in Facebook users

A study by David Luxton titled ‘Artificial Intelligence in Behavioural and Mental Healthcare’ published in ScienceDirect has explored the ability to predict depression in Facebook users.

The study used an artificial intelligence algorithm to detect vocabulary usage with negative connotations and said depression can be detected up to 3 months before a social media user is officially diagnosed.

Since the inception of social media studies have shown that there is an inextricable link between social media usage and social anxiety disorder, depression, an increase in suicide in young people, as well as an erosion to short term working memory.

The study suggests it is now possible to use an AI algorithm, to turn what many describe as a ‘catalyst’ for mental illness into a diagnostic tool through analysis of linguistic patterns.

The study claims that the AI predictive model is able to detect future depression as early as 3 months before first documentation onto medical records.

How does it work?

The study is a more sophisticated form of literary analysis – focusing on individual lexicon through frequency, etymology, context of occurrence, colloquialisms and emotive language (amongst other factors) to identify markers and patterns in text to discover an author’s motives and intentions.

It uses an AI algorithm to efficiently and accurately (relatively) identify markers of depression within Facebook posts.

The model takes into account the textual content of the post, length of post, frequency of posting, temporal posting patterns and demographics.

The study shows the 10 topics which are ‘most strongly associated with future depression status’, combined with using ’73 prespecified dictionaries’ to identify markers of depression within Facebook post text.

Findings and accuracy

Markers of a depression were lexicon embedded in negatively connotated language registers associated with a depressed mood, loneliness, hostility and sadness.

Interestingly, the study shows that first person pronouns such as ‘I’, ‘my’ and ‘me’ when appearing more frequently correlate to a more likely diagnosis of depression.

As well as the first-person pronoun marker, another strong identifier of depression was frequent reference to other medical conditions as well as ‘somatic complaints’ such as ‘hurt’, ‘bad’ and ‘pain’. Also, medical references in general are seen to be indicative of a depressive state.

The study is described by its creators as being ‘relatively modest’ in predicting depression in Facebook users, although do state that it could be ‘used in conjunction with other digital screening methods’.