"I just might kill myself": Suicide expressions on Twitter

Public Surveillance of Social Media for Suicide Using Advanced Deep Learning Models in Japan: Time Series Study From 2012 to 2022

BACKGROUND

Social media platforms have been increasingly used to express suicidal thoughts, feelings, and acts, raising public concerns over time. A large body of literature has explored the suicide risks identified by people's expressions on social media. However, there is not enough evidence to conclude that social media provides public surveillance for suicide without aligning suicide risks detected on social media with actual suicidal behaviors. Corroborating this alignment is a crucial foundation for suicide prevention and intervention through social media and for estimating and predicting suicide in countries with no reliable suicide statistics.

OBJECTIVE

This study aimed to corroborate whether the suicide risks identified on social media align with actual suicidal behaviors. This aim was achieved by tracking suicide risks detected by 62 million tweets posted in Japan over a 10-year period and assessing the locational and temporal alignment of such suicide risks with actual suicide behaviors recorded in national suicide statistics.

METHODS

This study used a human-in-the-loop approach to identify suicide-risk tweets posted in Japan from January 2013 to December 2022. This approach involved keyword-filtered data mining, data scanning by human efforts, and data refinement via an advanced natural language processing model termed Bidirectional Encoder Representations from Transformers. The tweet-identified suicide risks were then compared with actual suicide records in both temporal and spatial dimensions to validate if they were statistically correlated.

RESULTS

Twitter-identified suicide risks and actual suicide records were temporally correlated by month in the 10 years from 2013 to 2022 (correlation coefficient=0.533; P<.001); this correlation coefficient is higher at 0.652 when we advanced the Twitter-identified suicide risks 1 month earlier to compare with the actual suicide records. These 2 indicators were also spatially correlated by city with a correlation coefficient of 0.699 (P<.001) for the 10-year period. Among the 267 cities with the top quintile of suicide risks identified from both tweets and actual suicide records, 73.5% (n=196) of cities overlapped. In addition, Twitter-identified suicide risks were at a relatively lower level after midnight compared to a higher level in the afternoon, as well as a higher level on Sundays and Saturdays compared to weekdays.

CONCLUSIONS

Social media platforms provide an anonymous space where people express their suicidal thoughts, ideation, and acts. Such expressions can serve as an alternative source to estimating and predicting suicide in countries without reliable suicide statistics. It can also provide real-time tracking of suicide risks, serving as an early warning for suicide. The identification of areas where suicide risks are highly concentrated is crucial for location-based mental health planning, enabling suicide prevention and intervention through social media in a spatially and temporally explicit manner.

Differences in Suicide-Related Twitter Content According to User Influence

Background: The content of suicide-specific social media posts may impact suicide rates, and putatively harmful and/or protective content may vary by the author's influence. Aims: This study sought to characterize how suicide-related Twitter content differs according to user influence. Method: Suicide-related tweets from July 1, 2015, to June 1, 2016, geolocated to Toronto, Canada, were collected and randomly selected for coding (n = 2,250) across low, medium, or high user influence levels (based on the number of followers, tweets, retweets, and posting frequency). Logistic regression was used to identify differences by user influence for various content variables. Results: Low- and medium-influence users typically tweeted about personal experiences with suicide and associations with mental health and shared morbid humor/flippant tweets. High-influence users tended to tweet about suicide clusters, suicide in youth, older adults, indigenous people, suicide attempts, and specific methods. Tweets across influence levels predominantly focused on suicide deaths, and few described suicidal ideation or included helpful content. Limitations: Social media data were from a single location and epoch. Conclusion: This study demonstrated more problematic content vis-à-vis safe suicide messaging in tweets by high-influence users and a paucity of protective content across all users. These results highlight the need for further research and potential intervention.

Identifying Depression-Related Behavior on Facebook—An Experimental Study

Depression is one of the major mental health problems in the world and the leading cause of disability worldwide. As people leave more and more digital traces in the online world, it becomes possible to detect depression-related behavior based on people’s online activities. We use a novel Facebook study to identify possible non-textual elements of depression-related behavior in a social media environment. This study focuses on the relationship between depression and the volume and composition of Facebook friendship networks and the volume and temporal variability of Facebook activities. We also tried to establish a link between depression and the interest categories of the participants. The significant predictors were partly different for cognitive-affective depression and somatic depression. Earlier studies found that depressed people have a smaller online social network. We found the same pattern in the case of cognitive-affective depression. We also found that they posted less in others’ timelines, but we did not find that they posted more in their own timeline. Our study was the first to use the Facebook ads interest data to predict depression. Those who were classified into the less interest category by Facebook had higher depression levels on both scales.

Self-harm and suicidal content online, harmful or helpful? A systematic review of the recent evidence

Purpose

There are calls for greater regulation of online content related to self-harm and suicide, particularly that which is user-generated. However, the online space is a source of support and advice, including an important sharing of experiences. This study aims to explore what it is about such online content, and how people interact with it, that may confer harm or offer benefit.

Design/methodology/approach

The authors undertook a systematic review of the published evidence, using customised searches up to February 2021 in seven databases. The authors included empirical research on the internet or online use and self-harm or suicide content that had been indexed since 2015. The authors undertook a theoretically driven narrative synthesis.

Findings

From 4,493 unique records, 87 met our inclusion criteria. The literature is rapidly expanding and not all the evidence is high quality, with very few longitudinal or intervention studies so little evidence to understand possible causal links. Very little content online is classifiable as explicitly harmful or definitively helpful, with responses varying by the individual and immediate context. The authors present a framework that seeks to represent the interplay in online use between the person, the medium, the content and the outcome.

Originality/value

This review highlights that content should not be considered separately to the person accessing it, so online safety means thinking about all users. Blanket removal or unthinking regulation may be more harmful than helpful. A focus on safe browsing is important and tools that limit time and diversify content would support this.

Distant Supervision for Mental Health Management in Social Media: Suicide Risk Classification System Development Study

Background

Web-based social media provides common people with a platform to express their emotions conveniently and anonymously. There have been nearly 2 million messages in a particular Chinese social media data source, and several thousands more are generated each day. Therefore, it has become impossible to analyze these messages manually. However, these messages have been identified as an important data source for the prevention of suicide related to depression disorder.

Objective

We proposed in this paper a distant supervision approach to developing a system that can automatically identify textual comments that are indicative of a high suicide risk.

Methods

To avoid expensive manual data annotations, we used a knowledge graph method to produce approximate annotations for distant supervision, which provided a basis for a deep learning architecture that was built and refined by interactions with psychology experts. There were three annotation levels, as follows: free annotations (zero cost), easy annotations (by psychology students), and hard annotations (by psychology experts).

Results

Our system was evaluated accordingly and showed that its performance at each level was promising. By combining our system with several important psychology features from user blogs, we obtained a precision of 80.75%, a recall of 75.41%, and an F1 score of 77.98% for the hardest test data.

Conclusions

In this paper, we proposed a distant supervision approach to develop an automatic system that can classify high and low suicide risk based on social media comments. The model can therefore provide volunteers with early warnings to prevent social media users from committing suicide.

A machine learning approach predicts future risk to suicidal ideation from social media data

Machine learning analysis of social media data represents a promising way to capture longitudinal environmental influences contributing to individual risk for suicidal thoughts and behaviors. Our objective was to generate an algorithm termed "Suicide Artificial Intelligence Prediction Heuristic (SAIPH)" capable of predicting future risk to suicidal thought by analyzing publicly available Twitter data. We trained a series of neural networks on Twitter data queried against suicide associated psychological constructs including burden, stress, loneliness, hopelessness, insomnia, depression, and anxiety. Using 512,526 tweets from N = 283 suicidal ideation (SI) cases and 3,518,494 tweets from 2655 controls, we then trained a random forest model using neural network outputs to predict binary SI status. The model predicted N = 830 SI events derived from an independent set of 277 suicidal ideators relative to N = 3159 control events in all non-SI individuals with an AUC of 0.88 (95% CI 0.86-0.90). Using an alternative approach, our model generates temporal prediction of risk such that peak occurrences above an individual specific threshold denote a ~7 fold increased risk for SI within the following 10 days (OR = 6.7 ± 1.1, P = 9 × 10-71). We validated our model using regionally obtained Twitter data and observed significant associations of algorithm SI scores with county-wide suicide death rates across 16 days in August and in October, 2019, most significantly in younger individuals. Algorithmic approaches like SAIPH have the potential to identify individual future SI risk and could be easily adapted as clinical decision tools aiding suicide screening and risk monitoring using available technologies.