Leveraging Twitter to better identify suicide risk

Samah Fodeh, Joseph Goulet, Cynthia Brandt, Al-Talib Hamada
Proceedings of The First Workshop Medical Informatics and Healthcare held with the 23rd SIGKDD Conference on Knowledge Discovery and Data Mining, PMLR 69:1-7, 2017.

Abstract

While many studies have explored the use of social media and behavioral changes of individuals, few examined the utility of using social media for suicide detection and prevention. The study by Jashinsky et al, in particular, identified specific language patterns associated with a set of twelve suicide risk factors. We utilized their findings to assess the significance of the language used on Twitter for suicide detection. We quantified the use of Twitter to express suicide related language and its potential to detect users at high risk of suicide. First, we evaluated the presence of language related to twelve different suicide risk factors on Twitter using a list of terms/statements published by Jashinsky et al and searched Twitter for tweets indicative of 12 suicide risk factors. Using network analysis, for each suicide risk factor we established a subnetwork of users and their tweets related to that suicide risk factor. We computed the density of each subnetwork to estimate the presence of the language of that suicide risk factor. Second, we investigated relationships between suicide risk factors, using associated language patterns, In two groups “high risk” and “at risk”. We divided Twitter users into “high risk” and “at risk” based on two of the risk factors (“self-harm” and “prior suicide attempts”) and examined language patterns by computing co-occurrences of terms in tweets. We identified relationships between suicide risk factors in both groups using co-occurrences. We found that users within a subnetwork used similar language to express their feeling/thoughts. Stratifying users into “high-risk” and “at-risk”, we found stronger relationships between pairs of risk factors such as (“depressive feelings”, “drug abuse”), (“suicide around individual”, “self-harm”), and (“suicide ideation”, “drug abuse”) in the “high-risk” group relative to the “at-risk” group. In addition, the presence of social-related suicide risk factors including “gun ownership”, “suicide around individual”, “family violence”, and “prior suicide attempts” was more pronounced in the “high-risk” group.

Cite this Paper


BibTeX
@InProceedings{pmlr-v69-fodeh17a, title = {Leveraging Twitter to better identify suicide risk}, author = {Fodeh, Samah and Goulet, Joseph and Brandt, Cynthia and Hamada, Al-Talib}, booktitle = {Proceedings of The First Workshop Medical Informatics and Healthcare held with the 23rd SIGKDD Conference on Knowledge Discovery and Data Mining}, pages = {1--7}, year = {2017}, editor = {Fodeh, Samah and Raicu, Daniela Stan}, volume = {69}, series = {Proceedings of Machine Learning Research}, month = {14 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v69/fodeh17a/fodeh17a.pdf}, url = {https://proceedings.mlr.press/v69/fodeh17a.html}, abstract = {While many studies have explored the use of social media and behavioral changes of individuals, few examined the utility of using social media for suicide detection and prevention. The study by Jashinsky et al, in particular, identified specific language patterns associated with a set of twelve suicide risk factors. We utilized their findings to assess the significance of the language used on Twitter for suicide detection. We quantified the use of Twitter to express suicide related language and its potential to detect users at high risk of suicide. First, we evaluated the presence of language related to twelve different suicide risk factors on Twitter using a list of terms/statements published by Jashinsky et al and searched Twitter for tweets indicative of 12 suicide risk factors. Using network analysis, for each suicide risk factor we established a subnetwork of users and their tweets related to that suicide risk factor. We computed the density of each subnetwork to estimate the presence of the language of that suicide risk factor. Second, we investigated relationships between suicide risk factors, using associated language patterns, In two groups “high risk” and “at risk”. We divided Twitter users into “high risk” and “at risk” based on two of the risk factors (“self-harm” and “prior suicide attempts”) and examined language patterns by computing co-occurrences of terms in tweets. We identified relationships between suicide risk factors in both groups using co-occurrences. We found that users within a subnetwork used similar language to express their feeling/thoughts. Stratifying users into “high-risk” and “at-risk”, we found stronger relationships between pairs of risk factors such as (“depressive feelings”, “drug abuse”), (“suicide around individual”, “self-harm”), and (“suicide ideation”, “drug abuse”) in the “high-risk” group relative to the “at-risk” group. In addition, the presence of social-related suicide risk factors including “gun ownership”, “suicide around individual”, “family violence”, and “prior suicide attempts” was more pronounced in the “high-risk” group.} }
Endnote
%0 Conference Paper %T Leveraging Twitter to better identify suicide risk %A Samah Fodeh %A Joseph Goulet %A Cynthia Brandt %A Al-Talib Hamada %B Proceedings of The First Workshop Medical Informatics and Healthcare held with the 23rd SIGKDD Conference on Knowledge Discovery and Data Mining %C Proceedings of Machine Learning Research %D 2017 %E Samah Fodeh %E Daniela Stan Raicu %F pmlr-v69-fodeh17a %I PMLR %P 1--7 %U https://proceedings.mlr.press/v69/fodeh17a.html %V 69 %X While many studies have explored the use of social media and behavioral changes of individuals, few examined the utility of using social media for suicide detection and prevention. The study by Jashinsky et al, in particular, identified specific language patterns associated with a set of twelve suicide risk factors. We utilized their findings to assess the significance of the language used on Twitter for suicide detection. We quantified the use of Twitter to express suicide related language and its potential to detect users at high risk of suicide. First, we evaluated the presence of language related to twelve different suicide risk factors on Twitter using a list of terms/statements published by Jashinsky et al and searched Twitter for tweets indicative of 12 suicide risk factors. Using network analysis, for each suicide risk factor we established a subnetwork of users and their tweets related to that suicide risk factor. We computed the density of each subnetwork to estimate the presence of the language of that suicide risk factor. Second, we investigated relationships between suicide risk factors, using associated language patterns, In two groups “high risk” and “at risk”. We divided Twitter users into “high risk” and “at risk” based on two of the risk factors (“self-harm” and “prior suicide attempts”) and examined language patterns by computing co-occurrences of terms in tweets. We identified relationships between suicide risk factors in both groups using co-occurrences. We found that users within a subnetwork used similar language to express their feeling/thoughts. Stratifying users into “high-risk” and “at-risk”, we found stronger relationships between pairs of risk factors such as (“depressive feelings”, “drug abuse”), (“suicide around individual”, “self-harm”), and (“suicide ideation”, “drug abuse”) in the “high-risk” group relative to the “at-risk” group. In addition, the presence of social-related suicide risk factors including “gun ownership”, “suicide around individual”, “family violence”, and “prior suicide attempts” was more pronounced in the “high-risk” group.
APA
Fodeh, S., Goulet, J., Brandt, C. & Hamada, A.. (2017). Leveraging Twitter to better identify suicide risk. Proceedings of The First Workshop Medical Informatics and Healthcare held with the 23rd SIGKDD Conference on Knowledge Discovery and Data Mining, in Proceedings of Machine Learning Research 69:1-7 Available from https://proceedings.mlr.press/v69/fodeh17a.html.

Related Material