Historic Twitter Mining: A Case Study Aiming to Identify and Capture the Social Media Network Activities of People who Died by Suicide

Social media data is used by researchers, industries and governments alike for a diverse array of activities. For many, the appeal of social media is its real time nature and the global adoption. In the case of Twitter, a single tweet can be sent in real time and consumed by potentially billions of Twitter users. Twitter provides a streaming and search API that can be accessed and used by anyone. These APIs provide ways to programmatically access tweets. The default behaviour of these APIs is to release tweets made within the last 10 days. What if older tweets are required? Whilst a limited number of historic tweets can be collected from a given user account timeline, this assumes that the Twitter user accounts are known. What if the user accounts are not known but the details of the individual are? In this paper we present a case study that aims to tackle this issue directly. Specifically, we try to track the Twitter user accounts and associated social media networks of a known set of individuals that died by suicide in 2014/2015. The ultimate goal was to identify the social media activity posted about their death by people in their network. The hypothesis was that the social media activity following the death of someone who was part of a suicide cluster would be more extensive and potentially more harmful in tone than that following the death of a ‘singleton’ suicide. We also utilize Facebook suicid-related groups to manually identify users and their networks/clusters. We present the approach taken in automatic analysis of Twitter data and the lessons learnt in what transpired to be a not completely successful effort.