Today, people receive information through online sources such as social networks and news feeds. This constant feed bears sentiment and has an impact on our emotions. The big-data revolution has provided techniques to assess aggregate sentiment from massive amounts of textual sources.
For example, in the context of environmental stressors, Ueda et al. have shown that Twitter can act as a medium to transfer a general emotion of angst. However, the effects of sentiment conveyed by online feeds on a person’s emotional regulation have never been examined (except for cat-video viewing habits). Also, geolocation may influence individual emotions.
We intend to extend the above approach to examine co-variation between personal geo-context and the use of adaptive versus maladaptive emotion regulation strategies, examining factors such as distance from home/work and being in stressful environments (e.g., at the participant’s mother in law), the supportive nature of a persons social network, and its ad-hoc behavior during expressions of distress.
In exploratory work, Keren Segal is examining how responses to individuals expressing distress affect their future expressions of stress. As a first step, we have obtained a dataset of more than 600 million Tweets and Re-tweets in English belonging to 10M Twitter users that were collected during 2015 in the USA to use as our main dataset in this work. From this dataset, we are using an ML model trained using state-of-the-art methods [1] on the UMD suicidal ideation dataset [2] to retrieve suicidal ideation tweets and their subsequent responses. The textual content of each response will be paired with that of the original post and presented as a task to be tagged according to the type (or lack thereof) of support offered. The results will be used to explore how social support affects the subsequent online behavior of distressed individuals. We are currently in the process of refining the ML-method to work on a tweet-level to identify the suicidal tweets themselves, as the training data only identifies suicidal users.
Keren's work is generously supported by the Haifa University Data Science Research Center (DSRC).
[1] S. Ji, C. P. Yu, S.-f. Fung, S. Pan, and G. Long. Supervised learning for suicidal ideation detection in online user content. Complexity, 2018.
[2] H.-C. Shing, S. Nair, A. Zirikly, M. Friedenberg, H. Daum´e III, and P. Resnik. Expert, crowdsourced, and machine assessment of suicide risk via online postings. In Proceedings of the Fifth Workshop on Computational Linguistics and Clinical Psychology: From Keyboard to Clinic, pages 25–36, 2018.