[Paper Review] Tending Unmarked Graves: Classification of Post-mortem Content on Social Media

Jialun Aaron Jiang, Jed R. Brubaker|arXiv (Cornell University)|Mar 22, 2019

Mental Health via Writing39 references4 citations

TL;DR

This paper presents machine learning classifiers to automatically detect post-mortem social media content—both memorialized user profiles and individual comments—using linguistic features. Trained on 870,326 MySpace comments, the models achieve high accuracy in identifying deceased users soon after death and generalize well to Facebook Memorial Groups and obituaries, enabling compassionate platform design and improved data science by accounting for mortality.

ABSTRACT

User-generated content is central to social computing scholarship. However, researchers and practitioners often presume that these users are alive. Failing to account for mortality is problematic in social media where an increasing number of profiles represent those who have died. Identifying mortality can empower designers to better manage content and support the bereaved, as well as promote high-quality data science. Based on a computational linguistic analysis of post-mortem social media profiles and content, we report on classifiers developed to detect mortality and show that mortality can be determined after the first few occurrences of post-mortem content. Applying our classifiers to content from two other platforms also provided good results. Finally, we discuss trade-offs between models that emphasize pre- vs. post-mortem precision in this sensitive context. These results mark a first step toward identifying mortality at scale, and show how designers and scientists can attend to mortality in their work.

Motivation & Objective

• To address the lack of automated methods for identifying deceased users and memorial content in social media.
• To reduce reliance on manual reporting by survivors, which is slow, inconsistent, and emotionally taxing.
• To enable platform designers and data scientists to account for mortality in system design and longitudinal studies.
• To develop classifiers that detect both post-mortem profiles and individual memorial comments with high accuracy.
• To explore the ethical trade-offs between pre- and post-mortem precision in sensitive contexts.

Proposed method

• Trained supervised machine learning classifiers (e.g., SVM, logistic regression) on 2,688 public MySpace profiles with 870,326 comments.
• Used three feature sets: lexical, syntactic, and semantic (including BERT embeddings) to capture linguistic markers of post-mortem content.
• Classified profiles at the user level and comments at the individual message level using these features.
• Evaluated model performance using precision, recall, and F1 scores on held-out data and cross-platform test sets.
• Applied the trained model to two external datasets: Facebook Memorial Groups and newspaper obituaries to test generalizability.
• Explored model interpretability by analyzing key linguistic features associated with post-mortem content, such as references to death, grief, and past-tense narratives.

Experimental results

Research questions

• RQ1 Can machine learning models accurately classify social media profiles as post-mortem based on linguistic content?
• RQ2 Can individual comments be reliably classified as memorial content even when detached from the user profile?
• RQ3 How soon after death can post-mortem profiles be detected using early comment patterns?
• RQ4 How well do the classifiers generalize across different platforms (e.g., MySpace, Facebook, obituaries) and content types?
• RQ5 What are the ethical trade-offs between maximizing pre-mortem and post-mortem precision in this sensitive domain?

Key findings

• The classifiers achieved high accuracy in detecting post-mortem profiles and comments, with F1 scores exceeding 0.85 on the MySpace dataset.
• Post-mortem content could be reliably detected after only a few messages, indicating early detection is feasible.
• The models generalized well to Facebook Memorial Groups and newspaper obituaries, demonstrating cross-platform robustness.
• Linguistic features such as past-tense references, expressions of grief, and third-person narratives were strong indicators of post-mortem content.
• Models emphasizing post-mortem precision outperformed those prioritizing pre-mortem accuracy, especially in minimizing false positives for living users.
• The study highlights that even conservative labels like 'possibly post-mortem' can inform compassionate design without irreversible actions.