# Conversation Modeling on Reddit using a Graph-Structured LSTM

arXiv: https://arxiv.org/abs/1704.02080 · HTML: https://mlsanity.com/paper/1704.02080/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 7 Apr 2017.

## Built on by (most cited first, 12 of 12)

- [A Survey of Graph Neural Networks for Social Recommender Systems](https://mlsanity.com/paper/2212.04481/) (2022)
- [Graph Neural Networks for Molecules](https://mlsanity.com/paper/2209.05582/) (2022)
- [Graph Neural Network and Spatiotemporal Transformer Attention for 3D Video Object Detection from Point Clouds](https://mlsanity.com/paper/2207.12659/) (2022)
- [Characterization of Time-variant and Time-invariant Assessment of Suicidality on Reddit using C-SSRS](https://mlsanity.com/paper/2104.04140/) (2021)
- [Bayesian graph convolutional neural networks via tempered MCMC](https://mlsanity.com/paper/2104.08438/) (2021)
- [Search For Deep Graph Neural Networks](https://mlsanity.com/paper/2109.10047/) (2021)
- [Anger Breeds Controversy: Analyzing Controversy and Emotions on Reddit](https://mlsanity.com/paper/2212.00339/) (2022)
- [P4GCN: Vertical Federated Social Recommendation with Privacy-Preserving Two-Party Graph Convolution Network](https://mlsanity.com/paper/2410.13905/) (2024)
- [The Branch Not Taken: Predicting Branching in Online Conversations](https://mlsanity.com/paper/2404.13613/) (2024)
- [Traits of a Leader: User Influence Level Prediction through Sociolinguistic Modeling](https://mlsanity.com/paper/2501.04046/) (2025)
- [The Influence of Text Variation on User Engagement in Cross-Platform Content Sharing](https://mlsanity.com/paper/2505.03769/) (2025)
- [Benchmarking Multi-Modal Graph-based Social Media Popularity Prediction](https://mlsanity.com/paper/2606.27539/) (2026)

## Abstract (arXiv)

This paper presents a novel approach for modeling threaded discussions on social media using a graph-structured bidirectional LSTM which represents both hierarchical and temporal conversation structure. In experiments with a task of predicting popularity of comments in Reddit discussions, the proposed model outperforms a node-independent architecture for different sets of input features. Analyses show a benefit to the model over the full course of the discussion, improving detection in both early and late stages. Further, the use of language cues with the bidirectional tree state updates helps with identifying controversial comments.