Conversation Modeling on Reddit using a Graph-Structured LSTM

TL;DR

This paper introduces a graph-structured bidirectional LSTM for modeling Reddit discussions, capturing hierarchical and temporal aspects to improve comment popularity prediction and controversy detection.

Contribution

It presents a novel graph-structured LSTM architecture that effectively models threaded social media conversations, outperforming previous node-independent models.

Findings

The model outperforms baseline architectures in predicting comment popularity.

It improves detection accuracy throughout the entire discussion timeline.

Language cues enhance controversy identification within discussions.

Abstract

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.