Controversy Detection: A Text and Graph Neural Network Based Approach

Abstract

Controversial content refers to any content that attracts both positive and negative feedback. Its automatic identification, especially on social media, is a challenging task as it should be done on a large number of continuously evolving posts, covering a large variety of topics. Most of the existing approaches rely on the graph structure of a topic-discussion and/or the content of messages. This paper proposes a controversy detection approach based on both graph structure of a discussion and text features. Our proposed approach relies on Graph Neural Network (gnn) to encode the graph representation (including its texts) in an embedding vector before performing a graph classification task. The latter will classify the post as controversial or not. Two controversy detection strategies are proposed. The first one is based on a hierarchical graph representation learning. Graph user nodes are embedded hierarchically and iteratively to compute the whole graph embedding vector. The second one is based on the attention mechanism, which allows each user node to give more or less importance to its neighbors when computing node embeddings. We conduct experiments to evaluate our approach using different real-world datasets. Conducted experiments show the positive impact of combining textual features and structural information in terms of performance.

A SHAP-based controversy analysis through communities on Twitter

“…This subject could empower decision-makers in various fields to anticipate and mitigate potential controversies. The methodology could involve combining user interactions with the content of their messages and leveraging Graph Neural Network (GNN) techniques [3] to measure and quantify the evolution of the user graph over time. The integration of spatial and temporal dimensions is particularly compelling, as it allows for a more comprehensive tracking of controversy growth.…”

“…Some works attempted to overcome these limits by exploiting for instance named entities to infer the tendency nature (positive, negative, neutral) of users towards some given named entities [23], and user's vocabulary to cluster users with more similarities in their vocabularies [24]. Some recent works consider controversy detection as a graph classification problem [3]. Graph embedding techniques (GNN) and NLP techniques are used to combine the structure of users' interactions and text content of discussions by encoding the whole discussion graph (structure and texts) into low-dimensional and dense vector spaces.…”

“…Furthermore, researchers have explored the incorporation of textual content as a promising avenue for quantifying controversy. Recent studies have integrated textual information with structural data to enhance controversy classification tasks [3]. Notably, natural language processing (NLP) and deep learning techniques have been employed to enrich structural graph information and quantify controversy [4].…”

Supervised probabilistic latent semantic analysis with applications to controversy analysis of legislative bills

“…Many supervised learning approaches have been proposed for classifying texts into one of the two opposing opinions using annotated controversial corpora including sentences [61], documents [62] and document collections [61]. Some recent work addresses the task of identifying controversial contents on Wikipedia [63,64,65] and on social media [66,67,68].…”

Quantifying polarization in online political discourse

“…Indeed, this algorithm combines the advantages of both approaches to detect and quantize polarization intelligently. Another example of hybrid algorithm is Diffpool [47], a hybrid approach for polarization detection based on Deep Learning. More specifically, this approach is capable of representing a graph through graph convolutional neural networks, as well as content-based information through embeddings.…”