Encoding Hierarchical Schema via Concept Flow for Multifaceted Ideology Detection

TL;DR

This paper tackles Multifaceted Ideology Detection (MID) by leveraging label semantics within a hierarchical schema. It introduces Bidirectional Iterative Concept Flow (BICo) to propagate and aggregate concept information along a Root→Domain→Facet→Ideology tree, producing multi-granularity concept representations. Two complementary strategies, Concept Attentive Matching and Concept-Guided Contrastive Learning, align text with facet concepts and disentangle ideologies to improve detection. Experiments on the MITweet dataset show state-of-the-art MID performance and strong cross-topic generalization, demonstrating the practical impact of integrating label semantics into MID.

Abstract

Multifaceted ideology detection (MID) aims to detect the ideological leanings of texts towards multiple facets. Previous studies on ideology detection mainly focus on one generic facet and ignore label semantics and explanatory descriptions of ideologies, which are a kind of instructive information and reveal the specific concepts of ideologies. In this paper, we develop a novel concept semantics-enhanced framework for the MID task. Specifically, we propose a bidirectional iterative concept flow (BICo) method to encode multifaceted ideologies. BICo enables the concepts to flow across levels of the schema tree and enriches concept representations with multi-granularity semantics. Furthermore, we explore concept attentive matching and concept-guided contrastive learning strategies to guide the model to capture ideology features with the learned concept semantics. Extensive experiments on the benchmark dataset show that our approach achieves state-of-the-art performance in MID, including in the cross-topic scenario.

Figures (4)

• Figure 1: Upper: the multifaceted ideology schema and concepts of facets and ideologies liu-etal-2023-ideology. Lower left: the tree-like hierarchical structure of the schema. Lower right: an example of MID. "L" denotes Left, "R" denotes Right.
• Figure 2: Overview of our concept-enhanced multifaceted ideology detection framework. The blue box in the middle shows the proposed bidirectional iterative concept flow (BICo), which includes root-to-leaf concept metapath diffusion and leaf-to-root concept hierarchy aggregation. The concept representations are enriched gradually by bidirectional iteration, and are then used to enhance the two subtasks of MID through concept attentive matching and concept-guided contrastive learning.
• Figure 3: T-SNE visualization of text representations learned by different model variants in the Ideology Analysis subtask. CGCL denotes our Concept-Guided Contrastive Learning. CL denotes the Contrastive Learning without concept anchors. Red, green and blue dots represent Left, Center and Right samples, respectively.
• Figure 4: Results of different numbers of iterations.