Progressive learning improves explainable hateful meme detection accuracy
Learn Before You Judge: Progressive Knowledge-to-Decision Alignment for Explainable Hateful Meme Detection
Computation and Language
Summary
Online hateful memes spread harmful ideas by mixing images and text in subtle ways. The authors found that making a system explain its decisions at the same time it learns to spot hateful memes can confuse the system and reduce accuracy. To fix this, they created a method called ProKDA that learns in steps: it first gathers background knowledge, then learns to detect hatefulness, and finally aligns this knowledge to refine decisions. Their method works better on several tests and provides clearer explanations for why memes are flagged.
What this means in practice
- •For online content moderation teams: Enhance hateful meme detection tools with a stepwise learning method that produces both accurate labels and clear explanations to support moderation decisions.
- •For social media platform developers: Integrate a progressive knowledge alignment approach to improve automated identification and explanation of abusive meme content in user feeds.
Authors
Bo Xu, Chenyuan Wang, Xinyu Chen, Quanhao Zhu, Rui Lin, Liang Zhao, Hongfei Lin, Feng Xia
Abstract
Hateful memes spread abusive content through implicit interactions between images and text, posing serious threats to the safety of online communities. In recent years, multimodal large language models have been widely used for hateful meme detection and are increasingly adopted to generate explainable detection results. However, we find that existing explain-then-detect methods often couple explanation generation and label prediction within the same training process. This coupling causes interference between task objectives, leading to limited detection performance and even worse results than simple SFT baselines. To address these challenges, we propose ProKDA, a progressive knowledge-to-decision alignment method for explainable hateful meme detection. Inspired by the human annotation training process, ProKDA first uses an agentic background knowledge construction pipeline to obtain external knowledge related to meme understanding. It then adopts a three-stage training strategy that sequentially performs background knowledge learning, hatefulness detection learning, and hatefulness boundary alignment. Unlike prior explain-then-detect methods that jointly optimize both tasks, ProKDA focuses on a single training objective at each stage. This design reduces interference between the two tasks and progressively transforms background knowledge into robust detection decisions. Experiments on three public hateful meme benchmarks show that ProKDA achieves state-of-the-art detection performance and provides accurate, explainable, and evidence-supported decisions for hateful meme moderation. Project page: https://meizhiyuan88666.github.io/prokda.