In this paper, we address the challenge of detecting hateful memes in the low-resource setting where only a few labeled examples are available. Our approach leverages the compositionality of Low-rank adaptation (LoRA), a widely used parameter-efficient tuning technique. We commence by fine-tuning large language models (LLMs) with LoRA on selected tasks pertinent to hateful meme detection, thereby generating a suite of LoRA modules. These modules are capable of essential reasoning skills for hateful meme detection. We then use the few available annotated samples to train a module composer, which assigns weights to the LoRA modules based on their relevance. The model's learnable parameters are directly proportional to the number of LoRA modules. This modularized network, underpinned by LLMs and augmented with LoRA modules, exhibits enhanced generalization in the context of hateful meme detection. Our evaluation spans three datasets designed for hateful meme detection in a few-shot learning context. The proposed method demonstrates superior performance to traditional in-context learning, which is also more computationally intensive during inference.
User attribute extraction on social media has gain considerable attention, while existing methods are mostly supervised which suffer great difficulty in insufficient gold standard data. In this paper, we validate a strong hypothesis based on homophily and adapt it to ensure the certainty of user attribute we extracted via weakly supervised propagation. Homophily, the theory which states that people who are similar tend to become friends, has been well studied in the setting of online social networks. When we focus on age attribute, based on this theory, online friends tend to have similar age. In this work, we take a step further and study the hypothesis that the age gap between online friends become even smaller in a larger friendship clique. We empirically validate our hypothesis using two real social network data sets. We further design a propagation-based algorithm to predict online users' age, leveraging the clique-based hypothesis. We find that our algorithm can outperform several baselines. We believe that this method could work as a way to enrich sparse data and the hypothesis we validated would shed light on exploring the proximity of other user attributes such as education as well.