Reinforcement Learning from Human Feedback (RLHF) is effective for aligning Large Language Models (LLMs) with human preferences. However, RLHF’s complex process limits its ability to continually learn human feedback, making it impractical for real-world applications where the deployed model continuously receives feedback from users. The non-RL-based method, such as Direct Preference Optimization (DPO), is not primitively favorable for Continual Learning (CL). We observe that when combined with Experiment Relay (ER) for CL, DPO tends to significantly widen the gap in the probability of human-preferred and dispreferred responses. Consequently, this diminishes the diversity in model generation, potentially leading to model collapse. To overcome the above challenges, we propose the Continual Optimal Policy Regularization (COPR), a novel non-RL offline method to convert the historical optimal policies into optimization constraints when continually learning new preferences. We first derive a moderate reward function from the pairwise ranking loss and then use the moderate reward to calculate a new sampling distribution to construct novel learning objectives and constraints. We also provide formal proof of the learnability of COPR. The experimental results show that COPR outperforms strong CL baselines on our proposed benchmark, in terms of reward-based, GPT-4 evaluations and human assessment.

Nowadays, social media has become a popular platform for companies to understand their customers. It provides valuable opportunities to gain new insights into how a person’s opinion about a product is influenced by his friends. Though various approaches have been proposed to study the opinion formation problem, they all formulate opinions as the derived sentiment values either discrete or continuous without considering the semantic information. In this paper, we propose a Content-based Social Influence Model to study the implicit mechanism underlying the change of opinions. We then apply the learned model to predict users’ future opinions. The advantages of the proposed model is the ability to handle the semantic information and to learn two influence components including the opinion influence of the content information and the social relation factors. In the experiments conducted on Twitter datasets, our model significantly outperforms other popular opinion formation models.