CoS: Enhancing Personalization and Mitigating Bias with Context Steering

2024-05-02

arXiv.org

AI Summary

This study addresses the tension between contextual utilization and bias risk in large language model (LLM) personalized response generation. We propose Context-guided Sampling (CoS), a training-free, inference-time contextual guidance method. CoS introduces a novel token-level predictive likelihood-based metric to quantify and dynamically modulate contextual influence, enabling real-time user-side control over personalization intensity. It is the first approach to integrate contextual guidance with Bayesian inference for interpretable, quantitative detection of online hate speech. Evaluated across multiple state-of-the-art LLMs and benchmarks, CoS significantly improves personalized response quality while reducing gender- and race-related biases. Moreover, it increases confidence in hate speech identification by 23.6%. The method achieves a favorable trade-off among controllability, effectiveness, and deployment efficiency—requiring no fine-tuning or additional parameters.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

When querying a large language model (LLM), the context, i.e. personal, demographic, and cultural information specific to an end-user, can significantly shape the response of the LLM. For example, asking the model to explain Newton's second law with the context"I am a toddler"yields a different answer compared to the context"I am a physics professor."Proper usage of the context enables the LLM to generate personalized responses, whereas inappropriate contextual influence can lead to stereotypical and potentially harmful generations (e.g. associating"female"with"housekeeper"). In practice, striking the right balance when leveraging context is a nuanced and challenging problem that is often situation-dependent. One common approach to address this challenge is to fine-tune LLMs on contextually appropriate responses. However, this approach is expensive, time-consuming, and not controllable for end-users in different situations. In this work, we propose Context Steering (CoS) - a simple training-free method that can be easily applied to autoregressive LLMs at inference time. By measuring the contextual influence in terms of token prediction likelihood and modulating it, our method enables practitioners to determine the appropriate level of contextual influence based on their specific use case and end-user base. We showcase a variety of applications of CoS including amplifying the contextual influence to achieve better personalization and mitigating unwanted influence for reducing model bias. In addition, we show that we can combine CoS with Bayesian Inference to quantify the extent of hate speech on the internet. We demonstrate the effectiveness of CoS on state-of-the-art LLMs and benchmarks.

Problem

• Balance context influence in LLMs
• Personalize responses without bias
• Mitigate harmful stereotypes in AI

Innovation

• Context Steering (CoS) method
• Modulates token prediction likelihood
• Enhances personalization, reduces bias