🤖 AI Summary
Current large language models (LLMs) face two key bottlenecks in persona control: fine-tuning methods are costly and poorly generalizable, while prompt engineering lacks fine-grained trait modulation capability. This paper proposes UBPL—a fully unsupervised, decoding-time, plug-and-play persona control framework. Its core contributions are threefold: (1) unsupervised construction of a contextualized persona lexicon without annotated data; (2) automatic extraction of trait-associated tokens via Situation Judgment Tests for LLMs (SJT4LLM); and (3) dynamic, logits-level token probability reweighting to enable fine-grained, training-free, and model-agnostic persona intervention. Extensive experiments demonstrate that UBPL improves multi-dimensional persona alignment by 32.7% while incurring negligible BLEU degradation (<0.8), and it is compatible with any open-source LLM.
📝 Abstract
Personality plays a pivotal role in shaping human expression patterns, thus regulating the personality of large language models (LLMs) holds significant potential in enhancing the user experience of LLMs. Previous methods either relied on fine-tuning LLMs on specific corpora or necessitated manually crafted prompts to elicit specific personalities from LLMs. However, the former approach is inefficient and costly, while the latter cannot precisely manipulate personality traits at a fine-grained level. To address the above challenges, we have employed a novel Unsupervisedly-Built Personalized Lexicons (UBPL) in a pluggable manner during the decoding phase of LLMs to manipulate their personality traits. UBPL is a lexicon built through an unsupervised approach from a situational judgment test dataset (SJTs4LLM). Users can utilize UBPL to adjust the probability vectors of predicted words in the decoding phase of LLMs, thus influencing the personality expression of LLMs. Extensive experimentation demonstrates the remarkable effectiveness and pluggability of our method for fine-grained manipulation of LLM's personality.
