🤖 AI Summary
Personalized recommendation systems often fail to effectively identify and filter “inappropriate content” due to subjective user definitions and opaque algorithmic decision-making, undermining user experience and trust. To address this, we propose the first LLM-based framework for editable user preference profiling and conversational inappropriate-content filtering. Our method leverages prompt engineering to construct dynamic, human-intervenable, fine-grained preference representations, and introduces a lightweight, plug-and-play conversational filtering module that jointly ensures user controllability, model interpretability, and system compatibility. Evaluated on a 3.8B-parameter open-weight LLM, our approach achieves proxy-task performance on par with state-of-the-art commercial models. A one-week user study with 24 participants demonstrates statistically significant improvements: +27.4% in filtering accuracy, +41.2% in user trust, and measurable enhancement in platform-level recommendation distribution fairness.
📝 Abstract
Personalized algorithms can inadvertently expose users to discomforting recommendations, potentially triggering negative consequences. The subjectivity of discomfort and the black-box nature of these algorithms make it challenging to effectively identify and filter such content. To address this, we first conducted a formative study to understand users' practices and expectations regarding discomforting recommendation filtering. Then, we designed a Large Language Model (LLM)-based tool named DiscomfortFilter, which constructs an editable preference profile for a user and helps the user express filtering needs through conversation to mask discomforting preferences within the profile. Based on the edited profile, DiscomfortFilter facilitates the discomforting recommendations filtering in a plug-and-play manner, maintaining flexibility and transparency. The constructed preference profile improves LLM reasoning and simplifies user alignment, enabling a 3.8B open-source LLM to rival top commercial models in an offline proxy task. A one-week user study with 24 participants demonstrated the effectiveness of DiscomfortFilter, while also highlighting its potential impact on platform recommendation outcomes. We conclude by discussing the ongoing challenges, highlighting its relevance to broader research, assessing stakeholder impact, and outlining future research directions.