Content Moderation in TV Search: Balancing Policy Compliance, Relevance, and User Experience

2025-05-22

AI Summary

In TV search, conventional retrieval systems frequently return non-compliant or irrelevant results due to model uncertainty, erroneous metadata, or architectural limitations—undermining regulatory compliance, result relevance, and user trust. To address this, we propose the first large language model (LLM)-based post-retrieval intent alignment auditing framework tailored for TV search. Our method dynamically identifies and filters out non-intent-aligned content—i.e., violations or irrelevancies—using the user’s actual retrieval intent as the ground-truth criterion, thereby closing critical safety and quality gaps. Furthermore, it establishes an audit-feedback loop that drives iterative refinement of the underlying retrieval system. Experimental evaluation demonstrates significant improvements in compliance and relevance, reduced exposure risk of harmful or off-topic content, and establishment of a scalable, quality-assurance baseline. This work introduces a novel paradigm for multimodal search content governance grounded in intent-aware auditing.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Computer Vision: Image and Video Retrieval

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Millions of people rely on search functionality to find and explore content on entertainment platforms. Modern search systems use a combination of candidate generation and ranking approaches, with advanced methods leveraging deep learning and LLM-based techniques to retrieve, generate, and categorize search results. Despite these advancements, search algorithms can still surface inappropriate or irrelevant content due to factors like model unpredictability, metadata errors, or overlooked design flaws. Such issues can misalign with product goals and user expectations, potentially harming user trust and business outcomes. In this work, we introduce an additional monitoring layer using Large Language Models (LLMs) to enhance content moderation. This additional layer flags content if the user did not intend to search for it. This approach serves as a baseline for product quality assurance, with collected feedback used to refine the initial retrieval mechanisms of the search model, ensuring a safer and more reliable user experience.

Problem

• Enhancing content moderation in TV search systems
• Balancing policy compliance, relevance, and user experience
• Mitigating inappropriate or irrelevant search results

Innovation

• Uses LLMs for content moderation layer
• Flags unintended search results automatically
• Refines search models via feedback loop