Mind the Gap - Aligning Knowledge Bases with User Needs to Enhance Mental Health Retrieval

Aug 20, 2026

Access to reliable mental health information is vital for early help-seeking, yet expanding knowledge bases is resource-intensive and often misaligned with user needs. Such gaps reduce retrieval quality when user concerns appear in informal, contextual, or non-covered language. The work introduces an AI gap-informed corpus augmentation framework that detects underrepresented topics from naturalistic user data, then prioritizes expansions by coverage and usefulness, validated across multiple RAG pipelines in a controlled case study.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages26

Reading Time1 h 6 min

AccessFree to download

Introduction

Retrieval-augmented generation for mental health information

Knowledge base coverage gaps and misalignment with user needs

Related work and limitations

Mind the Gap: Aligning Knowledge Bases with User Needs to Enhance Mental Health Retrieval Amanda Chan Princeton University USA James Liu National University of Singapore Singapore arXiv :2509 . 13626v2 [ cs .IR] 21 Nov 2025 He Kai National University of Singapore Singapore Onno P. Kampman MOH Office for Healthcare Transformation Singapore Abstract Access to reliable mental health information is vital for early help-seeking, yet expanding knowledge bases is resource-intensive and often misaligned with user needs. This results in poor performance of retrieval systems when presented concerns are not covered or expressed in informal or contextualized language. We present an AI-based gap-informed framework for corpus augmentation that authentically identifies underrepresented topics (gaps) by overlaying naturalistic user data such as forum posts in order to prioritize expansions based on coverage and usefulness. In a case study, we compare Directed (gap-informed augmentations) with Non-Directed augmentation (random additions), evaluating the relevance and usefulness of retrieved information across four retrieval-augmented generation (RAG) pipelines. Directed augmentation achieved near-optimal performance with modest expansions–requiring only a 42% increase for Query Transformation, 74% for Reranking and Hierarchical, and 318% for Baseline–to reach ∼95% of the performance of an exhaustive reference corpus. In contrast, Non-Directed augmentation required substantially larger and thus practically infeasible expansions to achieve comparable performance (232%, 318%, 403%, and 763%, respectively) . These results show that strategically targeted corpus growth can reduce content creation demands while sustaining high retrieval and provision quality, offering a scalable approach for building trusted health information repositories and supporting generative AI applications in high-stakes domains. 1 Introduction Access to reliable and timely mental health information is essential for early help-seeking and psychological support [26] . Kalckreuth et al. [15] found that 70.9% of psychiatric patients used the Internet for mental health purposes, most often to learn about disorders, medications, or services. With the growing reliance on self-help platforms, forums, chatbots, and apps [10], retrieving accurate and trusted psychoeducational resources has become increasingly important–both when accessed directly or when surfaced in chatbot responses through retrieval-augmented generation (RAG) [16] . RAG is a natural language processing approach that bridges fluent generation and factual grounding by combining the semantic strengths of large language models (LLMs) with evidence from trusted knowledge bases [23, 18] . By conditioning outputs on retrieved documents rather than relying solely on pretrained parameters, RAG reduces hallucinations and produces more accurate, contextually appropriate, and trustworthy responses [4, 5, 30, 12] . 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: The Second Workshop on GenAI for Health: Potential, Trust, and Policy Compliance. In mental health, knowledge bases that power retrieval systems usually contain articles, exercises, and therapeutic guides curated by trained professionals. These are often national platforms tailored to local contexts (e.g., seasonal affective disorder is absent in tropical countries like Singapore), but they are not designed for automated retrieval and are costly to expand–especially with evolving user needs. As a result, they often lack coverage across the full spectrum of user needs, both in content and in alignment with user style, language, and context [14, 21, 11] . For instance, a systematic review inventory by the Swedish Agency for Health Technology Assessment (SBU) identified over 2,000 evidence gaps in mental health from 2005–2020, highlighting severe content gaps where no systematic review exists or where ev

FAQs

Why do mental health retrieval systems underperform when user needs are not covered in the knowledge base?

Because user concerns may be expressed in informal, contextual, or topic-specific language that the current corpus does not represent, leading to weaker retrieval results.

What is the proposed gap-informed framework and how does it augment the corpus?

It uses naturalistic user data to identify underrepresented topics (gaps), then prioritizes corpus expansions based on coverage and usefulness rather than random additions.

How does directed (gap-informed) augmentation compare with non-directed augmentation in the case study?

Directed augmentation achieves near-optimal retrieval and usefulness performance with modest corpus growth, while non-directed augmentation requires much larger expansions to reach comparable levels.

Aug 20, 2026

Upload by Emma Wilson