Keyphrase Generative Representation of Youth Crisis Conversations Beyond Static Taxonomies

2026-05-26

AI Summary

This study addresses the challenge of capturing the highly contextualized and dynamically evolving expressions of distress in adolescent mental health crisis conversations, which static labeling schemes often fail to represent accurately. Leveraging a dataset of over 700,000 help-seeking text messages, the authors expand an initial set of 19 problem categories into a hierarchical taxonomy of 39 classes and introduce a Keyword Generation Representation (KGR) method. KGR employs constrained large language models to automatically produce concise, dialogue-specific key phrases that enable dynamic identification of emerging and culturally embedded distress themes—such as immigrant-related struggles and caregiver burden. Experimental results demonstrate high inter-annotator agreement (0.96), with 81% of generated phrases deemed content-accurate and 74% significantly enhancing interpretability. Furthermore, topic retrieval accuracy improves substantially from 0.25 to 0.70.

Abstract

Crisis Responders (CRs) rapidly assess thousands of youth SMS conversations each year to identify mental health concerns and guide support. Yet youth distress is increasingly expressed through evolving and context-specific language that often does not fit fixed-label taxonomies. This work analyzed 703,975 de-identified Kids Help Phone conversations (2018-2023) and expanded KHP's 19-label issue taxonomy into a 39-label hierarchical schema. We then introduce Keyphrase Generative Representation (KGR), a constrained LLM generating concise, conversation-specific keyphrases, evaluated across 129 conversations and 387 expert annotations. The expanded taxonomy achieved expert consensus reliability, with an accuracy of 0.96, and expert review found that 81% of keyphrases accurately reflected content and 74% improved clarity. KGR surfaced identity-linked themes absent from the fixed taxonomy, including immigration problems and caregiver burden, and supported a topic-retrieval workflow that increased accuracy from 0.25 to 0.70 (+0.45) over the manual analyst process. KGR marks a shift toward hybrid, interpretable generative representations that extend crisis response beyond static taxonomies to surface emerging and culturally grounded patterns of youth distress.

Problem

• youth crisis conversations
• static taxonomies
• mental health concerns
• evolving language
• context-specific distress

Innovation

• Keyphrase Generative Representation
• crisis response
• youth mental health
• constrained LLM
• dynamic taxonomy