Georgia Institute of Technology
Private Messages, Public Health: Detecting Emotional Support with Long-Context LLMs
Abstract
dc:description.abstractThis paper introduces a method for detecting emotional support in private adolescent conversations using large language models (LLMs) capable of handling long contexts. We identify limitations in prior work relying on forum-style public data and short-context BERT models, and show how newer LLMs, coupled with structured prompting and clinically informed evaluations, can drastically improve classification performance on modern chat platforms like Instagram. We describe the processing of participant-donated Instagram data, detail the methodology for evaluating LLMs using a labeled subset, and explore the potential of high-accuracy models for predicting real-world mental health outcomes using a larger unlabeled dataset linked to participant well-being measures.
Degree
thesis:*- Level thesis:degree_level
- Masters
- Department dc:contributor.department
- Computer Science
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Dankhara, Chinar
- Advisor dc:contributor.advisor
-
- De Choudhury, Munmun
- Committee members dc:contributor.committeemember
-
- Kim, Jennifer
- Arriaga, Rosa
Subjects
dc:subject × 11Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81558