Back to search

Georgia Institute of Technology

Emotions in Large Language Models

Abstract

dc:description.abstract

Large language models (LLMs) through their training process learn and store a lot of general knowledge. However, the distribution of knowledge has a long-tail pattern of many infrequently used facts. This poses a challenge for LLMs when a query requires information that is on the long-tail. It is on such queries that LLMs have a tendency to hallucinate. Retrieval-augmented generation (RAG) improves an LLM's ability to answer such questions by retrieving the needed information and adding it to the LLM's context. Part of this thesis proposal looks at this retrieval algorithm and analyzes how it works. A crucial part of RAG is the retrieval corpus itself. Most RAG benchmarks use Wikipedia or Wikipedia-like texts as their retrieval corpus. These texts are written in a neutral and factual tone. However, when RAG systems retrieve internet-based content, they encounter text with diverse tones and linguistic styles, introducing challenges for downstream tasks. This thesis addresses this problem by constructing and validating datasets that introduce sarcasm and emotional variation into retrieved passages, and by developing methods that enable LLMs to better comprehend such pragmatically inflected inputs. In doing so, it explores both prompt-based and translation-based approaches for adapting text tone, and analyzes how emotions are represented in LLMs’ latent spaces, showing how these insights can be leveraged to improve RAG reading performance.

Degree

thesis:*
Name thesis:degree_name
Machine Learning, PhD
Department thesis:degree_department
Electrical and Computer Engineering
Grantor
Georgia Institute of Technology
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Reichman, Benjamin
Advisor dc:contributor.advisor
  • Heck, Larry
Committee members dc:contributor.committeemember
  • Kira, Zsolt
  • Goyal, Kartik
  • Anderson, David
  • Wang, May
ORCID person:identifier.orcid

Rights

Language dc:language.iso
English

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/81698

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Reichman, Benjamin. Emotions in Large Language Models. Georgia Institute of Technology, 2026. https://hdl.handle.net/1853/81698