2026-05-20T19:24:11Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-1991532025-02-20T18:43:36Zoai_pmh:alloai_pmh:repo_items

199153 https://hdl.handle.net/2286/R.2.N.199153 http://rightsstatements.org/vocab/InC/1.0/ http://creativecommons.org/licenses/by-nc-sa/4.0 2024-12 53 pages Sharma, Yash Elena Chavez-Echeagaray, Maria Gali, Mahesh Suthapalli, Vikash Barrett, The Honors College Text Maintaining consistency in written communication is crucial - especially in professional and academic contexts - where cohesive ‘writing styles’ reinforce credibility and clarity. With the advent of advanced large language models (LLMs), generating such coherent text from prompts has become more accessible to the general public; these models can make achieving a distinct and uniform writing style for the everyday user a simpler task. This thesis proposes a robust system that not only extracts nuanced writing style features from a moderately sized amount of user-provided texts, but that can also synthesize new text that aligns with these features. This is done by leveraging OpenAI’s GPT-4o model in specific and detailed ways, such as specific prompt instructions and guardrails; formatting information concisely and giving clear examples; and leveraging a ‘structured dialogue’ format. Additionally, a study was conducted to use the system and analyze writing styles from a variety of participants, in order to provide insight into its adaptability and scope for end-users. The results point to few-shot examples being more effective in generating text that is consistent with the users’ writing styles, and recommend adding detail while controlling certain aspects of text generation, such as word choice and use of transitions. Moreover, some of the lessons that were learned during this process - such as the impact of certain features over others, as well as the importance of specificity in the prompting process - helped shape the scope and direction for future work on this system. Computer Science Generative AI Writing Style Text Style Transfer Evaluating Text Style Extraction and Transfer using Prompt Based Learners