Date of Graduation

5-2026

Document Type

Thesis

Degree Name

Master of Science in Computer Science (MS)

Degree Level

Graduate

Department

Electrical Engineering and Computer Science

Advisor/Mentor

Gauch, Susan

Committee Member

Ngan Le, Thi Hoang

Second Committee Member

Pan, Yanjun

Keywords

Artificial Intelligence; Machine Learning; Natural Language Processing; Reviews; Video Games

Abstract

LLMs (Large Language Models) are powerful tools for engaging with textual data, carrying many advantages over classical NLP (Natural Language Processing) and ML (Machine Learning) approaches. However, a classical ML model can still be faster, more efficient to run, and accessible than an LLM. We seek to gain the benefits of LLM text comprehension and preserve them in a classical ML model, a hybrid approach. The LLM operates on text to surface relevant information and associations in our problem space, then the ML model trains on the LLM output. The model may learn from the LLM and provide a more efficient alternative to querying the LLM directly for future data. Using review data from video game marketplace Steam, we conduct a series of experiments toward this end. We prompt the LLM to surface various information from the raw data and train RNNs (Recurrent Neural Networks) to predict a single genre of the games, "RPG" (Role-Playing Game). We then evaluate performance of the trained RNN models on the raw data, checking for generalizability and performance loss/improvement. Results are promising. At baseline, using raw review data, a balanced (50% RPG, 50% non-RPG) dataset, and no LLM assistance, a shallow RNN can predict an average 64.1% accuracy for the genre under test. The maximum accuracy of the LLM on this same dataset is 84.1% Our other models under test lie between these two bounds and demonstrate merits from engaging the LLM during their training.

Download

Share

COinS