Jianing (Oscar) Xie

Supervisors: Alan Wang, Wenqi Shen

• Framed the study as design-science research, grounding it in an encoding-variability mechanism that links the linguistic diversity of training text to a model's robustness against collapse under recursive training.
• Designed a Writeprints linguistic-diversity framework that isolates which dimensions of training text drive the deepest diversity loss; the longer-term aim is to guide data interventions that reduce both diversity loss and long-tail knowledge loss.
• Implemented a Writeprints feature pipeline in Python computing 19 lexical, syntactic, structural, and content dimensions per document, so each diversity construct can be measured and disentangled separately.
• Fine-tuned a Llama-3.2-1B model with LoRA across five generations of recursive training on Wikipedia corpora, reproducing the iterative self-training loop that drives model collapse.

Supervisors: Richard A. Hunt, Joseph Simpson

• Conducted a science-of-science integrative review of how NHST (Null Hypothesis Significance Testing) became an institutionalized epistemic rule in management research rather than a purely statistical technique.
• Designed a reproducible search and screening protocol on Web of Science in Python; predefined inclusion criteria narrowed 492 candidate articles from FT50 and leading review journals to a final sample of 165 peer-reviewed papers.
• Scored each article's stance toward NHST through LLM-based polarity analysis and mapped papers onto management sub-fields; built the sentiment heat maps and publication-trend figures in the review.
• Proposed one of the review's six core insights: the risk of epistemic inversion under generative AI, where AI workflows can reinforce p-hacking by reasoning backward from detectable patterns to seemingly theory-driven claims.

Supervisors: Rong Du, David M. Townsend

• Used a quasi-natural experiment with Difference-in-Differences (DID) to study the impact of AI applications (recommendation, review summarization, knowledge Q&A) on consumer purchasing behavior and perceived quality.
• Developed web-scraping solutions with the Scrapy framework, extracting more than 100,000 user reviews.
• Conducted sentiment analysis with LLM APIs and NLP techniques (word2vec, BERT) and linguistic analysis with LIWC to identify key features in user reviews.
• Cleaned and processed over 1 million transaction records with ETL pipelines using NumPy, Pandas, and DuckDB.
• Theoretically, first to combine cue-utilization theory with generative AI in e-commerce, analyzing long-tail vs. winner-take-all dynamics; introduced Baidu Index as an objective brand-awareness metric with real sales data.

Supervisors: Rong Du, Richard A. Hunt

• Pioneered a diagnostic framework for AI-Generated Content (AIGC) trustworthiness by integrating Nomology theory with a dual cognitive perspective — a multi-level, nine-dimension framework built via Grounded Theory and validated through large-scale text mining.
• Conducted semi-structured interviews with over 20 users, including domestic and international undergraduates, doctoral students, and industry professionals.
• Performed text mining and sentiment analysis on social-media data using topic modeling (e.g., BERTopic) to validate the framework and quantify trust across the nine dimensions.
• Diagnosed a core "representation–substance imbalance": positive perceptions of AIGC's surface experience are undermined by distrust in its core substance (information quality, transparency).

Supervisors: Rong Du, Andrew Burton-Jones

• Used a quasi-natural experiment with DID to study how product traceability information (traditional and blockchain-based) affects consumer purchase behavior, with brand reputation and eWOM as moderators.
• Applied LDA topic modeling and text mining to 14,000 product reviews, constructing an attention index for user topic focus.

Rank 1 / 93

• B.Mgmt. in Big Data Management and Application (Xidian) — GPA 3.9 / 4.0
  Minor: Artificial Intelligence & Large Language Model Application (Xidian)
• B.S. in Business, Management: Entrepreneurship, Innovation & Technology Management (Virginia Tech) — GPA 3.78 / 4.0

Selected core courses

• Business Intelligence Analysis (Data Mining) — A−
• Introduction to Data Science — A
• Management Information System — A−
• Database Principle and Application — A
• Business Statistics, Analytics & Modeling — A
• Statistical Machine Learning & Text Mining — A
• Natural Language Processing — A
• Principles of Large Models and Industry Applications — A