SprottSchool of BusinessLearn + interrogate
The instructor develops the method, shows its business applications, and opens the focal paper’s claim and design.
You leave with a claim map and reconstruction targets.BUSI 6306 · doctoral research seminar
Artificial Intelligence Methodsin Business Research
Don’t just read AI research. Rebuild it.
Across four research cycles, you will learn a method, audit a published business study, reconstruct its evidence, and use what you discover to design a stronger next study.
This semester, you build
4research
dossiers
dossiers
1defensible
proposal
13 classes · four research cycles · no conventional exam
proposal
The course journey Learn applied AI methods in business research, reconstruct published implementations, then build and defend your own research.
One rhythm, repeated four cycles
First understand it. Then execute it. Then make it yours.
You will never meet a method as detached machinery. Every cycle stays anchored to a consequential business problem and a published paper.
The 3-Class Cycle
Every cycle occupies three consecutive classes and follows the same learning sequence.
Execute + audit
We run the reconstruction together, manipulate consequential choices, and interpret what recovers—and what does not.
You leave with executable evidence and a calibrated verdict.Defend + extend
You reconstruct a faculty-curated paper, defend your choices, and propose the next study with a rigorous methods section.
You leave with a research dossier and extension.The four cycles
Four ways AI turns business traces into evidence.
Select a cycle to see its business problem, focal study, methods, and what you will produce.
Your methods library
Know what to buy—and what Carleton already gives you.
The journal articles show the methods at work in business research. These books teach the methods systematically across the four cycles. A laptop capable of running Python, Jupyter, and Git is required in every class.
Student purchaseCarleton Library · O’ReillyFree official access
Research-design spine
Text as Data frames corpus construction, representation, measurement, validity, prediction, and inference.
Representation
Hands-On Large Language Models plus Jurafsky and Martin connect tokens, embeddings, classification, clustering, transformers, and semantic interpretation.
Measurement + LLMs & Agents
AI Engineering supplies evaluation, retrieval, agent design, model-as-judge limits, monitoring, and reliability.
Networks
Graph Neural Networks in Action and Network Science supply computation; Rawlings et al. supply network measurement, research design, and inference.
Text as Data ↗
Justin Grimmer, Margaret E. Roberts & Brandon M. Stewart · Princeton University Press, 2022
Assigned across Cycles 1 and 2.Hands-On Large Language Models ↗
Jay Alammar & Maarten Grootendorst · O’Reilly, 2024
Selected chapters in Cycles 1 and 3.AI Engineering ↗
Chip Huyen · O’Reilly, 2025
Evaluation chapters in Cycle 2; core system chapters in Cycle 3.Graph Neural Networks in Action ↗
Namid Stillman & Keita Broadwater · Manning, 2025
Cycle 4 computational methods text.Network Science ↗
Carlos Andre Reis Pinheiro · Wiley, 2022
Selected Cycle 4 graph-analysis chapters.Network Analysis with R companion ↗
Craig M. Rawlings, Jeffrey A. Smith, James Moody & Daniel A. McFarland
Cycle 4 network measurement, research design, and inference reference.The free website provides runnable R tutorials and data. The Python implementations used in class are supplied in the course repository.Speech and Language Processing ↗
Daniel Jurafsky & James H. Martin · 3rd-edition online manuscript
Selected chapters in Cycles 1 and 3. Assigned chapters will identify the version used for the current offering.Purchase rule: Text as Data is the only required student purchase. Access the four O’Reilly books through Carleton Library at no additional cost; use the free Jurafsky and Martin manuscript and Rawlings R companion online. Assigned chapters will appear in Brightspace.
Your semester
Thirteen classes form one research lifecycle.
Filter the sequence, then open any class to see how to arrive and what you are expected to carry forward.
How your work is assessed
Assessment rewards evidence, judgment, and a defensible next claim.
Each cycle research dossier carries equal weight. Workshop contributions reward documented peer-review work, while the final written research proposal and its Class 13 presentation and oral defense assess your individual research design.
15%Cycle 1 · Representation research dossier
Your research development across the four cycles
Each cycle gives you more responsibility for the research design.
The progression is about the decisions you will learn to make—not about four different grading standards.
Cycle 1 · RepresentationLearn the reconstruction process
→
The instructor supplies the data path, code sequence, and checkpoints. You run the analysis, compare it with the paper, identify discrepancies, and assess whether the representation measures the intended concept.
Cycle 2 · MeasurementMake and justify analytical choices
→
You choose and defend the estimand, estimator, uncertainty analysis, and decision-value test within a shared reconstruction framework.
Cycle 3 · LLMs & AgentsDesign the AI research procedure
→
You specify and freeze the model, prompts, retrieval and tools, agent rules, evaluator, and stopping conditions; preserve outputs and logs; and test whether the finding survives system changes.
Cycle 4 · NetworksConduct a near-independent reconstruction
You define the nodes and edges, build the network, choose the analyses, test alternative graph definitions, and develop the research extension with instructor review.
What counts as a reconstruction
Say exactly what you checked and what you found.
For every paper, begin with one specific published target: a table, figure, coefficient, prediction score, concept, ranking, or network result. Record the data and code you used, rerun the relevant analysis, and place your result beside the published result.
If the two results differ, investigate the data, sample, code, software version, and analytical choices. Explain the discrepancy when you can; label it unresolved when you cannot. A mismatch can still be strong research work. Claiming that you reproduced an analysis you did not execute cannot.
The four levels describe how far your evidence reaches. They are not four different grades.
1Check the paper’s reported tablesRecalculate a published comparison using numbers in the article or appendix. You have audited the claim, but you have not rerun the underlying observations.
2Rerun the authors’ prepared datasetExecute the analysis using the cleaned or intermediate files supplied with the paper. Report what earlier data preparation you still cannot verify.
3Rebuild the analysis from raw source dataStart with the original public records, reproduce the sample and variables, run the model, and compare the result with the paper.
4Repeat the study with new dataApply the published research design to a different period, market, organization, or sample and test whether the finding holds.
Every dossier must show this chainPublished target → your reconstructed result → difference → best explanation → conclusion the evidence supports
Before the course begins
Arrive ready to investigate, not already expert.
Your progress is saved in this browser. The course supplies methodological scaffolding, but it is not an introductory programming course.
Questions students usually ask
Know what kind of course you are entering.
By Class 13