Construct and inspect vocabulary and TF-IDF representations of a marketing corpus, and select and justify the number of topics and the use of hard or soft membership.

Tune topic-model parameters, preserve reproducible model outputs, and evaluate topic quality using within-topic similarity, between-topic separation, diagnostic sweeps, and exemplar documents.

Construct adjacency matrices and directed or weighted graphs from text and social-media interactions.

Interpret reciprocity, centrality, attributes, and bridge structures while documenting filtering and threshold choices.

Explain how next-token prediction, masked-language modeling, and contextual embeddings support LLM classification.

Translate a marketing research question into a closed label set with definitions and borderline-case rules, and design prompts that return exactly one valid, machine-readable classification label.

Validate a sample, execute efficient batch inference through an API or vLLM workflow, and select an appropriate fine-tuning workflow when prompt-based classification is insufficient.

Evaluate prompting and fine-tuning with accuracy, macro F1, class-level errors, and audits of disputed gold labels.