Data Scientist (Audience Research Projects)

The work We are a London audience intelligence and strategy agency working with major record labels, film distributors and brands. As part of our work we run large-scale campaign research projects: scraping and analysing millions of posts, comments and follow graphs to work out who an artist's audience actually is, what they respond to, and where the next audience is. We have multiple research projects live right now, on tight deadlines, and are looking for a competent data scientist with experience in research projects. to add additional capacity to our team. You would own the analytical layer of these projects. What you would own - Sample design. Stratified sampling across artists and time windows on scraped social data. Deciding what the sample can and cannot support. - Community detection. Clustering follow graphs (typically 500 to 10,000 seed accounts, tens of thousands of edges) into audience segments. Leiden, Louvain or your preferred approach. Resolution selection, stability testing across seeds, and telling us honestly when a cluster is an artefact. - Cluster characterisation. Over-index and lift analysis against stated base rates. Multiple comparison correction. - Theme and sentiment work. Embedding-based topic clustering on short social text. UMAP/HDBSCAN or better. Handling the noise cluster honestly rather than dropping it. - Classifier evaluation. We run LLM-based labelling for sentiment, intent, values and topic across large comment sets. You would report per-class precision and recall as a stated limitation. - Data modelling. Turning multi-platform social and streaming data into structured, tested, analysis-ready tables that persist across projects rather than living in one-off notebooks. You define the grain and the metric definitions; our developer builds and maintains. - Findings. Writing the first draft of what the data says, with every claim traceable and every n stated. Required - Available immediately - Strong applied probability and statistics on large, messy, non-random data - Experience in graph community detection: Leiden, - - Louvain or similar, including validation and stability testing - Embedding-based clustering and NLP on short-form social text - Fluent Python, strong SQL - Experience with marketing data (or similar web data), turning raw multi-platform data into well-structured analysis-ready datasets - LLM-assisted labelling, coding and enrichment of large datasets, including how to evaluate it Nice to have - dbt or similar tested-transformation tooling - Marketing attribution, incrementality and lift measurement (relevant to a later phase of our work, not this one) - An interest in music or culture Not this role This is not an ML engineering role. There are no models to train or serve. This is applied research where the deliverable is a defensible finding. Terms Freelance, starting immediately. Roughly 2 weeks intensive, then ongoing project work. Potential to convert to a senior permanent role in London or remote for the right person, though remote freelance long-term also works. Rate is negotiable for the right person.