My store has accumulated a healthy backlog of product-level customer reviews and a recent round of post-purchase survey responses. I want to turn this text into clear, reliable insight on how shoppers feel about specific items and overall brand experience, so I can adjust merchandising, messaging and support workflows with data rather than hunches. Scope The job centres on sentiment analysis only. You’ll ingest two data sources—historical customer reviews pulled from the site backend and structured open-ended answers from surveys—and produce an interpretable report of sentiment patterns at the product, category and global store level. Social media is out of scope for now. Preferred approach I’m flexible on tools as long as the method is defensible. If you normally lean on Python with libraries such as NLTK, spaCy or transformers, great. R with tidytext or any comparable stack is fine too. What matters is accuracy, clear explanation of methodology and easy-to-digest results. Deliverables • Cleaned, well-documented code or notebook that can be rerun on future data • Visual and written summary outlining overall sentiment distribution, top drivers of positive vs. negative emotion and any notable trends by product or timeframe • CSV/Excel file with sentiment scores appended to each original review/response Acceptance criteria The model should achieve at least 80 % precision and recall on a manually verified sample I’ll provide, and the final report must be understandable to a non-technical stakeholder. Timeline is flexible within two weeks once you receive the raw CSV exports. If this pilot goes smoothly there will be ongoing analytics projects around purchase behaviour and review frequency next.