Review sentiment scores for sellers

Design a feature for an online marketplace that shows each seller the sentiment of their product reviews: a positive, neutral or negative label on each review and a weekly trend chart. The data science team has a trained text classifier packaged as a Python model file. Buyers write about 40,000 reviews a day across 60,000 sellers, buyers can edit a review, sellers check the dashboard a few times a week, and the model is retrained about once a quarter. Decide whether to score reviews online, in a batch job, or both, and describe how scores are stored, served and recomputed when a new model ships.