For example, there are thousands of businesses categories, and each business can fall into several categories. To that end, there is a tremendous variety of users and businesses in the dataset. Knowing the specifics of your product and its users is key to designing and building an effective recommendation system for all subsets of users and businesses in the dataset. Business Problem and High-Level Objectivesīuilding an effective recommendation system is challenging because, in addition to requiring a familiarity with algorithmic recommendation techniques, making good predictions depends on business domain knowledge. We present our business objective and model evaluation framework below. Depending on the business objective, there are a number of ways to define a successful prediction (and a successful model). To evaluate the performance of our models, we will compare the ratings our models predict against the actual rating the user left for that business. We’ll use this data to develop an array of models that aim to predict the last review of a business for each user in the dataset (on a scale of 1 to 5 stars). More precisely, the Yelp Dataset is a vast dataset of business reviews left by users of the Yelp platform. In the scope of this report we present and compare several solutions for the specific challenge of building a production-grade recommendation system using the Yelp Open Dataset from the 2019 challenge. Recommendation tasks appear in a wide range of business contexts, from e-commerce to social networking. Specifically, these recommendation systems typically aim to maximize the chance that a user will engage with a recommended product. Recommendation systems are quickly gaining popularity with companies seeking a way to predict user preferences based on past experience.

Yelp raw data set