|
Chris Pollett >
Students > [Bio] [Blog] [Initial mockup with API - PDF] |
Prepare AI Model to be Trained and ObservedDescription:This deliverable is for training sample dataset from Yelp using Bayesian Personalized Recommender. Unlike the usual approach that can be found in the Netflix Prize challenge in which the model is optimized to predict an element S in the list as 1 and 0 for the rest of the elements. BPR helps avoid the issue where the rest of the elements which will be ranked in the future are presented to the model as negative feedback (0 values). BPR proposes a different approach by using item pairs as training data and optimize for correctly ranking item pairs instead of scoring single items because the model would just replace the missing item with negative values. In order to do so, BPR uses training dataset which consists of both positive and negative pairs as well as missing values, but in this case the missing values between two non-observed items are exactly the item pairs that have to be ranked in the future. This makes the training data and the test data disjoint from each other.Implementation:
1. Data is obtained from Yelp Challenge
|