1. Built a neural network without tuning and got accuracy .597 2. Built a neural network with default parameters (clean sweep) in hyper parameter tuning and got accuracy .62 With custom definition script using one layer we got: .638 With parameter range and random sweep got accuracy: .622 3. From hyper parameter tuning with entire grid using 2 hidden layers in auto mode, got an accuracy of 0.643 In this, using parameter range and custom script of 2 hidden layers and in entire grid got accuracy of .637 4. For Recommender system: • First we change the numeric values of reviews to categorical variables • Then we select columns in database in the order like: reviewerid, asin, overall in this particular order • Split the data in recommender mode • Train the matchbox recommender on number of traits • Use the test data to score matchbox recommender • Evaluate recommender 5. Word cloud that came out to be prominent are: can, the, like, just one, will, get, like, etc.