ML process
What is the very first step of the machine learning process?
Preprocessing/ filtering/ cleaning the data
How does collaborative filtering work?
If we want to recommend a product/class to a target user, we can use the ratings of other similar users to predict what the target user would like.
What math function do we use to compute the similarity between users?
Cosine similarity
What value did we replace for movies that did not have a rating for a particular user?
0.
What is a weakness of collaborative filtering?
New user doesn't have any data to work with. Can't compute similarity score.
How do we remove bias from our formula that calculates the predicted rating for a particular movie?
Subtract by the mean of the other user's rating and add the target user's average rating back.
How did we filter the total user ratings? After filtering and converting the table into matrix form, how many columns does the matrix now contain?
We selected ratings of movies that the target user has watched. 50 columns since the user only rated 50 movies.
What is the test set that includes 20 movie ratings for the target user used for?
This is the ground truth of what the target user actually rates a movie.
What are the final dimensions of the other user's ratings matrix for TESTING before we can predict ratings for the target user? Assume we take the top 50 similar users and 20 movies in our test set.
Since we have the top 50 users, there should be 50 rows and since we have 20 movies in the test set, there should be 20 columns.
How did we split the target user's ratings for training? How did we split the selected users' ratings for training?
We used the first 30 movies that the target user rated. The selected users' ratings also only contains the same 30 movies.
What is the test set that includes other user's movie ratings used for?
It is used to predict the rating for the target user.
How do we evaluate the model performance?
Compare the predicted rating and the ground truth test set we reserved earlier.
What 2 pieces of data does the training set include? How is the training data used?
The ratings for the 30 movies from the target user and all other users' ratings for the same movies. The cosine similarity is computed for the target user to the other users.
What is the range that the cosine similarity between any two vectors will return (hint: cos(theta))? What do the max and min values represent?
1 represents most similar, vectors are the same. 0 represents most dissimilar users and vectors are orthogonal (90 degrees).
What are the necessary pieces of information to calculate the predicted rating for a particular movie?