Collaborative Filtering
Questions to Answer:
What is a Recommender System?
What is Collaborative Filtering?
What are the Collaborative Filtering Techniques?
Recommender System
-
Automated systems that make recommendations based on the preference of users
-
Examples:
- Amazon or any other online store always makes recommendations of products to buy
- Netflix or Spotify
Collaborative Filtering
- is the tech behind most recommender systems, which is the process of filtering information by soliciting judgements from others to overcome the information overload problem.
Collaborative Filtering Techniques
- Collaborative Filtering Techniques are used to predict how well a user will like an item that he/she has not rated given a set of historical preference judgements for a community of users.
Nearest Neighbor
-
For Nearest Neighbor we need to define the similarity measure and neighborhood size
User-Based Nearest Neighbor (we’ve seen this many times before!)
- The process is given a user u, generate a prediction for an item i by using the ratings for i from users in u’s neighborhood
- Where Neighbor is equal to users with similar interests
[“Insert Equation for User-Based Nearest Neighbor]
Item-Based Nearest Neighbour
- Given a user u, generate a prediction for an item i by using a weighted sum of the users u’s rating for items that are most similar to i.
[“Insert Equation for Item-Based Nearest Neighbor]
Similarity Measure
- Numerical measure of how alike two data instances are, the higher they’re the more alike the instances are
Examples of Similarity Measures include:
- Jaccard Similarity
- Cosine Similarity
- Correlation Similarity
- Gaussian RBF Similarity
Jaccard Similarity
[“Insert more Notes Here”]
Cosine Similarity
[“Insert more Notes Here”]
Correlation Similarity
[“Insert more Notes Here”]
Gaussian RBF Similarity
[“Insert more Notes Here”]
Python Example of User-Based Similarity
import pandas as pd
data = pd.read_csv('ratings.csv',header='infer')
data
Output:
| Mission Impossible | Over the Hedge | Back to the Future | Harry Potter | |
|---|---|---|---|---|
| 0 | 5 | 3 | 4 | NaN |
| 1 | 5 | 4 | 5 | 5.0 |
| 2 | 2 | 2 | 4 | 5.0 |
| 3 | 3 | 1 | 1 | 2.0 |
Using from the sklearn libary!
from sklearn.metrics import pairwise
import pandas as pd
import numpy as np
X = data.to_numpy()
user_similarity = pairwise.rbf_kernel(X[:,:3],gamma=0.2)
usim = pd.DataFrame(user_similarity)
usim
Output:
| 0 | 1 | 2 | 3 | |
|---|---|---|---|---|
| 0 | 1.000000 | 0.670320 | 0.135335 | 0.033373 |
| 1 | 0.670320 | 1.000000 | 0.060810 | 0.003028 |
| 2 | 0.135335 | 0.060810 | 1.000000 | 0.110803 |
| 3 | 0.033373 | 0.003028 | 0.110803 | 1.000000 |
avg_ratings = data.mean(axis=1) # average ratings for each user
avg_ratings
Output:
0 4.00 1 4.75 2 3.25 3 1.75 dtype: float64
import numpy as np
ratings = (data['Harry Potter'][1:] - avg_ratings[1:])*usim[0].iloc[1:]
predicted = avg_ratings[0] + (ratings.sum()*1.0/usim[0].iloc[1:].sum())
predicted
Output:
4.4919499466890604
Python Example of Item-Based Similarity
item_similarity = pairwise.rbf_kernel(X[1:,:].T,gamma=0.2)
isim = pd.DataFrame(item_similarity)
isim
Output:
| 0 | 1 | 2 | 3 | |
|---|---|---|---|---|
| 0 | 1.000000 | 0.367879 | 0.201897 | 0.135335 |
| 1 | 0.367879 | 1.000000 | 0.367879 | 0.110803 |
| 2 | 0.201897 | 0.367879 | 1.000000 | 0.670320 |
| 3 | 0.135335 | 0.110803 | 0.670320 | 1.000000 |
import numpy as np
ratings = data.iloc[0][:3].to_numpy()
simval = isim[3][:3]
prediction = (simval*ratings).sum()/simval.sum()
prediction
Output:
4.026768397265431
Another Technique: Matrix Factorization
- Matrix factorization algorithms work by decomposing the user-item interaction matrix into the product of two lower dimensionality rectangular matrices.
