Description
Collaborative Filtering
Description
Implement a movie recommending system using collaborative filtering.
The following files are provided and their detailed descriptions are in README.txt:
(1) ratings.csv: contains ratings matrix
(2) users.csv: contains users’ information
(3) movies.csv: contains movies’ information
(4) toBeRated.csv: contains the cells that you need to fill
For similarity, you must implement all three types of similarity below:
(1) Jaccard similarity
(2) Pearson correlation similarity
(3) Cosine similarity
Results:
(1) Use the user-based method and ratings.csv only to fill the ratings matrix to predict
ratings for required cells in toBeRated.csv.
(2) Use the item-based method and ratings.csv only to fill the ratings matrix to predict
ratings for required cells in toBeRated.csv
(3) Combing users’ and items’ information, using the best similarity type you think, to fill
the ratings matrix to predict ratings for required cells in toBeRated.csv.
Testing:
You should use multi-fold cross validation to find which type of similarity is the best. Use RMSE
as the metric.
Output Format
20,000 lines, line i has the rating for the cell denoted by line i in toBeRated.csv
For example, line 1 should contain the rating for user 3374 and movie 673.
Deliverable
The deliverable should contain two folders with 7 files, please put all the files in to a directory
named “HW5-{YOUR FIRST NAME}-{YOUR LAST NAME}”:
HW5-{YOUR FIRST NAME}-{YOUR LAST NAME}
\ — code (50%)
| — recommender.py/ recommender.cpp/ recommender.java (code) (40%)
| — README.txt (showing how to run your code, your code should take types of similarity
and input files as input arguments.) (10%)
\ — results (50%)
| — result1.csv (10%) (Results with best RMSE from three types of similarity)
| — result2.csv (10%) (Results with best RMSE from three types of similarity)
| — result3.csv (10%) (Results with best RMSE from three types of similarity)
| — results.pdf (10%)
| — a. Use a bar char to show RMSE for each method (2 item/user based method * 3
similarity measures + 1 your method = 7 methods). Describe your observations. (10%)
| — b. Briefly explain your result1/result2. Explain how you get the result3 using what
methods and which type of similarity and what information. Explain why using these
settings will give the best result. (10%)
Note: Your can submit two version of your program, one for cross validation and the other for
predication. Or, you can put them all in one program without triggering cross validation
by default. Or handle this in anyway you feel comfortable about, as long as you explain
it in README.txt
Please archive the folder and name it as “HW5-{YOUR FIRST NAME}-{YOUR LAST NAME}.zip”.
and upload it to T-square.
