CSE 158/258, Fall 2022: Assignment 1

Instructions

In this assignment you will build recommender systems to make predictions related to book reviews from Goodreads.

Submissions will take the form of prediction files uploaded to gradescope, where their test set performance will be evaluated on a leaderboard.  Most of your grade will be determined by ‘absolute’ cutoffs; the  leaderboard  ranking  will  only  determine  enough  of your  assignment  grade  to  make  the assignment FUN.

The assignment is due Monday,  Nov  14, though make sure you upload solutions to the leaderboard regularly.

You should submit two files:

writeup .txt  a brief, plain-text description of your solutions to each task; please prepare this adequately in advance of the submission deadline; this is only intended to help us follow your code and does not need to  be detailed.

assignment1 .py A python file containing working code for your solutions.  The autograder will not  execute your code; this file is required so that we can assign partial grades in the event of incorrect solutions, check for plagiarism, etc.  Your solution should clearly document which sections correspond to each question and answer. We may occasionally run code to confirm that your outputs match submitted answers, so please ensure that your code generates the submitted answers.1

Along with two files corresponding to your predictions:2

predictions Read .csv,predictions Category .csv,predictions Rating .csv Files containing your predic- tions for each (test) instance (you should submit two of the above three files).  The provided baseline code demonstrates how to generate valid output files.

This assignment should be completed individually. To begin, download the files for this assignment from: http://cseweb.ucsd.edu/classes/fa22/cse258-a/files/assignment1.tar.gz

Files

train Interactions.csv.gz 200,000 ratings to be used for training.  This data should be used for the ‘read prediction’ (both classes) and ‘rating prediction’ (CSE258 only) tasks.  It is not necessary to use all ratings for training, for example if doing so proves too computationally intensive.

userID The ID of the user. This is a hashed user identifier from Goodreads.

bookID The ID of the book. This is a hashed book identifier from Goodreads.

rating The star rating of the user’s review.

train Category.json.gz Training data for the category prediction task (CSE158 only).  This file is json formatted, and contains the following fields:

n votes The number of‘likes’this review has received.

review id A hashed identifier for this review.

user id A hashed identifier for the user.

review text Text of the review.

rating Rating of the book.

genreID A numeric label associated with the genre.

genre A string version of the genre.

test Category.json.gz Test data associated with the category prediction task. This data has the same format as above, with the‘genre’and ‘genreID’labels hidden.

pairs Read.csv Pairs on which you are to predict whether a book was read (both classes).

pairs Category.csv Pairs (userID and reviewID) on which you are to predict the category of a book (CSE158 only).

pairs Rating.csv Pairs (userIDs and bookIDs) on which you are to predict ratings (CSE258 only). baselines.py A simple baseline for each task, described below.

Please do not try to collect these reviews from Goodreads, or to reverse-engineer the hashing function I used to anonymize the data.  Doing so will not be easier than successfully completing the assignment.  We will execute code for any solution suspected of violating the competition rules to confirm that it generates valid output; all code will be run through a plagiarism detector.

Tasks

You are expected to complete the following tasks:

Read prediction  (both classes) Predict given a (user,book) pair from ‘pairs Read.csv’ whether the user would read the book (0 or 1). Accuracy will be measured in terms of the categorization accuracy (fraction of correct predictions). The test set has been constructed such that exactly 50% of the pairs correspond to read books and the other 50% do not.

Category prediction  (CSE158 only) Predict the category of a book from a review. Five categories are used for this task, which can be seen in the baseline program, namely Children’s, Comics/Graphic Novels, Fantasy, Mystery/Thriller, and Romance.   Performance will be measured in terms of the fraction of correct classifications.

Rating prediction  (CSE258 only) Predict people’s star ratings as accurately as possible, for those (user,item) pairs in‘pairs Rating.txt’. Accuracy will be measured in terms of the mean-squared error  (MSE).

A competition page has been set up on gradescope to keep track of your results compared to those of other members of the class. The leaderboard will show your results on half of the test data, but your ultimate score will depend on your predictions across the whole dataset.

Grading and Evaluation

This assignment is worth 25% of your grade.  You will be graded on the following aspects.  Each of the two tasks is worth 10 marks (i.e., 10% of your grade), plus 5 marks for the written report.

• Your ability to obtain a solution which outperforms the leaderboard baselines on the unseen portion  of the test data (5 marks for each task).  Obtaining full marks requires a solution which is substantially better than baseline performance.

• Your ranking for each of the tasks compared to other students in the class (3 marks for each task).

• Obtain a solution which outperforms the baselines on the seen portion of the test data (i.e., the leader- board). This is a consolation prize in case you overfit to the leaderboard. (2 mark for each task).

Finally, your written report should describe the approaches you took to each of the tasks. To obtain good performance, you should not need to invent new approaches (though you are more than welcome to!)  but rather you will be graded based on your decision to apply reasonable approaches to each of the given tasks (5 marks total).

Baselines

Simple baselines have been provided for each of the tasks. These are included in‘baselines.py’among the files above.  They are mostly intended to demonstrate how the data is processed and prepared for submission to gradescope. These baselines operate as follows:

Read prediction Find the most popular books that account for 50% of interactions in the training data. Return‘1’whenever such a book is seen at test time,‘0’otherwise.