GuideDataCF.md

A quick tutorial to create your own data for General Collaborative Filtering

The general collaborative filtering scenarios include three data files:

• train_mat.pkl: a matrix of shape [n_user, n_item] containing training interactions
• test_mat.pkl: a matrix of shape [n_user, n_item] containing testing interactions
• val_mat.pkl: a matrix of shape [n_user, n_item] containing validation interactions

All three matrices are in coo_matrix format, where each entry has a value of either 0 or 1. An entry $(i, j)=1$ indicates that there is a recorded interaction between user_i and item_j.

Suppose you have an interaction file with the following format:

[1, 1, 2, 3, 4] # user_ids
[1, 2, 1, 1, 1] # item_ids

We can read the interaction file into two Python list objects: user_ids and item_ids. Let n_user and n_item be the total number of users and items, respectively. Then we can create the coo_matrix as follows:

import numpy as np
mat = coo_matrix((np.ones(len(user_ids)), (user_ids, item_ids)), shape=[n_user, n_item])

We can also save the coo_matrix as a pickle file for later use. Here's an example of how to do it:

import pickle
with open('mat.pkl', 'wb') as fs:
	pickle.dump(mat, fs)

Note that the datasets from some other scenarios in SSLRec also use a sparse matrix to store interaction data in a similar way.