Datasets

In this document, we provide detailed descriptions of the data files for each scenario.

General Collaborative Filtering

• train_mat.pkl This file contains a SciPy sparse matrix of shape [n_user, n_item]. It represents the training set interactions, where each entry with a value greater than 1 indicates an interaction.
• valid_mat.pkl This file contains a SciPy sparse matrix of shape [n_user, n_item]. It represents the validation set interactions, where each entry with a value greater than 1 indicates an interaction.
• test_mat.pkl This file contains a SciPy sparse matrix of shape [n_user, n_item]. It represents the test set interactions, where each entry with a value greater than 1 indicates an interaction.

Sequential Recommendation

• train.tsv: This file contains the training sessions. Each line of record follows the format [session_id, item_id_seq, item_id]. The item_id_seq is the observed sequence and the item_id is the next item for supervision.
• test.tsv This file contains the testing sessions. Each line of record follows the format [session_id, item_id_seq, item_id]. The item_id_seq is the observed sequence and the item_id is the ground truth next item for testing.

Social Recommendation

• trn_mat.pkl This file contains a SciPy sparse matrix of shape [n_user, n_item]. It represents the training set interactions, where each non-zero entry indicates an interaction.
• trn_time.pkl This file contains a SciPy sparse matrix of shape [n_user, n_item] contains timestamps (matrix entry values) of interactions in the training set.
• trust_mat.pkl This file contains a SciPy sparse matrix of shape [n_user, n_user] represents social relationships, where non-zero entries indicate that two users have a social relationship.
• test_mat.pkl This file contains a SciPy sparse matrix of shape [n_user, n_item]. It represents the test set interactions, where each entry with a value greater than 1 indicates an interaction.
• category.pkl This file contains a SciPy sparse matrix of shape [n_item, n_category] represents item categories, where entries mark the categories that an item belongs to (possibly multiple).

Knowledge Graph-enhanced Recommendation

• train.txt This file contains the observed user-item interactions for training. Each line starts with a User ID, followed by the IDs of items that the user has interacted with.
• test.txt This file contains the user-item interactions for testing. Each line starts with a User ID, followed by the IDs of items that the user has interacted with.
• kg_final.txt This file contains the KG tripltes. Each line of the record follows the format [item_id, relation_id, entry_id].
• item_list.txt This file contains the item ID information, which includes org_id, remap_id and freebase_id.
• entity_list.txt This file contains the entity ID information, which includes freebase_id and remap_id.
• user_list.txt This file contains the user ID information, which includes org_id and remap_id
• relation_list.txt This file contains the relation ID information, which includes org_id and remap_id

Multi-behavior Recommendation

• train_mat_{behavoir}.pkl (e.g., train_mat_buy.pkl) This file contains a SciPy sparse matrix with a shape of [n_user, n_item], where each non-zero entry represents an interaction under a specific behavior.
• train_mat_{meta_path}.pkl (e.g.,train_mat_pv_buy.pkl) This file contains a SciPy sparse matrix with a shape of [n_user, n_item]. Each non-zero entry in the matrix indicates an interaction between a user and an item. The interactions recorded in this matrix are the intersection of multiple behaviors.
• test_mat.pkl This file contains a SciPy sparse matrix with a shape of [n_user, n_item]. It represents the interactions for evaluation in the test set, where each non-zero entry represents an interaction.
• meta_multi_single_beh_user_index_shuffle This file is a list that contains the IDs of users who exhibit specific behaviors used for the meta-training of the CML model (e.g., active users, to avoid model overfitting on noise data).
• kg.txt (from datapreprocessing.py) This file contains records of item-item relationships for the KMCLR model, where each interaction is associated with a specific behavior. Each line of the record follows the format [item_id, behavior_id, item_id], indicating the interaction between two items with the corresponding behavior.