🚀 Semantic Codebook Learning for Dynamic Recommendation Models (MM 2024)

---

PyTorch implementation of Semantic Codebook Learning for Dynamic Recommendation Models on Sequential Recommendation task based on DIN, GRU4Rec, SASRec, BERT4Rec, APG, DUET.

Data and Data Description

Preprocessed dataset Arts can be found in link

/id:

• item(id): train.txt and test.txt are followed by conventional data processing method in sequential recommendation task
• semantic(id): Classification information obtained by the four models based on item id

/id_image:

• semantic(id + image)

/id_text:

• semantic(id + text)

/id_text_image:

• semantic(id + text + image)

/image:

• semantic(image)

/text:

• semantic(text)

/text_image:

• semantic(text + image)

Data Preprocessing（such as Amazon_arts_subset）

• Get the semantic id classified by item id embedding
  • Execute python generate_dataset.py under the path ./data/Amazon_arts_subset to get train.txt and test.txt
  • Execute bash _0_0_train.sh under ./DUET_full/scripts to obtain best_auc.pkl (semantic parameter in codebook) corresponding to the four models, type should be changed to _0_func_duet_train
  • Copy the folders obtained above to ./DC-DRP/checkpoint/SAVE/amazon_arts_subset_model/, and rename the folder to duet
  • Execute bash _2_duet_grad_pred.sh under ./DC-DRP/scripts to save the corresponding item id embedding
  • Execute python _func_item_to_category.py under ./DC-DRP/scripts to preprocess to obtain the train.txt and test.txt of the semantic information of the item id
• Get the semantic id of the id+text+image semantic
  • Execute python _func_multimodel_category.py under the path ./DC-DRP/scripts to get the train.txt and test.txt corresponding to the four models
  • Copy the results into ./data/Amazon_arts_subset/id_text_image
• semantic(id + image), semantic(id + text), semantic(text), semantic(image) are similar to above

semantic to parameter

• Add the configuration json file of the semantic data set under ./config_fusion, where id_vocab needs to be changed to the number of categories
• Execute bash _0_0_train.sh under ./DUET_full/scripts, change type to _0_func_duet_train

Training Fusion + VQ-VAE

• Add the configuration json file of the data set under ./config_fusion, where item_id_vocab is the number of items and category_id_vocab is the number of categories
• Execute bash _0_0_train.sh under ./DUET_full/scripts, and change type to _0_func_duet_fusion_vqvae_train