Multi-TAP

Main Process

Step 0. User specific information Pre-generation

profiles/user_info_gen.py Needs following:

Amazon review dataset

Amazon item meta dataset

https://amazon-reviews-2023.github.io/

we also provided sampled review and item meta data ./data/amazon/

Step 1. Domain Description Generation

should use Openai API Key python step1_domain_desc_gen.py

Step 2. Persona Sentence Generation

should use Openai API Key

python step2_persona_gen.py
--review_dir ./data/amazon/{domain_pair}/filtered_data/f_usr_reviews
--meta_dir ./data/amazon/{domain_pair}/filtered_data/f_item_meta
--domain_desc_dir ./profiles/domain_desc/
--user_info_dir ./profiles/user_info
--system ./system_prompt/category_persona_gen.txt
--tvt_root ./data/amazon/{domain_pair}
--out_dir ./profiles/persona_sentences/

Step 3. text embedding

Object is following: User Persona Sentences text2emb/t2e_persona.py Item meta data (Domain Description || Domain Description Keywords || Item's specific category) text2emb/t2e_itm.py

Implementation

Structure of Alt_TAP recommendation system

Multi-TAP/
├── Multi_TAP_main.py
├── step1_domain_desc_gen.py
├── step2_persona_gen.py
├── config.py ## <- needs API KEY
└── model/
    └── Multi_TAP/
        └── trainer/
            ├── Multi_TAP.py
            └── trainer.py

Backbone Model

LightGCN (SIGIR, 2020) from opensource repository: https://github.com/Coder-Yu/QRec