코드 구조
- data/
- preprocess/
- knowledge_encoding/
- RS/
모델 돌리기
- Dataset 다운로드 : MovieLens 1M dataset 다운로드받아
data/amz/raw_data/
에 넣어줌.
- Knowledge generation ←—Gemini사용
- 따로 공개하지 않음. 이미 prompt만들어둔거(prompt.hist, prompt.item) LLM에 넣어주면 됨.
- 이미 만들어놓은 결과물 :
data/m1-1m/knowledge
(LLM prompt와 answer pair)
data/ml-1m/knowledge/item.klg
: factual knowledge about movies in MovieLens-1M.
data/ml-1m/knowledge/user.klg
: preference knowledge about users in MovieLens-1M.
-
- Knowledge encoding
- RS: in folder
RS
- Run
python run_ctr.py
for ctr task
- Run
python run_rerank.py
for reranking task
<Prompt넣어서 answer Generation>