ColdNAS: Search to Modulate for User Cold-Start Recommendation. WebConf 2023

Usage:

Data source:

- For Last.FM, we use the data provided by TaNP (https://github.com/IIEdm/TaNP).(Already provided in .data.zip)

- For MovienLens-1M, we use the data provided by MeLU (https://github.com/hoyeoplee/MeLU).

- For BookCrossing, we use the data downloaded from (http://www2.informatik.uni-freiburg.de/~cziegler/BX/)

Requirements:

 --python 3.7.0 --torch 1.7.1 --cuda 11.0 --numpy 1.19.3

Run on Last.FM:

- Unzip data 'data.zip'

- Search by running 'search.py'

- Select top-K by output alpha and change the model in ‘model_lfm4.py’

- Train and evaluate the searched model by running 'evaluate.py'

update 2024.12.4

- 1-normalize during search