Hi there👋.
This repository is the official implementation of the paper FIRE: Fast Incremental Recommendation with Graph Signal Processing
, which is accepted by The Web Conference 2022.
If you use the code in your work, please cite our paper.
Step 1: Check the compatibility of your python packages, we recommend you to use the following setting that has passed our test if you want to reproduce the results.
numpy == 1.19.4
pandas == 1.1.4
python == 3.6.11
scipy == 1.5.3
sklearn == 0.23.2
sparsesvd == 0.2.2
-
If you use the default datasets (
Movielens 1M
orDouban Movie
), please unzip the dataset under the directorydataset/
first; -
If you use your own dataset, you should make sure that the format of your dataset is compatible with the setting (i.e. the following variables) in
dataloader.py
sep
: the separation symbol between columns in dataset (default: '\t')header_name
: the header name of data frame generated from dataset (default: ['u', 'i', 'r', 't'])pos_type
: The type of positive interactions (default: [4.0, 5.0]).
Note: You should disable the comment of Line 23-27 in
dataloader.py
to generate a new column named 'm' when loading data if you use your own dataset.
-
For
Movielens 1M
dataset:python main.py --dataset ml1m --use_user_si --use_item_si
-
For
Douban Movie
dataset:python main.py --dataset douban_movie --num_his_month 9 --num_cur_month 1 --decay_factor 1e-8 --pri_factor 128 --alphas '[0.2,1.0,0,0.2]' --use_item_si --user_threshold 0 --item_threshold 0.6