Code for Private Recommender Systems: How Can Users Build Their Own Fair Recommender Systems without Log Data? (SDM 2022)

Last update: Aug 21, 2022

Related tags

Deep Learning private-recsys

Overview

Private Recommender Systems: How Can Users Build Their Own Fair Recommender Systems without Log Data? (SDM 2022)

We consider how a user of a web service can build their own recommender system. Many recommender systems on the Internet are still unfair/undesirable for some users, in which case the users need to leave the service or unwillingly continue to use the system. Our proposed concept, private recommender systems, provides a way for the users to resolve this dilemma.

Paper: https://arxiv.org/abs/2105.12353

💿 Dependency

$ pip install -r requirements.txt
$ sudo apt install wget unzip

🗃️ Download and Preprocess Datasets

You can download and preprocess data by the following command. It may take time.

$ bash download.sh

hetrec.npy is the Last.fm dataset. home_and_kitchen.npy is the Amazon dataset. adult_*.npy and adult_*.npz are the Adult dataset.

🧪 Evaluation

$ python evaluate.py --data 100k --prov cosine --sensitive popularity
$ python evaluate.py --data 100k --prov bpr --sensitive popularity
$ python evaluate.py --data 100k --prov cosine --sensitive old
$ python evaluate.py --data 100k --prov bpr --sensitive old
$ python evaluate.py --data hetrec --prov bpr --sensitive popularity
$ python evaluate.py --data home --prov bpr --sensitive popularity
$ python evaluate_adult.py

100k is the MovieLens 100k dataset. hetrec is the LastFM dataset. home is the Amazon Home and Kitchen dataset.
--prov specifys the algorithm of the service provider's recommender system.
--sensitive specifyies the sensitive attribute. old is available only for the MovieLens datasets.

These scripts compute the sums of recalls, NDCGs, least ratios, and entropies for all users. Be sure to divide these values by the number of users to obtain the average values.

When your environment supports multi-processing, run, for example, the following commands to speed up the computation (with background executions):

$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 0
$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 1
$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 2
$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 3
$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 4
$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 5
$ python evaluate.py --data 100k --prov cosine --sensitive popularity --split 7 --block 6
$ python summary.py 7

🖋️ Citation

@inproceedings{sato2022retrieving,
  author    = {Ryoma Sato},
  title     = {Private Recommender Systems: How Can Users Build Their Own Fair Recommender Systems without Log Data?},
  booktitle = {Proceedings of the 2022 {SIAM} International Conference on Data Mining, {SDM}},
  year      = {2022},
}

Code for Private Recommender Systems: How Can Users Build Their Own Fair Recommender Systems without Log Data? (SDM 2022)

Related tags

Overview

Private Recommender Systems: How Can Users Build Their Own Fair Recommender Systems without Log Data? (SDM 2022)

💿 Dependency

🗃️ Download and Preprocess Datasets

🧪 Evaluation

🖋️ Citation

Owner

joisino

PyTorch implementation for our NeurIPS 2021 Spotlight paper "Long Short-Term Transformer for Online Action Detection".

Simple tool to combine(merge) onnx models. Simple Network Combine Tool for ONNX.

"MST++: Multi-stage Spectral-wise Transformer for Efficient Spectral Reconstruction" (CVPRW 2022) & (Winner of NTIRE 2022 Challenge on Spectral Reconstruction from RGB)

Key information extraction from invoice document with Graph Convolution Network

Makes patches from huge resolution .svs slide files using openslide

Explanatory Learning: Beyond Empiricism in Neural Networks

Pytorch code for paper "Image Compressed Sensing Using Non-local Neural Network" TMM 2021.

ICLR 2021 i-Mix: A Domain-Agnostic Strategy for Contrastive Representation Learning

Source code for CAST - Crisis Domain Adaptation Using Sequence-to-sequence Transformers (Accepted to ISCRAM 2021, CorePaper).

Human Pose Detection on EdgeTPU

ICCV2021 Paper: AutoShape: Real-Time Shape-Aware Monocular 3D Object Detection

The Few-Shot Bot: Prompt-Based Learning for Dialogue Systems

Additional environments compatible with OpenAI gym

Face recognition project by matching the features extracted using SIFT.

This package implements the algorithms introduced in Smucler, Sapienza, and Rotnitzky (2020) to compute optimal adjustment sets in causal graphical models.

This MVP data web app uses the Streamlit framework and Facebook's Prophet forecasting package to generate a dynamic forecast from your own data.

The Fundamental Clustering Problems Suite (FCPS) summaries 54 state-of-the-art clustering algorithms, common cluster challenges and estimations of the number of clusters as well as the testing for cluster tendency.

git《Tangent Space Backpropogation for 3D Transformation Groups》(CVPR 2021) GitHub:1]

An implementation of the methods presented in Causal-BALD: Deep Bayesian Active Learning of Outcomes to Infer Treatment-Effects from Observational Data.