coder-yu / qrec

QRec: A Python Framework for quick implementation of recommender systems (TensorFlow Based)

Introduction

QRec is a Python framework for recommender systems (Supported by Python 3.7.4 and Tensorflow 1.14+) in which a number of influential and newly state-of-the-art recommendation models are implemented. QRec has a lightweight architecture and provides user-friendly interfaces. It can facilitate model implementation and evaluation.
Founder and principal contributor: @Coder-Yu
Other contributors: @DouTong @Niki666 @HuXiLiFeng @BigPowerZ @flyxu
Supported by: @AIhongzhi (A/Prof. Hongzhi Yin, UQ), @mingaoo (A/Prof. Min Gao, CQU)

We also provide Pytorch implementations of some models in another library. Please click here

What's New

31/03/2022 - SimGCL proposed in our SIGIR'22 paper has been added.
12/10/2021 - BUIR proposed in SIGIR'21 paper has been added.
30/07/2021 - We have transplanted QRec from py2 to py3.
07/06/2021 - SEPT proposed in our KDD'21 paper has been added.
16/05/2021 - SGL proposed in SIGIR'21 paper has been added.
16/01/2021 - MHCN proposed in our WWW'21 paper has been added.
22/09/2020 - DiffNet proposed in SIGIR'19 has been added.
19/09/2020 - DHCF proposed in KDD'20 has been added.
29/07/2020 - ESRF proposed in my TKDE paper has been added.
23/07/2020 - LightGCN proposed in SIGIR'20 has been added.
17/09/2019 - NGCF proposed in SIGIR'19 has been added.
13/08/2019 - RSGAN proposed in ICDM'19 has been added.
09/08/2019 - Our paper is accepted as full research paper by ICDM'19.
20/02/2019 - IRGAN proposed in SIGIR'17 has been added.
12/02/2019 - CFGAN proposed in CIKM'18 has been added.

Architecture

Workflow

Features

• Cross-platform: QRec can be easily deployed and executed in any platforms, including MS Windows, Linux and Mac OS.
• Fast execution: QRec is based on Numpy, Tensorflow and some lightweight structures, which make it run fast.
• Easy configuration: QRec configs recommenders with a configuration file and provides multiple evaluation protocols.
• Easy expansion: QRec provides a set of well-designed recommendation interfaces by which new algorithms can be easily implemented.

Requirements

• gensim==4.1.2
• joblib==1.1.0
• mkl==2022.0.0
• mkl_service==2.4.0
• networkx==2.6.2
• numba==0.53.1
• numpy==1.20.3
• scipy==1.6.2
• tensorflow==1.14.0

Usage

There are two ways to run the recommendation models in QRec:

• 1.Configure the xx.conf file in the directory named config. (xx is the name of the model you want to run)
• 2.Run main.py.

Or

• Follow the codes in snippet.py.

For more details, we refer you to the handbook of QRec.

Configuration

Essential Options

Entry	Example	Description
ratings	D:/MovieLens/100K.txt	Set the file path of the dataset. Format: each row separated by empty, tab or comma symbol.
social	D:/MovieLens/trusts.txt	Set the file path of the social dataset. Format: each row separated by empty, tab or comma symbol.
ratings.setup	-columns 0 1 2	-columns: (user, item, rating) columns of rating data are used.
social.setup	-columns 0 1 2	-columns: (trustor, trustee, weight) columns of social data are used.
mode.name	UserKNN	name of the recommendation model.
evaluation.setup	-testSet ./dataset/test.txt	Main option: -testSet, -ap, -cv (choose one of them) -testSet path/to/test/file (need to specify the test set manually) -ap ratio (ap means that the ratings are automatically partitioned into training set and test set, the number is the ratio of the test set. e.g. -ap 0.2) -cv k (-cv means cross validation, k is the number of the fold. e.g. -cv 5) -predict path/to/user list/file (predict for a given list of users without evaluation; need to mannually specify the user list file (each line presents a user)) Secondary option:-b, -p, -cold, -tf, -val (multiple choices) -val ratio (model test would be conducted on the validation set which is generated by randomly sampling the training dataset with the given ratio.) -b thres （binarizing the rating values. Ratings equal or greater than thres will be changed into 1, and ratings lower than thres will be left out. e.g. -b 3.0） -p (if this option is added, the cross validation wll be executed parallelly, otherwise executed one by one) -tf (model training will be conducted on TensorFlow (only applicable and needed for shallow models)) -cold thres (evaluation on cold-start users; users in the training set with rated items more than thres will be removed from the test set)
item.ranking	off -topN -1	Main option: whether to do item ranking -topN N1,N2,N3...: the length of the recommendation list. *QRec can generate multiple evaluation results for different N at the same time
output.setup	on -dir ./Results/	Main option: whether to output recommendation results -dir path: the directory path of output results.

Memory-based Options

similarity	pcc/cos	Set the similarity method to use. Options: PCC, COS;
num.neighbors	30	Set the number of neighbors used for KNN-based algorithms such as UserKNN, ItemKNN.

Model-based Options

num.factors	5/10/20/number	Set the number of latent factors
num.max.epoch	100/200/number	Set the maximum number of epoch for iterative recommendation algorithms.
learnRate	-init 0.01 -max 1	-init initial learning rate for iterative recommendation algorithms; -max: maximum learning rate (default 1);
reg.lambda	-u 0.05 -i 0.05 -b 0.1 -s 0.1	-u: user regularizaiton; -i: item regularization; -b: bias regularizaiton; -s: social regularization

Implement Your Model

• 1.Make your new algorithm generalize the proper base class.
• 2.Reimplement some of the following functions as needed.

          - readConfiguration()
          - printAlgorConfig()
          - initModel()
          - trainModel()
          - saveModel()
          - loadModel()
          - predictForRanking()
          - predict()

For more details, we refer you to the handbook of QRec.

Implemented Algorithms

       

Rating prediction	Paper
SlopeOne	Lemire and Maclachlan, Slope One Predictors for Online Rating-Based Collaborative Filtering, SDM'05.
PMF	Salakhutdinov and Mnih, Probabilistic Matrix Factorization, NIPS'08.
SoRec	Ma et al., SoRec: Social Recommendation Using Probabilistic Matrix Factorization, SIGIR'08.
SVD++	Koren, Factorization meets the neighborhood: a multifaceted collaborative filtering model, SIGKDD'08.
RSTE	Ma et al., Learning to Recommend with Social Trust Ensemble, SIGIR'09.
SVD	Y. Koren, Collaborative Filtering with Temporal Dynamics, SIGKDD'09.
SocialMF	Jamali and Ester, A Matrix Factorization Technique with Trust Propagation for Recommendation in Social Networks, RecSys'10.
EE	Khoshneshin et al., Collaborative Filtering via Euclidean Embedding, RecSys'10.
SoReg	Ma et al., Recommender systems with social regularization, WSDM'11.
LOCABAL	Tang, Jiliang, et al. Exploiting local and global social context for recommendation, AAAI'13.
SREE	Li et al., Social Recommendation Using Euclidean embedding, IJCNN'17.
CUNE-MF	Zhang et al., Collaborative User Network Embedding for Social Recommender Systems, SDM'17.

                       

Item Ranking	Paper
BPR	Rendle et al., BPR: Bayesian Personalized Ranking from Implicit Feedback, UAI'09.
WRMF	Yifan Hu et al.Collaborative Filtering for Implicit Feedback Datasets, KDD'09.
SBPR	Zhao et al., Leveraing Social Connections to Improve Personalized Ranking for Collaborative Filtering, CIKM'14
ExpoMF	Liang et al., Modeling User Exposure in Recommendation, WWW''16.
CoFactor	Liang et al., Factorization Meets the Item Embedding: Regularizing Matrix Factorization with Item Co-occurrence, RecSys'16.
TBPR	Wang et al. Social Recommendation with Strong and Weak Ties, CIKM'16'.
CDAE	Wu et al., Collaborative Denoising Auto-Encoders for Top-N Recommender Systems, WSDM'16'.
DMF	Xue et al., Deep Matrix Factorization Models for Recommender Systems, IJCAI'17'.
NeuMF	He et al. Neural Collaborative Filtering, WWW'17.
CUNE-BPR	Zhang et al., Collaborative User Network Embedding for Social Recommender Systems, SDM'17'.
IRGAN	Wang et al., IRGAN: A Minimax Game for Unifying Generative and Discriminative Information Retrieval Models, SIGIR'17'.
SERec	Wang et al., Collaborative Filtering with Social Exposure: A Modular Approach to Social Recommendation, AAAI'18'.
APR	He et al., Adversarial Personalized Ranking for Recommendation, SIGIR'18'.
IF-BPR	Yu et al. Adaptive Implicit Friends Identification over Heterogeneous Network for Social Recommendation, CIKM'18'.
CFGAN	Chae et al. CFGAN: A Generic Collaborative Filtering Framework based on Generative Adversarial Networks, CIKM'18.
NGCF	Wang et al. Neural Graph Collaborative Filtering, SIGIR'19'.
DiffNet	Wu et al. A Neural Influence Diffusion Model for Social Recommendation, SIGIR'19'.
RSGAN	Yu et al. Generating Reliable Friends via Adversarial Learning to Improve Social Recommendation, ICDM'19'.
LightGCN	He et al. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation, SIGIR'20.
DHCF	Ji et al. Dual Channel Hypergraph Collaborative Filtering, KDD'20.
ESRF	Yu et al. Enhancing Social Recommendation with Adversarial Graph Convlutional Networks, TKDE'20.
MHCN	Yu et al. Self-Supervised Multi-Channel Hypergraph Convolutional Network for Social Recommendation, WWW'21.
SGL	Wu et al. Self-supervised Graph Learning for Recommendation, SIGIR'21.
SEPT	Yu et al. Socially-Aware Self-supervised Tri-Training for Recommendation, KDD'21.
BUIR	Lee et al. Bootstrapping User and Item Representations for One-Class Collaborative Filtering, SIGIR'21.
SimGCL	Yu et al. Are Graph Augmentations Necessary? Simple Graph Contrastive Learning for Recommendation, SIGIR'22.

Related Datasets

   

Data Set	Basic Meta					User Context
	Users	Items	Ratings (Scale)		Density	Users	Links (Type)
Ciao [1]	7,375	105,114	284,086	[1, 5]	0.0365%	7,375	111,781	Trust
Epinions [2]	40,163	139,738	664,824	[1, 5]	0.0118%	49,289	487,183	Trust
Douban [3]	2,848	39,586	894,887	[1, 5]	0.794%	2,848	35,770	Trust
LastFM [4]	1,892	17,632	92,834	implicit	0.27%	1,892	25,434	Trust
Yelp [5]	19,539	21,266	450,884	implicit	0.11%	19,539	864,157	Trust
Amazon-Book [6]	52,463	91,599	2,984,108	implicit	0.11%	-	-	-

Reference

[1]. Tang, J., Gao, H., Liu, H.: mtrust:discerning multi-faceted trust in a connected world. In: International Conference on Web Search and Web Data Mining, WSDM 2012, Seattle, Wa, Usa, February. pp. 93–102 (2012)

[2]. Massa, P., Avesani, P.: Trust-aware recommender systems. In: Proceedings of the 2007 ACM conference on Recommender systems. pp. 17–24. ACM (2007)

[3]. G. Zhao, X. Qian, and X. Xie, “User-service rating prediction by exploring social users’ rating behaviors,” IEEE Transactions on Multimedia, vol. 18, no. 3, pp. 496–506, 2016.

[4]. Iván Cantador, Peter Brusilovsky, and Tsvi Kuflik. 2011. 2nd Workshop on Information Heterogeneity and Fusion in Recom- mender Systems (HetRec 2011). In Proceedings of the 5th ACM conference on Recommender systems (RecSys 2011). ACM, New York, NY, USA

[5]. Yu et al. Self-Supervised Multi-Channel Hypergraph Convolutional Network for Social Recommendation, WWW'21.

[6]. He et al. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation, SIGIR'20.

Acknowledgment

This project is supported by the Responsible Big Data Intelligence Lab (RBDI) at the school of ITEE, University of Queensland, and Chongqing University.

If our project is helpful to you, please cite one of these papers.

@inproceedings{yu2021socially,
title={Socially-aware self-supervised tri-training for recommendation},
author={Yu, Junliang and Yin, Hongzhi and Gao, Min and Xia, Xin and Zhang, Xiangliang and Viet Hung, Nguyen Quoc},
booktitle={Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining},
pages={2084--2092},
year={2021}
}

@inproceedings{yu2021self,
title={Self-Supervised Multi-Channel Hypergraph Convolutional Network for Social Recommendation},
author={Yu, Junliang and Yin, Hongzhi and Li, Jundong and Wang, Qinyong and Hung, Nguyen Quoc Viet and Zhang, Xiangliang},
booktitle={Proceedings of the Web Conference 2021},
pages={413--424},
year={2021}
}

qrec's Issues

您好，想请教下关于CFGAN的问题

您好，我看到您CFGAN实现中，在predictForRanking函数里对于测试集中那些没有出现在训练集的用户预测项目排名时，直接使用的是用户全局平均评分作为预测值，我想问一下这样做是不是不太合理啊，还是大部分的推荐算法都是这样做的呢？求指教。
我对测试集的使用有点不太懂，测试集里的用户应该大部分是没有出现在训练集中吧，如果直接使用用户的全局评分作为预测值，那算法不是相当于对大部分测试数据没有起到作用吗？

SocialMF update rule for P

I was looking at code for SocialMF and I saw that for updating self.P[u] the following rule is used: self.P[u] += self.lRate * (error * q - self.regU * p - self.regS * relationLoss), however I checked the paper and saw that another term is missing

SocialMF

Hello,
I'm looking at the code of SocialMF.

I think that there is an error in this line:
relationLoss = p - fPred / denom
since p is the copy of the vector P[u] in the loop before.

Should it be corrected with
relationLoss = P[u] - fPred / denom
?

请问在python3.6下可以使用吗？

fold这个参数似乎没用过，一直都是默认的[1]

在所有的Recommender中的初始化函数中都有fold参数
def init(self,conf,trainingSet=None,testSet=None,relation=list(),fold='[1]'):

只有在RecQ中使用到了recommender = self.config['recommender']+'(self.config,self.trainingData,self.testData,self.relation)'

似乎没必要使用fold

您好，在使用新的数据集（Epinions）训练PMF 的tf版本时，效果很差

训练过程中loss一直没有收敛，最后的rmse值到了3.多（但是使用相同的数据集，用Python版本训练时，rmse值只有一点多），请问是哪里的原因呢？

随机梯度下降 socialmf

我觉得按照论文公式应该是下面注释的 但是 你的更好 为啥呢？

saveModel在所有的Recommender中都没有实现，是都没有实现保存功能吗

When i use BasicMF,it has a bug,NameError: global name 'exit' is not defined

please enter the num of the algorithm to run it:3
loading training data...
preprocessing...
Algorithm: BasicMF
Ratings dataset: D:\Documents\GitHub\RecQ\dataset\FilmTrust\trainset.txt
Training set size: (user count: 1466, item count 1822, record count: 25437)
Test set size: (user count: 1228, item count 837, record count: 6414)

Reduced Dimension: 10
Maximum Iteration: 30
Regularization parameter: regU 0.050, regI 0.050, regB 0.100

Initializing model [1]...
Building Model [1]...
BasicMF [1] iteration 1: loss = 4504.4702, delta_loss = -4504.47023 learning_Rate = 0.10000 MAE:0.94468 RMSE:1.26997
BasicMF [1] iteration 2: loss = 3045.5222, delta_loss = 1458.94800 learning_Rate = 0.10000 MAE:0.82719 RMSE:1.10244
BasicMF [1] iteration 3: loss = 2488.2282, delta_loss = 557.29403 learning_Rate = 0.10500 MAE:0.79605 RMSE:1.05366
BasicMF [1] iteration 4: loss = 2083.2779, delta_loss = 404.95032 learning_Rate = 0.11025 MAE:0.76532 RMSE:1.01478
BasicMF [1] iteration 5: loss = 1695.7569, delta_loss = 387.52094 learning_Rate = 0.11576 MAE:0.74591 RMSE:0.99706
BasicMF [1] iteration 6: loss = 1451.4868, delta_loss = 244.27015 learning_Rate = 0.12155 MAE:0.75029 RMSE:1.00164
BasicMF [1] iteration 7: loss = 1379.8544, delta_loss = 71.63238 learning_Rate = 0.12763 MAE:0.77166 RMSE:1.02828
Loss = NaN or Infinity: current settings does not fit the recommender! Change the settings and try again!
D:\Documents\GitHub\RecQ\algorithm\rating\BasicMF.py:16: RuntimeWarning: overflow encountered in double_scalars
self.loss += error2
D:\Documents\GitHub\RecQ\algorithm\rating\BasicMF.py:21: RuntimeWarning: overflow encountered in multiply
self.P[u] += self.lRateerrorq
D:\Documents\GitHub\RecQ\algorithm\rating\BasicMF.py:16: RuntimeWarning: invalid value encountered in double_scalars
self.loss += error2
Traceback (most recent call last):

File "", line 1, in
runfile('D:/Documents/GitHub/RecQ/main/main.py', wdir='D:/Documents/GitHub/RecQ/main')

File "C:\ProgramData\Anaconda2\lib\site-packages\spyder\utils\site\sitecustomize.py", line 710, in runfile
execfile(filename, namespace)

File "C:\ProgramData\Anaconda2\lib\site-packages\spyder\utils\site\sitecustomize.py", line 86, in execfile
exec(compile(scripttext, filename, 'exec'), glob, loc)

File "D:/Documents/GitHub/RecQ/main/main.py", line 117, in
recSys.execute()

File "RecQ.py", line 121, in execute
eval(recommender).execute()

File "D:\Documents\GitHub\RecQ\baseclass\Recommender.py", line 265, in execute
self.buildModel()

File "D:\Documents\GitHub\RecQ\algorithm\rating\BasicMF.py", line 26, in buildModel
if self.isConverged(iteration):

File "D:\Documents\GitHub\RecQ\baseclass\IterativeRecommender.py", line 80, in isConverged
exit(-1)

你好，使用数据epinions时，出现错误？-header参数要怎么设置？？

支持python 3.x吗?

可以在py 3.6+跑吗?

Typle Error

Hello professor!thanks for your code. when I run main.py and choose the algorithm . it always has the same question:
Traceback (most recent call last):
File "C:/Users/Administrator/Desktop/新建文件夹/RecQ/RecQ-master/main/main.py", line 67, in
recSys.execute()
File "C:\Users\Administrator\Desktop\新建文件夹\RecQ\RecQ-master\main\RecQ.py", line 100, in execute
self.measure = [dict.setdafault(m)[i] for i in range(1,k+1)]
File "C:\Users\Administrator\Desktop\新建文件夹\RecQ\RecQ-master\main\RecQ.py", line 100, in
self.measure = [dict.setdafault(m)[i] for i in range(1,k+1)]
AttributeError: type object 'dict' has no attribute 'setdafault'
How can I solve it?looking for your reply.

您好，请问SBPR的代码是您复现的还是有源码啊？

如题，我在论文里好像没有发现提供源码链接，所以是您自己复现的工作吗？

CFGAN结果复现

请问作者在做CFGAN试验时，有没有复现论文的结果？

问个问题 关于tensorflow的

看到有些代码支持tensorflow了 没咋接触过 请问tensorflow怎么debug呢?
意思就是说关于tensorflow的代码写错了 我怎么去发现啊

我用cofactor测试 evalRanking 得到的recall MAP跟论文相比好小 请问是怎么回事啊

ModuleNotFoundError: No module named 'evaluation.dataSplit'; 'evaluation' is not a package

ModuleNotFoundError: No module named 'evaluation.dataSplit'; 'evaluation' is not a package

NDCG measurement

Hi,
About the NDCG method in your code, the base of log is 'e'.
I have checked some papers, the base they used was ‘2’. Will this difference make any influence?
Thank you!

Reference: https://docs.python.org/2/library/math.html

如何安装运行

您好，我想请教一下如何安装运行这个库，试过在anaconda上安装，但是失败了，希望能解答一下

On the implementation of CFGAN

Dear researchers:

Thank you so much for your interest in my paper, "CFGAN: A Generic Collaborative Filtering Framework based on Generative Adversarial Networks"
I just have found your implementation of my work, and it is really good.
However, I found several differences from my implementation of CFGAN.
I summarized it at CFGAN's project page: [http://www.agape.hanyang.ac.kr/cfgan]

Thank you again for your interest and very good implementation.
Have a good day!

Best,
Dong-Kyu

ImportError

你好~,我在anoconda2.7下进入main文件夹运行main.py,选择ItemKNN,发现没法import ItemKNN,
ImportError: No module named ItemKNN
大意是在main文件夹运行没法执行 from algorithm.rating.ItemKNN import ItemKNN
请问正确的运行姿势是什么...

修改可视化配置后，display.py中self.dao = RatingDAO(conf, ratingData)报错

感谢大佬们的工作，但是弱弱的问一句，你们可视化这个模块真的是测试过了吗，或在是修改之后忘了测试？display.py中self.dao = RatingDAO(conf, ratingData)，而rating中的RatingDAO这个类需要config, trainingSet, testSet三个参数，谢谢

支持 膜拜 能实现一下 Hidden Factors and Hidden Topics: Understanding Rating Dimensions with Review Text(ACM recsys 13)

Is there a bug in ITEMKNN?

I think in RecQ/algorithm/rating/ItemKNN.py 61rows about " if i1 <> i2" ? what does this means?

KeyError: True when use euclidean method in qmath.py

In similarity method and euclidean method, the parameters x1 and x2 are all dicts but not arrays, when they pass to common method, they will get wrong, I think it is a bug.

Graph Neural Network for social recommendaiton

As you updated Deep Graph Collaborative Filtering recently, have you tried to implement another state-of-the-art algorithm GraphRec from the paper Graph Neural Network for Social Recommendation in RecQ?

I met some problems when implementing GraphRec as getting worse results than reported in the paper. I'm wondering if it's because my implementation is wrong or GraphRec doesn't work as reported.

数据

你好，数据集是不是要预先处理后才能使用你们的项目？还是项目有相关的代码划分数据集？

请问一下agr是哪一篇论文的模型呢

LOCABAL

Hello,
I can execute all the algorithms but not locabal. When I select locabal, it says to me "no module named LOCABAL".

Can you help me, please?

Question about IF-BPR: parameter batch_size is invalid!

I downloaded the ratings_data.txt and trust_data.txt from HERE, deleted the first line and placed it in dir dataset.

Run a4(IF-BPR) with this configuration file:

ratings=../dataset/epinions/ratings_data.txt
social=../dataset/epinions/trust_data.txt
ratings.setup=-columns 0 1 2
social.setup=-columns 0 1 2
recommender=IF_BPR
evaluation.setup=-cv 5 -b 1
item.ranking=on -topN 10
num.factors=20
num.max.iter=30
IF_BPR=-T 2 -L 20 -l 20 -w 5 -k 150 -a 0.2 -ep 5 -neg 5 -r 0.03
learnRate=-init 0.05 -max 1
reg.lambda=-u 0.01 -i 0.01 -b 0.01 -s 0.2
output.setup=on -dir ../results/

But I have encountered this problem：

please enter the num of the algorithm to run it:a4
loading training data...
loading social data...
preprocessing...
parameter batch_size is invalid!
parameter batch_size is invalid!
parameter batch_size is invalid!
parameter batch_size is invalid!
parameter batch_size is invalid!
Traceback (most recent call last):
  File "main.py", line 150, in <module>
    recSys.execute()
  File "/home/name/Desktop/RecQ/main/RecQ.py", line 100, in execute
    self.measure = [dict(m)[i] for i in range(1,k+1)]
KeyError: 1

Excuse me, may I ask which step did I not do well??

Epinions数据集运行IF-BPR算法时出现如下错误

Computing precision and recall of LOCABAL

Hello,
I'm trying to test LOCABAL. If I compute MAE and RMSE I obtain reasonable results. But if I ask to the framework to compute precision@10, recall@10, etc. I obtain very low values like the following ones:

The result of LOCABAL [1]:
Top 10
Precision:2.36053190652e-05
Recall:4.52903641589e-05
F1:3.10351373706e-05
MAP:1.20828946366e-05
NDCG:3.38163865285e-05

The result of LOCABAL [1]:
MAE:0.843245316305
RMSE:1.07930190638

Can you help me, please?

请问没有完整的实用指南吗？

有个小bug

config目录下的SoReg.conf第6行testSet.txt应该写成testset.txt。
另外，SoReg.py里的迭代公式只考虑了Followee� ,但没有考虑Followeer，是不是缺少一项，我看原论文中迭代公式考虑了两种情况。
还有由SoReg衍生的LOCABAL算法，看到你们有一个commit实现了，但后来又删除了，为什么不保留啊？

SoRec的TF版本

您好，我自己在尝试写SoRec的TF版本的遇到一个问题，因为损失函数涉及到评分矩阵和用户关系两个矩阵，不知道按批优化的时候能不能同时从这两个矩阵里抽取样本来进行优化

how to generate the data?

Hi,

Is there any tool to generate the data in the required format?
in particular, how to create the relationship between users?

Thanks,

Run 的时候有bug

LOCABAL梯度下降式子疑问

您好，我看到您写的LOCABAL算法了，我对您用的梯度下降公式有点疑问，您在哪儿找到的这个公式，我在原论文里没找到啊，然后我在其他论文里找到了部分，a synthetic approach for recommendation combining ratings social relations and reviews. 这个论文里使用了LOCABAL算法，他给出的迭代公式和您代码里写的还是有很大不同的。这里我还是有点疑惑的。

To apply ItemKNN to ranking problem.

Excuse me,
I run the main.py，and select the ItemKNN model.In my experiment，i want to apply itemKNN to ranking problem，but the output i see is a rating problem.what should i do ?

Temporal Factor in RS in general?

Was there any temporal dimension or consideration in your SVD model implementation?

Y. Koren, Collaborative Filtering with Temporal Dynamics, SIGKDD 2009.

不错 看到你们更新了

Soreg 算法

您好!请问在实现Soreg 算法时,通过0.00001 0.00001 0.001 等改变社会正则化项的权重为什么得到RMSE的值变化不大,与论文里的不一样呢?

使用filmtrust运行CUNE_BPR时，出现RuntimeError: you must first build vocabulary before training the model怎么解决呢？

您好，想请教下关于CFGAN损失函数的问题

您好，我发现CFGAN的程序运行后G_loss的值为负值，然后我对比了下论文，您实现的是论文的CFGAN_ZP方法吧，发现您定义的损失函数也没什么问题，跟论文是一致的，所以很迷惑，想向您请教下是什么原因造成生成器的损失函数值为负？麻烦您了

ImportError: No module named mkl

你好，在Macbook上运行main.py出现了一个问题，下载了依赖mkl，还是出现这个：

/usr/local/bin/python2.7 /Users/feanlau/Desktop/RecQ/main/main.py
Traceback (most recent call last):
  File "/Users/feanlau/Desktop/RecQ/main/main.py", line 3, in <module>
    from RecQ import RecQ
  File "/Users/feanlau/Desktop/RecQ/main/RecQ.py", line 9, in <module>
    import mkl
ImportError: No module named mkl

Process finished with exit code 1

请问一下，这个问题如何解决

Question about Top N > 100

I got a new problem again.
I study about your code,and in baseclass/recommendation.py "evalRanking" function, I note that the TOP-N problem "N" must be less than 100 and greater than 0,and the indicators MAP,NDCG are computed as MAP@N,[email protected] i need to calculete them on all dataset with WRMF method, I modified your "evalRanking" function like this:

        if self.ranking.contains('-topN'):
            top = self.ranking['-topN'].split(',')
            top = [int(num) for num in top]
            N = len(self.data.item)

and i also modeified the RankMeasure function in measure.py.
But now I got the error:

  File "../baseclass/baseclass/Recommender.py", line 181, in evalRanking
    if recommendations[r] < itemSet[item]:
IndexError: list index out of range

So,How should i do?thanks.

请教

你好，请教个问题，元路径的设计有什么特别要求吗？？怎样才是合理的？？

Question about computing loss function

First of all thanks for this awesome library, I have one question about computing loss value of various algorithms, why you didn't use a separate function for computing loss?
For example in PMF for every training tuple we are adding self.regU * p.dot(p) + self.regI * q.dot(q) to loss value which is much more than what we have in proposed objective functions.

I was wondering using a function to calculate loss value effect algorithm final value?

implementation of RSTE

I have one question about implementation of RSTE algorithm, why there is no difference between update rule of this model and PMF? how we have exploited the social data in updating parameters?

EE模型

不好意思,又打扰您了,同样的参数设置.为什么EE模型得出来的结果和加入社交信息的模型没多大区别呢?难道又优化了吗?希望有空解答