ahxt / neuacf

Implementation of paper Aspect-Level Deep Collaborative Filtering via Heterogeneous Information Networks

NeuACF

This is an implementation of paper (Aspect-Level Deep Collaborative Filtering via Heterogeneous Information Networks).

Please refer our paper if you use this code and the bibtex of this paper is:

@inproceedings{han2018aspect,
   title={Aspect-Level Deep Collaborative Filtering via Heterogeneous Information Networks.},
   author="Han, Xiaotian and Shi, Chuan and Wang, Senzhang and Philip, S Yu and Song, Li",
   booktitle={IJCAI},
   pages={3393--3399},
   year={2018}
 }

Requirements

• Python 3.6
• Tensorflow 1.2.1
• docopt 0.6.2
• numpy 1.13.3
• sklearn 0.18.1
• pandas 0.20.1
• scipy 1.0.0

How to Run

1. unzip dataset.7z
2. Compute the aspect-level similarity matrix with the matlab code
3. Run the model with the python code acf.py

example:

 python ./acf.py ../dataset/amazon/ amovie --mat "U.UIU,I.IUI,U.UICIU,I.ICI" --epochs 40 --last_layer_size 64 --batch_size 1024 --num_of_neg 10 --learn_rate 0.00005 --num_of_layers 2 --mat_select median

Parameters

Parameter	Note
--mat	sim_mat [default: ""]
--epochs	Embedding size [default: 40]
--last_layer_size	The number of iterations [default: 64]
--num_of_layers	The number of layers [default: 2]
--num_of_neg	The number of negs [default: 2]
--learn_rate	The learn_rate [default: 0.00005]
--batch_size	batch_size [default: 1024]
--mat_select	mat select type [default: median]
--merge	batch_size [default: attention]

Link

For more information, visit the webpage http://www.shichuan.org

neuacf's Issues

some error in code

Thanks for your code sincerely，but now there are some errors in the code you had shared，for example，no definition about len（mat_list)==4，maybe you delete some code.Can you upload or send the corre code？Because my senior Student Projec is based on your work！Thank you sincerely！！
This is my email：[email protected]

NameError

Hello,
Thank you for your work，when I run the code following your steps, I meet the following question, can you tell me how to solve the problem？
(NeuACF) F:\NeuACF\src>python ./acf.py ../dataset/amazon/ amovie --mat "U.UIU,I.IUI,U.UICIU,I.ICI" --epochs 40 --last_layer_size 64 --batch_size 1024 --num_of_neg 10 --learn_rate 0.00005 --num_of_layers 2 --mat_select median

--mat_select median

--mat U.UIU,I.IUI,U.UICIU,I.ICI

--last_layer_size 64

--epochs 40

--merge attention

--num_of_neg 10

--learn_rate 0.00005

--batch_size 1024

--num_of_layers 2

<dataset_dir> ../dataset/amazon/

amovie

U_feature1 shape: (6170, 6170)

I_feature1 shape: (2753, 2753)

Traceback (most recent call last):
File "./acf.py", line 109, in
U_feature2 = pd.read_csv( U_feature_dir2, sep=",", header=None ).fillna( 0 ).as_matrix()
NameError: name 'U_feature_dir2' is not defined

which file shoud be used ?

(.venv) mldl@ub1604:~/ub16_prj/NeuACF/src$ python3.6 ./acf.py ../dataset/amazon/ --mat "U.UIU,I.IUI,U.UITIU,I.ITI,U.UIVIU,I.IVI,U.UICIU,I.ICI" --epochs 40 --last_layer_size 64 --batch_size 1024 --num_of_neg 10 --learn_rate 0.00005 --num_of_layers 2 --mat_select median
/home/mldl/v_pub_py36/.venv/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
/home/mldl/v_pub_py36/.venv/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
Usage:
acf.py [options] <dataset_dir>

(.venv) mldl@ub1604:/ub16_prj/NeuACF/src$ python3.6 ./acf.py ../dataset/amazon/amovie.train.rating --mat "U.UIU,I.IUI,U.UITIU,I.ITI,U.UIVIU,I.IVI,U.UICIU,I.ICI" --epochs 40 --last_layer_size 64 --batch_size 1024 --num_of_neg 10 --learn_rate 0.00005 --num_of_layers 2 --mat_select median
/home/mldl/v_pub_py36/.venv/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
/home/mldl/v_pub_py36/.venv/lib/python3.6/importlib/_bootstrap.py:219: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88
return f(*args, **kwds)
Usage:
acf.py [options] <dataset_dir>
(.venv) mldl@ub1604:/ub16_prj/NeuACF/src$

Would you mind telling me how to derive the UI.data and IC.data from original dataset?

Thanks for your code sincerely, I have already run it successfully on amazon movie dataset by the four similarity matrixes you provided. But running the matlab code requires the UI.data which is not in the amazon.7z. So would you mind telling me how to get these data(UI IA IC etc.) from original dataset?

How to run matlab code?

When I try to run matlab gen_PathSimMat_MUM_amovie.m, I got two errors.

Unable to read file 'amovie'. No such file or directory.
Undefined function or variable 'UI'.

Then, I create amovie folder and put training and testing files in it, but I got another error.

Unable to read file 'amovie'. No such file or directory.

So, I think maybe this amovie should be a file. But what should be in this file? What is the format of this file?
How can I solve the undefined variable 'UI' error?

关于相似矩阵计算

您好，我发现直接使用您提供的data文件（维度:n×2）,确实会出现相似度矩阵为2×2的情况，不知道你是否先将data文件转换为邻接矩阵再进行相似度计算呢

about the movielens dataset

Thanks for your codes and data sincerely!And in your paper,the addtional information such as actors of movie is crawled from IMDB,would you mind to provide the movielens with it?

HR and NDCG

Hello, can you provide matlab code for HR and NDCG? thank you very much!

Could you please tell me how many epochs do you set in the paper?

Is it a specific number or the number when the model converges?
Thank you very much.