Personalized Hashtag Recommendation for Micro-videos, ACM MM2019

This is our Pytorch implementation for the paper:

Yinwei Wei, Zhiyong Cheng, Xuzheng Yu, Zhou Zhao, Lei Zhu, and Liqiang Nie(2019). Personalized Hashtag Recommendation for Micro-videos. In ACM MM`19, NICE, France,Oct. 21-25, 2019

Introduction

GCN_PHR is a novel graph convolutional networks (GCN) based hashtag recommendation method, which attempts to design a personalized hashtag recommendation method formicro-videos.

Environment Requirement

The code has been tested running under Python 3.5.2. The required packages are as follows:

• Pytorch == 1.1.0
• torch-cluster == 1.4.2
• torch-geometric == 1.2.1
• torch-scatter == 1.2.0
• torch-sparse == 0.4.0
• numpy == 1.16.0

Example to Run the Codes

The instruction of commands has been clearly stated in the codes.

• YFCC100M dataset
  python train.py --l_r=0.001 --weight_decay=0.1 --batch_size=1024 --dim_latent=64 --num_workers=30 --aggr_mode='mean' --scoring_mode='cat'
• Instagram dataset
  python train.py --l_r=0.001 --weight_decay=0.1 --batch_size=1024 --dim_latent=64 --num_workers=30 --aggr_mode='mean' --scoring_mode='cat'

Some important arguments:

• aggr_mode
  It specifics the type of aggregation layer. Here we provide three options:
  1. mean (by default) implements the mean aggregation in aggregation layer. Usage --aggr_mode 'mean'
  2. max implements the max aggregation in aggregation layer. Usage --aggr_mode 'max'
  3. add implements the sum aggregation in aggregation layer. Usage --aggr_mode 'add'
• scoring_mode:
  It indicates the implementation of user-specific micro-video/hashtag representation. Here we provide two options:
  1. cat(by default) concatenates the representation of user and micro-video/hashtag, and then feed into a MLP to calculate the representations. Usage --scoring_mode 'cat'
  2. fully_con calculates the user-specific micro-video/hashtag representations with a fully connected layer. Usage --concat 'fully_con'

Baselines:

• UTM proposed in User conditional hashtag prediction for images, SIG KDD2015.
• ConTagNet proposed in ConTagNet: Exploiting usercontext for image tag recommendation , ACM MM2016.[CODE]
• CSMN proposed in Attend to You: Personalized Image Captioning with Context Sequence Memory Networks, CVPR2017.[CODE]
• USHM proposed in Separating self-expression and visual content in hashtag supervision, CVPR2018.

Dataset

We provide two processed datasets: YFCC100M and Instagram.

• You can find the full version of dataset via YFCC100M and raw data Instagram.
• We select some users and micro-videos and extract the visual, acoustic, and textual features of micro-video.

	#Micro-video	#User	#Hashtag	Visual	Acoustic	Textual
YFCC100M	134,992	8,126	23,054	2,048	128	100
Instagram	48,888	2,303	12,194	2,048	128	100

-train.npy
Train file. Each line is a user with her/his hashtag towards the micro-video: (userID, Hashtag ID and micro-video ID)
-val.npy
Validation file. Each line is a user with her/his 1,000 negative hashtags and several positive hashtags for a micro-video: (userID, Neg_Hashtag ID, Pos_Hashtag ID, and micro-video ID)
-test.npy
Test file. Each line is a user with her/his 1,000 negative hashtags and several positive hashtags for a micro-video: (userID, Neg_Hashtag ID, Pos_Hashtag ID, and micro-video ID)

Copyright (C) Shandong University

This program is licensed under the GNU General Public License 3.0 (https://www.gnu.org/licenses/gpl-3.0.html). Any derivative work obtained under this license must be licensed under the GNU General Public License as published by the Free Software Foundation, either Version 3 of the License, or (at your option) any later version, if this derivative work is distributed to a third party.

The copyright for the program is owned by Shandong University. For commercial projects that require the ability to distribute the code of this program as part of a program that cannot be distributed under the GNU General Public License, please contact weiyinwei@hotmail.com to purchase a commercial license.