Twitter RecSys Challenge 2021

1. 👨‍👦‍👦 Members

• YoungIn Kwon(권영인) - Master degree student in AIGS(Aritificial Intelligence Graduate School) of UNIST(Ulsan National Institute of Science and Technology; South Korea)
• JoonHyung Kim(김준형) - Master degree student in AIGS(Aritificial Intelligence Graduate School) of UNIST(Ulsan National Institute of Science and Technology; South Korea)
• YeongHo Lee(이영호) - Undergraduate student in Industrial Engineering of UNIST(Ulsan National Institute of Science and Technology; South Korea)

2. 🗺️ Competition Overview

• Goal:

  • To predict the probability of different engagement types of a target user for a set of tweets based on heterogeneous input data while providing fair recommendations.

• schedule

  index	date	content
  1	19 March, 2021	Open to registration
  2	22 March, 2021	Dataset release & RecSys challenge starts
  3	30 March, 2021	Validation server open
  4	June 9, 2021	Validation data released, test server opens
  5	June 23, 2021	RecSys challenge ends
  6	June 30, 2021	Announcement of the final leaderboard and winners Paper submission for RecSys Challenge Workshop
  7	July 13, 2021	Paper submissions due
  8	Aug 10, 2021	Paper acceptance notifications
  9	Aug 23, 2021	Camera-ready papers due
  10	Sep 27 - Oct 1, 2021	Workshop taking place as part of the ACM RecSys conference.

• Data

  • Tweet Features
    • Text tokens: Ordered list of Bert ids corresponding to Bert tokenization of Tweet text
    • Hashtags:Tab separated list of hastags (identifiers) present in the tweet
    • Tweetd id:Tweet identifier
    • Present media:Tab separated list of media types. Media type can be in (Photo, Video, Gif)
    • Present links:Tab separeted list of links (identifiers) included in the Tweet
    • Present domains:Tab separated list of domains included in the Tweet (twitter.com, dogs.com)
    • Tweet type:Tweet type, can be either Retweet, Quote or Toplevel
    • Language:Identifier corresponding to the inferred language of the Tweet
    • Timestamp:Unix timestamp, in sec of the creation time of the Tweet
  • Engaged With User Features
    • User id:User identifier
    • Follower count:Number of followers of the user
    • Following count:Number of accounts the user is following
    • Is verified?:Is the account verified?
    • Account creation time:Unix timestamp, in seconds, of the creation time of the account
  • Engaging User Features
    • User id:User identifier
    • Follower count:Number of followers of the user
    • Following count:Number of accounts the user is following
    • Is verified?:Is the account verified?
    • Account creation time:Unix timestamp, in seconds, of the creation time of the account
  • Engagement Features
    • Engagee follows engager?:Does the account of the engaged tweet author follow the account that has made the engagement?
    • Reply engagement timestamp:If there is at least one, unix timestamp, in s, of one of the replies
    • Retweet engagement timestamp:If there is one, unix timestamp, in s, of the retweet of the tweet by the engaging user
    • Retweet with comment engagement timestamp:If there is at least one, unix timestamp, in s, of one of the retweet with comment of the tweet by the engaging user
    • Like engagement timestamp:If there is one, Unix timestamp, in s, of the like

3. 🔥 Models

< Model Structure >

<STEP 1> Pretrain - Text Token Embedding

• Glove
  • Input: text_tokens
  • Ouput: Embedding vector of words
• DAN
  • Input: Embedding vector of GloVe from text_tokens
  • Ouput: d dimension vector

<STEP 2> OverSampling

• GAN
  • Input: Output of DAN
  • Ouput: Fake inputs with Label

<STEP 3> Train

• ANN
  • Input: Output of DAN or GAN
  • Ouput: Classifier

<STEP 4> Inference(predict)

• Input: Output of DAN or GAN
• Ouput: Classifier

4. 📊 Performance

index	model	RCE	AP	Selected
1	ANN_reply_dan_not_gan	11.5176	0.1092	O
2	ANN_retweet_dan_not_gan	9.0952	0.2455
3	ANN_retweet_c_dan_not_gan	3.0766	0.0174	O
4	ANN_like_dan_not_gan	9.2520	0.5930
5	ANN_reply_dan_gan	11.0893	0.1109
6	ANN_retweet_dan_gan	9.2546	0.2485	O
7	ANN_retweet_c_dan_gan	3.1725	0.0166
8	ANN_like_dan_gan	9.1365	0.5881	O

5. 🛠️ Requirements

1) Docker Environment(Submission)

kwonyoung234/dgann_submission_env

2) Train Environment

• OS: Windows 10
• CPU: i5-10600K
• GPU: GTX 1660 SUPER
• RAM: 64GB

3) Libraries

Package	Version	Installation
dask	2021.4.0	conda
glove-python-binary	0.2.0	pip
numpy	1.20.1	conda
pandas	1.2.2	conda
pytorch	1.7.1	conda
scikit-learn	0.24.1	conda
scipy	1.6.1	conda
torchvision	0.8.2	conda
tqdm	4.58.0	conda
pickle5	0.0.11	pip

6. 📂 File and Folder

├── imgs
├── input
│   ├──model
|       GloVe and etc preprocess object
│   	├── glove_10_25_0.05_20.model: GloVe model
│   	├── dict_label_encoder.pkl: Dictionary for result of label encoder in data preprocessing
│   	├── robust_scaler.pkl: Robust scaler for data preprocessing
|       DAN pretrained model
│   	├── dan_reply_over.pth: DAN model for reply
│   	├── dan_retweet_over.pth: DAN model for retweet
│   	├── dan_retweet_c_over.pth: DAN model for retweet_with_comment
│   	├── dan_like.pth: DAN model for like
|       GAN pretrained model
│   	├── D_dan_reply.pth: Discrimiator of dan_reply used for classifier's input dataset
│   	├── G_dan_reply.pth: Generator of dan_reply used for classifier's input dataset
│   	├── D_dan_retweet.pth: Discrimiator of dan_retweet used for classifier's input dataset
│   	├── G_dan_retweet.pth: Generator of dan_retweet for classifier's input dataset
│   	├── D_dan_retweet_c.pth: Discrimiator of dan_retweet_c used for classifier's input dataset
│   	├── G_dan_retweet_c.pth: Generator of dan_retweet_c used for classifier's input dataset
│   	├── D_dan_like.pth: Discrimiator of dan_like used for classifier's input dataset
│   	├── G_dan_like.pth: Generator of dan_like used for classifier's input dataset
|       ANN pretrained model
│   	├── ANN_reply_dan_not_gan.pth: ANN for reply non using gan 
│   	├── ANN_reply_dan_gan.pth: ANN for reply using gan
│   	├── ANN_retweet_dan_not_gan.pth: ANN for retweet non using gan
│   	├── ANN_retweet_dan_gan.pth: ANN for retweet using gan
│   	├── ANN_retweet_c_dan_not_gan.pth: ANN for retweet_with_comment non using gan
│   	├── ANN_retweet_c_dan_gan.pth: ANN for retweet_with_comment using gan
│   	├── ANN_like_dan_not_gan.pth: ANN for like non using gan
│   	└── ANN_like_dan_gan.pth: ANN for like using gan
├── custom_dataset.py: datasets for neural network
├── run.py: inference result using trained weight by train.py
├── model.py: neural network models used in this project
├── processor.py: process data preprocess / train / predict for all models of project
└── utils.py: utilities(metrics and earlystopping module) for implementing code

7. 🏃 Run the code

python run