TUBE: Embedding Behavior Outcomes for Predicting Success

Description: This repository contains the C++ implementation of the TUBE model proposed in paper TUBE: Embedding Behavior Outcomes for Predicting Success accepted by KDD 2019

Usage

1. Make

To make the executable file, please change into the project folder and run:

make all

Notes: This program uses gnu++11 standard. If you need advanced control over options when compiling the program, please look into the ./Makefile file.

2. Execute

After the executable file ./tube is generated, run:

./tube --input_behaviors_file data/synthetic-skill03.txt --output_goal_embs_file goals.txt --output_context_embs_file contexts.txt --dims 16 --negative 2 --threads 4 --samples 1

List of parameters:

--input_behaviors_file: The input file of training behaviors. Each line follows format <goal>\t<context_1>[,<context_2>,...]
--output_goal_embs_file: The output file of goal embeddings. First line is header: <#goal>\t<#dimension>. Then, each line follows format <goal>\t<dimension_1>\t<dimension_2>\t...\t<dimension_n>
--output_context_embs_file: The output file of context embeddings. First line is header: <#context>\t<#dimension>. Then, each line follows format <context>\t<dimension_1>\t<dimension_2>\t...\t<dimension_n>
--dims: The number of dimensions of the embedding; default is 8.
--threads: The number of threads used for training; default is 2.
--samples: The number of behavior samples used for traninig in millions; default value equals to the number of behaviors in --input_behaviors_file X 500.
--negative: The number of negative samples; default is 1.
--rate: The initial value of learning rate; default is 0.005.

Data

A pre-processed demo dataset is included.

./data/synthetic-skill03.txt: The synthetic behavior dataset with k=3 as described in the paper.

Other datasets can be found at:

Complete synthetic datasets: https://bit.ly/2Uyx2Io
Real publication dataset with 2k venues: https://bit.ly/2GgbqfP
Real publication dataset with 5k venues: https://bit.ly/2TtJatW
Real publication dataset with 10k venues: https://bit.ly/2UuOfSI

Example

Other examples are provided in the ./demo.sh file.

Miscellaneous

If you find this code package to be useful, please consider cite us:

@inproceedings{wang2019tube,
  title={Tube: Embedding behavior outcomes for predicting success},
  author={Wang, Daheng and Jiang, Tianwen and Chawla, Nitesh V and Jiang, Meng},
  booktitle={Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  pages={1682--1690},
  year={2019}
}

DM2-ND/TUBE