python codes for PreNets
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
code
figure_draw
model code
paper
result
CIKM-ppt.pdf
CIKM_cover.png
LICENSE
README.md
intensity_function.png

README.md

PreNets - CIKM 2018

The 27th ACM International Conference on Information and Knowledge Management (CIKM2018), Lingotto, Turin, Italy. [Accepted]

Title: Adversarial Training Model Unifying Feature Driven and Point Process Perspectives for Event Popularity Prediction
Author: Qitian Wu, Chaoqi Yang, Hengrui Zhang, Xiaofeng Gao, Paul Weng and Guihai Chen

Abstract: This paper targets a general popularity prediction problem for event sequence, which recently has aroused great attention due to its extensive applications in various domains. Feature driven method and point process method are two basic thinking paradigms to tackle the prediction problem but both of them suffer from respective limitations. In this paper, we pioneeringly propose PreNets unifying two thinking paradigms in an adversarial manner. On one side, the feature driven model, like a ‘critic’, aims to discriminate the predicted popularity from the real one based on a set of temporal features from the sequence. On the other side, the point process model, like an ‘interpreter’, recognizes the dynamic patterns in the sequence to generate a predicted popularity to fool the ‘critic’. Through a Wasserstein learning based two-player game, the training loss of the ‘critic’ would guide the ‘interpreter’ to better exploit the sequence patterns and enhance the prediction, while the ‘interpreter’ would push the ‘critic’ to select the effective early features that can help the discrimination. This mechanism enables the framework to absorb the strengths of both feature driven and point process methods. Empirical results show that PreNets can achieve 21.1% MAPE improvement for Twitter cascade prediction and 6.3% for Amazon review prediction.


Edited on 9th August, 2018 by Chaoqi. Any problems, please contact at ycqsjtu@gmail.com