Presentation 2017-09-19
Model Ensemble for Failure Event Detection using Multiple User Activity Data on the Web
Motoyuki Oki, Koh Takeuchi, Yukio Uematsu,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Mobile network service providers aim to maintain stable operation and improve service performance using multiple user's activity data. The data is sometimes corrupted by a number of unusual behavior such as service failure occurrence. In this paper, we address the detection and forecasting problems of finding and extracting failure event. We collect tweets related to the service, search queries on a search box, and web access logs from official web pages. We present the feature extraction for text classification problems and informative representation of the time series. We propose ensemble event detection and forecasting methods using supervised classification models and unsupervised anomaly detection models. Experiments to detect and predict multiple failure events in a content service provider demonstrate the effectiveness of the proposed method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Failure Prediction / Event Detection / Social Network / Machine Learning / Ensemble Model
Paper # DE2017-20
Date of Issue 2017-09-11 (DE)

Conference Information
Conference Date 2017/9/18(3days)
Place (in Japanese) (See Japanese page)
Place (in English) Ochanomizu University
Topics (in Japanese) (See Japanese page)
Topics (in English) Big Data Management, Information Retrieval, Knowledge Discovery, etc.
Chair Akiyo Nadamoto(Konan Univ.)
Vice Chair Koji Eguchi(Kobe Univ.) / Shingo Otsuka(Kanagawa Inst. of Tech.)
Secretary Koji Eguchi(Kogakuin Univ.) / Shingo Otsuka(Univ. of Marketing and Distrbution Science)
Assistant Kazuo Goda(Univ. of Tokyo) / Yuroaki Shiokawa(Tsukuba Univ.)

Paper Information
Registration To Technical Committee on Data Engineering / Special Interest Group on Database System / Special Interest Group on Information Fundamentals and Access Technologies
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Model Ensemble for Failure Event Detection using Multiple User Activity Data on the Web
Sub Title (in English)
Keyword(1) Failure Prediction
Keyword(2) Event Detection
Keyword(3) Social Network
Keyword(4) Machine Learning
Keyword(5) Ensemble Model
1st Author's Name Motoyuki Oki
1st Author's Affiliation NTT Communications Corporation(NTT Communications)
2nd Author's Name Koh Takeuchi
2nd Author's Affiliation NTT Communication Science Laboratories(NTT)
3rd Author's Name Yukio Uematsu
3rd Author's Affiliation NTT Communications Corporation(NTT Communications)
Date 2017-09-19
Paper # DE2017-20
Volume (vol) vol.117
Number (no) DE-212
Page pp.pp.123-128(DE),
#Pages 6
Date of Issue 2017-09-11 (DE)