Video Big Data Retrieval Over Media Cloud: A Context-aware Online Learning Approach

Yinan Feng, Pan Zhou, Jie Xu, Shouling Ji, Dapeng Oliver Wu

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Online video sharing (e.g., YouTube, YouKu) has emerged as one of the most important services in the current Internet, where billions of videos on the cloud are awaiting people for exploration. Hence, it is necessary for a personalized video retrieval system helping users to find interesting videos from such big data contents. One of the main challenges is how to process the increasing video big data and resolve the accompanying "cold start" issue efficiently. Another challenge is how to satisfy the users' need for the personalized retrieval results, which makes the accuracy of results unknown. In this paper, we formulate the personalized video big data retrieval problem as an interaction between user and system with a stochasticprocess (SP), not just a similarity matching, accuracy (feedback) model of the retrieval, introduce users' real-time context into the retrieval system, and propose a general framework of this problem. By using a novel contextual multi-armed bandit-based algorithm to balance the accuracy and efficiency, we propose a context-based online big-data-oriented personalized video retrieval system. It can support dynamically increasing datasets and has the nature of cross-modal retrieval. Our approach provides accurate retrieval results with sublinear regret and linear storage complexity and significantly improves the learning speed. Furthermore, by learning together for a cluster of similar contexts, we can reach sublinear storage complexity with the same regret but a slightly poorer performance on the "cold start" issue than the previous one. We validate our theoretical results by experiments on a tremendously large dataset, which show that the proposed algorithms outperform existing bandit-based online learning methods in terms of accuracy and efficiency, and the adaptation from bandit framework offers further benefits.

Original languageEnglish (US)
JournalIEEE Transactions on Multimedia
DOIs
StateAccepted/In press - Jan 1 2018

Fingerprint

Internet
Feedback
Big data
Experiments

Keywords

  • Big Data
  • Big data
  • Cloud computing
  • Clustering algorithms
  • Complexity theory
  • contextual bandit
  • Media
  • media cloud
  • online learning
  • online learning
  • Real-time systems
  • Streaming media
  • video retrieval

ASJC Scopus subject areas

  • Signal Processing
  • Media Technology
  • Computer Science Applications
  • Electrical and Electronic Engineering

Cite this

Video Big Data Retrieval Over Media Cloud : A Context-aware Online Learning Approach. / Feng, Yinan; Zhou, Pan; Xu, Jie; Ji, Shouling; Wu, Dapeng Oliver.

In: IEEE Transactions on Multimedia, 01.01.2018.

Research output: Contribution to journalArticle

@article{d6ef9067d878434dadfb7ae169406032,
title = "Video Big Data Retrieval Over Media Cloud: A Context-aware Online Learning Approach",
abstract = "Online video sharing (e.g., YouTube, YouKu) has emerged as one of the most important services in the current Internet, where billions of videos on the cloud are awaiting people for exploration. Hence, it is necessary for a personalized video retrieval system helping users to find interesting videos from such big data contents. One of the main challenges is how to process the increasing video big data and resolve the accompanying {"}cold start{"} issue efficiently. Another challenge is how to satisfy the users' need for the personalized retrieval results, which makes the accuracy of results unknown. In this paper, we formulate the personalized video big data retrieval problem as an interaction between user and system with a stochasticprocess (SP), not just a similarity matching, accuracy (feedback) model of the retrieval, introduce users' real-time context into the retrieval system, and propose a general framework of this problem. By using a novel contextual multi-armed bandit-based algorithm to balance the accuracy and efficiency, we propose a context-based online big-data-oriented personalized video retrieval system. It can support dynamically increasing datasets and has the nature of cross-modal retrieval. Our approach provides accurate retrieval results with sublinear regret and linear storage complexity and significantly improves the learning speed. Furthermore, by learning together for a cluster of similar contexts, we can reach sublinear storage complexity with the same regret but a slightly poorer performance on the {"}cold start{"} issue than the previous one. We validate our theoretical results by experiments on a tremendously large dataset, which show that the proposed algorithms outperform existing bandit-based online learning methods in terms of accuracy and efficiency, and the adaptation from bandit framework offers further benefits.",
keywords = "Big Data, Big data, Cloud computing, Clustering algorithms, Complexity theory, contextual bandit, Media, media cloud, online learning, online learning, Real-time systems, Streaming media, video retrieval",
author = "Yinan Feng and Pan Zhou and Jie Xu and Shouling Ji and Wu, {Dapeng Oliver}",
year = "2018",
month = "1",
day = "1",
doi = "10.1109/TMM.2018.2885237",
language = "English (US)",
journal = "IEEE Transactions on Multimedia",
issn = "1520-9210",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Video Big Data Retrieval Over Media Cloud

T2 - A Context-aware Online Learning Approach

AU - Feng, Yinan

AU - Zhou, Pan

AU - Xu, Jie

AU - Ji, Shouling

AU - Wu, Dapeng Oliver

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Online video sharing (e.g., YouTube, YouKu) has emerged as one of the most important services in the current Internet, where billions of videos on the cloud are awaiting people for exploration. Hence, it is necessary for a personalized video retrieval system helping users to find interesting videos from such big data contents. One of the main challenges is how to process the increasing video big data and resolve the accompanying "cold start" issue efficiently. Another challenge is how to satisfy the users' need for the personalized retrieval results, which makes the accuracy of results unknown. In this paper, we formulate the personalized video big data retrieval problem as an interaction between user and system with a stochasticprocess (SP), not just a similarity matching, accuracy (feedback) model of the retrieval, introduce users' real-time context into the retrieval system, and propose a general framework of this problem. By using a novel contextual multi-armed bandit-based algorithm to balance the accuracy and efficiency, we propose a context-based online big-data-oriented personalized video retrieval system. It can support dynamically increasing datasets and has the nature of cross-modal retrieval. Our approach provides accurate retrieval results with sublinear regret and linear storage complexity and significantly improves the learning speed. Furthermore, by learning together for a cluster of similar contexts, we can reach sublinear storage complexity with the same regret but a slightly poorer performance on the "cold start" issue than the previous one. We validate our theoretical results by experiments on a tremendously large dataset, which show that the proposed algorithms outperform existing bandit-based online learning methods in terms of accuracy and efficiency, and the adaptation from bandit framework offers further benefits.

AB - Online video sharing (e.g., YouTube, YouKu) has emerged as one of the most important services in the current Internet, where billions of videos on the cloud are awaiting people for exploration. Hence, it is necessary for a personalized video retrieval system helping users to find interesting videos from such big data contents. One of the main challenges is how to process the increasing video big data and resolve the accompanying "cold start" issue efficiently. Another challenge is how to satisfy the users' need for the personalized retrieval results, which makes the accuracy of results unknown. In this paper, we formulate the personalized video big data retrieval problem as an interaction between user and system with a stochasticprocess (SP), not just a similarity matching, accuracy (feedback) model of the retrieval, introduce users' real-time context into the retrieval system, and propose a general framework of this problem. By using a novel contextual multi-armed bandit-based algorithm to balance the accuracy and efficiency, we propose a context-based online big-data-oriented personalized video retrieval system. It can support dynamically increasing datasets and has the nature of cross-modal retrieval. Our approach provides accurate retrieval results with sublinear regret and linear storage complexity and significantly improves the learning speed. Furthermore, by learning together for a cluster of similar contexts, we can reach sublinear storage complexity with the same regret but a slightly poorer performance on the "cold start" issue than the previous one. We validate our theoretical results by experiments on a tremendously large dataset, which show that the proposed algorithms outperform existing bandit-based online learning methods in terms of accuracy and efficiency, and the adaptation from bandit framework offers further benefits.

KW - Big Data

KW - Big data

KW - Cloud computing

KW - Clustering algorithms

KW - Complexity theory

KW - contextual bandit

KW - Media

KW - media cloud

KW - online learning

KW - online learning

KW - Real-time systems

KW - Streaming media

KW - video retrieval

UR - http://www.scopus.com/inward/record.url?scp=85058652851&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85058652851&partnerID=8YFLogxK

U2 - 10.1109/TMM.2018.2885237

DO - 10.1109/TMM.2018.2885237

M3 - Article

AN - SCOPUS:85058652851

JO - IEEE Transactions on Multimedia

JF - IEEE Transactions on Multimedia

SN - 1520-9210

ER -