23-01-2013, 10:54 AM
Application and Research of Large-scale Parallel Computing in Analysis of Mobile Users’ Online Behavior
1Application and Research.pdf (Size: 237.7 KB / Downloads: 134)
Abstract
This paper introduces the requirements and methods
of analyzing the data of mobile users, and analyses the
application of large-scale parallel computing in analysis of the
online behavior of mobile users and focuses on how to solve the
technical problems of massive data analysis.
INTRODUCTION
One of the key measures to win over the competitors is
providing differentiated and precision service for information
service providers whether Internet service providers or telecom
service providers. The prerequisite and guarantee of useroriented
precision marketing is user behavior analysis. We can
understand the real needs of the user and provide individuate
and fine service, enhance customer loyalty and build a better
profit model at the same time through in-depth analysis of user
behavior.
User behavior analysis is analyzing the characteristics of
the users themselves based on the continuous operation, use
behavior of users and related content of information, which
requires the analysis of massive data mining in order to form
the model. These vast amounts of user-related behavior and
logs are a big challenge to the technology of data filtering,
analysis and mining. In this paper, we analyze the user
behavior and the content which is being browsed by the user
when the mobile terminal user access to the mobile Internet,
and identify the user's interests and hobbies applied in the
commercial advertising. The key point is how to solve the
technical problems of massive data analysis using the largescale
parallel computing.
The parallel distribution of mining algorithm
The most difficult and troublesome is the parallel
distribution of mining algorithm in the course of analysis of
user online behavior. Of course, paralleling the mining
algorithm not refers to the division of mass data and parallel
processing independent each other, but rather parallels the logic
and flow of the mining algorithm itself. For many algorithms,
the more complete data requests the higher precision
algorithms, especially when the algorithm is related to the
probability, so the data cannot be segmented. Therefore, this
chapter focuses on parallel distribution of key algorithms in the
course of analysis of user online behavior.
Summary and Outlook
In general, we know the key technologies and related
solutions of mass data analysis through research of analysis of
mobile users’ online behavior based on cluster computing.
Based on large-scale clusters computing and MapReduce
parallel framework, it has become a new trend to filter,
analysis, and mining of mass data, and more and more
applications of mining developed in this direction. In this trend,
not only the open source community, but also academics have
made many contributions and support, including open source
software project, theoretical analysis and so on. There are still
some issues need to improve, such as the reliability of
distributed storage, transaction of parallel computing tasks and
fault tolerance and so on, which will be solved gradually and
some already have the realization of optimum methods and
mechanisms.