Web Usage Mining

MapReduce-based Emerging Pattern Mining in Hypergraph Learning

Xiuming Yu, Meijing Li and Keun Ho Ryu

Database/Bioinformatics Lab, Chungbuk National University, Cheongju, South Korea

1 RESEARCH PROBLEM

Web usage mining is a popular research area in data

mining. As the rapid growth of internet, more and

more log information is collected by the web servers

around the world. It becomes difficult to extract

useful information from these huge web log data.

Classic techniques of web usage mining are

performed with low efficency in large number of

web log data, because a lot of system resource is

needed to deal with large computation.

Most techniques of web mining are performed

based on assosiation rule mining or frequent pattern

mining, which aim to find relationships among web

pages or predicting the behavoir of web users. That’s

difficult to find some favourite web pages in

different web users.

2 OUTLINE OF OBJECTIVES

In this research, we aim to find some favourite web

pages in different web users in large web log data.

We propose an efficient approach based on

hypergraph learning by considering the

programming model of MapReduce and the

techniques of emerging pattern mining.

Comparison with traditional pagerank method in

web mining, our proposed method which is based on

hypergraph learning can save a lot of cost in

computation by considering fewer nodes in

generated graphs of page links.

The programming model of MapReduce is used

to improve our proposed approach by counting

visiting time which accessed by web users in one

web page. It can greatly improve the efficiency of

computation.

Emerging patterns in web log data can be

represented as favourite web pages of different web

users, we can obtain useful information by

performing the techniques of emerging pattern

mining in web log data.

3 STATE OF THE ART

Web Usage Mining (WUM), also known as web

access, the web access pattern tracking can be

defined as web page history, the mining task is a

process of extracting interesting patterns in web

access logs. There are so many techniques of web

using mining have been proposed (Yu, 2011, Yu,

2012, Li, 2014). It is still a hot topic in the research

area of data mining.

A hypergraph is a specific graph whose edges

can connect more than two vertices or nodes.

Because of this characteristic, hypergraph is suitable

to solve the problem of high-order relation. In this

research, we consider that it can reduce the number

of vertices or nodes in generated graphs of page

links, to save the cost of computation in the process

of mining task. Hypergraph has been widely used in

the area of data mining, and there are many articles

(Xie, 2014, Pliakos, 2014, Chen, 2014, Yu, 2014)

about hypergraph learning have been published.

Hadoop-MapReduce (Narayanan, 2014, Ghit,

2014, Doulkeridis, 2014) is a programming model

and an associated implementation for processing and

generating large datasets, which processes large

datasets in parallel fashion. It allows users to not

worry about the nuisance details of coordinating

parallel sub-tasks and maintaining distributed file

storage. It greatly increases user productivity while

users process a large amount of parallel data. A

MapReduce program consists of two parts: a map

function, which processes a key/value pair to

generate a set of intermediate key/value pairs, and a

reduce function, which merges all of the

intermediate values associated with the same

intermediate key.

The discovery of class comparison or

discrimination information is an important problem

in the field of data mining. Recently, a novel kind of

knowledge pattern, called EPs (emerging patterns),

is introduced in (Dong, 1999). Emerging patterns,

defined as multivariate features whose supports

change significantly from one class to another, are

Yu X., Li M. and Ryu K..

Web Usage Mining - MapReduce-based Emerging Pattern Mining in Hypergraph Learning.

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

very useful as a mean of discovering distinctions

between different classes of data. By aggregating the

differentiating power of EPs, the constructed

classifiers are usually more accurate than other

existing state-of-the-art classifiers (Dong, 1999), (Li,

2015). EP-based classifiers also use this concept in

analysis of high-dimensional biomedical data for

rule discovery, diagnosis, prognosis and for better

understanding of mechanics of the disease.

Emerging patterns are patterns whose supports

change significantly from one dataset to another. By

using the techniques of emerging pattern mining, we

can find emerging web pages in web log data. This

technique has been applied in many articles

(Sherhod, 2014, Yu, 2014) and it is still a hot topic

in computer science.

4 METHODOLOGY

In this research, we get large pages as vertices in

hypergraph from processed web log data, transform

original data set into hypergraphs grouping by web

users, and find emerging patterns in these

hypergraphs based on the programming model of

mapreduce. Our work flow is shown in Figure 1.

Figure 1: Work flow of our proposed approach.

4.1 Data Pre-Processing

Web log data is automatically recorded in web log

files on web servers when web users access the web

server through their browsers. Not all of the records

sorted into the web log files are legal or necessary

for the mining task, so before analysis of the web log

data, the data cleaning phase needs to be

implemented.

4.1.1 Removing Records with Missing Value

Data

Some of the records sorted in web log file are not

complete because some of the parameters of the

records are lost. For example, if a click-through to a

web page was executed while the web server was

shut down, then in the log file only the IP address,

user ID and access time are recorded; the method,

URL, referrer and agent will be lost. This kind of

record is illegal for our mining task, so these records

must be removed.

4.1.2 Removing Illegal Records with

Exception Status Numbers

Some illegal records are caused by errors in the

requests or by the server. Although the records are

intact, the activity did not execute normally. For

example, records with the status numbers of 400 or

404 are caused by HTTP client errors, bad requests

or a request not found. Records with status numbers

500 or 505, are caused by HTTP server errors, when

the internal server cannot connect, or when the

HTTP version is not supported. These kinds of data

are illegal for our task, so the records must be

removed.

4.1.3 Removing Irrelevant Records with No

Significant URLs

Some URLs in the records consist of .txt, .jpg, .gif or

.js extensions, which are automatically generated

while a web page is requested. These records are

irrelevant to our mining task, so they must be

removed.

Figure 2: Common log format for web log data.

4.1.4 Selecting the Essential Attributes

As shown in the common log format of web log data

in Figure 2, there are many attributes in one record,

but for web-usage mining, not all the attributes are

necessary. In this paper, the attributes for IP address,

time, URL and referrer are essential attributes to our

task, so they should remain but the rest of the

attributes should be discarded.

WebUsageMining-MapReduce-basedEmergingPatternMininginHypergraphLearning

4.2 Generate Large Page

A large page set is a set of frequent web pages. We

define frequent web pages as whose support

thresholds are greater than or equal to user specified

minimum support threshold.

In this paper, a web log file denotes a data set,

and one web page is defined as an event and LP

denotes the set of web pages that are accessed by

web users with enough frequency over a period of

time. An important definition for generating LP is

user session. Here the user session time is defined as

one hour for simplicity. Then, the example data is

grouped by one hour for each web user. According

to the experimental data, candidate event types are

extracted and their supports are calculated. To

calculate the support count for each candidate, we

need to count the visit times that are accessed by

different web users. Finally, a user specified

Minimum Support threshold for Large Page (MSLP)

must be defined. MSLP denotes a kind of abstract

level that is a degree of generalization. The support

value will be determined by the proportion of web

users accessing times of web pages. Selecting MSLP

is very important; if it is low, then we can get a

detailed event. If it is high, then we can get general

events. In this example, MSLP is defined as 75%. In

other words, if a web page is accessed by greater

than 75% web users, then this web page can be

denoted as a large page.

4.3 Generate Hypergraph

After generating large web pages, all of the large

web pages are defined as vertices in hypergraph. In

web log data, each web user is represented as a class

which may contain hundreds or thousands of

accessed records. For example, for one web user U1,

he or she visited 20 web pages {p1, p2... p20} in a

user session, we can get generalized graph of page

links shown in Figure 3. We assume that large page

set in Figure 3 is {p6, p12, p16, p19}, then for web

user U1, his or her hypergraph of accessed pages can

be described in Figure 4.

4.4 Emerging Pattern Ming in

Generated Hypergraphs

After generating hypergraphs for all of web users,

we aim to find emerging patterns in these

hypergraphs. An example of hypergraphs of some

web users is shown in Figure 5.

Figure 3: Generalized graph of page links of U1.

Figure 4: Hypergraph of user U1.

In the process of emerging pattern mining, we

will use the idea of



-EP and JEP (Jumping

Emerging Pattern). An example of emerging pattern

mining is presented using the example data shown in

Figure 5. We make the hypergraph in web user U1

Figure 5: Example hypergraphs of some web users.

ICPRAM2015-DoctoralConsortium

as class 1 and other hypergraphs in other web users

as class 2. Table 1 shows that the items in these two

classes. Table 2 lists out all possible EPs, their

support and the growth rates of all EPs. If we set the

minimum growth rate threshold



is =1.5, there are

six EPs: Class 1 has four EPs ({p6}, {p12}, {p14},

{p16}) and five JEPs ({p6, p12}, {p6, p16}, {p12,

p14}, {p6, p12, p14} and {p12, p6, p16}).

Table 1: A sample dataset with two classes.

Class 1 Class 2

p6 p3

p12 p4

p14 p6

p16 p7

p6, p12 p10

p6, p16 p11

p12, p14 p12

p6, p12, p14 p14

p12, p6, p16 p16

p6, p12, p14, p16 p17

p18

p19

p7, p12

…

Table 2: Discovering Eps for

1.5





Items Support Growth Rate of EPs

Class 1 Class 2 Class 1 Class 2

p6 0.56 0.08 7 -

p12 0.56 0.17 3.3 -

p14 0.33 0.33 - -

p16 0.33 0.08 4.1 -

p6, p12 0.33 0

∞

p6, p16 0.22 0.08 2.75 -

p12, p14 0.22 0.08 2.75 -

p6, p12, p14 0.11 0

∞

p12, p6, p16 0.11 0

∞

p6, p12, p14,

p16

0 0 - -

However, JEPs ({p6, p12}, {p6, p12, p14} and

{p12, p6, p16}) are JEPs of class 1 with support of

not zero values in class 1 and zero in class 2. It can

be seen that these JEPs, appearing one in class 1 but

zero time in class 2, are not useful for classification,

especially when there are much noise present in the

data.

5 EXPECTED OUTCOME

In this section, we will perform our proposed

approach to web log data and do some experiments

to proof the efficiency of our proposed approach.

5.1 Experiment Data

In the experiments, we used weblog data come from

a NASA website (http://ita.ee.lbl.gov/html/contrib/

NASA-HTTP.html), which is cited in [18], because

that web log is large-scale temporal data with time

points. The event file can be generated from a web

log file including the URL requested, HTTP method

requested, the IP address from which the request

originated, and a timestamp.

Two web log files, NASA_access_log_Jul95 and

NASA_access_log_Aug95, were used as our two

experiment datasets. The first dataset was collected

from 00:00:00 July 1, 1995, through 23:59:59 July

31, 1995 (a total of 31 days). The second dataset was

collected from 00:00:00 August 1, 1995, through

23:59:59 August 31, 1995. The uncompressed

content of the first dataset is 205.2 MB, and it

contains 1,891,715 records. The uncompressed

content of the second dataset is 167.8 MB, and it

contains 1,569,898 records.

5.2 Analysis of Our Proposed

Approach

In this section, we want to evaluate our proposed

approach by three ways.

 Algorithm based on mapreduce programming

model vs. traditional algorithm;

 Prediction based on hypergraph learning vs.

generalized graph mining;

 Accuracy of emerging pattern mining;

For the way of Algorithm based on mapreduce

programming model vs. traditional algorithm, we

want to get expected outcome which our proposed

approach performed with less time than traditional

algorithm.

For the way of Prediction based on hypergraph

learning vs. generalized graph mining, we want to

see our proposed approach can save a lot of time in

computation.

6 STAGE OF THE RESEARCH

In this research, we propose a process of getting

large pages from processed web log data, define

these large web pages as vertices in hypergraph, then

transform original data set into hypergraphs

grouping by web users, and find emerging patterns

WebUsageMining-MapReduce-basedEmergingPatternMininginHypergraphLearning

in these hypergraphs based on the programming

model of mapreduce. The main stage of the research

is that how to apply the techniques of web usage

mining, hypergraph learning and emerging pattern

mining to web usage mining. And in the stage of

experiment, we try to use right way to evaluate our

proposed approach, and do sufficient experiments to

prove our research points.

ACKNOWLEDGEMENTS

This research was supported by Export Promotion

Technology Development Program, Ministry of

Agriculture, Food and Rural Affairs (No.114083-3),

Basic Science Research Program through the

National Research Foundation of Korea (NRF)

funded by the Ministry of Science, ICT and Future

Planning (No.2013R1A2A2A01068923) and (No.

2008-0062611).

REFERENCES

Xie, Y., Yu, H. M., and Hu, R. 2014. Probabilistic

hypergraph based hash codes for social image search.

Journal of Zhejiang University SCIENCE C, 15(7),

537-550.

Pliakos, K., and Kotropoulos, C. 2014. Simultaneous

image tagging and geo-location prediction within

hypergraph ranking framework. In Acoustics, Speech

and Signal Processing (ICASSP), 2014 IEEE

International Conference on (pp. 6894-6898). IEEE.

Chen, X., Peng, Q., Han, L., Zhong, T., and Xu, T. 2014.

An effective haplotype assembly algorithm based on

hypergraph partitioning. Journal of Theoretical

Biology.

Yu, J., Hong, C., Tao, D., and Wang, M. 2014. Semantic

embedding for indoor scene recognition by weighted

hypergraph learning. Signal Processing.

Narayanan, A. H., Krishnakumar, U., and Judy, M. V.

2014, An Enhanced MapReduce Framework for

Solving Protein Folding Problem Using a Parallel

Genetic Algorithm. In ICT and Critical Infrastructure:

Proceedings of the 48th Annual Convention of

Computer Society of India-Vol I (pp. 241-250).

Springer International Publishing.

Ghit, B., Yigitbasi, N., Iosup, A., and Epema, D. 2014.

Balanced Resource Allocations Across Multiple

Dynamic MapReduce Clusters. In ACM SIGMETRICS.

Doulkeridis, C., and Nørvåg, K. 2014. A survey of large-

scale analytical query processing in MapReduce. The

VLDB Journal, 23(3), 355-380.

Dong, G., Zhang, X., Wong, L., and Li, J. 1999. CAEP:

Classification by aggregating emerging patterns. In

Discovery Science (pp. 30-42). Springer Berlin

Heidelberg.

Li, G., Law, R., Vu, H. Q., Rong, J., and Zhao, X. R. 2015.

Identifying emerging hotel preferences using

Emerging Pattern Mining technique. Tourism

Management, 46, 311-321.

Sherhod, R., Judson, P., Hanser, T., Vessey, J., Webb, S.

J., and Gillet, V. 2014. Emerging Pattern Mining to

Aid Toxicological Knowledge Discovery. Journal of

chemical information and modeling.

Yu, Y., Yan, K., Zhu, X., Wang, G., Luo, D., and Sood, S.

2014. Mining Emerging Patterns of PIU from

Computer-Mediated Interaction Events. In Agents and

Data Mining Interaction (pp. 66-78). Springer Berlin

Heidelberg.

Yu, X., Li, M., Kim, H., Lee, D. G., Park, J. S., and Ryu,

K. H. 2011. A novel approach to mining access

patterns. In Awareness Science and Technology

(iCAST), 2011 3rd International Conference on (pp.

350-355). IEEE.

Li, M., Yu, X., and Ryu, K. H. 2014. MapReduce-based

web mining for prediction of web-user navigation.

Journal of Information Science, 0165551514544096.

Yu, X., Li, M., Lee, D. G., Kim, K. D., and Ryu, K. H.

2012. Application of closed gap-constrained sequential

pattern mining in web log data. In Advances in

Control and Communication (pp. 649-656). Springer

Berlin Heidelberg.