Integrated Measurement for Pre-Fetching in Mobile Environment

Roziyah Darus, Hamidah Ibrahim, Mohamed Othman and Lilly Suryani Affendey

Faculty of Computer Science and Information Technology, University Putra Malaysia, 43400 Serdang, Selangor, Malaysia

Keywords: User Interestingness, Integrated Measurement, Pre-Fetching.

Abstract: Pre-fetching is used to predict next query of data items before any problems occur due to network

congestion, delays, and latency problems. Lately, pre-fetching strategies become more complicated in which

to support new types of application especially for mobile devices. Sometime the pre-fetched data items are

not interested to the users. Due to this complication, an intelligent technique is introduced where an

integrated measurement using data mining with Bayesian approach is proposed to improve the query

performance. In previous study, the pre-fetched data items were filtered using data driven measurement. The

data was generated based on the data frequency metrics whereby the structure of the query pattern is

quantified using statistical methods. The measurement is not good enough to solve sequence query in

mobile environment. In this paper, a new technique is proposed to generate new and potential pre-fetching

set for the users. A subjective measurement is used to determine the pre-fetching set based on user

interestingness. The integrated measurement generates strong and weak association rules based on the data

and user interestingness criterions. The result shows that the performance is significantly improved whereby

the technique managed to quantify the uncertainty of users' expectation in the next possible query.

1 INTRODUCTION

In mobile environment, pre-fetching is very

tremendous in its functions in terms of providing an

effective technique to improve the availability of the

required data items or to predict next possible

queries for users. Pre-fetching process will pre-fetch

the predicted data items from memory into cache

before any miss hit problems occur due to network

congestion, delays, and latency (Hui and Guohong,

2004; Liqiang and Howard, 2006, Liu et al., 2000).

Pre-fetching prevents mobile users from any delay

of on-going job or any potential of terminating job

due to the disconnection problems. The predicted

data items must be useful and really can help users

to react on their advantage. Users will feel confident

towards the pre-fetched data items that can benefit

them in finishing their job successfully especially

during disconnection.

In order to pre-fetch the most useful data items,

an intelligent technique must consider data driven

and user driven criterion (Osmar, 1999). In this

research, both criterions of data interestingness and

user interestingness are used in determining the

predicted data items. User interestingness criterions

will be taking into consideration after data

interestingness criterions is treated. In previous

research (Wang et al., 2002, Hui and Guohong,

2004) the pre-fetched data items were measured

based on data interestingness only. The predicted

data items are depend on the frequently used items

whereby the structure of (A→B) is quantified using

statistical methods. Another word the quantity and

quality of the data items are controlled by data

frequency using support and confidence metrics.

The generated rule from data interestingness is

more concerned on data driven criterion rather than

user driven criterion. The measurement is not good

enough to solve rule quality problems that involve

users as in answering sequence query problem in

mobile environment (Mary et al., 2009). According

to (Hyeoncheol and Eun, 2005), frequency based

measures can generate uninteresting or incorrect

association rules if the dataset includes un-

informative instances.

However, according to (Hampton, Moore &

Thomas, 1973) a combination of human judgements

and Bayes' Theorem will process the information

more effectively than either one alone. An intelligent

technique is used for data prediction in the pre-

fetching process. The association rules in data

mining technique are used to generate the candidate

sets for the pre-fetching sets. By using the objective

measurement, the candidate set is filtered based on

205

Darus R., Ibrahim H., Othman M. and Suryani Affendey L..

Integrated Measurement for Pre-Fetching in Mobile Environment.

DOI: 10.5220/0004989802050212

In Proceedings of 3rd International Conference on Data Management Technologies and Applications (DATA-2014), pages 205-212

ISBN: 978-989-758-035-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

strong association rules using support and

confidence metrics. The criterion of the candidate

set is towards data driven in which it considers the

frequent number of items used only rather than the

user involvement in the transaction.

In our approach, the confidence value will be

treated as the prior probability to get the posterior

probability using Bayesian approach. A prior

probability will be assigned as an initial degree of

belief to the users as their general knowledge

towards the transaction. The pre-fetching set will be

generated from an integration of the objective

measurement and the subjective measurement of

data interestingness and user interestingness. In this

research,

we will prove that the integrated

measurement can give better query performance

than the previous work.

2 RELATED WORK

There are many pre-fetching strategies that have

been introduced in previous works. The trend in pre-

fetching strategy started with semi-automated

hoarding set where they rely on user intervention to

some extent (Geoffrey et al., 1997, James and

Mahadev, 1992). Their works were based on

observing past accessed patterns of users and

involved in maintaining the user’s profiles. Another

pre-fetching approach was designed and operated in

specific environment and focused was on semi-

structured data such as for location dependent web

pages and location dependent data services (Ho and

Hwan, 2004; Mariano and Ana, 2006; Shi et al.,

2005).

Lately, with the advancement of mobile

computing technology the pre-fetching techniques in

mobile systems become more complicated in which

to support new types of applications such as in

mobile environments (Shi et al., 2005, Darus and

Ibrahim, 2010, Darus and Ibrahim, 2011). Due to

this complication, researchers started to introduce

new intelligent technique which requires data

mining process (Yucel et al., 2000). Existing

automated pre-fetching techniques did not focus on

the granularity of mobile databases. They focused on

the files such as web pages (Wang et al., 2002, Hui

and Guohong, 2004). The pre-fetching set was

measured by data driven measurement in which the

data generated were based on the frequent data items

in structure of query patterns. The quantity and the

quality of generated data items are controlled by the

support and confidence metric only.

Other researchers concentrated on mining tuples

in a mobile relational DBMS environment, however

their approach can still be further improved to get

optimal solutions (Abhinav, 2005, Li and Lv, 2007).

They also used data driven measurement where the

pre-fetched data items (A→B) were computed based

on support and confidence in which the items are un-

logically related and sometime meaningless for

certain mobile users.

There are many pre-fetching strategies that have

been introduced but from our knowledge, none of

them are concerned on including user interestingness

by using Bayesian approach to contribute additional

criterion for the pre-fetching set.

3 INTEGRATED

MEASUREMENT

By using a-priori algorithm (Agrawal et al., 1993),

an association rule is used to produce an initial

pattern (X

→Y) for data interestingness. A list of

strong rules is generated in which to satisfy certain

threshold for minimum support (min_sup) and

minimum confidence (min_conf) values. The strong

rules in the objective measurement become the

candidate set in determining the potential of pre-

fetching set.

The candidate sets then are treated in the

subjective measurement. According to (Avi &

Alexander, 1996), they represented α, ξ and E as

degree of belief, previous evidence and new

evidence of knowledge in a belief system, B,

respectively. P (α | ξ) is defined as a conditional

probability that α holds, with some previous

evidence ξ supporting that belief. Then we compute

the subjective probability of the users.

Equation (3.1) refers to the confidence values, α

with previous evidence and will be treated as the

prior probability in the measurement. The

confidence value is assigned as the initial degree of

belief in a belief system concept (Liqiang &

Howard, 2006).

 Confidence (α

): Initial degree of Belief = d (α

| ξ) = P (α | ξ) (3.1)

Equation (3.2) is introduced as a normalized

weighted support values, w

, in which to reduce any

bias occur, as in (Hampton, Moore & Thomas,

1973).

 Weight: w

= Support (α

)/Σ Support (α

) (3.2)

According to (Liqiang and Howard, 2006), an

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

206

equation (3.3) is used in which the weight, w

, and

the confidence values are used to find the value of

user interestingness, I (α

, B, ξ).

 User Interestingness: I (α

, B, ξ) = d (α

| α

’, ξ) -

d (α

| ξ) = ǀ Confidence (α

’) – Confidence

(α

) ǀ = ǀ P (E | α, ξ) - P (α | ξ) ǀ (3.3)

A new degree of belief denoted as P (α| E, ξ) in

which the belief, α is based on the new evidence E in

the context of the old evidence ξ. It can be computed

using Bayes’ rule, as given in equation (3.4).

 New degree of Belief: P (α | E, ξ) = P (E |α, ξ) P

(α | ξ) / {P (E | α, ξ) P (α | ξ) + P (E|

￢

α, ξ)

P (

(

￢

α | ξ) (3.4)

An interestingness of pattern p, relative to

previous evidence ξ can be determined by P (α | p, ξ)

whereby it represents the confidence of rule p, given

belief α as in equation (3.5). It also can be defined as

the user interestingness relative to the difference

between the prior and posterior probabilities in the

belief system.

 User Interestingness measurement pattern p: I

(p, B, ξ) =

∑



{| P (α | p, ξ) – P (α | ξ) |}

/ P (α | ξ) (3.5)

By taking the confidence values as the initial

degree of belief in a belief system, a new pattern

using user interestingness values, I (α

, B, ξ) will be

generated and the patterns for pre-fetching set and

candidate set will be compared and analysed.

Let us consider Table 1 as a sample of online

customer order information. In this example, we

want to show that the pre-fetched items or products

using the integrated measurement really meaningful

for the user.

First, we generate the list of products from Table

1 for the candidate set using a-priori algorithm. We

specify the minimum support value (min_sup) = 4

and the minimum confidence value (min_conf) =

0.5, to produce the list of strong association rules as

in Table 2. A list consists of the most frequently

ordered products which has been filtered using

support and confidence metrics. In this case, the

objective measurement in which the data

interestingness of the products has been considered.

From Table 2, the highest confidence value is for

products

B#3→B#4, i.e. B#3 and B#4. Then it

followed by B#4 and B#5 and so on. It means that

these are the products which are the most frequently

ordered products by the customer. Based on data

interestingness, these are the products that will be

pre-fetched as the pre- fetching set. This approach

has been used in the previous work

Then we extend the previous work by

introducing our approach called integrated

measurement.

After the objective measurement process has been

carried out, a subjective measurement is introduced

to filter according to the user interestingness

products. After the products from the strong rules

are treated for the candidate set, we then treat the

highest confidence value as the initial belief of user

in a belief system for the transaction. This belief is

important in determining the future products to be

ordered by the user. The initial belief represents

general knowledge of ordering behaviour of the

user.

By using equation (3.1) to (3.5), we compute the

results to determine the pre-fetching set from strong

rules as in Table 3. It consists of nine beliefs, in

which support is used to calculate weights for non-

bias values and confidence value is treated as the

initial confidence for each belief.

Table 1: Sample of a Customer Order information.

Cust

_ID

Order_Date Product_Query

Cust

_ID

Order_Date Product_Query

CO1

2-May-07 B#6,B#8,B#16

CO1

3-October-08 B#12

26-May-07 B#4,B#10,B#15,B#20,B#5,B#8 10-November-08 B#2

2-Jun -07 B#2,B#5,B#3 29-Disember-08 B#8,B#6,B#5,B#4

6-Jul-07 B#6,B#9,B#5,B#4,B#25,B#10 27-Jan-09 B#4,B#12,B#6

3-Jan-08 B#2,B#20,B#6 5-Feb-09 B#3

12-Apr-08 B#5,B#4,B#6,B#3,B#15,B#16 12-Apr-09 B#20,B#2,B#8

9-May-08 B#8,B#2,B#15 15-May-09 B#2,B#1

7-Jun-08 B#4,B#25,B#6,B#5 15-Jun-09 B#8,B#3,B#4,B#5,B#2,B#16,B#5

19-July-08 B#16,B#12,B#4,B#20,B#10 11-July-09 B#9

1-August-

08 B#8,B#4,B#15,B#3,B#12,B#2 14-August-09 B#9,B#20,B#4,B#3,B#15,B#1

15-Sept-08 B#8,B#25,B#3,B#4,B#5,B#20 15-September-09 B#9,B#8,B#3,B#5,B#4,B#15

IntegratedMeasurementforPre-FetchinginMobileEnvironment

207

Table 2: Strong Association Rules as the candidate sets.

Strong Rule Expression Support Confidence

B#2→B#8 4 4/8 = 0.5

B#3→B#5 5 5/8 = 0.6

B#3→B#4 - Based on data

interestingness this is the most

interesting products

6 6/8 = 0.8 - Highest confidence value so it becomes the

initial degree of belief of the user

B#3→B#8 4 4/8 = 0.5

B#4→B#5 8 8/12 = 0.7

B#4→B#8 6 6/12 = 0.5

B#4→B#15 6 6/12 = 0.5

B#5→B#6 5 5/9 = 0.6

B#5→B#8 5 5/9 = 0.6

∑ Support (α

) 48

Table 3: Computing User Interestingness for Strong Association Rules.

Belief Expression

Weight, w

Support(α

)

/ Σ

Support(α

)

Initial

confidence

for each

belief

d (α

| ξ) =

P (α

| ξ)

Local

confidence,

d (α

|α

i’,

ξ) =

P (E

|α

ξ)

I (α

’, Y, ξ) =

d (α

| α

i’,

ξ) -

d (α

| ξ) =

ǀ Confidence (α

’)

–Confidence (α

) ǀ

Interestingness

for each belief

I (α

’, Y, ξ)

B#2→B#8 4/48 = 0.083 4/8 = 0.5 1/3=0.3 (ǀ0.33 – 0.5ǀ) =

0.17

0.01411

B#3→B#5 5/48 = 0.104 5/8 = 0.6 3/4=0.8 (ǀ0.75 – 0.625ǀ) =

0.125

0.013

B#3→B#4 6/48 = 0.125 6/8 = 0.8 2/4=0.5 (ǀ0.5– 0.75ǀ) =

0.25

0.03125

B#3→B#8 4/48 = 0.083 4/8 = 0.5 2/4=0.5 (ǀ0.5– 0.5ǀ) = 0 0

B#4→B#5 8/48 = 0.167 8/12 = 0.7 4/5=0.8 (ǀ0.8-0.67ǀ) = 0.13 0.02171

B#4→B#8

Based on user

interestingness,

i.e. the most

interesting

products

6/48 = 0.125 6/12 = 0.5 4/5 = 0.8 (ǀ0.8-0.5ǀ) = 0.3 0.0375 - The

highest

Interestingness

value

B#4→B#15 6/48 = 0.125 6/12 = 0.5 3/5 = 0.6 (ǀ0.6-0.5ǀ) = 0.1 0.0125

B#5→B#6 5/48 = 0.104 5/9 = 0.6 1/5 = 0.2 (ǀ0.2-0.56ǀ) = 0.3 0.0312

B#5→B#8 5/48 = 0.104 5/9 = 0.6 4/5=0.8 (ǀ0.8– 0.556ǀ) =

0.244

0.025376

New Degree of Belief : P (α | E, ξ) = (0.8x0.8) / (0.8x0.8) + (0.2x(1-0.8)) = 0.9412

Relative Interestingness Measurement: I (p, B, ξ) = |0.9412 – 0.8 |/ 0.8 = 0.1765

By referring to the results from Table 3, the new

degree of belief towards the new pre-fetching set

B#4→B#8 with the initial belief for the new evidence

is 94.12%. Products B#4 and B#8 have the highest

interestingness value then followed by B#3 and B#4

and so on.

It means that these are the products which are the

most interesting products to be ordered by the user

based on user interestingness.

The result also shows that, a relative difference

in terms of interestingness measurement of pattern

towards the new pre-fetching set is 17.65%. Then

we also work on the weak association rules. In this

case we want to show that even though there are

products which are not chosen as the candidate set,

but still the products are useful to users.

In this example, we want to identify, what are the

other possible products that can be chosen as the

potential to be the pre-fetching set for the user in

case of insufficient of pre-fetching set. In this case,

we specify lower minimum values for minimum

support and minimum confidence where min_sup =

2 and min_conf = 0.4 from Table 1 to produce a list

of weak association rules as in Table 4. Again, by

using the equation of (3.1) to (3.5), we compute the

results to determine the other possible to be pre-

fetching set from weak rules as in Table 4

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

208

Table 4: The Weak Association Rules - Parameters and their values in a belief system.

Belief

(α

)

Rule

Expression

Support

Confidence

value for

each belief,

d (α

| ξ) =

P (α

| ξ)

Weight, w

Support (α

) /

∑Support (α

)

Local

Confidence

for each belief,

d (α

| α

i’,

ξ) =

P (E

| α

ξ)

I (α

, Y, ξ) =

d (α

| α

i’,

ξ) -

d (α

| ξ) =

ǀ Confidence (α

’)

– Confidence (α

) ǀ

Interestingne

ss for each

belief =

I (α

’, Y, ξ)

B#4→B#6 5 5/12=0.421 0.05 4/6 = 0.67 ǀ0.67 – 0.421ǀ =

0.249

0.01245 (The

highest

interestingnes

s value

B#9→B#1

2 2/4=0.5 0.02 2/2 = 1 ǀ1 – 0.5ǀ = 0.5 0.01

B#9→B#2

2 2/4=0.5 0.02 1/2 = 0.5 ǀ0.5 – 0.5ǀ = 0 0

B#10→B#

2 2/3=0.67

(The highest

confidence

value)

0.02 1/1 = 1 ǀ1 – 0.67ǀ = 0.333 0.00666

New Degree of Belief: P (α | E, ξ) = (1x0.67) / (1x0.67) + (0.5x(1-0.67)) = 0.8024

Relative Interestingness Measurement: I (p, B, ξ) = |0.8024 – 0.67 |/ 0.67 = 0. 1976

Based on the results from Table 4, the new

degree of belief towards the new potential to be pre-

fetching set with an initial degree of belief as the

new evidence is 80.2 %. Products B#4 and B#6 have

the highest interestingness value then followed by

B#9 and B#15 and so on. It means that these are

among the products that can be considered as part of

the pre-fetching sets for the user. The relative

difference in terms of Interestingness measurement

pattern towards the new potential to be pre-fetching

set is 19.76%.

In Table 5, we summarize all the pre-fetching set

based on data interestingness and user

interestingness values. First we refer to products

B#4 and B#8 in which these products are very

meaningful from user perspective but not as

meaningful as from data interestingness point of

view. In fact some of the products can be discarded

from the pre-fetching set as in this case we refer to

products B#3 and B#8. These products are

meaningful after the products were treated in

objective measurement but then the products are

meaningless after being treated in subjective

measurement. Lastly, there are many other products

that are not considered as the candidate set, or in

weak association rules, can also be part of the

interesting products and potential to be the pre-

fetching set as for products B#4 and B#6.

4 EXPERIMENTS

We carried out an experiment to identify whether the

pre-fetching set and also the potential to be the pre-

fetching set from strong and weak association rules

can contribute for better query performances. The

integrated measurement is performed by applying

the objective measurement and then the subjective

measurement process. The data set used in the

experiment is taken from TPC-D data schema as in

(Shi et al., 2005). The data set was generated using

random generator given by Transaction Processing

Council, for Decision Making as in the TPC-D

database schema. The data set consists of four main

entities, i.e. regions, nations, customers and

products/items. For analysis purposes we choose a

sample of customers and products from Vietnam.

We treated data region=Asia as global data set and

data nation=Vietnam as local data set.

In this experiment, sequence query pattern is

used as in (Shi, Binshan and Qun, 2005) to identify

the query performance. Comparisons of results by

using pre-fetching set and potential to be the pre-

fetching set from four different cache sizes have

been carried out. We apply the a-priori algorithm

and the equation of (3.1) to equation (3.5) to

generate the pre-fetching set. A sequence query

pattern is used to identify the query performance.

Results are shown in Figure 1 and Figure 2.

IntegratedMeasurementforPre-FetchinginMobileEnvironment

209

Table 5: Results of Pre-fetching Set using Objective Measurement (OM) and Integrated Measurement (IM) for Strong

Association Rules and Weak Association Rules.

Figure 1: Comparison of Average for Query Performance in Pre-fetching Set from Strong Association Rules between

Objective Measurement (OM) and Integrated Measurement (IM).

5 RESULTS AND DISCUSSION

By using the integrated measurement process, Figure

1 shows that the pre-fetching set from strong rules

contributes a significant difference in query

performance, i.e. by an average of 35%. Even

though the pre-fetched data items are limited in

terms of availability at lower cache size but the pre-

fetched data items are increased tremendously at

higher cache size for the two measurements process.

From the result, it shows that by using IM, the

sequence query pattern performed better than OM

approach.

In Figure 2, the result shows that the pre-

fetching set from weak association rules contributes

slightly difference in query performance, i.e. by an

average of 9.2 % for the two types of measurements.

The pre-fetched data items are limited in terms of

availability at lower cache size compared to higher

cache size in the two measurements process. Again

the results show that by using IM approach, the

sequence pattern query can still performed better

than OM even though from the potential to be pre-

fetching set of weak association rules.

6 CONCLUSIONS

The integrated measurement technique manages to

generate many interesting data items for pre-fetching

set based on user interestingness values. By taking

into consideration the confidence value from data

interestingness as the initial belief of users, we

managed to generate an un-expected pattern for

0,00

20,00

40,00

60,00

80,00

100,00

5% 10% 15% 20%

Cache Hit Ratio (%)

Cache Size

OM IM

Pre-fetching Set from Strong

Association Rules using

Objective Measurement for

Data Interestingness

Pre-fetching Set from Strong

Association Rules using Subjective

Measurement for User

Interestingness

Pre-fetching Set from Weak

Association Rules using Subjective

Measurement for User

Interestingness

Pre-fetching

Set

Confidence

Values, C

Pre-fetching

Set

User

Interestingness

Values, I

Pre-fetching Set

User Interestingness

Values, I

1. B#3→B#4 0.8

1. B#4→B#8 0.038

1. B#4→B#6

0.012

2. B#4→B#5 0.7 2. B#3→B#4 0.031 2. B#9→B#15 0.010

3. B#5→B#6 0.63 3. B#5→B#6 0.031 3. B#2→B#3 0.008

4. B#3→B#5 0.6 4. B#5→B#8 0.025 4. B#10→B#15 0.007

5. B#5→B#8 0.6 5. B#4→B#5 0.022 5. B#10→B#20 0.007

6. B#4→B#8

0.5

6. B#2→B#8 0.014 6. B#4→B#25 0.007

7. B#4→B#15 0.5 7. B#3→B#5 0.013 7. B#3→B#9 0.005

8. B#2→B#8 0.5 8. B#4→B#5 0.013 8. B#4→B#9 0.003

9. B#3→B#8 0.5

9. B#3→B#8

0 9. B#4→B#10 0.003

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

210

Figure 2: Comparison of Average for Query Performance in Pre-fetching Set from Weak Association Rules between

Objective Measurement (OM) and Integrated Measurement (IM).

REFERENCES

Abhinav, A. V. 2005. Data Stashing Strategies for

Disconnected and Partially Connected Mobile

Environments. In Doctoral Thesis, RMIT University.

Doi=10.1.1.112.308

Agrawal R., T. Imielinski & A. N. Swami 1993. Mining

Association Rules between Sets of Items in Large

Databases, In Proceedings of the 1993 ACM SIGMOD

Conference, pp. 207–216.Doi>10.1145/170036. 170072

Avi, S. & Alexander,T. (1996). What Makes Patterns

Interesting in Knowledge Discovery Systems. IEEE

Transactions on Knowledge and Data Engineering,

Vol.8, No. 6, pp. 970-974. Doi:10.1109/69.553165

Darus R. & Ibrahim H. 2010. New Prediction Model for

Pre-fetching in Mobile Database. In Proceedings of

the 12th International Conference on Information

Integration and Web-based Applications & Services

(iiWAS), pp. 938-942. Doi>10.1145/1967486.1967653

Darus R. & Ibrahim H. 2011. User Interestingness for Pre-

fetching in Mobile Environment. In Proceedings of the

International Conference on Information Technology

and Multimedia (ICIM), pp. 1-6. Doi:10.1109/

ICIMU.2011.6122738

Geoffrey, H., Kuenning & Gerald, J. P. 1997. Automated

Hoarding for Mobile Computers, SOSP-16 10197

ACM, pp. 264-275. Doi:10.1.1.15.9270

Hampton, J. M., Moore & P. G, Thomas, H. 1973.

Subjective Probability and Its Measurement. In

Journal of the Royal Statistical Society. Series A, Vol.

136, No. 1, pp.21-42. http://links.jstor.org/sici?sici=00

35-238%281973%29136%3AI%3C2I%3ASPAIM%3

E2.0.CO%3B2-P

Ho, S. K. & Hwan, S. Y. 2004. Association Based Pre-

fetching Algorithm in Mobile Environments. In

Proceedings of the International Conference on

Embedded Software and Systems, (ICESS), Springer-

Verlag Berlin Heidelberg, 2005, pp. 243–250.

Doi:10.1007/11535409_34

Hui, S. & Guohong, C. 2004. Cache-Miss-Initiated Pre-

fetch in Mobile Environments. In Proceedings of the

International Conference on Mobile Data Mana-

gement, IEEE. Vol. 28, Issue 7, pp. 741-753. Doi:ieee

computersociety.org/10.1109/MDM.2004.1263086

Hyeoncheol, K. & Eun, Y. K. 2005. Information-Based

Pruning for Interesting Association Rule Mining in the

Item Response Dataset. In Proceedings of the

International Conferences in Knowledge-Based and

Intelligent Engineering & Information Systems (KES

378. Doi:10.1007/11552413_54

James J. K. & Mahadev, S. 1992. Disconnected Operation

in the Coda File System. ACM Transactions on

Computer Systems, Vol. 10, Issue 1, pp. 3-25. Doi=10.

1.1.12.448

Liqiang Geng & Howard J. Hamilton. 2006.

Interestingness Measures for Data Mining: A Survey,

Journal ACM Computing Surveys (CSUR), Vol. 38,

Issue 3, pp.1-32. Doi:1145/1132960.1132963

Li Yi-jun & Lv Ying-jie. 2007. Research on Different

Customer Purchase Pattern Based on Subjective

Interestingness. In Proceedings

of the International

Conference on Management Science & Engineering

(ICMSE 2007), pp. 3-8. Doi:10.1109/ICMSE.2007.

4421816

Liu B., Hsu W., Chen S. & Ma Y. 2000. Analyzing the

Subjective Interestingness of Association Rules. In

Journal Intelligent Systems and their Applications,

IEEE, Sep/Oct 2000, pp. 47-55. Doi:10.

1109/5254.889106

Mariano, C. T. N. & Ana, C. S. 2006. Hoarding and Pre-

fetching for Mobile Databases. In Proceedings of the

5th IEEE/ACIS International Conference on Computer

and Information Science and 1st IEEE/ACIS

International Workshop on Component-Based

Software Engineering, Software Architecture and

Reuse, IEEE, pp. 219-224. Doi:10.1109/ICIS-

COMSAR.2006.44

0,0

20,0

40,0

60,0

80,0

100,0

5% 10% 15% 20%

Cache Hit Ratio (%)

Cache Size

OM IM

IntegratedMeasurementforPre-FetchinginMobileEnvironment

211

Mary, M. J. F., Ilayaraja N. & Nadarajan R. 2009. Cache

Pre-fetch and Replacement with Dual Valid Scopes for

Location Dependent Data in Mobile Environments. In

Proceedings of the 11th International Conference on

Information Integration and web-based Applications

and Services. pp. 364-371. Doi:10.1145/1806338.

1806404

Osmar R. Zaiane, 1999. CMPUT690 Principles of

Knowledge Discovery in Databases.

Shi, M. H., Binshan, L. & Qun, S. D. 2005. Intelligent

Cache Management for Mobile Data Warehouse

Systems. Journal of Database Management, Vol. 16,

No. 2, pp. 46-65. Doi:10.4018/jdm.2005040103

Wang, K., Zhou, S. & Han, J. 2002. Profit Mining: From

Patterns to Actions. In Proceedings of the 8th

Conference on Extending Database Technology, pp.

70–87. Doi=10.1.1.12.7328

Yucel, S., Ozgur. U. & Ahmed, K. E. 2000. Association

Rules for Supporting Hoarding in Mobile Computing

Environments. In Proceedings of the 10

International

Workshop on Research Issues in Data Engineering,

IEEE Computing Society Press, pp. 389-401. Doi:ieee

computersociety.org/10.1109/RIDE.2000.836502

DATA2014-3rdInternationalConferenceonDataManagementTechnologiesandApplications

212