Big Data Knowledge Service Framework based on Knowledge Fusion

∗

Fei Wang

, Hao Fan

1†

and Gang Liu

School of Information Management, Wuhan University, Luojiasan Road, Wuhan, P.R. China

School of Information Management, Central China Normal University, Luoyu Road, Wuhan, P.R. China

Keywords:

Knowledge Fusion, Knowledge Service, Process Model, Implementation Pattern, System Framework.

Abstract:

In big data environments, knowledge fusion is the necessary prerequisite and effective approach to implement

knowledge service. This paper ﬁrstly analyses the requirements of big data knowledge service and the contents

of knowledge fusion, constructs a multi-level architecture of knowledge service based on knowledge fusion.

Then, this paper presents a design of a knowledge fusion process model and analyses its implementation

patterns. Finally, a system framework of big data knowledge service is proposed based on knowledge fusion

processes, in which processes of both knowledge fusion and knowledge service are organically combined

together to provide an effective solution to achieve personalized, multi-level and innovative knowledge service.

1 INTRODUCTION

Knowledge is awareness and understanding about

people or things in the objective world, which is gen-

erated by feeling, communicating and logic reasoning

activities in the course of practice and education and

maybe facts, information or skills. With the devel-

opment of data creating, releasing, storing and pro-

cessing technologies, data is showing a rapid growth

trend in all society areas. Of all the data available

to the human civilization, 90% were produced in the

past two years (Meng and Chi, 2013). Big data gives

rise to the emergence of large scale knowledge bases.

Famous knowledge base research projects, e.g. DB-

pedia, KnowItAll, NELL and YAGO, use informa-

tion extraction techniques acquiring knowledge from

high quality network data sources (e.g. Wikipedia),

and automatically realize its construction and man-

agement (Suchanek and Weikum, 2014). Facts in

knowledge bases, including entity names, semantic

classes and their relationships, are derived from tex-

tual data in Web.

Meanwhile, big data brings about information

overload and pollution too, in which knowledge also

presents characteristics of heterogeneity, diversityand

independence. In the era of data, with rapidly increas-

∗

This paper is supported by the Chinese NSFC Inter-

national Cooperation and Exchange Program, Research on

Intelligent Home Care Platform based on Chronic Diseases

Knowledge (71661167007).

†

Corresponding Author

ing of information and knowledge, knowledge dis-

covery has become the research focus in various dis-

ciplines, including data science and information sci-

ence(Ye and Ma, 2015). Therefore, in order to im-

prove the efﬁciency and quality of knowledge service,

issues of analysing and utilizing knowledge existing

in big data, eliminating the inconsistency between dif-

ferent sources, and extracting, discovering and induc-

ing the potential valuable connotations, have become

important in knowledge management researches.

Knowledge Service (KS) is to meet user needs, by

analysing the knowledge requirements in the domain,

and applying knowledge acquisition, analysis, reor-

ganization and implementation processes via service

procedures helping users to ﬁnd and form solutions

(Zhang, 2000). In big data environments, KS is no

longer limited to traditional literature and information

services, but turns its perspective into massive frag-

mented information, user behaviors and relationships,

and the resulting real-time, unstructured and machine

data (Qin et al., 2013). Features of multi-source, het-

erogeneous, real-time and low value density data have

brought unprecedented challenges to KS.

Knowledge Fusion (KF) is to acquire and utilize

knowledge aiming at the KS problems. Operated by

KF activities, implicate and undiscovered valuable

knowledge is mined from various distributed and het-

erogeneous data sources. KF converts autonomous

knowledge into a new one with higher levels of in-

tension and reliability, while KS helps users to ﬁnd

potential associations between knowledge and fact,

116

Wang, F., Fan, H. and Liu, G.

Big Data Knowledge Service Framework based on Knowledge Fusion.

DOI: 10.5220/0006036301160123

In Proceedings of the 8th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2016) - Volume 3: KMIS, pages 116-123

ISBN: 978-989-758-203-5

improve decision-making levels by making more efﬁ-

cient, objective and scientiﬁc judgments. KF becomes

a new growth point for KS (Tang and Wei, 2015).

2 RELATED WORK

2.1 Processes of KF

KF is a new concept developed on Information Fu-

sion. There are many intersections between the two

research areas. The early deﬁnition of KF is given

by Preece in the KRAFT project (Preece et al., 2001),

refers to a process locating and extracting knowledge

from multiple, heterogeneous on-line sources, and

transforming it so that the union of the knowledge

can be applied in problem-solving. Smirnov studies

on patterns for context-based KF in decision support

system (Smirnov et al., 2015), provides several KF

patterns including Simple, Flat, Extension, Conﬁg-

ured, Instantiated, Historical and Adaptation fusion,

speciﬁcally for decision-making demands.

Hou (Hou et al., 2006) and Xu (Xu et al., 2010)

believe that KF is to acquire new knowledge via intel-

ligently processing distributed databases, knowledge

bases and data warehouses by transforming, integrat-

ing and fusing knowledge. Xu gives a KF frame-

work based on ontology to reduce the fused knowl-

edge scale and improve its validity and accuracy.

The framework is composed by functional modules

such as constructing meta-knowledge set, determin-

ing knowledge measure indicators, designing fusion

algorithms and handling post-fusion knowledge.

Gou (Gou and Wu, 2006) proposes a method to

share and integrate distributed knowledge sources by

using the fusion process to map knowledge objects

into the ontology base and construct the knowledge

set, in which KF algorithms based on Genetic Algo-

rithm and Semantic Rule are used, and a feedback

mechanism is considered to optimize the fusion pro-

cess and knowledge space.

Qiu (Qiu and Yu, 2015) and Guo (Guo et al., 2012)

review and evaluate research trends and theoretical

developments of KF, point out that, there is not yet

a general framework for KF systems, as well as di-

rectly applicable KF algorithms and standardized pro-

cedures. Existing researches mainly focus on speciﬁc

KF frameworks, algorithms, and practical theories.

2.2 Framework of Big Data KS

Big data KS is produced in processes of acquiring,

storing, organizing and analysing data for decision-

making, which is a new model of information service

used to solve demands of handling structured, semi-

structured and unstructured data multi-dimensionally.

It has features of intelligent services oriented, auton-

omy demands, uncertainty and customer driven, and

is based on processes of sharing knowledge, ability

and resources (Qin et al., 2013). Kawtrakul indicates

that intelligently applying KS promotes the demand

of innovative and service-oriented economy, and the

key function of innovative service is to provide per-

sonalized KS (Kawtrakul, 2010). Research on big

data KS management needs to solve several key is-

sues such as representable, treatable, combinable and

reliable abilities of big data.

In terms of research on KS framework, Gao (Gao

et al., 2016) studies a library KS model based on asso-

ciated data, divides the KS process into three levels:

publishing resource, integrating resource and apply-

ing resource. Zhou (Zhou, 2008) provides a frame-

work of personalized KS based on SOA, considering

generation and acquisition mechanisms of the frame-

work and its implementation methods, and realizing

personalized KS according to user feedback infor-

mation and different requirements. Zhu (Zhu et al.,

2010) proposes a distributed KS framework based on

cloud-shadowmodel, which is supposed to be applied

in knowledge management systems.

Guan (Guan, 2015) provides a factor-relationship

model of big data KS based on the knowledge sup-

ply chain, “acquisition-processing-storage-service-

transfer”. It uses knowledge bases to store contents

required by KS processes, which is limited in terms of

massive data migration and dynamic knowledge up-

dating. Li (Li et al., 2013) analyses an architecture

of big data KS platform by partitioning it into sev-

eral hierarchical layers, systematically expounds the

key technologies required to build the platform. How-

ever, without analysing module relationships and log-

ical structures in layers, it lacks method analysis for

KS realization.

As discussed above, there is no uniform deﬁnition

of KF concepts and research categories, and a general

KF framework oriented to KS requirements has not

yet formed. It is necessary to carry out research on a

general applicable KF process model, and propose a

system framework of KS based on KF processes.

3 REQUIREMENT ANALYSIS ON

BIG DATA KS

In big data environments, data represents information

assets characterized by such a high volume, velocity,

variety and veracity to require speciﬁc technologies

and analytical methods for transforming it into value,

Big Data Knowledge Service Framework based on Knowledge Fusion

117

which brings new challenges to KS activities such as

knowledge acquisition, analysis and storage. The 4Vs

of big data puts forward new requirements of KS real-

ization such as heterogeneous, multi-source, implicit,

dynamic and veriﬁability demands.

KF is a process of applying knowledge extraction,

analysis, reorganization and integration activities over

multiple heterogeneous knowledge sources, mining

the implied valuable knowledge and information, and

forming new knowledge oriented to user needs. This

section discusses the KF contents oriented to the KS

requirements, and proposes a multi-level architecture

of KS based on KF processes.

3.1 KF Contents for KS Requirements

3.1.1 Heterogeneous Demand for KS

The accessing methods and querying results of het-

erogeneous knowledge sources are different from

each other, and so do its structures and contents.

Implementing KS over heterogeneous knowledge

sources, ﬁrst of all, needs to access, deﬁne and de-

scribe the required contents. Knowledge objects from

different sources may be deﬁned in different ways,

which needs to be transformed via a uniﬁed descrip-

tion method to carry out structural comparison and

content analysis among them. Then, subsequent KF

activities can be applied to discover and correct in-

consistency among the knowledge objects, to elimi-

nate redundant or repeated ones. Thus, oriented to

KS demand of heterogeneity, KF requires activities

of knowledge extraction, knowledge representation,

knowledge trans-formation and knowledge cleansing.

Knowledge extraction is built for extracting em-

bedded contents from sources to form knowledge ob-

jects, through activities of identiﬁcation, comprehen-

sion, selection and induction. Knowledge represen-

tation is a process of deﬁning and describing orga-

nizational structures, association rules and content-

handling mechanisms of knowledge objects. Knowl-

edge transformation transfers heterogeneous knowl-

edge into homogeneous one with uniform structure

represented in the same method, which is the key

issue solving the heterogeneity problem. Knowl-

edge cleansing is a process of reviewing and check-

ing knowledge objects, removing duplicate contents

and correcting errors, providinga uniﬁed and accurate

knowledge source for following KS activities, which

ensures correctness and consistency of the source.

3.1.2 Multi-source Demand for KS

Knowledge existing in multiple sources might be ho-

mogeneous on its structures and contents, but also re-

sults in the multi-source demand of KS, which needs

to select, aggregateand reorganize fragmented knowl-

edge, and so called knowledge integration. The con-

cepts of knowledge integration and knowledge fusion

have overlaps in term of dealing with multi-sources

and multi-structures knowledge objects, both of them

have connections and differences.

Literally, integration is the process of aggregat-

ing multiple individual objects to form a whole one,

while fusion is the process of recombining multiple

individual objects, and splitting and dismantling it

into a complete one. Integration emphasizes aggre-

gation and combination, while fusion is more empha-

sis on merging and reorganizing. After fusion pro-

cesses, knowledge objects should have new emerging

features relative to the original ones. Therefore, this

paper argues that KF is the advanced stage of knowl-

edge integration. KF applies fusion algorithms and

matching rules over the result of knowledge integra-

tion to implement deduction, discovery and innova-

tion of knowledge.

Furthermore, KF is also different from knowledge

aggregation, in which KF has no need to keep and

remain all knowledge concepts, relationships and in-

stances from the original sources, but need to con-

struct the required objects meeting KS demands.

3.1.3 Implicit Demand for KS

The implicit demand refers to identifying effective,

novel, potentially valuable knowledge objects and as-

sociations in services. Knowledge mining is a pro-

cess of ﬁnding and discovering implicit knowledge

objects from knowledge collections, linking the ob-

jects within a more explicit and effective way, and re-

sulting in new knowledge objects and associations.

In KF processes, knowledge mining can not

be completed individually by itself, but work with

knowledge integration together in manners of com-

plementary, interdependent, mutually supportive and

iterative. Knowledge integration constructs founda-

tion sources to support knowledge mining, and the

derivativeknowledgegenerated by knowledgemining

is further used to support recombining and reorganiz-

ing knowledge as well as eliminating redundancies,

by which the result of knowledge integration is opti-

mized and reﬁned progressively. The two processes,

carrying out alternately, help users to obtain solutions

efﬁciently from massive, low value and complicated

data sources, to meet the implicit demand.

3.1.4 Dynamic Demand for KS

In big data environments, rapidly growing and contin-

uously updating data involves the dynamic demand

KMIS 2016 - 8th International Conference on Knowledge Management and Information Sharing

118

#$%&'%()*+

,&-*.)(/0

!+*1&23-2

42(.)$2%&

5%%6

7%%6

8%%6

6%%6

"%%6

9%3**: ;2<=62.$)>2 620%+()>=,+%&?@)@A:2+,>>2@@

95"6BC"6 9D%@2 E,4F 9((:B"G8B6HG8 ;65I H)33&21%.2 A+(*&*-?

Customized 62.$)>2

8.2>)@)*+=62.$)>2

62.$)>2

H%+%-202+(

J%&'2K%3323=

62.$)>2

6(.%(2-?

!+*1&23-2

42:.2@2+(%()*+=

!+*1&23-2

G.%+@L*.0%()*+=

!+*1&23-2=

M&2%+@)+-

!+*1&23-2=

#N(.%>()*+

!+*1&23-2=H%(>/)+- !+*1&23-2=7+(2-.%()*+ !+*1&23-2=H)+)+-

F2(1*.O=!+*1&23-2=6*'.>2@

,@@*>)%(23

5%(%

I)+O23

!+*1&23-2

")2&3=!+*1&23-2=D%@2@

I)(2.%>('.2

D'@)+2@@=

42@2%.>/

92(2.*-2+2*'@=5%(%

6>)Q=

5%(%

Product

Data

8%(2+(

5%(%

!+*1&23-2

H%)+(2+%+>2

!+*1&23-2

8.*$2+%+>2

Figure 1: Muti-Level Architecture of Big Data Knowledge Service based on Knowledge Fusion.

of KS, i.e. the result of KS is constantly changing

with the updates of the knowledge sources. Knowl-

edge maintenance is the activity used to solve the dy-

namic demand. When the source content is modiﬁed

and changed, the fused knowledge needs to be up-

dated as well to ensure the consistency, reliability and

real-time abilities.

According to speciﬁc requirements of users, main-

tenance strategies might be either periodically or im-

mediately. In addition, due to the high cost of mi-

grating large data, incremental maintenance is a fea-

sible approach to implement knowledge maintenance.

Namely, only the changes of the knowledge sources

are captured and used to compute the updates of fused

knowledge, i.e. the fused knowledge is refreshed with

the updates rather than recomputed from scratch.

3.1.5 Veriﬁability Demand for KS

The veriﬁability demand is used to guarantee the cor-

rectness and validity of KS results. Both processes

generating results and source references for making

decisions, need to be traceable and veriﬁable. Knowl-

edge provenance is a process of ﬁnding and tracing

the lineages and the producing procedures of fused

knowledge, which results in either metadata record-

ing and annotating knowledge migration, conversion,

cleansing and integration procedures, or the original

information and initial data content of a speciﬁc fused

knowledge object, in the knowledge sources.

In addition to knowledge provenance, knowledge

evaluation is required for evaluating KF activities and

results, in terms of measuring the standardization, ef-

fectiveness and logicality of the activities, and ap-

praising the correctness, accuracy and completeness

of the results. Knowledge provenance can be used to

evaluate correctness and accuracy, to meet the veriﬁ-

ability demand of KS.

3.2 Multi-level Architecture of KS

As discussed above, KF is the necessary prerequi-

site and effective means to implement big data KS.

To meet KS requirements, KF contents include activi-

ties of knowledge extraction, representation, transfor-

mation, cleansing, integration, mining, maintenance,

provenance and evaluation. Based on these KF activ-

ities, this paper proposes a multi-level architecture, as

shown in Figure 1, constructing the KS implementa-

tion procedure within ﬁve levels: Data-as-a-Service

(DaaS), Infrastructure-as-a-Service (IssA), Platform-

as-a-Service (PaaS), Fusion-as-a-Service (FaaS), and

Software-as-a-Service (SaaS).

Knowledge is widely distributed in various hetero-

geneous knowledge sources in big data environments,

the DaaS layer provides the underlying data sources

for KS activities. The IaaS layer provides support for

accessing data and extracting knowledge from the un-

derlying sources, via protocols, middleware tools or

APIs embedded within the big data infrastructure.

Parts of KF activities, such as knowledge extrac-

tion, representation, retrieval, maintenanceand prove-

nance, need to access knowledge sources and handle

the knowledge initially extracted from the sources.

The PaaS layer is speciﬁcally designed to provide

channels for these KF activities exchanging data with

the IaaS interfaces. The Fusion-as-a-service (FaaS)

layer, which is originally proposed in this paper, con-

tains the core activities of KF process that is used for

mainly solving demands of the upper KS activities.

The SaaS layer constitutes the business logic part

of big data KS directly interacting with users to

achieve the performance of personalized, multi-level

and innovative services. Customized services produce

personalized efforts to meet speciﬁc user demands in

accordance with requirement expressions; Precision

services produce different services for different users,

according to user types, preferences and behavior

Big Data Knowledge Service Framework based on Knowledge Fusion

119

!"#$%&'

()*+,-$

+,)

&."/0

!*)%&'

&."/1)23)

4-#)

5)$/"*6

7"8*9)

(-$-

7"8*9)

&':;,-18-$+".:<:=#)*:'))2>-96

4(:?.-1@#+#

A""1#

B.C*-#$*89$8*):

!1-$C"*D

E)>:(-$-

?!B

E)>:

7)*,+9):

!1-$C"*D

F$"1"3@

G81)#

?13"*+$HD

I+*$8-1

J-.21+.3

K-$)*-1+L)2

=#)*%

M"-1

N".$).$

7"18$+".

I-18)%

-22)2

=#)*:O8)*+)#:P:

G)Q8+*)D).$#

K-+.$).-.9)

;R$*-9

$+".

N1)-.#

+.3

B.$)3*-

$+.3

&."/0

E0H"8#)

G)S*)#

).$+.3

A*-.C"*

+.3

G)$*+),

+.3

&."/0

5)$/"*

Figure 2: Process Model of Big Data Knowledge Fusion.

characteristics. Value-added services are going be-

yond regular service ranges, adopting unconventional

methods to achieve valuable services, as a perfor-

mance of service innovation.

Service management covers quality standards, se-

curity speciﬁcations, evaluation indicators, etc., to en-

sure KS qualities. Service strategy includes service

optimization, distribution and identiﬁcation, etc., to

promote personalized and multi-level services.

4 PROCESSES OF BIG DATA KF

4.1 KF Process Model

KF activities can be treated as functional modules that

transform inputs into outputs, and used for composing

the KF process model. We roughly split the KF pro-

cess into three stages: Pre-KF, In-KF and Post-KF,

shown in Figure 2, in which the wide-white-line ar-

rows represent ﬂows of data and knowledge, and the

ﬁne-line ones represent ﬂows of control information.

In Pre-KF stage, massive knowledge exists in

the initial state, and is widely located in distributed

knowledge bases and network sources. The big data

infrastructure and Web service platforms provide sup-

ports for accessing the Pre-KF knowledge.

The In-KF stage contains the main activities of

KF, either migrating knowledge objects from the

sources and storing it in the knowledge base though

knowledge extraction, cleansing and integration ac-

tivities, or generating knowledge networks to deﬁne

knowledge structures and associates though knowl-

edge representation, transformation and retrieval ac-

tivities. Knowledge networks aggregate and reorga-

nize knowledge deﬁnitions, structures and associates

from the source. Meanwhile, the In-KF stage in-

cludes the part of constructing domain ontology, de-

signing fusion algorithms and rules according to user

requirements and its problem deﬁnitions, which can

be revised and validated by accepting user feedbacks.

Also, it is necessary to establish a measurementmech-

anism for evaluating KF activities and it results, in

order to measure the efﬁciency and effectiveness and

provide a basis for optimizing and improving KS.

KF is a necessary prerequisite and effective means

for the realization of big data KS. From literature

service to information service and knowledge ser-

vice, which is a gradually deepening and developing

process. Based on the fused and derivative knowl-

edge, KS activities are divided into four types: user-

goal oriented, knowledge-content oriented, problem-

solution oriented and value-added oriented.

The user-goal oriented KS is a target driven ser-

vice, which is focusing on “whether or not did the

service solve users problems?”, but not “does the

service provide the information users need?”. The

knowledge-content oriented KS is based on logic ac-

quisition, through information acquisition and inte-

gration to form knowledge products in line with the

needs of users. The problem-solution oriented KS is

committed to helping users ﬁnd or form a compre-

hensive solution through the continuous inquiry, anal-

ysis and reorganization of information and knowl-

edge. The value-added KS is concerned with repro-

cessing the existing services, realizing service values

by improving the efﬁciency of applying and innovat-

ing knowledge, so as to form new service products

with unique value.

4.2 KF Implementation Pattern

KS is supported both by knowledge that is rela-

tively stable, veriﬁed and frequently accessed, such as

the pathogenesis of a disease, clinical manifestation,

characteristic index, commonly used drugs, treatment

and pathological knowledge in chronic disease man-

agement situations; and by knowledge that is widely

KMIS 2016 - 8th International Conference on Knowledge Management and Information Sharing

120

existing in the network, real-time updated, usually

changing and unfrequently visited, such as the diag-

nosis treatment cases of the same disease type, and its

related research works, news reports, and so on.

Costs of big data accession, migration and stor-

age are very high. It is infeasible for realizing KS

by organizing and storing all source knowledge into a

single knowledge warehouse. Thus, this paper distin-

guishes KF implementation patterns into three types:

oriented to materialized knowledge warehouse, ori-

ented to virtual knowledge network, and based on the

mixed models.

4.2.1 Oriented to Materialized Knowledge

Warehouse

The KF implementation pattern oriented to material-

ized knowledge warehouse is to prepare knowledge

that is relatively stable and need to be frequently vis-

ited by KS activities, in terms of providing a local

organization and storage mode.

Knowledge warehouse is an integrated collection

facing topics, non volatile and changeable, which is

used to decision-support for knowledge service and

management processes. The so-called materialized is

to acquire the knowledge objects required by KS in

advance, and physically preserve it in warehouses, in

order to avoid repeatedly visiting and computing op-

erations over the sources.

Under this pattern, data analysing and processing

tools provided by the big data infrastructure are used

to support knowledge extraction activities. Appro-

priate knowledge objects are migrated and stored in

the warehouse by steps of knowledge cleansing, inte-

grating and loading. During the procedure, not only

knowledge objects, but also knowledge structures and

associates are handled to establish knowledge shar-

ing and reusing mechanisms. There is also implicit

knowledge in the warehouse needing to be excavated

and discovered, according to the types of user require-

ments, to form types of derivative knowledge.

4.2.2 Oriented to Virtual Knowledge Network

The KF implementation pattern oriented to virtual

knowledge network is used for handling knowledge

widely existing in the network and providing access

to methods for KS activities.

Knowledge networks might be divided into

three types: subject-subject, subject-knowledge and

knowledge-knowledge. In this paper, we refer to

knowledge-knowledge networks deﬁning knowledge

structures, associates and constraints among knowl-

edge objects. Virtual knowledge networks contain

the knowledge deﬁnitions, but do not contain speciﬁc

knowledge units and instances. The time of execut-

ing conventional data query and analysis technologies

will increase to be unacceptable for handling big data,

and the pattern oriented to virtual network is mainly

used for dealing with this high-costs.

In addition to containing deﬁnitions of structures,

associates and constraints of fused knowledge ob-

jects, virtual knowledge networks use metadata, i.e.

the query statements over the data sources, to describe

procedures generating the fused objects. When KS

users need to access speciﬁc knowledge contents and

instances of virtual knowledge networks, user queries

are decomposed, step by step, into various speciﬁc

sub-queries over data sources. Based on Web service

platforms and its data processing API, knowledge re-

trieval module acquires query results from the source

in manners of stream or batch processing. Knowl-

edge representation, transformation and integration

activities further use the results to produce fused and

derivative knowledge to meet the user requirements.

4.2.3 Based on Mixed Models

The KF implementation pattern based on the mixed

modes comprehensively handle knowledge types re-

quired for KS activities, in which both material-

ized warehouses and virtual networks are considered.

Since linked knowledge will change along with the

changes of network, it is necessary to apply dynamic

selection of data sources and evaluate its knowledge

quality, and design adaptable strategies for choosing,

maintaining and updating the fused knowledge mate-

rialized in the warehouse.

5 SYSTEM FRAMEWORK OF

BIG DATA KS

5.1 KS Framework Model

As discussed above, KF processes either gather

and transform various, distributed and heterogeneous

knowledge into a materialized warehouse, or use

virtual network to link massive remote knowledge

sources. Both of them feasibly provide available

knowledge sources for following-up KS activities.

In order to achieve personalized, multi-level and

innovative services, this paper further analyses the KS

process, decomposes its activities, and constructs a

system framework based on the KF process model,

as shown in Figure 3, to meet the KS requirements.

Personalized and multi-levels demands for pre-

cision services require to analyse user requirements

and decompose complex problems into smaller, more

Big Data Knowledge Service Framework based on Knowledge Fusion

121

!"#$%&'#$(#"%)%*#+'($#,#-."

/0%1$23#""

4567'6.(2-

/-287#9:#%;#$5(3#%<1=

;#$5(3#

>2,?2-#-.

;#,6-.(3

*#6"2-(-:

0(#79

<-67@A(-:

B-.272:@

C6"#

;#$5(3#

D#32,?2".

;#$5(3#

D#"3$(E(-:

;#$5(3#

*#".$'3.'$(-:

;#$5(3#

F6.3G(-:

;#$5(3#

=,?7#,#-.(-:

H#.I

;2'$3#

D6.6

;2'$3#

/-28I

C6"#

Pre-KF

F6.#$(67(A#9

L($.'67

/-28I%H#.I

Post-KF

Knowledge Service Stratigy

;#$5(3#

B?.(,(A(-:%

;#$5(3#

%<77236.(-:%

;#$5(3#%

=9#-.(M@(-:%

1$2E7#,

D#32,?2"(I%

/-28I

=-.#:$6.(2-

/-28I%

>7#6-"(-:

/-28I

*#.$63.(2-

/-28I

*#.$(5#67

/-28I

N$6-"M2$,6.(2-

/-28I

*#?$#"#-.6.(2-

In-KF

/-287#9:#

F6(-.#-6-3#%

/-287#9:#

1$25#-6-3#%

/;%F6-6:#,#-.

;#$5(3#

;#3'$(.@%

;#$5(3#

&'67(.@

;#$5(3#

4567'6.(2-

Figure 3: System Framework of Big Data KS based on KF.

manageable and tractable sub-problems. Then, the

service components needed to solve sub-problems are

compared with the ones provided by the KS system.

Throughservice matching and restructuring activities,

KS solutions for the entire problem are formed, and ﬁ-

nally implemented with the support of KF processes.

In the framework, the KS API is the interface in-

teracting with users. User queries and requirements

are analysed and resolved by semantic reasoning and

domain analysis activities based on domain ontol-

ogy. Then, activities of problem decomposition, ser-

vice matching, service restructuring and service im-

plementation are used to realize the requirements.

Semantic reasoning focuses on knowledge con-

cepts and their relevance in semantic levels, and do-

main analysis is an analytical activity of the concept

objects in a particular ﬁeld, such as domain deﬁni-

tion, commonality abstraction, characteristic descrip-

tion, concept description, relationship identiﬁcation,

and so on. Semantic reasoning and domain analy-

sis are necessary manners to decompose user require-

ments, and generally require the participation of do-

main experts for completion.

Service matching is a process of retrieving and

ﬁnding the service component satisfying user require-

ments from the base storing various service compo-

nents and speciﬁcations. Service matching may use

the Universal Description, Discovery, and Integration

(UDDI) standards to establish norms to complete ser-

vice releasing and discovering processes. Service re-

structuring is to test and verify the service compo-

nents found in service matching, remove redundan-

cies, and reassemble them into a feasible service solu-

tion. After service restructuring, the service solution

is converted to access requests over the fused knowl-

edge, and executed by service implementation activi-

ties, mainly including processes of request analysing

and decomposing, request rewriting and executing,

result assembling and integrating, etc.

Service description and service decomposition ap-

ply granularity analyses on services provided by KF

processes and its results, i.e. fused knowledge, re-

sulting in ﬁne-grained service components and spec-

iﬁcations, which are used for matching with the de-

composed ﬁne-grained user requirements, and re-

constructing service solutions further accurately and

speciﬁcally satisfying the user requirements.

The KS framework presented here reveals cor-

responding relationships and inﬂuence mechanisms

among KS activities and KF processes, and provides

a systematic solution for implementing personalized,

multi-level and innovative services for KS users.

5.2 KS Strategy

The KS strategy is concerned with the processes

of service identiﬁcation, allocation and optimiza-

tion. Service identiﬁcation is to abstract applicable

and functionalized service components from complex

business ﬂows, which is carrying out standardized

management on the service interfaces. The KF pro-

cess, as the basis of KS, includes knowledge repre-

sentation, extraction, cleansing, transformation and

integration modules. Different modules combined to-

gether form different KF implementation patterns ori-

ented to either materialized warehouses or virtual net-

works. Service identiﬁcation is mainly to standardize

the interface deﬁnition and component functionality

of the services provided by KF processes, and enable

them to be feasible and accessible for KS activities.

Service allocation is the process of assigning the

identiﬁed service to appropriate users and problem

domains according to the requirements, so as to real-

ize precision services. Among user requirements and

service components with different granularity, there

might be one-to-one, one-to-many or many-to-many

mapping associations. Based on analysing and clas-

sifying these associations, allocation principles and

mechanisms are developed to supply references for

service matching and restructuring.

In speciﬁc situations, one requirement might cor-

respond with a variety of service allocation strategies.

Different strategies produce different service match-

ing and restructuring schemes, so that service solu-

tions and implementation results are varied. Service

optimization is to analyse affecting effects of service

allocation strategies, adjust service matching and re-

structuring schemes, and optimize KF process results,

so as to ensure the realization of innovative KS.

KMIS 2016 - 8th International Conference on Knowledge Management and Information Sharing

122

5.3 KS Management

KS management is essential to improve the quality of

service (QoS) of the framework, in terms of standard-

izing and administrating security measures, quality

standards and evaluation indicators of KS. Security

measures include mechanisms of user authentication,

role administration and privilege management, gener-

ally working along with the service allocation strate-

gies, to assign appropriate KS solutions for various

types and levels of users, in which both the security

and the precision of the services are considered.

Quality standards and evaluation indicators are in-

terdependent and complementary. Combined with

the evaluation module of KF process, quality stan-

dards are used to determine and classify the corre-

spondence relationships between KF results and KS

requirements in various degrees and levels, which

is also the standard of satisfaction for KS demands,

while evaluation indicators are used to assess the QoS

in perspectives of user cognition and sensation, to in-

dicate the differences between user expectation and

user perception, between expected service results and

experienced service results, and between desired ser-

vice qualities and perceived service qualities.

6 CONCLUSIONS

KF is a new research topic rising to the challenges

of KS. The emergence of massive, various and het-

erogeneous knowledge in big data environments im-

plies new demands on both KF and KS implementa-

tions. This paper reviews current research on KF pro-

cesses and KS frameworks, and analyses big data KS

requirements in terms of KF contents. Then this pa-

per constructs the KF process model and implementa-

tion patterns, proposes a multi-level architecture and

a system framework of big data KS, organically com-

bining with KF processes together, to meet demands

of personalized, multi-level and innovative services.

In future work, we will consider to apply the KF

process model for empirical analysis in a speciﬁc do-

main, i.e. chronic diseases knowledge management.

REFERENCES

Gao, J., Li, K., and Liang, Y. (2016). Research on library

knowledge service model based on linked data. Infor-

mation Science, 34(05):64–68.

Gou, J. and Wu, Y. (2006). Knowledge fusion: A new

method to share and integrate distributed knowledge

sources. In 1st European Conference on Technology

Enhanced Learning, Greece, October, 1-4.

Guan, S. (2015). Study on key elements and implemen-

tation model of big data knowledge service platform.

Library Forum, 06:87–93.

Guo, Q., Guan, X., and Cao, X. (2012). Research progress

and trends of knowledge fusion. Journal of China

Academy of Electronics and Information Technology,

7(3).

Hou, J., Yang, J., and Jiang, Y. (2006). Knowledge fusion

algorithm based on metadata and ontology. Journal

of Computer-Aided Design and Computer Graphics,

18(06):819–813.

Kawtrakul, A. (2010). Beyond knowledge management:

knowledge service innovation. In Intl Conference on

Data Engineering and Management.

Li, Z., Cui, J., and Chen, C. (2013). Research on key tech-

nologies of big data knowledge service platform con-

struction. Information and Documentation Services,

02:29–34.

Meng, X. and Chi, X. (2013). Big data management: con-

cepts, technologies and challenges. Computer Re-

search and Development, 50(1):146–169.

Preece, A., Hui, K., and Gray, A. (2001). Kraft: An agent

architecture for knowledge fusion. International Jour-

nal of Cooperative Information Systems, 10(12):171–

195.

Qin, X., Li, C., and Mai, F. (2013). The connotation, typical

features and conceptual model of large data knowl-

edge service. Information and Documentation Ser-

vices, (02):18–22.

Qiu, J. and Yu, H. (2015). Research progress and trends of

knowledge fusion in perspectives of knowledge sci-

ence. Library and Information Service, 59(08):126–

132+148.

Smirnov, A., Levashova, T., and Shilov, N. (2015). Patterns

for context-based knowledge fusion in decision sup-

port systems. Information Fusion, (21):114–129.

Suchanek, F. and Weikum, G. (2014). Knowledge bases in

the age of big data analytics. In Proceedings of the

VLDB Endowment, volume 7, pages 1713–1714.

Tang, X. and Wei, W. (2015). The growth points of knowl-

edge service in big data age. Researches in Library

Science, (05):9–14.

Xu, C., Li, A., and Liu, X. (2010). Knowledge fusion archi-

tecture. Journal of Computer-Aided Design and Com-

puter Graphics, 22(07).

Ye, Y. and Ma, F. (2015). The rise of data science and its

relation with information science. Journal of Informa-

tion Science, 34(6):575–580.

Zhang, X. (2000). Towards knowledge service: searching

for growth points of library and information work in

the new century. Journal of Library Science in China,

26(5):32–37.

Zhou, L. (2008). Personalized knowledge service architec-

ture based on soa. In Proceeding of 22nd National

Symposium of Computer Information Management.

Zhu, J., Shen, S., and Zhao, D. (2010). Knowledge ser-

vice architecture based on ontological cloud shadow

model. In Academic annual conference of information

society of China Electronic Association Proceeding.

Big Data Knowledge Service Framework based on Knowledge Fusion

123