A Formal Model for Intellectual Relationships among Knowledge Workers
and Knowledge Organizations
Mao-Lin Li1 and Shi-Kuo Chang1
Department of Computer Science, University of Pittsburgh, Pittsburgh, PA, USA
{ moli, chang }@cs.pitt.edu
1
Abstract—Academic learning network is consisted of multiple
knowledge organizations and knowledge workers. The degree
of relationship can be derived from the interaction within them.
In this paper, we propose a formal ownership model to
decribe the interactions within academic learning network and
further provide an evaulation process to quantify the degree of
relationship. Proposed approach is also integrated with
realistic social platform SMNET.
Keywords- Social-Network, Formal Model, Knowledge Agents
I.
INTRODUCTION
The analysis of social-network relationships becomes a critical
issue in recent years. People build their own network and
connect to others by sharing their life (e.g. Photos, Videos,
Articles, etc…) on specific platforms. (e.g. Google+,
Facebook, etc…). These platforms analyse the information
provided by users and further discover potential relationships
among users. Numerous approaches including graph theory,
information retrieval or machine learning techniques are used
to evaluate the relationships among users and groups in
conventional social-networks. Such relationships will also be
significant metrics for commercial purpose (e.g. in
advertisement and in recommendation systems).
An academic learning network can be considered as another
type of social network. An academic learning network consists
of knowledge workers and knowledge organizations.
Workers/Organizations
publish/upload
their
papers,
view/download others’ publications. Workers can cooperate
with others in an organization for the same research goal or
compete with other research organization. However, there are
few discussions about how to model and analyse these
intellectual relationships among different academic
organizations.
In this paper, we propose a formal model to describe the
intellectual relationships of completed academic network.
Combined with existed social graph platform, the relationships
within different people/groups in academic network can be
visualized and further provide more information about
academic social network.
Our main contributions consist of three parts:
1.
Propose a formal ownership model to describe
completed academic networks.
2.
3.
Combine with social graph to visualize intellectual
relationships.
Provide significant and flexible characteristic
matrices to evaluate relationships without using
sophisticated algorithm.
The intellectual relationships would allow us to minimally
recognize who are the collaborators and who are the
competitors in an academic learning network.
Slow
Intelligence principles [2] can then be applied so that the
academic learning network can achieve its goals.
The paper is organized as follows: In Section 2 we present the
ownership model in academic learning network. The social
graph will be presented in Section 3.
II.
RELATED WORK
Social network analysis use different metrics (measure) as a
reference for evaluating the relationships within various
individuals in social network, e.g. Centrality, degree and
closeness [6]. Most of these metrics are based on the situation
of nodes and link in social network and then using graph
theory to analyze the relationships. In proposed model, in
addition to above static metrics, we further consider various
dynamic actions within different situations in academic
learning network.
Some analysis software like UCInet [9] helps research to
explore various size of social network with visualization and
further investigation. Our ownership model is applied in a
realistic academic learning network SMNET [4], SMNET
integrated with Intellectual Property Rights model (IPR)
model to formalize an institutional aggregator for metadata
and content. In addition to traditional social graph, SMENT
can represent hierarchical level in academic learning network,
from individual worker to academic organization Users can
describe their network with our formal descriptions, and
upload their contents to the academic learning network, the
content could be academic publications, technical reports the
completed social graph can be generated automatically.
Relationships evaluation in academic learning network. Many
researches [7,8] use bibliometric approach like citation and cocitation analysis to estimate the relationships in the network,
the result can be used for finding strategic alliance or
identifying the structure of invisible colleges. Our formal
model also can include citation and co-citation metrics, since
the source of above two metrics can be obtained by
“references” part in a publication. Furthermore, out model can
204 evaluate the degree of cooperative and competitive in
academic learning network.
III.
OWNERSHIP MODEL


In this section, we first introduce the concept of academic
learning network. Then we present the definition of proposed
ownership model and demonstrate a completed academic
network with proposed ownership model.

Permission := {ContentKind X Right X UserKind} is the
set of permission depend on ContentKind and UserKind
to map corresponding action rights.
Access History := {Who, When, What, Frequency} is
the set of access history of each publication.
Relationship := {author, member, leader… } is set of
relationships within workers, organizations and
publications.
A. Academic Leaning Network
ii.
Filter Functions
Fig. 1 shows an example of academic learning network.
Knowledge workers (KW1 and KW2) publish their academic
To represent the ownership in academic learning network
papers (P1 to P3) or view/download academic paper from
precisely, we further propose filter functions to achieve
others’ publications (KW3). A knowledge organization (KO1)
permission purpose. Knowledge workers will configure the
can have multiple knowledge workers (KW1 and KW2).
permission constraints of each publication according to
Workers have the ownership of their publications and workers
ContentKind and UserKind of publication. Every non-author
can configure specific permission for their publications. There
must access publication through corresponding filter. The
will be some actions and relationships within different objects.
representation of filter function as follows:
The academic learning network will be established through
WIDi .permission(PID j ),i, j int
multiple objects and their interaction.
, which means the permission for workeri’s publicationj and
Download
Publish
filter function will return corresponding Right to requesters.
P1
Publish
Publish
P2
P3
KW: Knowledge Worker
KO: Knowledge Organization
Pi: ith publication
KW2
Figure 1 A sketch of academic learning network
B. Formal Descriptions
Here we introduce our ownership model. DISIT[1] defines a
set of formal definition for Intellectual Property Right (IPR)
model. We construct our model based upon their definition
and further provide a light-weighted but sufficient model for
academic learning network. It is a flexible model because all
the parameters in ownership model can be configured
according to the requirement. The sets for the ownership
models are defined as follows:
i.
Basic Attributes







IV.
KW3
KW1
SOCIAL GRAPH
The formal definitions described above helps us to model our
knowledge workers, organizations publications and filter. Now
we further analyze the relationships within these participants
in
academic
learning
network.
We
consider
workers/organizations/publications/filter in academic learning
network as individual objects (Nodes) and the relationships
within these participants can be represented by the links
connecting the nodes. With this visualization, we can quantify
the features in graph, e.g. the number of node, the number of
link to evaluate the relationship within participants in an
academic learning network.
A.
Object Representation
Each object is comprised of various attributes. We visualize
these objects with social graph concept introduced in [1].
Table 1 lists all visual icons we use in academic learning
networks, the detailed explanations are in the following
sections.
Table 1 Visual icons for Social Graph
Name
Graph
Attributes
Knowledge Worker
{WID, OID, PID,
Permission, Filters}
WID := {WID1…WIDn} is the set of worker IDs, each
knowledge worker has its unique ID as its identification.
OID := {OID1…OIDn}is the set of organization IDs,
each organization has its unique ID as its identification.
PID := {PID1…PIDn}is the set of publications IDs, each
publication has its unique ID as its identification.
Description := { keywords, abstract …}is the set of
publication descriptions.
ContentKind := { journal, conference, workshop…} is
the set of publication categories.
Right := {publish, viewAll, viewPartial, download,…} is
the set of actions of workers/organizations.
UserKind := {author, lab member,…} is the set of user
kinds.
Knowledge
Organization
Publication
Action
Filter
205 {WID, OID, Leader}
{WID, PID,
ContentKind,
Description,
Permission }
{Right}
{WID, PID,
Permission,
ContentKind,
UserKind}
WID:{“Mike”, “Jack”}
PID:{“P3”}
Description:{…}
Author
Co-Author
P3
WID:{“Jack”}
PID:{“P3”}
…
WID:{“Mike”}
PID:{“P1”, “P3”}
…
Member
Leader
G
View
OID:{“G”}
Leader:{“Mike”}
WID:{“Mike”, “Jack”}
Author
WID:{“Mike”}
PID:{“P1”}
Description:{…}
Filter1 (Mike.permission(P1))
P1
PartialView
Filter1 (Mike.permission(P1))
WID:{“Ben”}
PID:{“P5”}
…
Download
Relationship
{Relationship}
publications. Actions we use here are download, viewAll and
viewPartial.
Figure 2 Academic Learning Network with Social Graph
v.
i.
Knowledge Worker
A knowledge worker object includes its unique ID (WID) and
their publication ID (PID) in academic learning network.
Workers configure filters for their publications with specific
vi.
permission to control the access right of a publication.
ii.
Knowledge Organization
An organization comprises multiple members (e.g. Professors
and Students in a University). Knowledge organization object
includes its unique ID (OID) and knowledge workers (WID)
in this organization. Each organization will be hosted by a
leader. Leader will manage the relationships within members
in the same organization.
iii.
Publication
A publication object consists of its unique ID (PID), authors’
ID (WID), type (ContentKind) and the description about this
publication.
iv.
Action
Actions in academic learning network are defined in Right in
our ownership model, which means the actions to access
Filter
A filter controls the access rights for a publication, the
permission constraints are configured by author. In academic
learning network, every publication should be access through
its filter except its author.
Relationship
There are at least three kinds of relationships in academic
learning network:
1.
Workers/Organizations V.S. Workers/Organizations
Workers and organizations in networks have some
relationships with others. E.g. Colleague, members in
same organization.
2.
Workers V.S. Publications
Workers have relationship with publications. E.g.
Author, Co-Author.
3.
Publication V.S. Publications
There exist some relationships within publications. E.g.
Reference.
B.
Academic Learning Network with Social Graph
With above social graph, now we explain how to combine our
ownership model with social graph to establish an academic
learning network. Fig. 2 depicts a simple academic learning
206 network with our social graph. We will go through this graph
to explain our idea in detail.
i.
Sub-Graph
shows a spectrum to represent our evaluation reference. Here
we consider positive weight represent the degree of
cooperative relationship and negative weight the degree of
competitive relationship.
Each knowledge worker can be represented by a sub-graph. In
Fig. 2, Mike is a knowledge worker and he has a publication
P1. Hence, the object Mike has attributes: WID:{Mike} and
PID:{“P1”, “P3”}. Each publication object also has their
attributes, e.g. WID, PID and Descriptions, the relationship
between authors and their publication are represented with
Author relationship graphs. According to defaulted rule in our
ownership model, authors can directly access their
publications without filter, so Mike can perform an action
View to his publication P1. Mike also sets a filter1 for P1, Any
non-author have to access P1 through filter1, and the
permission of filter1 depends on the ContentKind of P1 and the
UserKind of accessors.
ii.
Competitive
Cooperative
0
+
Figure 3 The spectrum of relationship evaluation
A. Evaluation Metrics
The matrices we use have two categories, baseline weight and
link weight. The evaluation process will accumulate the
weights from these two parts.
i.
Baseline Weight
The baseline weight represents a base relationship within
different knowledge workers in the academic learning network.
Here we use the similarity in their publications’ description.
Since we can obtain abstract and keyword information in their
publications’ description. This information can give us an
initial evaluation for knowledge workers in academic learning
network. We choose the number of the same words in their
descriptions as a reference.
The connection within sub-graphs
In addition to the representation of individual knowledge
worker, there will be some connections within different
knowledge workers (sub-graphs), the connections can be
categorized into three types:
1. Relationship link:
Workers cooperative with each other.
2. Filter-action link:
One want to access the publication of others.
3. Group link:
Workers in the same knowledge organization.
Baseline weight (BW<a,b>)
= # of the same words in publications’ descriptions of
KWa and KWb.
ii.
In Fig. 2, Mike and Jack are in the same knowledge
organization G, and Mike is the leader in G, so there are two
relationship graphs Leader and Member links <Leader,
Member> within Mike, Jack and G, this connection is called
Group link. Mike has one cooperative publication P3 with Jack,
so both Mike and Jack have Author and Co-Author
relationships with P3, this connection is called Relationship
link and we use <Author, Co-Author> to represent it. Now if
Jack and Ben want to access P1, since both of them are not
author of P1, the request will be checked by the filter Filter1
configured by Mike. The permission here lets non-author
knowledge worker can partially view P1. This connection is
called Filter-action link and denoted with <Filter1,
PartialView>.
Link Weight (LW<a,b>)
As we mentioned in previous section, the connections within
knowledge workers are established with relationship link,
filter-action link and group link. Hence, we can assign a
specific weight to each link. The value can be configured
according to different situations. Here we just propose some
examples in our evaluation.
This social graph is flexible and can be expanded with
arbitrary connections, in next section, we will introduce how to
utilize the features in social graph to evaluate the relationships
in academic learning network.

Weights for relationship link (Weight_RL<a,b>)
The weight of relationship link depends on the type of
relationship. In our ownership, the relationship within
knowledge worker can be established through objects
(cooperative publication) or knowledge organization
(members, colleagues). We can assign different weights
according to various relationship links. E.g. we can set the
weight of <Author, CoAuthor> relationship link will larger
than the weight of <Author, Advisor> relationship link.
Weight_RL<a,b>
V.
n
RELATIONSHIP EVALUATION
After establishing academic learning network with proposed
social graph. Now we explain how to utilize our social graph
to evaluate the relationship within academic learning networks.
In this paper, we focus on the degree of cooperative and
competitive relationship as our evaluation reference. Fig. 3
207 =

weight_relationshipi within KWa and KWb ,
i
i: Relationship in Typei

Weights for filter-action link (Weight_FL<a,b>)
= C + 20
The weight of filter-action link depends on the information
in access history of publications. Since filter-action links
will be established when non-author knowledge workers
trying to access other knowledge workers’ publications, we
can obtain these information from publications’ access
history, e.g. number of download, view and accessed by
who. Hence, we can use these information as a reference
for evaluation.
And then we evaluate the relationship between Mike and Ben.
We also assume their baseline weight (BW<Mike, Ben>) is a
constant C and there is a filter-action link < Filter1, Download
>. According to the formula, the numeric relationship between
Mike and Ben is:
Weight_Relationship<Mike, Ben> :
= BW + (Weight_FL)
= C + (W<Filter1, Download>)
= C + 10
n
Weight_RL<a,b> =

# of Actioni * weight_Actioni
i

Weights for group link (Weight_GL<a,b>)
The weight of group link depends on the connection in a
knowledge organization, since in a knowledge organization,
there are various type of connections, e.g. <Leader,
Member> or <Colleague, Colleague>, we can assign
different weights to these connections.
n
Weight_GL<a,b> =

connectioni within KWa and KWb
i
in organization
iii.
Total Evaluation
VI.
After defining baseline weight and link weight, we can
evaluate the relationship within any knowledge worker.
Note that all weights in our evaluation approach can be
configured, the type of relationship link and the connection in
an organization can also be extended for different academic
learning network.
A. The SMNET Scenario
Here we describe how to map our formal description to
SMNET platform. Table 3 lists the mapping between SMENT
metadata configuration and proposed formal description, once
users provide their information, the corresponding social graph
will be generated automatically. SMNET has versatile
configuration to generate intent social network. Here we only
list part of attributes that describing our academic learning
network.
B.
An Example
Now we demonstrate our evaluation approach with an
example. Table 2 lists exemplary weights for different links.
We use these weights to evaluate the relationship among
knowledge workers as shown in Fig. 2.
Category of Link
Relationship Link
Filter-Action Link
Group Link
Table 3 The mapping between SMNET and Formal Description
SMNET
Formal Description
Title
PID
Creators
{WID1, WID2,…}
Descriptions
{ Keywords, Abstract …}
Taxonomy Classification
ContentKind
->ContentKind
Group Section
Permission
Weight
+20
-5
-10
+5
First, we evaluate the relationship between knowledge workers
Mike and Jack. We assume their baseline weight (BW<Mike, Jack>)
is a constant C, and there are three links connecting them
together, relationship link <Author, Co-Author>, group link
<Leader, Member> and filter-action link <Filter1,
PartialView>. According to the formulae given in 4.1, the
numeric relationship between Mike and Jack is:
Weight_Relationship<Mike, Jack> :
= BW + (Weight_RL+ Weight_FL+ Weight_GL)
= C + (W<Author, Co-Author> + W<Filter1, PartialView> + W<Leader, Member>)
= C + (20 + (-5) +5)
CASE STUDY
We apply our formal model into a realistic academic learning
network – SMNET [4], developed by Dr. Paolo Nesi and
DISIT Lab in Florence, Italy, as the test bed. In the following,
we first explain the scenario of SMNET with a realistic case as
an example 5.
Weight_relationship(KWa, KWb) = BW<a,b> + LW<a,b>
Table 2 Metrics for Links
Kind
<Author, Co-Author>
<ViewPartial>
<Download>
<Leader, Member>
C. Analysis
From above example, we can observe the relationship
evaluation depends on the type of links and their user-defined
parameters. In general case, we can consider the group link
(static) and filter-action link (dynamic) will occupy the major
part in relationship evaluation. Since the knowledge workers
in the same knowledge organization, the probability of their
cooperation will be increased. For an example, the relationship
link within a group will be more positive (e.g. <Leader,
Member>) than the relation link within different knowledge
organization. Furthermore, if there are number of filter-action
link between two knowledge workers, we should evaluate
their relationship according to the dynamic actions.
Fig. 4 shows the scenario of using our ownership model with
SMNET platform, when knowledge workers want to upload
their publications, the information should be provided
according to the format of Table 3. After filling publications'
information, SMENT generates corresponding social graph,
then the relationships within academic learning can be
calculated. Fig. 5 shows the screen of the initial SMNET.
208 Upload Content
SMNET Platform
KWs
Social Graph
Generation
Upload Content
Upload Content
Figure 4 Scenario of SMNET Platform
Note that the filters and dynamic actions (View, Download,...)
are invisible in SMNET social graph, so we use dotted line to
represent it and all dynamic actions should be recorded in out
access history. In Fig. 5, Mike, Ben and Jack (Knowledge
Worker) are the members in Intellectual Network group
(Knowledge Organization). Mike is the admin (Leader) and the
creator (Author) of Publication A in Intellectual Network. Jack
and Ben are cited names of Publication A (Co-authors). As an
author of Publication A, Mike set the permission of filter that
only the member in Intellectual Network group can view
Publication A. So John (Non-Intellectual network group
member) cannot view publication A but Ben and Jack
(Intellectual network group member) can view it. All dynamic
actions will be recorded in access history of each publication
for evaluating relationships. After obtaining social graph, the
weights can then be computed. Currently the weights are
computed off-line.
In the next version, the weights
computation
will
be
included
in
SMNET.
Figure 5 Social Graph in SMNET
209 VII. CONCLUSION & FUTURE WORK
REFERENCES
Academic learning network can be seen as a subset of social
network, but there are few specific approaches to clearly
define it and analysis the relationships in it with low-cost
algorithm.
In this paper, we propose a formal ownership model that
capable to describe academic learning network and can be
easily integrated with existed social network platform, the
intelligent
relationships
within
Knowledge
Worker/Organization can be represented by social graph.
[1] Pierfrancesco Bellini, Ivan Bruno, Paolo Nesi, Michela
Paolucci, “Institutional Services and Tools for Content,
Metadata and IPR Management”. DMS, 2013.
[2] Shi-Kuo Chang, "A General Framework for Slow
Intelligence Systems", International Journal of Software
Engineering and Knowledge Engineering, Volume 20,
Number 1, February 2010, 1-16.
[3] Shi-Kuo Chang, Yingze Wang and Yao Sun, “Visual
Specification of Component-based Slow Intelligence Systems”,
Proceedings of 2011 International Conference on Software
Engineering and Knowledge Engineering, Miami, USA, July
7-9, 2011, 1-8.
[4] Sentient Multimedia Network, http://smnet.disit.org
[5] European Library of Artistic Performance, ECLAP,
http://www.eclap.eu/
[6] Wiki book, “Social Network Analysis: Theory and
Applications”, 2011.
[7] Hyunjung Kim and George A. Barnett, “Social Network
Analysis Using Author Co-Citiation Data”, Proceedings of
Americas Conference on Information Systems (AMCIS), 2008.
[8] Tsai-Yuan Lin and Yun-Yao Cheng. “Exploring the
Knowledge Network of Strategic Alliance Research: CoCitation Analysis”, International Journal of Electronic
Business Management, Vol. 8, No. 2, pp. 152~160, 2010.
[9] Borgatti, S.P., Everett, M.G. and Freeman, L.C., “Ucinet
for Windows: Software for Social Network Analysis”,
Harvard, MA: Analytic Technologie, 2002.
With quantified metrics, the degree of relationship can be
easily evaluated and further help us to analyze academic
learning network. In addition to static factors in social network,
we further consider various dynamic actions as our evaluation
reference.
More sophisticate relationship and evaluation metrics will be
extended to measure more complex social network in the
future.
ACKNOWLEDGMENT
Thanks for Dr. Paoli Nesi in University of Florence and his
student Michela Paolucci for providing SMNET as our
experimental platform and helpful advice to complete this work.
210