Strategic Management of Intellectual Property: R&D Investment Appraisal Using Dynamic Bayesian Network

International Journal of Computer Applications Technology and Research
Volume 4– Issue 9, 640 - 647, 2015, ISSN: 2319–8656
www.ijcat.com 640
Strategic Management of Intellectual Property: R&D
Investment Appraisal Using Dynamic Bayesian
Network
L . O. Nwobodo
Enugu State University of Science and
Technology
Enugu State, Nigeria
H.C. Inyiama
Nnamdi Azikiwe University[NAU]
Awka, Anambra State, Nigeria
Abstract :
Company executives are under increasing pressure to proactively evaluate the benefits of the huge amounts of
investment into intellectual property (IP). The main goal of this paper is to propose a Dynamic Bayesian Network
as a tool for modeling the forecast of the distribution of Research and Development (R&D) investment efficiency
towards the strategic management of IP. Dynamic Bayesian Network provides a framework for handling the
uncertainties and impression in the qualitative and quantitative data that impact the effectiveness and efficiency
of investments on R&D. This paper specifies the process of creating the graphical representation using impactful
variables, specifying numerical link between the variables and drawing inference from the network.
KEY TERMS: IP Dynamic Bayesian Network, R&D Investment Efficiency.
1 INTRODUCTION
Intellectual property (IP), which is now being
regarded [1][2] as constituting the core assets of
companies, is centered around innovation. The very
basis for innovation is developing new products and
services, which logically means that more IP will be
generated and thus increases the need for protection
[3].
The research and development (R&D) Investment of
major manufactures has reached an annual level of
several billion dollars [4]. Questions are being raised
as to whether this investment amount is efficient from
the perspective of its effectiveness, such as whether
such investment is made efficiently and whether the
results of R&D are appropriately contributing to
company profits.
Essentially, activities for R&D and intellectual
properties must be dealt with in the management
cycle shown in figure 1. It is important to realize that
R&D strategy and the IP strategy exactly represent
the business strategy of a company. There are many
companies that spend amounts for R&D that exceed
their amount of capital investment. In response to
these moves, institutional investors have started to
show an interest in the content of R&D investment
and the effectiveness and efficiency of such
investment. However, due to uncertainties and
imprecise data, it is extremely difficult to proactively
demonstrate explicitly the efficiency and
effectiveness of R&D investment.
In an automated Bayesian Network decision support
system, the probability judgment analysis make over
time can be captured, compared to actual data when
it becomes available, and then provide feedback on a
timely basis [5]. Based on the work by AIpart and
Raiffa [6], this feedback should improve analysts’
probability assessment which should lead to
improved future performance.
[7] observed: “whether or not they recognize it,
virtually all decisions that investors make are
exercises in probability: For them to succeed, it is
critical that their probability statement combines the

www.ijcat.com 641
historical record with the most recent data variable.
And that is Bayesian analysis in action”.
Hence this paper proposes a Bayesian Network model
for the forest of the distribution of R&D investment
efficiency.
Figure 1: Ideal management cycle for activities for
R&D and intellectual properties [4]
2. BAYESIAN NETWORKS
Bayesian networks provide a formalism for reasoning
about partial beliefs under conditions of uncertainty.
These parameters are combined and manipulated
according to the rules of probability theory [8][9]. Let
us consider n discrete random variables x1, x2---- xn,
a discrete acyclic graph with n nodes, and suppose the
jth
node of the graph is associated to the xj variable.
Then the graph is a Bayesian network, representing
the variables x1,x2 ……….xn, if
P(x1, x2, ---------xn) = π p(xj / parents(xj)),
Where parents (xj) denotes the set of all variables Xi
such that there is an arc from node xi to xj in the
graph. The probability terms in the product are
described by conditional probability Tables (CPT)
which may be set by hand or learned from data.
Standard algorithms such as junction tree [9] [10]
exist to perform inference networks.
Dynamic Bayesian Networks (DBNs) allow the
modeling of entities in a changing environment where
the values of variables change over time [10] [11].
Functionally, DBNs capture the process of variable
values changing over time by representing multiple
copies of network modes with one copy for each time
step [10]. Visually, they may be displayed using two
copies of each recurrent node representing the current
t, and previous, t-1 states. A Bayesian network is a
tool to help expert represent uncertainty, ambiguous
or incomplete knowledge. Bayesian networks use
probability theory to represent uncertain knowledge.
A Bayesian network consists of two parts-a
qualitative graphical structure of the relationships in
the model and a quantitative structure represented by
the probability distributions that are indicated by the
graph. In a Bayesian network, historical information
can provide a framework or baseline model to
develop prior distribution. New quantitative
information, qualitative information, or evidence can
be added to the network as appropriate to develop
posterior probabilities.
Decision makers in many different contexts combine
qualitative data and qualitative information. Bayesian
networks have been applied in a wide variety of
decision-making contexts. Some examples are
venture capital financing [12], auditing [13], medical
diagnosis [14], and software design [15], among
many.
j

www.ijcat.com 642
3 DEVELOPING THE BAYESIAN
NETWORK
There are three important steps in building the
Bayesian network. The first step is developing the
graphical model. This step includes the relevant
variables and specifying whether they are
independent, or not. The second step is the
specification of the numerical relationship between
the variables that are not independent. The third step
is making inferences or decisions based on new
evidence.
3.1 Graphical Representation
As mentioned in the previous section, the first step in
construction the Bayesian network is the graphical
model. The graphical model is a directed, acyclic
graph were nodes represent variables and directed arc
(arrows) represent the conditional probability
relationship assumed in the model. The variables and
the arcs between the variables are the inputs to the
graph.
In this paper, as part of R&D strategies, the forecast
of R&D investment efficiency is to be made using the
variables: infrastructure, capability of science and
technology, Risk, R&D efficiency management
performance, total R&D personal capability, total
R&D expenditure. These are variables that are
assumed by management (for the sake of analysis)
that impacts Return on Investment (ROI) on IP R&D.
The relationship among the IP variables maybe
represented by the Bayesian network of figure 2. The
network consists of four discrete variables:
Capability of Science and Technology (CST), Risk
(RSK), R&D Efficiency (RDE), Management
Performance (MP), Total R&D Personnel Capability
(PC),Total R&D Expenditure (RDEP), and
Investment Efficiency (IE).
Figure 2: Bayesian Network for forecasting R&D
Investment efficiency.
In the network, a node with arcs leading out only
indicates a marginal probability distribution. For
example CST, RDEXP, RSK are marginal
probability distribution. A node with arcs leading into
it indicate a conditional relationship. For example, the
factors (variables) that determine the value for
infrastructure (IF) is the capability of science and
technology (CST) and total R&D Personnel
Capability (PC). The node infrastructure (IF) is a
conditional node.
The absence of a directed arc from a node is also
meaningfully because the absence indicates
independence assumption. The absence of a directed
arc denotes conditional independence between nodes.
Thus the lack of an arc from CST to RSK signifies
that it is assumed that RSK is independent of CST;
C
P
R
M
I
R
IR

www.ijcat.com 643
the lack of an arc from IF to MP signifies that it is
assumed that MP is independent of IF.
The analyst benefits in several ways from the
graphical construction of a Bayesian network. First,
the construction of the graphical portion of the
network helps the analyst clarify and refine his view
of the relationships among the variables [5] Next, the
analyst may not always have a good understanding of
how a decision is reached. They fully understand
which variables are used, but how the variables are
combined and heighted to come up with a decision is
not always well understood or systematic [5]. In [16],
almost all analyst agree that qualitative information
was important, but when questioned about how it was
incorporated in a decision, most analyst could not be
specific.
3.2 Determining Numerical links between
variables
Each node in the Bayesian network is a variable that
is described either as a constant value, a probability
distribution, or as a function of other variables [17].
In a Bayesian network, the primary focus is on
determining the probability distribution of the
relevant nodes. A Bayesian network model is
represented at two levels, qualitative and quantitative.
At the qualitative level a directed graph is used (as
done here in figure 2) in which nodes represent
variables and directed arcs describe the conditional
independence relations embedded in the model. At
the level, conditional probability distributions are
specified for each variable in the network. Each
variable has a set of possible values called its state
space that consists of mutually exclusive and
exhaustive values of the variable.
There are two primary ways to find probability
distribution, for the nodes in the network. One way is
historical data. The other is to use subjective
probability judgments. The two methods can also be
combined.
For the model in this paper capability science and
technology (CST) is specified as having two states:
“High” and “Low; infrastructure (IF) has two states:
“Good” and “Bad”, total R&D Personnel Capability
(PC) has two states: “High” and “Low”, total R&D
Expenditure (RDEXP) has two states: “High” and
“Low”; Management Performance (MP) has two
states: “Good” and “Bad”; R&D efficiency (RDE)
has two stets: “High” and “Low”; Risk (RSK) has two
states: “High” and Low” and investment efficiency
(IE) has two states: “High” and “Low”.
A fundamental assumption of a Bayesian network is
that when the conditionals for each variable is
multiplied, the joint probability distribution for all
variables in the network is obtained.
Suppose a sequence of the variables in a Bayesian
network is picked such that for all directed arcs in the
network, the variable at the tail of each arc precedes
the variable at the head of the arc in the sequence.
Since the directed graph is acyclic, there always exist
one such sequence. In figure 1, one such sequence is
CST IF RDE MP RSK IE.
From figure 2, this shows that the model makes the
assumption:
P(CST, IF, RDE, PC, RDEXP, MP, RSK, IE) =
P(CST) P(IF/CST,PC) P(RDE/IF, PC, RDEXP,
MP) P(PC/MP) P(RDEXP)

www.ijcat.com 644
P(MP/RDEXP) P(RSK)
P(IE/RDE,MP,RSK),
Where denotes point wise multiplication of
tables.
For each variable, a table of conditional
probability distribution has to be specified, one
for each configuration of states of its parents.
For the model, tables 1(a) – (h) gives the tables
of conditional distributions-P(CST),
P(IF/CST,PC) P(RDE/IF, PC,RDEXP,MP),
P(PC/MP), P(RDEXP), P(MP/RDEXP),
P(RSK), P(IE/RDE,MP,RSK).
(a)
P(CST) High Low
0.25 0.75
(b)
(c)
P(RDE/IF, PC,RDEXP,MP) High Low
High, High, High, Good 0.85 0.35
High, High, High, Bad 0.60 0.40
High, High, Low, Good 0.80 0.20
High, High, Low, Bad 0.55 0.45
High, Low, High, Good 0.67 0.33
High, Low, High, Bad 0.70 0.30
High, Low, Low, Good 0.68 0.32
High, Low, Low, Bad 0.66 0.34
Low, High, High, Good 0.50 0.50
Low, High, High, Bad 0.49 0.51
Low, High, Low, Good 0.40 0.60
Low, High, Low, Bad 0.75 0.25
Low, Low, High, Good 0.30 0.70
Low, Low, High, Bad 0.20 0.80
Low, Low, Low, Good 0.10 0.90
Low, Low, Low, Bad 0.25 0.75
(d)
P(PC/MP) High Low
Good 0.15 0.85
Bad 0.70 0.30
(e)
P(IF/CST,)PC Good Bad
High, High 0.80 0.2
High, Low 0.06 0.40
Low, High 0.25 0.75
Low, Low 0.50 0.50
High, High 0.01 0.90
P(RDEXP) High Low
0.65 0.35

www.ijcat.com 645
(f)
(g)
P(RSK) High Low
0.70 0.30
Table 1: Tables of conditional probabilities for the
Bayesian network of figure 2.
P(IE/RDE, MP, RSK) High Low
High, Good High, 0.90 0.10
High, Good, Low 0.85 0.15
High, Bad, High 0.02 0.98
High, Bad, Low 0.65 0.35
Low, Good, High 0.70 0.30
Low, Good Low 0.55 0.45
Low, Bad, High 0.5 0.5
Low, Bad, Low 0.1 0.9
3.3 Making inference
The ultimate goal is to model the probability
distribution of the investment efficiency (IE) for IP
portfolio.
Once a Bayesian network is constructed, it can be
used to make inferences about the variables in the
model. The conditionals given in Bayesian network
representation specify the prior joint distribution of
the variables. If the values of some are observed (or
learnt), then such observations can be represented by
tables where 1 is assigned for observed values and 0
for unobserved values. Then the product of all tables
(conditionals and observations) gives the posterior
joins distribution of the variables. Thus the joint
distribution of variables changes each time new
information is learnt about item.
Often the interest is on some target variables. In this
case, inference is made by computing the marginal of
the posterior joint distributions for the variables of
interest. Consider the situation described by the
Bayesian network in figure1. The interest is in the
true state of the R&D investment efficiency. Given
the prior model (as per the probability tables given):
table 1(a) – (b)), the marginal distribution is
computed (giving probability values for “High” and
“Low”). Now suppose it is learnt that (i.e new
observation) risk (RSK) is “Low” and management
performance (MP) is “Good”. The posterior marginal
distribution of R&D investment efficiency changes.
4 CONCLUSION
The main goal of this paper is to propose a Bayesian
network as a tool for modeling the forecast of the
distribution of R&D investment efficiency towards
the strategic management of IP.
The improvement of investment efficiency by
reviewing R&D and the reorganization of a business
portfolio have become urgent issue to manage. Many
companies are realizing that if they are going to spend
any money on IP, it better be IP that has value to the
business. So companies are now developing strategic
plans as to (a) where they want inventions (b) what
P(MP/RDEXP) Good Bad
High
Low
0.10 0.90
0.70 0.30

www.ijcat.com 646
resultant IP they want to create and (c) the efficiency
of their R&D investment. Company executives now
face a variety of opportunities that require
sophisticated analysis and decisions. These
requirements are better met by advanced decision
support tools.
With the capability to model the forecast of the
efficiency of R&D investment given different
scenarios of the combination of resources (R&D
personnel, management performance, R&D budget),
risk and market demand, companies are able to make
effective decisions regarding their IP portfolio. This
enables companies to evaluate the business benefit of
any IP before its creation. This means for example
they can strengthen the few patents they will file by
focusing on “inventing around their own IP” before
filing it. An advanced analytic tool, such as the
Bayesian network, helps to work the process of
inventing around” to systematic and robust.
There are many off-the-shelf software systems that
allow Bayesian networks to be constructed
graphically by end-users, for example Bayesialab
(www.bayesi.com), Netica (www.norsys.com) and
Hugin(www.hugin.com). These tools allow the user
to enter the graph and specify numerical relationships
among the variables. The software in use calculates
the inference based on these inputs. The inference
results are shown graphically as probability
distribution for the network. The analysis would help
executives make better strategic decision regarding
their IP portfolio.
5. REFERENCES
[1] Dino Isa, Pter Blanchfield, Zh.yuan
2009“Intellectual Property Management System
for the super-capacitor Pilot plant “workshop on
advances on advances intelligent computing.
[2] Benintend, S. 2003 “Intellectual property
valuation one important step in a successful
asset management system “Payton: University
of Dayton School of Law, (2003) PP 12, 14; 16-
20
[3] John Cronin 2010, “The cause for Developing IP
an “Executable” IP strategy in 2010” IP Capital
Group, Inc, PP 1.
[4] Masayuki Miyake, Yuji Mune, and Keiichi
Himeno, Dec 2004 “Strategic Intellectual
Property Management: Technology Appraisal
by using the “Technology Heetmop”, NRI papers
No. 83.
[5] Riza Demiser, Roreld R. Mom, Catherine Shenoy,
June, 2005, “Bayesian Networks. A Decision
Tool to improve portfolio Risk Analysis”, PP 9-
11 .
[6] Alpert, M., and H. Raiffa, 1982 “A Progress
Report on the Training of Probability Assession.
Judgment under uncertainty: Heuristics and
Biases. ed. D. Kahneman, P. Slovic, A. Tversky.
New York: Cambridge University Press.
[7] Hagstrom, R.G. 1999 “The Warren Buffett
Portfolio. Wiley and Sons, New York.
[8] Pearl .J., 1988 “probability reasoning in
Intelligent systems: Networks of plausible
Inference. Morgan Kaufmann Publishers, 2nd
edition on.
[9] Mark Taylor, Charles fox “Inventory
Management with Dynamic Bayesian
Network Software systems” Adaptive
Behavior Research Corp, Sheffield, UK.
[10] Arbib M.A, editor 2003 “The handbook of brain
theory and neural networks, chapter
Bayesian networks. Press.
[11] K.P. Murphy 2002 “Dynamic Bayesian
networks: Representation, Inference and
learning “Technical report, University of
California, Berkeley.
[12] Kemmeser, B.S. Mishra, and PP. Shenoy 2001
“Bayesian Casual Maps as Decision Aids in
venture capital decision making: Methods
and Applications” University of kensas,
working Paper.
[13] Gillete, P and R.P. Sriverstava 2000. “Attribute
Sampling: A Belief Function Approach to
Statically Audit Evidence”. Auditing: A
journal of practice and theory vol.19, no 1.
(Spring) 145 – 155.
[14] Bielza, C., S. Rios-Insua, and M. Goniez 1999
“Influence Diagrams for Neorata Jaundice

www.ijcat.com 647
management” In lecturer notes in Artificial
Intelligence 1620, (ed. Werror Horn, Yural
Shahar, G. Lindberge, Steen Andresen, J.
Wyatt) springing-Verley Berlin Heride/berg
1999, 138-142.
[15] Horvitz, E., J. Breese, D. Heckerman, D. Hovel,
and K. Rommelse 1998. “The Lumiere
Project: Bayesian user modeling for
Inferring the Goals and Needs of software
Users” proceeding of the fourteenth
conference on uncertainty in Artifical
intelligence, July.
[16] Catherine Slonoy, Pracash P. Sloney 1998.
“Bayesian network models of portfolio risc
and return”, School of business working
paper No.279.
[17] Slonoy 2001. “ A Description of Security
Analysis” Decision making process.”
University of Kansas, working paper.

Strategic Management of Intellectual Property: R&D Investment Appraisal Using Dynamic Bayesian Network

More Related Content

What's hot (18)

Viewers also liked (11)

Similar to Strategic Management of Intellectual Property: R&D Investment Appraisal Using Dynamic Bayesian Network (20)

More from Editor IJCATR (20)

Recently uploaded (20)

Strategic Management of Intellectual Property: R&D Investment Appraisal Using Dynamic Bayesian Network