Professional Documents
Culture Documents
Knowledge, Ignorance, and Uncertainty: An Investigation From The Perspective of Some Differential Equations
Knowledge, Ignorance, and Uncertainty: An Investigation From The Perspective of Some Differential Equations
net/publication/356731255
CITATIONS READS
0 64
3 authors:
Juri Yanase
Complete Decisions, LLC
21 PUBLICATIONS 139 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Evangelos Triantaphyllou on 22 December 2021.
*: Corresponding co-author
Moreover, in order to proceed with the modelling approach proposed in this study the
following reasonable assumptions are made:
• Assumption I: People are in enthusiasm for acquiring knowledge to get to know the
world around them, and the process of acquiring knowledge is a continuous one.
• Assumption II: The more knowledge people possess, the less uncertainty their
recognition will have.
• Assumption III: The more the uncertainty of the people’s recognition is, the more
likely it is that the people will seek knowledge when making a better recognition.
The above assumptions are rational for ordinary people because they are consistent with
some famous quotes related to the concept of “knowledge”:
• "All men by nature desire knowledge" (Aristotle (384 BC - 322 BC), Metaphysics),
and "The desire of knowledge, like the thirst of riches, increases ever with the acquisition
of it" (Laurence Sterne (1713-1768)).
• "Knowledge is power" (Sir Francis Bacon (1561 - 1626)).
• "To be conscious that you are ignorant is a great step to knowledge" (Benjamin Disraeli
(1804 - 1881)), "Perplexity is the beginning of knowledge" (Kahlil Gibran (1883 - 1931)),
and "The beginning of knowledge is the discovery of something we do not understand"
(Frank Herbert (1920 - 1986)).
It is known that knowledge has the property of granularity (Wierman, 1999). However,
in order to conduct a study based on the proposed differential equations, we assume that the process
of acquiring knowledge by people is a continuous one. This consideration is reflected in
Assumption I.
2 Background Information
It is widely agreed that people employ knowledge to recognize objects. The term
‘knowledge’ can be defined in numerous ways (for a review, please refer to Alavi and Leidner
(2001)). In this paper, we adopt the view of Alavi and Leidner (2001). That is, “knowledge is the
result of cognitive processing triggered by the inflow of new stimuli.” We also take into account
the views of Carlsson et al., (1996) and Stanley (2002), in which knowledge is related to the
2
cognition capacity of human beings, with the potential for influencing people's future actions.
Obviously, sufficient knowledge always leads to good recognition. Otherwise, uncertainty arises.
Regarding this study, the most relevant literature on uncertainty and knowledge is
attributed to Shannon (1948) and Wierman (1999). Shannon’s information entropy has found
many applications in the field of expert systems. For instance, it has been used to examine the
objective weights in TOPSIS (Wang & Lee, 2009; Chen, 2021), and for measuring consensus in
group recommender systems (Yalcin et al., 2021). Shannon introduced his entropy formula for
measuring uncertainty based on three assumptions (Shannon, 1948). That is, the Shannon entropy
is not deduced from any mathematical model. This is also the case for Wierman’s work (Wierman,
1999). Wierman constructed his knowledge measure based on several axioms.
This paper investigates the issues of knowledge and ignorance from a novel perspective.
In particular, it establishes some differential equations, from which one can obtain Wierman’s
knowledge formula. Moreover, this paper shows that these results satisfy a crucial complementary
relationship.
where [𝑥𝑥𝑖𝑖 ] = {𝑥𝑥𝑗𝑗 |𝑥𝑥𝑗𝑗 ∈ 𝑋𝑋, 𝑥𝑥𝑗𝑗 ∼ 𝑥𝑥𝑖𝑖 }. Clearly, if we have full knowledge about the entities of the
set X, then we will differentiate any one from the others. In this study one may observe that when
knowledge increases, this does not necessarily indicate that the “correctness” of the assessments
will increase too.
In the above setting, each equivalence class will include only one entity as the entities are
assumed to be distinct of each other. On the contrary, if we have no knowledge at all, then we will
not be able to differentiate any entity from each other. In this case, all the entities will be included
in a single equivalence class. Therefore, the cardinality numbers of the equivalence classes reflect
a relationship of the knowledge for recognition and uncertainty. Mathematically speaking if we
define
𝑚𝑚
𝑚𝑚
𝑈𝑈 = ��|[𝑥𝑥𝑖𝑖 ]|,
𝑖𝑖=1
then we have 𝑈𝑈𝑚𝑚𝑚𝑚𝑚𝑚 = 1 and 𝑈𝑈𝑚𝑚𝑚𝑚𝑚𝑚 = 𝑚𝑚, and the larger the value of 𝑈𝑈 is, the less distinct the
entities will be to the person (analyst studying the entities of the set X). In this case the boundary
conditions of formula (3) will be as follows:
𝐾𝐾 = ln 𝑚𝑚 , if 𝑈𝑈 = 1,
�
𝐾𝐾 = 0, if 𝑈𝑈 = 𝑚𝑚.
Thus formula (3) can be transformed into the following:
−𝐾𝐾 + ln 𝑚𝑚 = ln 𝑈𝑈, or:
1
𝐾𝐾 = ln 𝑚𝑚 − ln 𝑈𝑈 = ln 𝑚𝑚 − 𝑚𝑚 ∑𝑚𝑚
𝑖𝑖=1 ln|[𝑥𝑥𝑖𝑖 ]|. (5)
The above analysis leads to the following theorem:
4
Theorem 1:
The knowledge level, denoted as K, which corresponds to a cognitive process on m entities
1
{𝑥𝑥1 , 𝑥𝑥2 , … , 𝑥𝑥𝑚𝑚 }, can be computed as follows: K = ln 𝑚𝑚 − 𝑚𝑚 ∑𝑚𝑚
𝑖𝑖=1 ln|[𝑥𝑥𝑖𝑖 ]|.
Similar to the previous considerations, the boundary conditions of equation (4) are as
follows:
𝐼𝐼 = 0, if 𝑈𝑈 = 1
� .
𝐼𝐼 = ln 𝑚𝑚 , if 𝑈𝑈 = 𝑚𝑚
Under these conditions, equation (4) can be rewritten as follows:
ln 𝑈𝑈 = 𝐼𝐼, or:
1
𝐼𝐼 = ln 𝑈𝑈 = 𝑚𝑚 ∑𝑚𝑚
𝑖𝑖=1 ln|[𝑥𝑥𝑖𝑖 ]|. (6)
By combining equations (5) and (6) we get:
𝐾𝐾 + 𝐼𝐼 = ln 𝑚𝑚. (7)
When the boundary conditions of equation (4) are considered, the following theorem can be
derived.
Theorem 2:
The sum of the knowledge level, denoted as K, and the ignorance level, denoted as I, which
correspond to a cognitive process on m entities {𝑥𝑥1 , 𝑥𝑥2 , … , 𝑥𝑥𝑚𝑚 }, can be computed as follows: K +
I = ln 𝑚𝑚.
Example 1:
Suppose that there is a set X comprised of five distinct objects. That is, 𝑋𝑋 =
{𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜1, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5}. Two persons, say Adam and Beth, were asked
to assess the weights of these objects. They reported their weighing results as shown in Table 1.
Table 1: Data for the illustrative example (assessed weights of the 5 objects)
𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 ∗ Adam Beth
𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜1 60 60.2
𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2 63 62.8
𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3 63 63.1
𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4 61 61.1
𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5 61 61.1
∗
Weighed in grams.
We use Theorem 1 to measure these two persons’ knowledge levels in their recognitions of
the objects’ weights.
The equivalence classes deduced from Adam’s weighing results are as follows:
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜1] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜1},
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3},
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3},
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5}, and
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5}.
The equivalence classes deduced from Beth’s weighing results are as follows:
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜1] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜1},
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜2},
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜3},
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5}, and
[𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5] = {𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜4, 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜5}.
Similarly with above, by using Theorem 1 Beth’s knowledge level can be calculated as follows:
1
𝐾𝐾 (Beth) = ln 5 − ln (1 × 1 × 1 × 2 × 2) ≐ 1.3322.
5
6
From the above one can see that Beth has a higher knowledge level than Adam. This result is
consistent with our intuitive perception of the data shown in Table 1.
The next question is to determine the relative difference of two knowledge levels, such as
the ones by Adam and Beth in the previous example. The previous values depend on the
cardinality m of the compared entities. Thus, the previous values need to be normalized so one
may compare them across various set sizes. Given that the maximum value of the knowledge
level is equal to ln 𝑚𝑚 (i.e., ln 5 in this example) while the minimum value is equal to ln 1 (i.e.,
0), one can express the relative knowledge difference of two decision makers X and Y, denoted as
DK(X, Y), as follows:
�𝐾𝐾(𝑋𝑋) −𝐾𝐾(𝑌𝑌) �
DK(X, Y) = ln 𝑚𝑚
.
Since m is always greater than 1, the previous ratio can always be defined.
In the previous illustrative example, the relative knowledge difference between Adam and
Beth is equal to 0.1723. Instead of weight, one can consider any other measurable aspect, such as
length, surface, volume, monetary value, temperature, and so on.
Remark: Similar to the practice in utility theory, as a knowledge measure, K’s maximal
value can be set arbitrarily (only if it is larger than its minimal value). However, as it is shown
in Section 5, if we set the K’s maximal value (i.e., when U=1) as ln 𝑚𝑚, the derived formula is
related to an existing fundamental development, i.e., Wierman’s knowledge granularity measure.
the cardinality of partition block 𝐵𝐵𝑗𝑗 , and thus 𝑏𝑏𝑗𝑗 = �𝐵𝐵𝑗𝑗 �. We have 𝑚𝑚 = |𝑋𝑋| = ∑𝑛𝑛𝑗𝑗=1�𝐵𝐵𝑗𝑗 � = ∑𝑛𝑛𝑗𝑗=1 𝑏𝑏𝑗𝑗 .
The knowledge granularity measure (in bits), denoted as G(B), is defined by Wierman (1999) as
follows:
𝑏𝑏 𝑏𝑏
𝐺𝐺(𝐁𝐁) = − ∑𝑛𝑛𝑗𝑗=1 𝑚𝑚𝑗𝑗 log 2 𝑚𝑚𝑗𝑗.
Let [𝑥𝑥𝑖𝑖 ] denote the partition block (i.e., equivalence class) which corresponds to 𝑥𝑥𝑖𝑖 .
Then, the above formula can be rewritten as follows:
𝑏𝑏 𝑏𝑏 1 𝑏𝑏 𝑏𝑏 1 |[𝑥𝑥𝑖𝑖 ]|
𝐺𝐺(𝐁𝐁) = − ∑𝑛𝑛𝑗𝑗=1 𝑚𝑚𝑗𝑗 log 2 𝑚𝑚𝑗𝑗 = − 𝑚𝑚 ∑𝑛𝑛𝑗𝑗=1 ∑𝑘𝑘=1
𝑗𝑗
log 2 𝑚𝑚𝑗𝑗 = − 𝑚𝑚 ∑𝑚𝑚
𝑖𝑖=1 log 2 𝑚𝑚
, or:
1
𝐺𝐺(𝐁𝐁) = log 2 𝑚𝑚 − 𝑚𝑚 ∑𝑚𝑚
𝑖𝑖=1 log 2 |[𝑥𝑥𝑖𝑖 ]|. (8)
One can see that equation (8) is quite similar to the formula of Theorem 1, which defines
the level of knowledge. This is described explicitly next as Theorem 3.
Theorem 3:
The knowledge level, denoted as K, of a partition on a finite set is equivalent to Wierman’s
knowledge granularity measure, denoted as G(B), up to a constant (exactly, ln 2). That is, the
following relation is true: 𝐾𝐾 = ln 2 × 𝐺𝐺(𝐁𝐁).
7
ln 𝑑𝑑
Proof: Consider Theorem 1 and equation (8). Next, by using the transformation log 𝑐𝑐 𝑑𝑑 = ln 𝑐𝑐
for changing the base of logarithmic expressions, the following result is derived:
1 1
𝐾𝐾 = ln 𝑚𝑚 − 𝑚𝑚 ∑𝑚𝑚 𝑚𝑚
𝑖𝑖=1 ln|[𝑥𝑥𝑖𝑖 ]| = ln 2 × �log 2 𝑚𝑚 − 𝑚𝑚 ∑𝑖𝑖=1 log 2 |[𝑥𝑥𝑖𝑖 ]|� = ln 2 × 𝐺𝐺(𝐁𝐁).
8
Definition 1 (Hou, 2015b):
A preference map (for short, PM) 𝑃𝑃𝑃𝑃 = [𝑃𝑃𝑃𝑃𝑖𝑖 ]𝑚𝑚×1 , which corresponds to a weak ordering over
the alternative set {𝐴𝐴1 , 𝐴𝐴2 , ⋯ , 𝐴𝐴𝑚𝑚 }, is defined by
𝑃𝑃𝑃𝑃𝑖𝑖 = {|𝑆𝑆𝑖𝑖 | + 1, |𝑆𝑆𝑖𝑖 | + 2, ⋯ , |𝑆𝑆𝑖𝑖 | + |𝑇𝑇𝑖𝑖 |}, (9)
where 𝑆𝑆𝑖𝑖 is the alternative 𝐴𝐴𝑖𝑖 ’s predominance set denoted by 𝑆𝑆𝑖𝑖 = {𝑘𝑘|𝑘𝑘 ∈ M, 𝐴𝐴𝑘𝑘 ≻ 𝐴𝐴𝑖𝑖 }, and 𝑇𝑇𝑖𝑖
is the alternative 𝐴𝐴𝑖𝑖 ’s indifference set denoted by 𝑇𝑇𝑖𝑖 = {𝑘𝑘|𝑘𝑘 ∈ M, 𝐴𝐴𝑘𝑘 ~𝐴𝐴𝑖𝑖 }.
Example 2:
To illustrate the above concept, let us consider two ordinal rankings 𝐴𝐴1 ~𝐴𝐴2 ≻ 𝐴𝐴5 ≻
𝐴𝐴3 ~𝐴𝐴4 and 𝐴𝐴1 ≻ 𝐴𝐴2 ≻ 𝐴𝐴3 ~𝐴𝐴4 ~𝐴𝐴5 . According to Definition 1, their corresponding PMs are:
{1,2} {1}
⎡ ⎤ ⎡ ⎤
{1,2} {2}
(1) ⎢ ⎥ (2) ⎢ ⎥
𝑃𝑃𝑃𝑃(1) = �𝑃𝑃𝑃𝑃𝑖𝑖 � = ⎢{4,5}⎥ and 𝑃𝑃𝑃𝑃(2) = �𝑃𝑃𝑃𝑃𝑖𝑖 � = ⎢{3,4,5}⎥,
5×1 5×1
⎢{4,5}⎥ ⎢{3,4,5}⎥
⎣ {3} ⎦ ⎣{3,4,5}⎦
respectively. The entries of a PM are sets whose elements indicate the alternatives’ possible ranking
positions. For instance, the first entry of 𝑃𝑃𝑃𝑃(1) (i.e., {1,2}) indicates the possible ranking
positions of alternative 𝐴𝐴1 , the second entry indicates the possible ranking positions of alternative
𝐴𝐴2 , and so on.
Because the entries of a PM are sets comprised of natural numbers, set operations such as
∩ and ∪ can be defined or operated in an entry-wise way for two PMs. Recall that the entry of a
PM is a set which indicates an alternative’s possible ranking positions. Therefore, if the two PMs
of two experts have some nonempty entry-wise intersections then the two experts agree on some
ranking positions for some of the alternatives. This leads to the definition of consensus as
described next.
(𝑗𝑗)
maps. These preference maps are said to be in consensus if and only if ∀𝑖𝑖�⋂𝑛𝑛𝑗𝑗=1 𝑃𝑃𝑃𝑃𝑖𝑖 ≠ ∅� or,
(𝑗𝑗) (𝑘𝑘)
equivalently, ∀𝑖𝑖, 𝑗𝑗, 𝑘𝑘�𝑃𝑃𝑃𝑃𝑖𝑖 ∩ 𝑃𝑃𝑃𝑃𝑖𝑖 ≠ ∅�. Moreover, if these preference maps are in consensus,
(𝑐𝑐) (𝑗𝑗)
then their consensus PM is defined as 𝑃𝑃𝑃𝑃(𝑐𝑐) = �𝑃𝑃𝑃𝑃𝑖𝑖 � = ⋂𝑛𝑛𝑗𝑗=1 𝑃𝑃𝑃𝑃(𝑗𝑗) = �⋂𝑛𝑛𝑗𝑗=1 𝑃𝑃𝑃𝑃𝑖𝑖 � .
𝑚𝑚×1 𝑚𝑚×1
For the purpose of an illustration, we consider the two PMs in Example 2. It is easy to
check by Definition 2 that they are in consensus. Moreover, their consensus ranking is as follows:
9
{1,2} {1} {1}
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
{1,2} {2} {2}
(𝑐𝑐) ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
𝑃𝑃𝑃𝑃(𝑐𝑐) = �𝑃𝑃𝑃𝑃𝑖𝑖 � = ⎢{4,5}⎥ ∩ ⎢{3,4,5}⎥ = ⎢{4,5}⎥,
5×1
⎢{4,5}⎥ ⎢{3,4,5}⎥ ⎢{4,5}⎥
⎣ {3} ⎦ ⎣{3,4,5}⎦ ⎣ {3} ⎦
which indicates that the consensus ranking is: 𝐴𝐴1 ≻ 𝐴𝐴2 ≻ 𝐴𝐴5 ≻ 𝐴𝐴3 ~𝐴𝐴4 .
We remark that, the consensus defined by Definition 2 can be applied or be generalized
to many relevant cases, e.g., the case of full consensus (i.e., all the people have consensus in terms
of all the alternatives), or the case of a partial consensus (i.e., only some people have consensus in
terms of some of the alternatives). The interested reader may refer to Hou et al. (2020) for more
details. Furthermore, the PMs and the consensus defined by Definition 2 can also be used to discuss
some other topics in group decision making, for example, group decision making under a fuzzy
environment (Hou, 2016) and a post-consensus analysis in group decision making (Triantaphyllou
et al., 2019; Triantaphyllou et al., 2020).
Example 3:
Suppose that there is an alternative set {𝐴𝐴1 , 𝐴𝐴2 , 𝐴𝐴3 , 𝐴𝐴4 , 𝐴𝐴5 } , and a group of experts,
{𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸1, 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸2, 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸3} . Assume that the experts provided their weak orderings on the
alternative set as follows:
We use Theorem 1 to measure the experts’ knowledge levels. We first write out their preference
maps:
𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸1 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸2 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸3
𝐴𝐴1 {1} {1} {1,2}
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
𝐴𝐴2 {2,3} {2,3,4} {1,2}
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
𝐴𝐴3 ⎢{2,3}⎥ , ⎢{2,3,4}⎥ , ⎢ {3} ⎥.
𝐴𝐴4 ⎢{4,5}⎥ ⎢{2,3,4}⎥ ⎢ {4} ⎥
𝐴𝐴5 ⎣{4,5}⎦ ⎣ {5} ⎦ ⎣ {5} ⎦
By using equation (5), the experts’ knowledge levels are computed as follows:
1
• Expert1: 𝐾𝐾 (𝐸𝐸1) = ln 5 − 5 ln (1 × 2 × 2 × 2 × 2) ≐ 1.0549.
1
• Expert2: 𝐾𝐾 (𝐸𝐸2) = ln 5 − 5 ln (1 × 3 × 3 × 3 × 1) ≐ 0.9503.
1
• Expert3: 𝐾𝐾 (𝐸𝐸3) = ln 5 − 5 ln (2 × 2 × 1 × 1 × 1) ≐ 1.3322.
10
In addition, when some experts are in consensus, the cardinality number of their consensus
PM will not be more than any cardinality numbers of the PMs of the individual experts. Let 𝐾𝐾 (𝑖𝑖) ,
for 𝑖𝑖 = 1,2, ⋯ , 𝑛𝑛, represent the knowledge levels of the PMs (experts) that are in consensus, and
𝐾𝐾 (𝑐𝑐) represent the knowledge level of their consensus PM. Then we have the following corollary:
Corollary 1:
When the experts’ PMs are in consensus, then the knowledge level of their consensus ranking is
not less than any individual knowledge levels of the experts’ rankings. That is, the following
relation is true: 𝐾𝐾 (𝑐𝑐) ≥ max 𝐾𝐾 (𝑖𝑖) .
𝑖𝑖
For the case in Example 3, the three rankings are in consensus and the consensus ranking is
𝐴𝐴1 ≻ 𝐴𝐴2 ≻ 𝐴𝐴3 ≻ 𝐴𝐴4 ≻ 𝐴𝐴5 , which takes a maximal knowledge value (exactly, ln 5 ) in any
decision problem with five alternatives.
Following Definition 1, we obtain the preference maps which correspond to the experts’
preferences. They are the columns in Table 2. In addition, the consensus preference map (according
to Definition 2) is listed as the last column.
Using Theorem 1, we obtain the experts’ knowledge levels as follows:
1
𝐾𝐾 (𝐸𝐸1 ) = ln 5 − ln (1 × 1 × 1 × 2 × 2) ≐ 1.3322,
5
1
𝐾𝐾 (𝐸𝐸2 ) = ln 5 − ln (2 × 3 × 3 × 3 × 2) ≐ 0.6730,
5
1
𝐾𝐾 (𝐸𝐸3 ) = 𝐾𝐾 (𝐸𝐸4 ) = ln 5 − ln (1 × 1 × 1 × 1 × 1) ≐ 1.6094.
5
11
Table 2: The preference Maps (PMs) that correspond to the experts’ rankings
Alternatives E1 E2 E3 E4 PM
From Definition 2, we know that the four experts are in consensus, and their consensus
preference map is the one listed in the last column of Table 2. Moreover, the knowledge level of
the consensus preference map is as follows:
1
𝐾𝐾 (𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐) = ln 5 − ln (1 × 1 × 1 × 1 × 1) ≐ 1.6094.
5
From the above computational results, one can observe that the higher the knowledge level
is, the more distinctive is the experts’ ranking order. The above results also show that the
knowledge level of the consensus preference map is not less than the knowledge level of any of
the four experts. This observation is a plausible result, and it is derived from Corollary 1. This
observation means that, reaching consensus (as defined in Definition 2) is a process that reduces
the ignorance, while the knowledge increases. Theorem 2 coincides with this observation.
One may use such information to characterize the performance levels of the individual
experts in group decision making. In this type of analysis decision making experts may be clustered
together based on their knowledge levels. This may provide potentially an actionable insight of the
group decision making process, especially if this process is repeated over time. A relevant
approach is the one discussed in (Triantaphyllou et al., 2020).
8 Conclusions
This paper proposes a quantification approach for two important concepts in information
science and systems research. These concepts are the knowledge and ignorance levels as they relate
to a number of cognitive functions.
The quantification approaches are based on some simple and highly intuitive differential
equations. This is the main theoretical contribution of this paper as it is a novel and important
aspect. The theoretical developments in this paper demonstrate that the derived formula for the
proposed knowledge level is related to an existing fundamental development, namely, Wierman’s
measure of uncertainty. The proposed approaches show that, in classification and ranking cases,
Wierman’s knowledge formula can be derived from a certain differential equation under certain
boundary conditions. This strengthens the validity of the proposed quantifications for the concepts
of knowledge and ignorance. A number of theorems and corollaries are derived to comprehensibly
establish the various theoretical results.
12
The results of this paper can be used in areas where it is important to apply the notion of
knowledge and ignorance in the context of cognitive tasks. An application area can be in group
decision making systems where one may be interested to record and monitor over time the
performance of the various experts in terms of their knowledge and ignorance aspects. The graph
theoretic and data mining approaches described in a recent paper by the authors (Triantaphyllou et
al., 2020) can be enhanced by using the new concepts about knowledge and ignorance.
Another potential application is in the area of data mining and classification systems. In
general, any area that deals with cognitive tasks that rely on the concepts of knowledge and
ignorance can benefit by using the theoretical results derived in this paper. Thus, the results of this
study can become an integral foundation for more important developments in the future in many
areas related to information science and expert / intelligent systems.
Acknowledgments: The first author was supported by the National Natural Science
Foundation of China (No. 71571019). The authors of this paper are also very appreciative for the
constructive comments provided by the three reviewers which helped them to improve this paper
significantly.
References
Alavi, M., & Leidner D. E. (2001). Knowledge management and knowledge management systems:
Conceptual foundations and research issues. MIS Quarterly, 25, 107-136.
https://doi.org/10.2307/3250961.
Anjaria, K. (2020). Negation and entropy: effectual knowledge management equipment for
learning organizations. Expert Systems with Applications, 157, Article 113497.
https://doi.org/10.1016/j.eswa.2020.113497.
Arrow, K. J. (1950). A difficulty in the concept of social welfare. Journal of Political Economy,
58(4), 328-346. https://doi.org/10.1086/256963.
Boroushaki, S., & Malczewski, J. (2010). Measuring consensus for collaborative decision-making:
a GIS-based approach. Computers, Environment and Urban Systems, 34(4): 322–332.
https://doi.org/10.1016/j.compenvurbsys.2010.02.006.
Carlsson, S. A., El Sawy, O., Eriksson, I. V., & Raven, A. (1996, July). Gaining Competitive
Advantage Through Shared Knowledge Creation: In Search of a New Design Theory for
Strategic Information Systems. In ECIS (pp. 1067-1076).
Chen, P. (2021). Effects of the entropy weight on TOPSIS. Expert Systems with Applications, 168,
Article 114186. https://doi.org/10.1016/j.eswa.2020.114186
Hou, F. (2015a). A consensus gap indicator and its application to group decision making. Group
Decision and Negotiation, 24(3), 415-428. https://doi.org/10.1007/s10726-014-9396-4.
Hou, F. (2015b). The prametric-based GDM selection procedure under linguistic assessments. In
13
2015 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) (pp. 1-8). IEEE.
https://doi.org/10.1109/FUZZ-IEEE.2015.7337821.
Hou, F., & Triantaphyllou, E. (2019). An iterative approach for achieving consensus when ranking
a finite set of alternatives by a group of experts. European Journal of Operational Research,
275(2), 570-579. https://doi.org/10.1016/j.ejor.2018.11.047.
Hou, F., Zhai, Y., & You, X. (2020). An equilibrium in group decision and its association with the
Nash equilibrium in game theory. Computers & Industrial Engineering, 139, Article
106138. https://doi.org/10.1016/j.cie.2019.106138.
Raymond, W. (1965). Introduction to the Foundations of Mathematics (2nd ed.). John Wiley &
Sons, Chapter 2-8: Axioms defining equivalence, pp. 48-50.
Stanley, C. (2002). Knowing and acknowledging, must we mean what we say. Cambridge
University Press, 238-266.
Triantaphyllou, E., Hou, F., & Yanase, J. (2020). Analysis of the final ranking decisions made by
experts after a consensus has been reached in group decision making. Group Decision and
Negotiation, 29(2): 271-291. https://doi.org/10.1007/s10726-020-09655-5.
Triantaphyllou, E., Yanase, J., & Hou, F. (2020). Post-consensus analysis of group decision making
processes by means of a graph theoretic and an association rules mining approach. Omega,
Article 102208. https://doi.org/10.1016/j.omega.2020.102208.
Wang, T. C., & Lee, H. D. (2009). Developing a fuzzy TOPSIS approach based on subjective
weights and objective weights. Expert Systems with Applications, 36(5), 8980-8985.
https://doi.org/10.1016/j.eswa.2008.11.035.
Yalcin, E., Ismailoglu, F., & Bilge, A. (2021). An entropy empowered hybridized aggregation
technique for group recommender systems. Expert Systems with Applications, Article
114111. https://doi.org/10.1016/j.eswa.2020.114111.
14