CHAPTER 3. MOTIVATION APPLICATION
4.6 Niche Assessment Module
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
49
image model similarity analysis, the fulfilled images can then be identified. Then, goal fulfilment score can be computed. The formula is showed in Figure 4.11. The denominator is the number of goal images and the numerator is the number of goal images which are fulfilled. The higher score indicates the higher level of goal fulfilment and its value will between 0 and 1. For more details of goal fulfilment analysis, reference Figure 4.12. Finally, the output of candidate generation module will be the candidates list with the highest goal fulfilment score.
Goal ful illment score = the number of goal images that can be ful illed the number of goal images
Figure 4.11 The formula of goal fulfilment score
4.6 Niche Assessment Module
Niche assessment module is built for evaluating the market potential of each likely partner composition. While the candidate generation process in the previous module is completed, several partner choices would be generated. It is beneficial if more information can be provided to make proper decision for partner selection. Recall from the research claim in chapter one, the proposed mechanism also can help users to increase the possibilities of building attractive and unique image. We certainly need to have a way of analyzing this part. This is the purpose of niche assessment module.
‧
Niche assessment here involves attractiveness analysis and uniqueness analysis.
Attractiveness analysis is to measure the consumer desirability and uniqueness analysis is to examine the degree of differentiation (Yu¨ ksel & Akgu¨ l 2007). The
Goal Fulfillment Component
Step1 : For each matched real businesses, compute its level of goal fulfillment.
Set GoalImg = the goal image j
Set CountFulfilledGoalImages= count the number of fulfilled goal images FOR i TO the total number of matched real businesses
Alliance= Predicted image configuration after cooperate with business CountFulfilledGoalImages = 0
Set AllianceImg = the image element k of Alliance FOR j TO the total number of the goal images
Maxsimilarity = 0
FOR k TO the total number of elements in Alliance
TempSimilarity = Compute the semantic similarity index between and AllianceImg and GoalImg
IF MaxSimilarity < TempSimilarity THEN MaxSimilarity = TempSimilarity
END NEXT
IF Maxsimilarity > specific threshold THEN Tag GoalImg with “fulfilled”
CountFulfilledGoalImages = CountFulfilledGoalImages +1 END
NEXT
IF CountFulfilledGoalImages/ the total number of the goal images> a specific
threshold THEN
Keep this business in the candidate business type list ELSE
Remove this business from the candidate business type list END
Figure 4.12. Goal fullfillment algorithm
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
51
true meanings behind the attractive and unique analysis components are to evaluate how attractive or unique the predicted image are for the possible new alliances. For the sake of assessing attractiveness and uniqueness, we have to predict the image configuration of a new alliance when a new partnership is built. Similar to goal fulfilment analysis, image model mixing is also adopted to generate a new image model of a new alliance. The following parts address the respective mechanisms about attractiveness analysis and uniqueness analysis
The notion of attractiveness here refers to the extent of allurement and capacity that can satisfy the needs of customers (Yu¨ ksel & Akgu¨ l 2007). As mentioned in Chapter 3, uVoyage system provides a destination recommendation service that offers an opportunity for users to uncover the tourism destinations and business entities fulfilled their emotional needs. The basic logic of this recommendation system is to match the customer emotional preferences and needs to the images of a destination or business entities. Simply put, if the images possessed by a destination or business entity can match the customer preference and needs, the matching relationship is built and a recommendation is generated. Therefore, when we want to measure how attractive images a new alliance possesses, we can compute how many emotional preferences can be matched to the predicted alliance images. The more customers can be matched, the higher level of attractiveness is measured.
Customer emotional preferences are presented in the form of customer image model. We prescribe that if the configuration of customer image model is similar to the configuration of alliance image model, then these two image models are matched.
This comparison can be easily accomplished by leverage image similarity analysis introduced in the previous sections. The higher image similarity score indicates higher
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
52
level of good match. This score is definitely between 0 and 1, but to which level can be determined “matched” ? This issue will be examined in the section 6.3.1.3 in chapter 6.
The algorithm of attractiveness component is then presented in Figure 4.13. For computation simplicity, the collected customer preferences information has been classified into several customer preference classes. The reason for classifying customer image model into several classes is that it may take too much time to find the best match by image similarity analysis if considering all the image models in the system. The alternative way is only to perform image similarity analysis for the representative image model of each class. The image models in the same group mean they have similar image configurations. If an image class is matched, that indicates all the image models in this class are all matched. The computation complexity of fining best match will be significantly reduced. More information regarding preference classification is provided in the next section.
Attractiveness analysis technically follows two steps. The first step is to use image model mixing API to predict the image model of new alliance. The second step is to compute attractiveness score. According to the formula of attractiveness score in Figure 4.14, the denominator indicates the total number of customer image models and the numerator refers to the total number of matched image model. In simpler terms, the system calculates how many percentage of preferences can be matched based on the new alliance image model.
‧
Figure 4.13 The algorithm of attractiveness analysis component
On the other hand, uniqueness here signifies the extent of image differences between a new alliance and existing business entities (Cracolici & Nijkamp 2009). An alliance image model is determined as “unique” when its image configuration is substantially varied from other image models. Similarly, uniqueness analysis also leverages pre-processed image clusters to reduce computation complexity. The only difference is that attractiveness analysis uses customer preference clusters and uniqueness analysis employs business image clusters.
To understand the extent of image difference between two image models, all need to do is to examine the differences of intensity values in two image models. The reasons are as follows. All of image models, no matter which kind, are composed of the same 122 image elements (e.g., pleasant, charming and gorgeous…etc.) with their quantity
Attractiveness Analysis Component
Step1: Generate the image model of a new alliance
Step2: Compute the image model similarity score between the alliance image model and the image model of each customer preference classes
Set score = the image similarity score of the new alliance and customer preference class
Set sumOfMatchedImgModels = the number of matched image models Initially, sumOfMatchedImgModels = 0
If (the score> threshold ) then
Set numOfImgModel = Get the number of image models belonging to customer preference class
sumOfMatchedImgModels = sumOfMatchedImgModels +numOfImgModel End if
Compute the attractiveness score
= "#$ %& '$( )* +",#$- +.$ )-$/0 (.$.,0& 4*5+",#$-6 .5)-$/0)
"#$ ")"+/ %& '$( )* ,&0") $( +.$ )-$/ % "#$ 080"$
‧
values and intensity values. The quantity value of each element denotes the number of people who think the target (i.e., business or destination) demonstrates the image elements and the intensity value refers to the percentage of quantity value in the image model. If people think a business or destination is not “charming”, then the quantity and intensity value of image element “charming” in the image model will be zero. Given that the image elements in a model are completely the same and the only difference rests on their quantity and intensity values, the comparison of image models in uniqueness analysis merely involves examining the differences of the quantity or intensity values among different models. However, we would like to treat image models equally without discrimination on the popularity of the target (i.e., business or destination) so only intensity value is considered in this analysis. Above descriptions can be verified in the following example.
Table 4.4 An example of two image models
Image element No.
Adjective Quantity Intensity
Image quantity and intensity value. For image model 1, 100 people think the target (business,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
55
alliance or destination) is charming, 500 people think it’s fascinating and 400 people think it’s enjoyable. Everyone can have multiple feelings against the target. The intensity value is computed according to each quantity value and the sum of them. For instance, the intensity value of charming in image model 1 is 0.1 (100/1000). Because the image elements in both models are the same and the measure for examining the image configurations should be independent of popularity of the target, only intensity value is considered.
The algorithm of uniqueness analysis is provided in Figure 4.14. Uniqueness analysis is designed for measuring how different the new alliance image model is. The initial step for uniqueness analysis is the same as attractiveness analysis- generate the image model of the new alliance. Next, to address the extent of difference between existing business image models and the new alliance image model, the dissimilarity indexes against different business image classes are computed. The dissimilarity index indicates the degree of dissimilarity between two given image models. The formula of dissimilarity index is that sum of intensity difference between two image models divided by 2, which is the maximum value of intensity difference in extreme case that the image elements with non-zero intensity values are completely different in any given two image models. The index will be hence limited in 0 to 1. For example, table 4.4 shows the sum of intensity difference is 0.4.
After all the dissimilarity indexes for each business class are computed, the minimum value of the dissimilarity indexes is selected as the final uniqueness score because the minimum value means the most conservative estimate. To put it more clearly, the new alliance image model should be compared with the representative image models of each business cluster. Each comparison generates one dissimilarity value. Assume that there are three image business clusters (e.g., BC1, BC2, BC3 ) and
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
56
therefore three dissimilarity values are generated (e.g., 0.2 for BC1, 0.4 for BC2 and 0.8 for BC3). The dissimilarity value 0.2 will be chosen as the final uniqueness score.
Hence, uniqueness score can be used to demonstrate the level of uniqueness of a given image model.
Figure 4.14 The algorithm of uniqueness analysis component
However, even though those scores can indicate the level of attractiveness and uniqueness of a new alliance, the users may be not able to understand whether it is good enough, especially compared to other business in the same destination.
Therefore, the implementation of the system would involve offering some reference
Uniqueness Analysis Component
Step1: Generate the image model of a new alliance
Step2: Compute the dissimilarity index between the alliance image model and the image model of each business classes
Set BusinessImgCluster = the representative image model of business image cluster i
Set AIntensity = the intensity value of image element j in image model of the new alliance
For each business cluster i
Calculate the uniqueness index on the basis of business cluster i
Set Intensity= the intensity value of image element j in image model of business image cluster i
Dissimilarity index = ∑@AB CDEFBG HI =EJKB BLBEBC@M|<6%"$%0"8=>6%"$%0"8=?|
=NO
P
Next
Step 3: Choose the minimum value of uniqueness index as the final uniqueness score Uniqueness Score = Min{Uniqueness Index}
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
57
points which can help user understand its position among the businesses in the same destination. For instance, after an attractiveness score is computed, the system than starts to examine how many percentage of business whose attractiveness score is below the newly generated score and the new alliance would fall into any of five ranks (i.e., A,B,C,D,E) (see Table 4.5). The users will be able to see the level of attractiveness in terms of the rank and the meanings of the ranks will be also elaborated. Then, it would be much clear to users when they appreciate the meanings of these indicators.
Table 4.5 The reference points for attractiveness and uniqueness score Rank (the level of
attractiveness or uniqueness)
How many percentage of business whose score is
below your score
A (high) 80~100%
B (relatively high) 60~80%
C (medium) 40~60%
D (relatively low) 20~40%
E (low) 0~20%
In sum, when the image of a new alliance is evaluated as attractive and unique, this new alliance would be considered to have a market potential. By appraising the niche of each possible partner composition, this module is able to identify the novel partnerships with high desirability and differentiation.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
58