Assumptions and Experimental Data

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

70

After validating the ability of the proposed mechanism to improve goal achievement, we can further test whether it can help SMEs to create their market niche. As mentioned earlier, there are two indicators regarding market niche assessment in this study (i.e., the level of uniqueness and level of attractiveness for a given alliance). We then have the following performance indicator as showed in Figure 6.1. If the coefficient is greater than 1, it means the proposed mechanism can effectively help SMEs to find appropriate partners for the sake of creating market niche. Moreover, the hypothesis for this issue is put forward.

WXYZ[\ ]^_`[ ab[cc^_^[d\ =

\`[ e[f[e bc gd^hg[d[ii + \`[ e[f[e bc X\\YX_\^f[d[ii (Xc\[Y _b_bk[YX\^dl m^\` b\`[Y nWoi )

\`[ e[f[e bc gd^hg[d[ii + \`[ e[f[e bc X\\YX_\^f[d[ii (p[cbY[ _b_bk[YX\^dl m^\` b\`[Y nWoi )

Figure 6.1. Market Niche coefficient formula

 Hypothesis 4: The proposed mechanism can effectively help SMEs to form a new alliance with market niche.

In the next section, it outlines the assumptions made for revealing the limitations of the mechanism as well as the experimental data set to test the hypotheses depicted in the previous section.

6.2 Assumptions and Experimental Data 6.2.1 Assumptions

Before the experiments are performed, several assumptions should be addressed.

 Assumption 1: When the system goes through partner discovering process, the search base for partners is limited in the same destination where the SME is located. In this research, we only try to tackle the problem of local tourism development. Nevertheless, it is still worth bearing in mind that the opportunities

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

71

for service innovation would become restricted if the search base is limited in a destination. Once this restriction is lifted, cooperation across destinations becomes possible. It is more likely to find the appropriate partners for goal achievement because more possibilities are opened up.

 Assumption 2: The image models of SMEs in the system are quite representative.

The proposed approach heavily relies on the image model data. If the quality of image models does not reach the acceptable level, then the outcomes of recommendations would not be reliable.

 Assumption 3: The user’s inputs (the goal statement) should at least include one noun. The goal statement should be in the form of metaphorical statement. Every metaphor needs to have a vehicle which is a noun.

 Assumption 4: The performance of language translation API is good enough. The adopted technique of metaphor comprehension was designed for English context.

In order to apply this technique to Chinese context, we need language translation API to translate Chinese statement into English. However, if the statement is too complex, then the quality of translation would not good enough.

 Assumption 5: The cultural difference is ignored. After the Chinese metaphorical statement has been translated into English. The context for metaphor comprehension would switch from eastern to western. That is not so promising in real application.

6.2.2 Experimental data

Given that it’s intrinsically impossible to acquire real image model of SMEs before the system is put into practice. Alternatively, we use the computer program to automatically generate the testing data. In total, 140 business image models and 140

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

72

customer image models are separately generated.

For the purpose of testing the hypotheses, two different levels (higher and lower) of image diversity context should be formulated. A destination with higher level image diversity indicates the SMEs within the destination give people diversified feelings. We can understand the level of image diversity by examining the number of image clusters and the distance between image clusters. The details are provided as below.

As mentioned in Chapter 4, every image model in the form of colors can be mixed into one RGB value. If an image model have similar configuration, it would be mixed into the similar RGB value. Through this conversion process, every image model can be converted to one RGB value and projected to a three-dimension (RGB) space. The cluster analysis then can be performed. In this regard, the RGB values in the same cluster means these image model have the similar image configuration.

In addition, the number of clusters implies the variety of image models. More image clusters reflects the wide difference in image models. On the other hand, if there is a tendency for the clusters to flock together, then it means the image models are relatively homogeneous(i.e., lower image diversity). If the clusters in the three-dimension space are located in a scattered fashion, it represents that the image models are relatively heterogeneous (i.e., higher image diversity). We believe it is strategically important to appreciate that healthy tourism ecosystem should exhibit higher image diversity, the ability to offer comprehensive experiences.

In order to automatically generate image models creating two different image diversity settings, the image model generation program should be able to manipulate the configuration of an image model. We know that the level of image diversity is decided by the number of clusters and the distance between each cluster. We also know that each cluster is formed from several nearby dots in the RGB space. If we

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

73

can ensure each generated image model can be converted to its designated RGB value, then the results of image cluster analysis can be deftly manipulated. In this regard, this program needs to have an input that is a RGB value. According to the given RGB value, the program will start to find the closest five colors of image elements. It will give these five image elements larger image quantity value and the rest of them get smaller value. The reason to do so is that the color mixing mechanism will only select the image elements that their intensity values are high enough to perform the color mixing. If the selected colors are those colors which are close to the given RGB value, then the color mixing result will be also close to it because the color mixing mechanism is basically to get the center of gravity of input colors.

After the program is developed, it can be used to generate image models, 140 for customers and 140 for businesses. The only problem is how to decide the input RGBs for generating image models. Regarding this issue, we adopt the random way to do so because the image model just is developed by our research team, and there are no practical image model data for us to reference. Therefore, if we want to generate the image models that fit image lower diversity setting, all we need to do is to input fewer number of colors to generate 140 image colors and vice versa. For example, if we input one color to generate 10 image models, then these image models would have similar image configurations; contrariwise, if we input ten different colors to separately generate 10 image models, then these image models would have different image configurations. According to this characteristic of image model generation program, the fewer colors are input and then the level of image diversity of image models will be smaller.

Figure 6.2 and 6.3 demonstrate the results of cluster analysis of 140 business image models for context setting of experiments. The blue circles represent the centers of each cluster. In a lower image diversity context, the number of clusters is

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

74

relative few (i.e., 5 image clusters) and the distance between the centers of each cluster is relatively short. Contrary to Figure 6.2, Figure 6.3 demonstrates the results for higher image diversity context (i.e, 11 image clusters) in similar manner. With different context settings, we are able to examine the impact of image diversity.

Figure 6.2 concentrated centers of image clusters (lower image diversity)

Figure 6.3 scattered centers of image clusters (higher image diversity)

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

75