On Ontology-Based Tourist Knowledge Representation and Recommendation

: In the rapid development of the information technology age, many travelers search for travel articles through the Internet. These travel articles include the experience and knowledge of traveler, which can be used as a reference for tourism planning and attraction selection. At present, the most travel experience and knowledge is available in online travel reviews (OTR). OTR and eWOM (electronic word-of-mouth) contain a lot of knowledge of consumers and travelers. Many travelers often look for OTR content through virtual communities, blogs


Introduction
Many travelers enjoy sharing their experiences or knowledge with others, and this can then be used as a reference by others when planning their trips, thus enabling them to learn more about the attractions they are going to visit [1]. Tourist knowledge is thus defined in this work as the opinions, experiences, and appraisals of travelers regarding target tourist attractions The most commonly used form of traveler knowledge is now online travel reviews (OTR), which are shared via virtual communities or blogs [2][3][4]. The value of tourist knowledge in OTR can be raised if it is first analyzed, was also used to design a method for recommending tourist attractions. The method put tourist knowledge into effective sharing and use. Ontology has been a well-developed technique. Constructive ontology, deductive ontology, and natural language analysis were applied in the development of the recommendation mechanism. Nonetheless, current tourist attraction recommendation systems on the market have not used constructive ontology, deductive ontology, and natural language analysis. The study utilized well-developed technology to share and manage tourist knowledge, and this is what set this study apart from others. OTR is a kind of tacit knowledge. There are few tourist attraction recommendation systems that adopted tacit knowledge. The reason is that the collection, analysis, management, and application of tacit knowledge is more complicated. Tacit knowledge is adopted in this study. This has shown the value of this study in contributing to the application of tacit knowledge in this field. Therefore, this study developed an ontology-based tourist knowledge representation and recommendation mechanism to carry out appraisal analysis of OTR content. Through the collection of positive appraisals related to tourist attractions, tourist knowledge structures were constructed based on ontology. These tourist knowledge structures were then used to develop a tourist knowledge representation and recommendation mechanism. This study thus has the following aims: (i) designing an OTR and eWOM management framework; (ii) constructing an ontology-based tourist knowledge representation and recommendation procedure; (iii) providing an ontology-based tourist knowledge representation and recommendation method; and (iv) developing an ontology-based tourist knowledge representation and recommendation mechanism.
This paper is organized as follows. In Section 2, the differences between OTR and ontology characteristics are analyzed, and the results are then used to develop an ontology-based tourist knowledge representation and recommendation model. In Section 3, the methodology used for designing and developing the ontology-based tourist knowledge representation and recommendation procedure is explained. In Section 4 the model is developed and tested to examine its practicality. Finally, the conclusions of this work are presented in Section 5, along with recommendations for future research.

OTR and Ontology Characteristics Analysis
In this section, the differences between OTR content and ontology characteristics are analyzed, and this is then used to develop an ontology-based tourist knowledge representation and commendation model.

Design of an OTR and eWOM Management Framework
eWOM is the user's feeling or experience after they use a product or service. Now, there are many types of eWOM. For example, many travelers often share online travel reviews (OTR) through social communities or blogs. OTR allow a traveler to express their own ideas and opinions about tourism, and eWOM shows the opinions and ideas of a production or service. Thus, ORT is a kind of eWOM. In order to effectively manage and analyze OTR content, the study will revise and propose a new application function of OTR and eWOM, which is based on   [6] OTR (online travel reviews) and (eWOM) (electronic word-of-mouth) management framework, as shown in Figure 1.
(1). eWOM collection: Since there is an overwhelming amount of eWOM available online, it is both difficult and inefficient to try and obtain all of it about a specific product or service. As a result, the eWOM collection process seeks to collect and filter eWOM-related content from blogs and web forums. (2). eWOM analysis: In order to understand the product or service appraisals contained in eWOM content, this study uses natural language processing and sentiment analysis to analyze it. The eWOM analysis can identify whether the product or service appraisals are positive or negative, and thus help eWOM knowledge representation and reasoning. The research method is eWOM collection and eWOM analysis, based on the  eWOM analysis method [24].
(3). Ontology-based SWOT analysis: The results of the eWOM analysis can be used to understand positive and negative appraisals of an enterprise as they will contain information about a firm's strengths, weaknesses, opportunities, and threats, thus helping strategic planning. Ontology-based SWOT analysis mainly uses computerized eWOM analysis techniques, the result of the related study will be reference to   [6]. (4). Consumer demands analysis: The content of eWOM and OTR contains a lot of opinions and appraisals from consumers and tourists. These opinions and appraisals can reflect the requirement of consumers and tourists. Therefore, we will develop eWOM and OTR appraisal as an analysis method of consumer's demand. The result of the related study will be reference to Lin et al. (2017) [25]. (5). Knowledge representation and recommendation: OTR is a kind of eWOM, and from the eWOM analysis results, we can find out the customer's tourism evaluation through eWOM analysis. These appraisal are fragmented. Thus, knowledge representation and recommendation is mainly to construct consumer appraisal as a knowledge structure, and then make inference through the knowledge structure. OTR is a kind of eWOM. Thus, OTR can transform the consumer's travel experience into a tourism knowledge structure which help other travelers find more tourist attractions. (6). Appraisal-based BCG matrix analysis: The result of eWOM analysis can realize the positive and negative appraisal of an enterprise. This appraisal would include the strength, weaknesses, opportunities, and threats about an enterprise, the production, and the service. The function is mainly to identify the performance of the business and product line through the analysis technology of the automated eWOM. The eWOM analysis result will help the enterprise adjust their business or product and make the strategic planning efficient.
Appl. Sci. 2019, 9, x FOR PEER REVIEW 4 of 27 strengths, weaknesses, opportunities, and threats, thus helping strategic planning. Ontologybased SWOT analysis mainly uses computerized eWOM analysis techniques, the result of the related study will be reference to   [6]. (4). Consumer demands analysis: The content of eWOM and OTR contains a lot of opinions and appraisals from consumers and tourists. These opinions and appraisals can reflect the requirement of consumers and tourists. Therefore, we will develop eWOM and OTR appraisal as an analysis method of consumer's demand. The result of the related study will be reference to Lin et al. (2017) [25]. (5). Knowledge representation and recommendation: OTR is a kind of eWOM, and from the eWOM analysis results, we can find out the customer's tourism evaluation through eWOM analysis. These appraisal are fragmented. Thus, knowledge representation and recommendation is mainly to construct consumer appraisal as a knowledge structure, and then make inference through the knowledge structure. OTR is a kind of eWOM. Thus, OTR can transform the consumer's travel experience into a tourism knowledge structure which help other travelers find more tourist attractions. (6). Appraisal-based BCG matrix analysis: The result of eWOM analysis can realize the positive and negative appraisal of an enterprise. This appraisal would include the strength, weaknesses, opportunities, and threats about an enterprise, the production, and the service. The function is mainly to identify the performance of the business and product line through the analysis technology of the automated eWOM. The eWOM analysis result will help the enterprise adjust their business or product and make the strategic planning efficient. As can be seen in Figure 1, the eWOM and OTR management framework is displayed. This study focuses on knowledge representation and recommendation, and the following section explains the procedures of knowledge representation and recommendation method development. The procedures, methods, and results are as follows:

Analysis of OTR Content Characteristics
In order to effective analyze OTR, this study first examined the characteristics of OTR content. The results were then used as the basis for developing the system. Related studies were reviewed to help organize the procedure for analyzing OTR content characteristics [5,6,24], as shown in Figure 2.

•
Appraisal function: there are many positive and negative appraisals in OTR for a travel attraction. Thus, OTR appraisal can be used as a reference for filtering popular attractions. • Popular attraction filtering: many studies show that online appraisals determine the user's willingness to purchase [6,24], which also means that the production with higher positive appraisals is the consumers' favorite. That is, the attraction gaining the higher positive appraisals will be tourists' preferences. Therefore, we can judge popular attractions through the total quantity of positive and negative appraisals. • Popular attraction changing: consumer demand for products would change over time. [25] In the same way, travelers' evaluation of attractions will also alter with the time. When the requirement of consumers or travelers changes, the quantity of positive and negative appraisal also becomes different. Therefore, it is necessary to continuously collect the positive and negative appraisal of the attractions, which help travelers understand the latest tourist attractions.  As can be seen in Figure 1, the eWOM and OTR management framework is displayed. This study focuses on knowledge representation and recommendation, and the following section explains the procedures of knowledge representation and recommendation method development. The procedures, methods, and results are as follows:

Analysis of OTR Content Characteristics
In order to effective analyze OTR, this study first examined the characteristics of OTR content. The results were then used as the basis for developing the system. Related studies were reviewed to help organize the procedure for analyzing OTR content characteristics [5,6,24], as shown in Figure 2.

•
Appraisal function: there are many positive and negative appraisals in OTR for a travel attraction. Thus, OTR appraisal can be used as a reference for filtering popular attractions. • Popular attraction filtering: many studies show that online appraisals determine the user's willingness to purchase [6,24], which also means that the production with higher positive appraisals is the consumers' favorite. That is, the attraction gaining the higher positive appraisals will be tourists' preferences. Therefore, we can judge popular attractions through the total quantity of positive and negative appraisals. • Popular attraction changing: consumer demand for products would change over time. [25] In the same way, travelers' evaluation of attractions will also alter with the time. When the requirement of consumers or travelers changes, the quantity of positive and negative appraisal also becomes different. Therefore, it is necessary to continuously collect the positive and negative appraisal of the attractions, which help travelers understand the latest tourist attractions. This study applied set theory to explain the relationships between different tourist attraction sets. Each piece of OTR content contains a tourist attraction set. There are different types of sets among different tourist attraction sets, and they are classified into two types: disjoint sets and intersections, as explained below:

Popular
• Disjoint sets: If a specific tourist attraction cannot be found in two tourist attraction sets, it is likely that the two OTR do not overlap, and thus there is no direct relationship between the two tourist attraction sets.

•
Intersection: If two tourist attraction sets share the same tourist attraction information, it means that the two OTR overlap. As shown in Figure 3, OTR 1 contains tourist attraction set 1, and the content includes K, L, M, N, O, P, Q, R, S, and T. As for OTR 2, it contains tourist attraction set 2, and the content has A, B, C, D, K, L, M, N, O, and P. OTR 1 and OTR 2 thus share some of the same tourist attraction information, namely K, L, M, N, O, and P. This indicates that author OTR1 and author OTR 2 may share similar appraisals regarding tourist attractions K, L, M, N, O, and P. In addition, OTR 1 content recommends not only "K, L, M, N, O, and P", but also "Q, R, S, and T". This means that there is a direct relationship among tourist attractions "K, L, M, N, and O" and "Q, R, S, and T". On the other hand, OTR 2 content recommends "K, L, M, N, O, and P" and "A, B, C, and D", which shows that there is a direct relationship among tourist attractions This study applied set theory to explain the relationships between different tourist attraction sets. Each piece of OTR content contains a tourist attraction set. There are different types of sets among different tourist attraction sets, and they are classified into two types: disjoint sets and intersections, as explained below: • Disjoint sets: If a specific tourist attraction cannot be found in two tourist attraction sets, it is likely that the two OTR do not overlap, and thus there is no direct relationship between the two tourist attraction sets.

•
Intersection: If two tourist attraction sets share the same tourist attraction information, it means that the two OTR overlap. As shown in Figure  Based on the notion that tourist attractions "K, L, M, N, O, and P" have direct relationships with "Q, R, S, and T" and "A, B, C, and D", there may be an indirect relationship between tourist attraction "Q, R, S, and T" (OTR1) and "A, B, C, and D" (OTR2). As seen in Figure 3, besides tourist attractions "K, L, M, N, O, and P", the authors of OTR1 and OTR2 content may also consider "Q, R, S, and T" and "A, B, C, and D" when they are recommending tourist attractions, When travelers want to visit tourist attractions "Q, R, S, and T", they may be recommended to visit tourist attractions "K, L, M, N, O, and P", due to the direct relationship. They may also be recommended to visit tourist attractions "A, B, C, and D" due to the indirect relationship.
There are thus latent relations between and among tourist attractions. In order to understand and effectively manage these, this study designed a tourist knowledge structure, as shown in Figure 4.  Based on the notion that tourist attractions "K, L, M, N, O, and P" have direct relationships with "Q, R, S, and T" and "A, B, C, and D", there may be an indirect relationship between tourist attraction "Q, R, S, and T" (OTR1) and "A, B, C, and D" (OTR2). As seen in Figure 3, besides tourist attractions "K, L, M, N, O, and P", the authors of OTR1 and OTR2 content may also consider "Q, R, S, and T" and "A, B, C, and D" when they are recommending tourist attractions, When travelers want to visit tourist attractions "Q, R, S, and T", they may be recommended to visit tourist attractions "K, L, M, N, O, and P", due to the direct relationship. They may also be recommended to visit tourist attractions "A, B, C, and D" due to the indirect relationship.

Design of the Tourist Knowledge Structure
There are thus latent relations between and among tourist attractions. In order to understand and effectively manage these, this study designed a tourist knowledge structure, as shown in Figure 4. Based on the notion that tourist attractions "K, L, M, N, O, and P" have direct relationships with "Q, R, S, and T" and "A, B, C, and D", there may be an indirect relationship between tourist attraction "Q, R, S, and T" (OTR1) and "A, B, C, and D" (OTR2). As seen in Figure 3, besides tourist attractions "K, L, M, N, O, and P", the authors of OTR1 and OTR2 content may also consider "Q, R, S, and T" and "A, B, C, and D" when they are recommending tourist attractions, When travelers want to visit tourist attractions "Q, R, S, and T", they may be recommended to visit tourist attractions "K, L, M, N, O, and P", due to the direct relationship. They may also be recommended to visit tourist attractions "A, B, C, and D" due to the indirect relationship.
There are thus latent relations between and among tourist attractions. In order to understand and effectively manage these, this study designed a tourist knowledge structure, as shown in Figure 4.

Design of the Tourist Knowledge Structure
In order to understand the tourist knowledge structure of OTR, the study used the research method of Chen and Chen (2012) [5], and   [6], to explain the definition of "tourist knowledge structure relations" and "design of concept schema for tourist knowledge structure", as follows:

Definition of Tourist Knowledge Structure Relations
Based on Figure 4, this section defines the tourist knowledge structure, based on the hierarchy relation, direct relation, and indirect relation. This is elaborated in Figure 5, and the detailed explanations are as follows:

•
Hierarchy relation: A hierarchy relation indicates OTR content that includes tourist attraction information. For example, through hierarchy relation, it is expected to be possible to find tourist attractions related to OTR content.

•
Direct relation: In addition to a hierarchy relation, there can also be a direct relation between attractions. There is a direct relation between a father node and a child node. As seen in Figure 5, tourist attraction C is the child node of tourist attraction D, which indicates that there is a direct relation between the two. • Indirect relation: Since there is a direct relation between a grandfather node and a father node, and a direct relation also exists between a father node and a child node, it may be said that there is an indirect relation between a grandfather node and a child node. For example, tourist attraction C is the father node of tourist attraction B, and tourist attraction B is the father node of tourist attraction A. Tourist attraction C is the grandfather node of tourist attraction A. There is thus an indirect relation between tourist attraction C and tourist attraction A. In order to understand the tourist knowledge structure of OTR, the study used the research method of Chen and Chen (2012) [5], and   [6], to explain the definition of "tourist knowledge structure relations" and "design of concept schema for tourist knowledge structure", as follows:

Definition of Tourist Knowledge Structure Relations
Based on Figure 4, this section defines the tourist knowledge structure, based on the hierarchy relation, direct relation, and indirect relation. This is elaborated in Figure 5, and the detailed explanations are as follows: • Hierarchy relation: A hierarchy relation indicates OTR content that includes tourist attraction information. For example, through hierarchy relation, it is expected to be possible to find tourist attractions related to OTR content. • Direct relation: In addition to a hierarchy relation, there can also be a direct relation between attractions. There is a direct relation between a father node and a child node. As seen in Figure  5, tourist attraction C is the child node of tourist attraction D, which indicates that there is a direct relation between the two. • Indirect relation: Since there is a direct relation between a grandfather node and a father node, and a direct relation also exists between a father node and a child node, it may be said that there is an indirect relation between a grandfather node and a child node. For example, tourist attraction C is the father node of tourist attraction B, and tourist attraction B is the father node of tourist attraction A. Tourist attraction C is the grandfather node of tourist attraction A. There is thus an indirect relation between tourist attraction C and tourist attraction A.

Design of Concept Schema for the Tourist Knowledge Structure
Based on the ideas presented above, this study used an object oriented design model to design a schema for the tourist knowledge structure, as shown in Figure 6. This schema is composed of tourist attractions, attributes, and relations, and is explained as follows.

Design of Concept Schema for the Tourist Knowledge Structure
Based on the ideas presented above, this study used an object oriented design model to design a schema for the tourist knowledge structure, as shown in Figure 6. This schema is composed of tourist attractions, attributes, and relations, and is explained as follows.

•
Tourist attraction: Tourist attractions are the basic units that make up the tourist knowledge ontology, and this part of the schema records tourist attractions related to OTR content. • Attribute: The tourist attraction attributes include (i) tourist attraction ID (number), (ii) tourist attraction name, (iii) tourist attraction district, (iv) tourist attraction review, and (v) distance. Tourist attraction district indicates the city or area in which the tourist attraction is located. Tourist attraction review indicates whether the tourist attraction has a positive or negative appraisal. Distance shows the distance between tourist attractions, as measured in kilometers. • Tourist attraction: Tourist attractions are the basic units that make up the tourist knowledge ontology, and this part of the schema records tourist attractions related to OTR content. • Attribute: The tourist attraction attributes include (i) tourist attraction ID (number), (ii) tourist attraction name, (iii) tourist attraction district, (iv) tourist attraction review, and (v) distance. Tourist attraction district indicates the city or area in which the tourist attraction is located. Tourist attraction review indicates whether the tourist attraction has a positive or negative appraisal. Distance shows the distance between tourist attractions, as measured in kilometers.

•
Relation: As indicated in the previous section, the relations among tourist attractions include the (i) hierarchy relation, (ii) direct relation, and (iii) indirect relation [26].

Analysis of Ontological Characteristics
Based on the results from Section 2.3, this study used ontology for the tourist knowledge structure, as shown in Figure 7. Ontology characteristics are analyzed as the basis to design the methods of recommendation for tourist attractions. Chen (2010) stated that in an ontology there is direct relation between a grandfather node and a father node [26], and a direct relation also exists between a father node and a child node, while there is an indirect relation between a grandfather node and a child node [26]. As can be seen in Figure 7, there is a direct relation between TA B and C, and a direct relation also exists between TA B and A, while there is an indirect relation between TA A and C. When travelers are located at TA G, the system will recommend attractions based on both direct and indirect relations.

Analysis of Ontological Characteristics
Based on the results from Section 2.3, this study used ontology for the tourist knowledge structure, as shown in Figure 7. Ontology characteristics are analyzed as the basis to design the methods of recommendation for tourist attractions. Chen (2010) stated that in an ontology there is direct relation between a grandfather node and a father node [26], and a direct relation also exists between a father node and a child node, while there is an indirect relation between a grandfather node and a child node [26]. As can be seen in Figure 7, there is a direct relation between TA B and C, and a direct relation also exists between TA B and A, while there is an indirect relation between TA A and C. When travelers are located at TA G, the system will recommend attractions based on both direct and indirect relations. As a result, travelers can get four recommendations for tourist attractions: (1) TA E and D, (2) TA E and F, (3) TA H, and (4) TA I. Travelers are then able to select their own tourist attractions based on their preferences.
As shown in Figure 8, the tourist knowledge ontology includes hierarchical relations, and tour planning can be carried out using these. After the first and last destinations are chosen, the travelers can find related tourist attractions based on hierarchical relations. For instance, the first stop is set at TA C, and the last destination is set at TA I. D, E, and G are the tourist attractions related to TA C and I. Finally, the analytical results can be used to help travelers conduct their tour planning.
Ontological characteristics can thus be applied to tourist attraction reasoning and tour planning. On the basis of direct and indirect relations, tourist attraction reasoning helps people find popular tourist attractions. As for tour planning, the travelers first decide the first and last destinations, and then tourist attractions between these are recommended by the system. node and a child node [26]. As can be seen in Figure 7, there is a direct relation between TA B and C, and a direct relation also exists between TA B and A, while there is an indirect relation between TA A and C. When travelers are located at TA G, the system will recommend attractions based on both direct and indirect relations. As a result, travelers can get four recommendations for tourist attractions: (1) TA E and D, (2) TA E and F, (3) TA H, and (4) TA I. Travelers are then able to select their own tourist attractions based on their preferences.  As shown in Figure 8, the tourist knowledge ontology includes hierarchical relations, and tour planning can be carried out using these. After the first and last destinations are chosen, the travelers can find related tourist attractions based on hierarchical relations. For instance, the first stop is set at TA C, and the last destination is set at TA I. D, E, and G are the tourist attractions related to TA C and I. Finally, the analytical results can be used to help travelers conduct their tour planning.
Ontological characteristics can thus be applied to tourist attraction reasoning and tour planning. On the basis of direct and indirect relations, tourist attraction reasoning helps people find popular tourist attractions. As for tour planning, the travelers first decide the first and last destinations, and then tourist attractions between these are recommended by the system.

Ontology-Based Tourist Knowledge Representation and Recommendation Method Development
Based on the OTR and ontology characteristics analysis results presented in Section 2, this study then designed an ontology-based tourist knowledge representation and recommendation procedure, which includes the following stages: (i) tourist attraction evaluation, (ii) tourist attraction ontology construction, and (iii) tourist attraction recommendation.

Procedure for Ontology-Based Tourist Knowledge Representation and Recommendation
Using the model presented in Section 2, this study designed a procedure for ontology-based tourist knowledge representation and recommendation, and this is expected to help travelers conduct their tour planning or tourist attraction selection. As shown in Figure 9, the procedure of ontologybased tourist attraction recommendation consists mainly of tourist attraction evaluation, tourist knowledge ontology construction, and tourist attraction recommendation. Tourist knowledge ontology construction involves concept set generation, hierarchy relationship generation, and distance calculation. Finally, tourist attraction recommendation involves tourist attraction matching, tourist attraction reasoning, tour planning, and tourist attraction sorting. These are discussed in more detail in the following section.

Ontology-Based Tourist Knowledge Representation and Recommendation Method Development
Based on the OTR and ontology characteristics analysis results presented in Section 2, this study then designed an ontology-based tourist knowledge representation and recommendation procedure, which includes the following stages: (i) tourist attraction evaluation, (ii) tourist attraction ontology construction, and (iii) tourist attraction recommendation.

Procedure for Ontology-Based Tourist Knowledge Representation and Recommendation
Using the model presented in Section 2, this study designed a procedure for ontology-based tourist knowledge representation and recommendation, and this is expected to help travelers conduct their tour planning or tourist attraction selection. As shown in Figure 9, the procedure of ontology-based tourist attraction recommendation consists mainly of tourist attraction evaluation, tourist knowledge ontology construction, and tourist attraction recommendation. Tourist knowledge ontology construction involves concept set generation, hierarchy relationship generation, and distance calculation. Finally, tourist attraction recommendation involves tourist attraction matching, tourist attraction reasoning, tour planning, and tourist attraction sorting. These are discussed in more detail in the following section.  Figure 9. Procedure for ontology-based tourist attraction recommendation.

Tourist Attraction Evaluation
This study used Pai et al.'s (2013) research method to conduct OTR analysis [24]; the analysis result will be the source of tourist attraction evaluation. People usually prefer to visit tourist attractions that have positive appraisals. In order to see if an attraction has mainly positive or negative appraisals, this study adopted the majority rule as the basis for the evaluation. This method is carried out using Equations (1), (2), and (3), which include the following variables: POTRk p is the number of positive appraisals that the kth tourist attraction has in the tourist appraisal database. p represents the positive appraisal set. NOTRk n is the number of negative appraisals that the kth tourist attraction has in the tourist appraisal database. n represents the negative appraisal set. Allsumk calculates whether tourist attraction k has an overall positive or negative appraisal. If tourist attraction k has a positive value, it means that it has a positive appraisal. In contrast, if tourist attraction k has a negative value, this means that it has a negative appraisal.
Based on Equations (1)-(3), this study designed a tourist attraction evaluation method, as shown in Figure 10, and described in more detail below.
Step 2. Calculate the number of negative appraisals that tourist attraction ki has in the database.
Step 3. Calculate the number of positive appraisals that tourist attraction ki has in the database.
Step 4. Allsumk calculates the overall appraisal of tourist attraction ki. Ck represents a temporary variable, and it stores the results of Allsumk.
Step 5. If the result of Ck is a positive value, step 6 should be implemented. On the other hand, if Ck is a negative value, step 7 should be implemented. Step 6. Ck is a popular tourist attraction.
Step 7. Ck is not a popular tourist attraction.
Step 8. Record the result that Ck is a popular tourist attraction.

Tourist Attraction Evaluation
This study used Pai et al.'s (2013) research method to conduct OTR analysis [24]; the analysis result will be the source of tourist attraction evaluation. People usually prefer to visit tourist attractions that have positive appraisals. In order to see if an attraction has mainly positive or negative appraisals, this study adopted the majority rule as the basis for the evaluation. This method is carried out using Equations (1)-(3), which include the following variables: POTR k p is the number of positive appraisals that the kth tourist attraction has in the tourist appraisal database. p represents the positive appraisal set. NOTR k n is the number of negative appraisals that the kth tourist attraction has in the tourist appraisal database. n represents the negative appraisal set. Allsum k calculates whether tourist attraction k has an overall positive or negative appraisal. If tourist attraction k has a positive value, it means that it has a positive appraisal. In contrast, if tourist attraction k has a negative value, this means that it has a negative appraisal.
Based on Equations (1)-(3), this study designed a tourist attraction evaluation method, as shown in Figure 10, and described in more detail below.
Step 1. Input the tourist attraction set, presented numerically as k i ∈{c 1 ,c 2 ,c 3 , . . . ,c j }, C is the number of tourist attraction sets (j = 1~K).
Step 2. Calculate the number of negative appraisals that tourist attraction k i has in the database.
Step 3. Calculate the number of positive appraisals that tourist attraction k i has in the database.
Step 4. Allsum k calculates the overall appraisal of tourist attraction k i . C k represents a temporary variable, and it stores the results of Allsum k .
Step 5. If the result of C k is a positive value, step 6 should be implemented. On the other hand, if C k is a negative value, step 7 should be implemented.
Step 6. C k is a popular tourist attraction.
Step 7. C k is not a popular tourist attraction.
Step 8. Record the result that C k is a popular tourist attraction. Step 1.
Step 8. Figure 10. Process of tourist attraction evaluation.
This study used this algorithm and the related formula (Equations (1)-(3)) to calculate the results of the tourist attraction evaluation, as shown in Table 1.

Tourist Knowledge Ontology Construction
Formal concept analysis (FCA) is widely applied to knowledge structuring and organization in relation to online content [5][6][7]27,28], and thus it was used in the current study as the basis of developing a tourist knowledge ontology. The FCA is used to structure the hierarchical relationships among tourist attractions, based on the OTR content set and tourist attraction set. The main procedure of the tourist knowledge ontology construction includes the following steps: (i) concept set generation, (ii) hierarchy relationship generation, and (iii) distance calculation, as explained in more detail below. This study used this algorithm and the related formula (Equations (1)-(3)) to calculate the results of the tourist attraction evaluation, as shown in Table 1.

Tourist Knowledge Ontology Construction
Formal concept analysis (FCA) is widely applied to knowledge structuring and organization in relation to online content [5][6][7]27,28], and thus it was used in the current study as the basis of developing a tourist knowledge ontology. The FCA is used to structure the hierarchical relationships among tourist attractions, based on the OTR content set and tourist attraction set. The main procedure of the tourist knowledge ontology construction includes the following steps: (i) concept set generation, (ii) hierarchy relationship generation, and (iii) distance calculation, as explained in more detail below.

Concept Set Generation
Based on the results reported in Section 3.2, the OTR content analysis, OTR content set (y∈{y j , j = 1,2,3, . . . ,N})), and tourist attraction set that has positive appraisals can be found (x∈{x i , i = 1,2,3, . . . ,M}). The OTR content set and tourist attraction set were used to build a binary matrix, shown in Equation (4): r(a m , b n ) indicates the relationship between the OTR content set and the tourist attraction set. If a tourist attraction appears in the OTR content set, then the value of r(a m , b n ) is 1, and otherwise the value of r(a m , b n ) is 0, as shown in Table 2.
Y is defined as the subset of OTR content set D. The relationship between Y and D is marked as Y⊆D. X is the subset of tourist attraction set W. The relationship between X and W is shown as X⊆W. If X satisfies Equation (5) and Y satisfies Equation (6), then X and Y is known as a concept c, which can be defined as c(x, y), All of the concepts c(x i , y j ) are collectively defined as C. Based on Table 2, Equations (4)-(6) are used to carry out tourist attraction set generation, as shown in Figure 11.

Concept Set Generation
Based on the results reported in Section 3.2, the OTR content analysis, OTR content set (y∈{yj, j = 1,2,3,…,N})), and tourist attraction set that has positive appraisals can be found (x∈{xi, i = 1,2,3,…,M}). The OTR content set and tourist attraction set were used to build a binary matrix, shown in Equation r(am, bn) indicates the relationship between the OTR content set and the tourist attraction set. If a tourist attraction appears in the OTR content set, then the value of r(am, bn) is 1, and otherwise the value of r(am, bn) is 0, as shown in Table 2.
Y is defined as the subset of OTR content set D. The relationship between Y and D is marked as Y⊆ D. X is the subset of tourist attraction set W. The relationship between X and W is shown as X⊆ W. If X satisfies Equation (5) and Y satisfies Equation (6), then X and Y is known as a concept c, which can be defined as c(x, y), All of the concepts c(xi, yj) are collectively defined as C. Based on Table 2, Equations (4)-(6) are used to carry out tourist attraction set generation, as shown in Figure 11.   Figure 11. Tourist attraction set generation. Figure 11. Tourist attraction set generation.

Hierarchy Relationship Generation
FCA obtains the concept hierarchy based on set operations related to mutual exclusiveness, intersections, and unions. Therefore, when tourist attractions X 1 and X 2 have similar OTR content and the former has more OTR content than the latter, this means tourist attraction X 1 is the super concept of tourist attraction X 2 . On the other hand, when tourist attractions X 1 and X 2 have similar OTR content and tourist attraction X 1 has less OTR content than tourist attraction X 2 , this means tourist attraction X 1 is the child concept of tourist attraction X 2 . This study thus applied set operations, as follows. If X 1 ⊆X 2 , c 1 (x 1 ,y 1 ) is the child concept of c 2 (x 2 ,y 2 ). The same tourist attraction may have many super concepts and child concepts to build the hierarchical relationships among tourist attractions. The study used Equation (7) to calculate the supremum of tourist attractions, and this can obtain the hierarchical relationships among tourist attraction words.
(x 1 , y 1 )∪(x 2 , y 2 ) = (τ(y 1 ∩y 2 ), y 1 ∩y 2 ) In addition to hierarchical relationships, tourist attractions also have mutual relations with one another. Take tourist attractions c 1 (x 1 ,y 2 ) and c 2 (x 2 ,y 2 ) as an example; if X 1 ⊂X 2 and X 2 ⊂X 1 , then c 1 and c 2 have a mutual relation. This study built the hierarchical relationships among tourist attractions based on Figure 11, as shown in Figure 12.

Hierarchy Relationship Generation
FCA obtains the concept hierarchy based on set operations related to mutual exclusiveness, intersections, and unions. Therefore, when tourist attractions X1 and X2 have similar OTR content and the former has more OTR content than the latter, this means tourist attraction X1 is the super concept of tourist attraction X 2. On the other hand, when tourist attractions X1 and X2 have similar OTR content and tourist attraction X1 has less OTR content than tourist attraction X2, this means tourist attraction X1 is the child concept of tourist attraction X2. This study thus applied set operations, as follows. If X1⊆ X2, c1(x1,y1) is the child concept of c2(x2,y2). The same tourist attraction may have many super concepts and child concepts to build the hierarchical relationships among tourist attractions. The study used Equation (7) to calculate the supremum of tourist attractions, and this can obtain the hierarchical relationships among tourist attraction words.

Distance Calculation:
The distance calculation measures the distance between tourist attractions in the tourist knowledge ontology. In order to calculate this automatically, this study applied Google Maps API. The main procedures in the distance calculation are as follows: (i) obtain the tourist attraction input, (ii) use the HTML parser, and (iii) carry out distance extraction, as shown in Figure 13, explained as follows: (1) Tourist attraction input: The names of tourist attractions from the tourist knowledge ontology were used as the data for Google Maps' API. As shown in Figure 13, if there is a relationship between tourist attractions A and B in tourist knowledge ontology, then they are found on Google Maps to obtain the distance between the two points. (2) HTML parser: Every Google Maps platform has its own HTML tag format. In order to get the accurate distance between two tourist attractions, the HTML parser was used to analyze the HTML structure of the Google Maps' platform. (3) Distance extraction: In this step, the HTML tags and web links were deleted, while the distance between tourist attractions was recorded and saved in the tourist knowledge ontology.

Distance Calculation
The distance calculation measures the distance between tourist attractions in the tourist knowledge ontology. In order to calculate this automatically, this study applied Google Maps API. The main procedures in the distance calculation are as follows: (i) obtain the tourist attraction input, (ii) use the HTML parser, and (iii) carry out distance extraction, as shown in Figure 13, explained as follows: (1) Tourist attraction input: The names of tourist attractions from the tourist knowledge ontology were used as the data for Google Maps' API. As shown in Figure 13, if there is a relationship between tourist attractions A and B in tourist knowledge ontology, then they are found on Google Maps to obtain the distance between the two points. (2) HTML parser: Every Google Maps platform has its own HTML tag format. In order to get the accurate distance between two tourist attractions, the HTML parser was used to analyze the HTML structure of the Google Maps' platform. (3) Distance extraction: In this step, the HTML tags and web links were deleted, while the distance between tourist attractions was recorded and saved in the tourist knowledge ontology.  Figure 13. Process of distance calculation.

Tourist Attraction Recommendation
Based on the discussions presented in Section 3.3, this study designed a process of tourist attraction recommendation. The main procedure includes tourist attraction matching, tourist attraction reasoning, tour planning, and tourist attraction sorting, as explained below.

Tourist Attraction Matching
When travelers are using tourist reasoning or tour planning, they usually search for the name of the target tourist attraction. Tourist attraction matching thus matches the attraction names from the travelers with those in the tourist knowledge ontology. In this way, the system may be able to find the tourist attraction information the travelers are looking for.
Tourist attraction matching applies the Jaccard Coefficient to make a comparison between tourist attraction names, with the calculation shown in Equation (8).

Tourist Attraction Recommendation
Based on the discussions presented in Section 3.3, this study designed a process of tourist attraction recommendation. The main procedure includes tourist attraction matching, tourist attraction reasoning, tour planning, and tourist attraction sorting, as explained below.

Tourist Attraction Matching
When travelers are using tourist reasoning or tour planning, they usually search for the name of the target tourist attraction. Tourist attraction matching thus matches the attraction names from the travelers with those in the tourist knowledge ontology. In this way, the system may be able to find the tourist attraction information the travelers are looking for.
Tourist attraction matching applies the Jaccard Coefficient to make a comparison between tourist attraction names, with the calculation shown in Equation (8).

Sub jectSimilarity(TA, TK
The process of subject matching is based on Equation (8), as shown in Figure 14. "TA" represents the tourist attraction travelers are searching for. TK i is the ith tourist attraction name set from the tourist knowledge ontology (i.e., TK i = {TK 1 ,TK 2 ,TK 3 , . . . ,TK i }). TS j is a set of tourist attraction similarity values (i.e., TS j = {TS 1 ,TS 2 ,TS 3 , . . . ,TS j }).

Tourist Attraction Reasoning:
The tourist attraction reasoning was based on Section 2.4., about the analysis of ontological characteristics. The location of travelers is used to find the direct and indirect relations in the tourist attraction reasoning. This process is explained below.
The tourist knowledge ontology is transformed into a relation matrix. (Xi, Xj) represents the relationship between two tourist attractions. If the value of (Xi, Xj) is 1, it means there is a relationship between tourist attractions Xi and Xj. If the value of (Xi, Xj) is 0, it means there is no relationship between tourist attractions Xi and Xj, as shown in Figure 15. As can be seen in Figure 15, tourist attraction X1 is set as the location for the traveler. Tourist attraction reasoning was then carried out to see the direct and indirect relations among the attractions. The results of the reasoning are (X2, X4), (X2, X4), (X3, X6), and (X3, X7), respectively, as shown in Table  3.

Tourist Attraction Reasoning
The tourist attraction reasoning was based on Section 2.4., about the analysis of ontological characteristics. The location of travelers is used to find the direct and indirect relations in the tourist attraction reasoning. This process is explained below.
The tourist knowledge ontology is transformed into a relation matrix. (X i , X j ) represents the relationship between two tourist attractions. If the value of (X i , X j ) is 1, it means there is a relationship between tourist attractions X i and X j . If the value of (X i , X j ) is 0, it means there is no relationship between tourist attractions X i and X j , as shown in Figure 15.

Tourist Attraction Reasoning:
The tourist attraction reasoning was based on Section 2.4., about the analysis of ontological characteristics. The location of travelers is used to find the direct and indirect relations in the tourist attraction reasoning. This process is explained below.
The tourist knowledge ontology is transformed into a relation matrix. (Xi, Xj) represents the relationship between two tourist attractions. If the value of (Xi, Xj) is 1, it means there is a relationship between tourist attractions Xi and Xj. If the value of (Xi, Xj) is 0, it means there is no relationship between tourist attractions Xi and Xj, as shown in Figure 15. As can be seen in Figure 15, tourist attraction X1 is set as the location for the traveler. Tourist attraction reasoning was then carried out to see the direct and indirect relations among the attractions. The results of the reasoning are (X2, X4), (X2, X4), (X3, X6), and (X3, X7), respectively, as shown in Table  3. As can be seen in Figure 15, tourist attraction X 1 is set as the location for the traveler. Tourist attraction reasoning was then carried out to see the direct and indirect relations among the attractions. The results of the reasoning are (X 2 , X 4 ), (X 2 , X 4 ), (X 3 , X 6 ), and (X 3 , X 7 ), respectively, as shown in Table 3. Table 3. Record the results of tourist attraction reasoning.

Steps Seat (S) Neighbor Node (N) Action
The sub nodes of X 2 and X 3 are searched 2 X 2 X 2 →X 4 , X 2 →X 4 Output the search results: (X 2 , X 4 ) and (X 2 , X 5 ) 3 X 3 X 3 →X 6 , X 3 →X 7 Output the search results: (X 3 , X 6 ) and (X 3 , X 7 ) A computer program was used to automatically carry out tourist attraction reasoning, based on the algorithm shown in Algorithm 1. if-end for-end } traversal (X i ){ while (!empty(Q))//it judges if queue is null )//In relation matrix, the value of (X i , X j ) is 1, and noted X j is not visited. Then, there is a relationship between X i and X j , and X j has not been searched by the program.
visit[X j ] = true; //The found node X j is set as visited and will not be searched later. output (X i , X j ); // Out put the result node, X j traversal(X i ); //It requests traversal() function and loads the value of node(X i ) if-end if (all visit[X j ])//If all the nodes (X j ) has been visited. dequeue(Q); //It delete the front node of queue. traversal(front(Q)); //It requests traversal() function and loads the first information (node) from queue.
if-end for-end while-end }

Tour Planning
Based on the starting point and destination set by the travelers and tourist knowledge ontology, the tourist attractions between the starting point and destination are found. The method for tour planning is explained as follows.
In Figure 15, the tourist attraction X 1 is set as the starting point of a tour and X 8 is set as the final destination. Using the tourist knowledge ontology, tourist attractions related to X 1 and X 8 are found. The results of the tour planning are X 1 , X 3 , X 6 , and X 8 , as shown in Figure 16.
information (node) from queue.

if-end
for-end while-end }

Tour Planning
Based on the starting point and destination set by the travelers and tourist knowledge ontology, the tourist attractions between the starting point and destination are found. The method for tour planning is explained as follows.
In Figure 15, the tourist attraction X1 is set as the starting point of a tour and X8 is set as the final destination. Using the tourist knowledge ontology, tourist attractions related to X1 and X8 are found. The results of the tour planning are X1, X3, X6, and X8, as shown in Figure 16.

Start Node End Node Search Process
Step 2 Step 3 Step 6 Step 5 Step 4 Step 7 Step 8 Step 9 Step 10 Figure 16. Process for tour planning.
In order to automatize the tour planning process, the study developed a tour planning algorithm, as detailed in Algorithm 2, below. In order to automatize the tour planning process, the study developed a tour planning algorithm, as detailed in Algorithm 2, below.

Algorithm 2. Algorithm for tour planning.
Tourist knowledge ontology transformation into a relation matrix [X i ] [X j ] Input relation matrix [X i ] [X j ] input start node (X s )//The user inputs the start node (X s ), X s is a node in the relation matrix. input end node (X e )//The user inputs end node (X e ), X e is a node in the tourist knowledge ontology. create stack(S)//It sets up stack. main(){ for X i = 1 toX A //X 1 to X A are the nodes in the tourist knowledge ontology. The nodes from X 1 to X A are set as not visited.
visit[X i ] = false; for-end visit [X s ] = true; //The starting point (X s ) is set as visited. push (S, X s ); //The starting point (X s ) is put into the stack. for //It searches the nodes that has not been visited (X i ) visit [X i ] = true; // It finds the node (X i ) and sets it as visited. push (S, X i ); //The node (X i ) is put into the stack traversal (X i ); //It requests traversal() function and loads node ( The value of the relation matrix (X i , X j ) is 1, and the node X j is set as not visited.
if ([X j ] == [X e ])//Node (X j ) equals to destination node (X e ) output stack(S); //It outputs all the nodes in the stack. else//Node (X j ) is different from destination node (X e ) visit [X j ] = true; //Node (X j ) is set as visited.
Push (S, X j ); //Node (X j ) is saved into the stack. traversal(X j ); //It requests formula, traversal(), and loads node (X j ) if-end if (all visit[X j ])//All the nodes (X j ) has been visited. pop(S); //The nodes of stack (pop) is deleted. traversal(top(S)); //It requests traversal() function and loads the first node (top(S)) on top of the stack.
if-end for-end }

Tourist Attraction Sorting
Based on Section 3.4.2, attraction reasoning, and Section 3.4.3, tour planning, there may be one or more than one results. The tourist attraction sorting ranks all the analytical results based on their distances. The nearest destination is ranked first and will be recommended to travelers. The method for tourist attraction sorting is explained as follows: Section 3.3.3 showed how to measure the distance between tourist attractions X i and X j . The relation matrix thus becomes a distance matrix. (X i , X j ) is used to represent the distance between tourist attractions, as displayed in Figure 17. or more than one results. The tourist attraction sorting ranks all the analytical results based on their distances. The nearest destination is ranked first and will be recommended to travelers. The method for tourist attraction sorting is explained as follows: Section 3.3.3 showed how to measure the distance between tourist attractions Xi and Xj. The relation matrix thus becomes a distance matrix. (Xi, Xj) is used to represent the distance between tourist attractions, as displayed in Figure 17. As Algorithm 2 showed, the results of tourist attraction reasoning are (X2, X4), (X2, X5), (X3, X6), and (X3, X7). The distance matrix reveals the distance between points. The distance of (X2, X4) is 4, the distance of (X2, X5) is 8, the distance of (X3, X6) is 64, and the distance of (X3, X7) is 23. These are then ranked by distance, and the one with the nearest distance is ranked first. The first recommended result is thus (X2, X4), followed by (X2, X5), (X3, X7), and (X3, X6). In order to automatize the tourist attraction sorting, this study developed an algorithm for tourist attraction sorting, as shown in Algorithm 3.  As Algorithm 2 showed, the results of tourist attraction reasoning are (X 2 , X 4 ), (X 2 , X 5 ), (X 3 , X 6 ), and (X 3 , X 7 ). The distance matrix reveals the distance between points. The distance of (X 2 , X 4 ) is 4, the distance of (X 2 , X 5 ) is 8, the distance of (X 3 , X 6 ) is 64, and the distance of (X 3 , X 7 ) is 23. These are then ranked by distance, and the one with the nearest distance is ranked first. The first recommended result is thus (X 2 , X 4 ), followed by (X 2 , X 5 ), (X 3 , X 7 ), and (X 3 , X 6 ). In order to automatize the tourist attraction sorting, this study developed an algorithm for tourist attraction sorting, as shown in Algorithm 3.

Prototype Implementation and System Evaluation
Based on the proposed tourist knowledge structure-based tourist attraction recommendation method, this section describes a prototype for implementing an analysis of the tourist attraction recommendation system using C#. The technology used for this implementation, as well as the results with an illustrative case study, are described below.

Implementation Environment
The technology used for implementing the prototype include an Intel Core TM i5-760-2.8GHz PC, Microsoft Windows 7 Professional, Internet Information Services (IIS), Microsoft SQL Server 2012, and Microsoft Visual Studio 2010. Figure 18 shows the tourist knowledge structure-based tourist attraction recommendation mechanism framework, which includes three layers of user interface, mechanism operation, and tourist knowledge repository, as explained below.
(1) User interface layer: The users of the program are travelers. The system provides two major functions: tourist attraction reasoning and tour planning. With regard to the former, after the travelers input their location, the program uses the direct and indirect relations to recommend nearby tourist attraction. For tour planning, the travelers first decide the starting and final tourist attractions. The program then uses the tourist knowledge ontology to find attractions between these two points. (2) Mechanism operation layer: There are five modules in this mechanism, as follows: application module, tourist attraction evaluation module, tourist knowledge ontology construction module, and tourist attraction recommendation module. (3) Tourist knowledge repository layer: This includes the following five repositories: OTR content repository, tourist attraction name repository, positive appraisal word repository, negative appraisal word repository, and tourist knowledge ontology (OWL) repository. The OTR content repository stores OTR content. The tourist attraction name repository stores tourist attraction names. The positive appraisal word repository stores positive appraisal words related to tourist attractions, while the negative appraisal word repository stores negative words. The tourist knowledge ontology (OWL) repository stores the relationships and attributes among nodes in the tourist knowledge ontology.
The technology used for implementing the prototype include an Intel Core TM i5-760-2.8GHz PC, Microsoft Windows 7 Professional, Internet Information Services (IIS), Microsoft SQL Server 2012, and Microsoft Visual Studio 2010. Figure 18 shows the tourist knowledge structure-based tourist attraction recommendation mechanism framework, which includes three layers of user interface, mechanism operation, and tourist knowledge repository, as explained below.  (1) User interface layer: The users of the program are travelers. The system provides two major functions: tourist attraction reasoning and tour planning. With regard to the former, after the

Implementation Results
Through the research of   [24], the system conducted OTR collection and appraisal analysis. The OTR source is from Taiwan PTT (Taiwan bulletin board system name) Forum, of which the content is related to Southern Taiwan (Pingtung County, Kaohsiung City, Tainan City, and Chiayi County). The study collated 9146 of OTR content found 720 attractions through the research method. Therefore, 720 attractions were constructed and identified as tourist knowledge. Finally, this study will be as a system, shown as Figures 19-23. Figure 19 shows the selection of the recommendation model. Figure 20 shows the input of the traveler's location, while Figure 21 shows the results of tourist attraction reasoning model. Figure 22 shows the input start and end nodes, and Figure 23 shows the results of the tour planning model. travelers input their location, the program uses the direct and indirect relations to recommend nearby tourist attraction. For tour planning, the travelers first decide the starting and final tourist attractions. The program then uses the tourist knowledge ontology to find attractions between these two points.
(2) Mechanism operation layer: There are five modules in this mechanism, as follows: application module, tourist attraction evaluation module, tourist knowledge ontology construction module, and tourist attraction recommendation module.
• (3) Tourist knowledge repository layer: This includes the following five repositories: OTR content repository, tourist attraction name repository, positive appraisal word repository, negative appraisal word repository, and tourist knowledge ontology (OWL) repository. The OTR content repository stores OTR content. The tourist attraction name repository stores tourist attraction names. The positive appraisal word repository stores positive appraisal words related to tourist attractions, while the negative appraisal word repository stores negative words. The tourist knowledge ontology (OWL) repository stores the relationships and attributes among nodes in the tourist knowledge ontology.

Implementation Results
Through the research of   [24], the system conducted OTR collection and appraisal analysis. The OTR source is from Taiwan PTT (Taiwan bulletin board system name) Forum, of which the content is related to Southern Taiwan (Pingtung County, Kaohsiung City, Tainan City, and Chiayi County). The study collated 9146 of OTR content found 720 attractions through the research method. Therefore, 720 attractions were constructed and identified as tourist knowledge. Finally, this study will be as a system, shown as Figures 19-23. Figure 19 shows the selection of the recommendation model. Figure  20 shows the input of the traveler's location, while Figure 21 shows the results of tourist attraction reasoning model. Figure 22 shows the input start and end nodes, and Figure 23 shows the results of the tour planning model.            Figure 23. Results of tour planning model.

System Evaluation
In order to evaluate the effectiveness of this method and help travelers quickly look for popular attractions and tourism planning, this study conducted a satisfaction survey about this system after use. Currently, the user's satisfaction survey is commonly used in information system assessments to assess the efficacy of the system. The user's satisfaction survey in this study was extracted from the research by Li and Chang (2009) [29],   [6], and Lin et al. (2017) [25]. The criteria of satisfactory assessment is based on the Likert scale, where 1 = very disagree, 3 = not determined, 5 = very agree.
The system evaluation of this study was conducted in control and experimental groups. The participation of control groups is mainly for evaluating the traditional search methods (ex:blog, BBS, search engine); the experimental groups participation is for the evaluation of the system. Then, based on the information from the participants of control and experimental groups experiences, t-test was used to understand what the significant difference between the traditional search methods and the research is. Therefore, 60 students were tested for the system evaluation, these students must have experience in using BBS (bulletin board system), Blog, and search engine to search for popular attractions and travel planning.
According to Table 4, the analysis result showed that the method of this study is able to quickly recommend popular attractions and travel planning, which is better than the traditional way of blog, BBS, search engine. However, this study is a short-term system assessment to help the user. A completed system evaluation is required to take 3 to 5 years, so the study advises that a long-term research is conducted in the future to improve the effectiveness of the system.

System Evaluation
In order to evaluate the effectiveness of this method and help travelers quickly look for popular attractions and tourism planning, this study conducted a satisfaction survey about this system after use. Currently, the user's satisfaction survey is commonly used in information system assessments to assess the efficacy of the system. The user's satisfaction survey in this study was extracted from the research by Li and Chang (2009) [29],   [6], and Lin et al. (2017) [25]. The criteria of satisfactory assessment is based on the Likert scale, where 1 = very disagree, 3 = not determined, 5 = very agree.
The system evaluation of this study was conducted in control and experimental groups. The participation of control groups is mainly for evaluating the traditional search methods (ex:blog, BBS, search engine); the experimental groups participation is for the evaluation of the system. Then, based on the information from the participants of control and experimental groups experiences, t-test was used to understand what the significant difference between the traditional search methods and the research is. Therefore, 60 students were tested for the system evaluation, these students must have experience in using BBS (bulletin board system), Blog, and search engine to search for popular attractions and travel planning.
According to Table 4, the analysis result showed that the method of this study is able to quickly recommend popular attractions and travel planning, which is better than the traditional way of blog, BBS, search engine. However, this study is a short-term system assessment to help the user. A completed system evaluation is required to take 3 to 5 years, so the study advises that a long-term research is conducted in the future to improve the effectiveness of the system.

Conclusions and Further Work
The study applied ontology into the development of a tourist knowledge representation and recommendation mechanism. It is hoped that through the application of Internet, tourist knowledge can be effective managed and used. It may be able to turn the OTR on various blogs and web forums into valuable information. Travelers can extract helpful tourist attraction information to assist their traveling decision-making. In addition, the method may also be able to help tourism business to understand current popular tourist attractions. The tourism business may use the information as the reference to design their tourism products. It may improve their competitiveness in the tourism market. The primary results and contributions of this study are summarized as follows: (1) Ontology-based tourist knowledge analysis: Based on the features of OTR and ontology, this study designed an ontology-based tourist knowledge representation and recommendation model to analyze OTR content. It is expected that this can help users to find knowledge to better plan tours. (2) Ontology-based tourist knowledge representation and recommendation method: Based on the ontology-based tourist knowledge representation and recommendation model, the study developed a method for ontology-based tourist knowledge representation and recommendation. The method includes an OTR collection module, OTR content analysis module, tourist knowledge ontology construction module, and tourist attraction recommendation module. (3) eWOM analysis mechanism: The study developed a method for ontology-based tourist knowledge representation and recommendation. Through the automatization of identifying and analyzing OTR, it is expected that this approach can provide the tourist knowledge that travelers need when planning their trips.  [25]. The results showed that the breadth search, the operation method, and the analysis results are better than the traditional ones.
After an interview with the user, in their opinion, the advantage of this system is to find the link and idea for each tourist attraction, and to provide the users with new tourists.  [30], text corpus-based tourism helps to observe and realize changes in the tourism market. OTR in this study is viewed as a kind of text corpus based tourism, which of the direction is confirmed to be correct, because the study by Gong, et al. (2018) [31] has the same opinion. However, the source information between Gong, et al. (2018) and this study is different. Gong, el al. (2018) is mainly for the tourist packages; this study is for OTR. In terms of data analysis, OTR collate the travelers' negative and positive appraisals, but tourist packages are from the enterprise's perspective. Therefore, the OTR data is more objective than the tourist packages.
Based on the proposed model and method in this study, the following future research directions are recommended: (1) This study extracts tourist knowledge from OTR content, which is then used as the basis for developing a tourist attraction reasoning and tour planning method. Nevertheless, tourist attractions are not the only factor that travelers take into consideration when the tourists are planning their tour. They may also consider issues such as time, budget, age, and preferences. These factors can thus be included in future studies to provide better recommendations. (2) In order to produce a tourist knowledge structure that meets the need of travelers, an ontology adapting and management mechanism can be considered in future research, as this could help to better understand and maintain the tourist knowledge structure. In addition, OTR content from tourist experts can also be collected to construct another knowledge structure. In this way, a multifaceted tourist knowledge management mechanism can be developed to provide travelers with a variety of recommendations.