國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
I. Introduction
1.1.
Background
Tourism is the most popular activity that people love to do during their holidays such as a long weekend after their busy working days or when their children have summer or winter vacation. According to the statistics by the United Nation World Tourism Organisation (UNWTO) 2013, 25 million tourists travelled overseas in the year 1950 but that number had increased dramatically by 1010 million to 1035 million tourists by 2012, which averaged about 66.7% per year. Domestic tourists also increased to 6 billion. UNWTO also forecasted that there will be 1.8 billion tourists travelling overseas by 2030 with an average annual rate of 3.3% from 2010 to 2030 (United Nation World Tourism Organisation, 2013).
Tourists often plan their trips in accordance with search results found on various travel agents’ websites after providing inputs such as the desired travel location and time. The tourism recommendation system we plan to develop is the tourists use to plan their trips on travel agents’ websites. It is a filtering system that utilizes users’
past behaviours and series of discretionary characteristics in recommending another or additional items to the users. Thus, different tourists with the same tourism experience or personal preference would receive different outcomes as they value different aspects of tourism differently.
1.2.
Problem Statement
From the recommendation system, the search results, depending on past individual behaviour such as previously searched key words, list information on all potential tourism locations. This causes people to feel that the tourism information
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
displayed on the results page is difficult to read and quickly identify their desirable places and time. Therefore, this study will collect and utilize tourists’ perception, their decision criteria, and the decision purpose of going on a trip on the tourism recommendation system in order to have more detailed and user oriented search results for the users.
In a rational decision making process, human brains use different frames to act with different decision types (Martino, Kumaran, Seymour & Dolan, 2006). The two-stage decision in this study adopts similar ideas as the Two-Stage Model of Free Will (Doyle, 2011; Dennett, 1981). The model describes that people are able to generate alternative possibilities “freely”, that could be caused or uncaused by prior events, and possibilities “will” be adequately evaluated to choose or select what is desired (Doyle, 2011). In addition, Dennett (1981) denotes the two-stage model as a reasonable idea for making a decision. He mentioned in the case when considering an important decision which an individual cannot decide immediately, one can reject irrelevant courses first and then the remaining options will always have more negligible bearings that will be considered at the second stage. The second stage is to analyse those considerations/options in detail and select the most reasonable and desirable idea in order to make a final decision that the decision maker likes without any regret.
The idea of the Two-Stage Model of Free Will is applied to the two-stage decision model for testing the combination of classifiers in this study. Using the two-stage decision model to simulate a decision with a goal of arousing specific affection by a decision behaviour will yield a better outcome. For example, the usable purpose of participating in a tour or a vocational activity is to get specific joyful
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(relaxing, educating, aesthetic) experiences. A recommendation system which combines the expected affection aroused, decision influencing variables and choices of engaged behaviour will conclude the usable purpose of the decision maker. While this kind of decision type appears to be the selection of a tour (behaviour) but the reality is that it is attempting to engage in certain behaviours which achieve the real purpose of getting specific joyful experiences.
To reach the purposes of recommending the decision target and the arousing of the expected enjoyment (experiences), according to the specific conditions, the hybrid data mining method combining the Decision Tree and K-nearest neighbour approaches (DTKNN) has been proposed in existing literature (Fu & Tu, 2011). The researchers suggested, first, building a database which classifies the data to different subsets that are based on affection aroused using Decision Tree. The algorithm will choose one subset based on the expected affection aroused and then filter the most suitable choices from one of the data subset by K-nearest neighbour (KNN). Decision trees can classify large amounts of data into specific classes based on the attributes of the data. It also possesses the advantage of less processing time. And the KNN methodology allows for greater flexibility in the classification work needed to find all the training examples that are relatively similar to the attributes of the test example. In the algorithm of DTKNN, Decision Trees can classify large amounts but yields less accuracy than the KNN methodology (precision rate 80%) in gift-giving application (Fu & Tu, 2011). This research aims to improve the algorithm in the applied in tourism recommendation system.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
1.3.
Goals and Objectives
There are three aspects that need to be improved from the DTKNN methodology.
The first aspect is to improve the effectiveness of the system by increasing the precision rate at the first stage of the classification. It is important as the precision rate of decision tree classification before combining with KNN to DTKNN is currently inaccurate as it produces unexpectedly low rates. In order to do so, this study will apply these two methodologies: Multi-staged Binary Tree and Back Propagation of Error Neural Network (BPNN) to examine whether the precision rate can be improved to exceed Decision Tree methodology. A higher precision rate will have a direct positive effect on the effectiveness. The reason to select Multi-staged Binary Tree and BPNN to compare with the Decision Tree algorithm is because both classifications methods are more accurate and have been applied on several applications than Decision Tree classification method (Shmueli, Patel & Bruce, 2010).
Secondly, to improve the efficiency by decreasing the amount of questions which the users are required to fill. As the previously proposed approach of DTKNN mentioned above, the system with Decision Tree methodology does not have path ways to generate results with adequate precision rate. Users have to enter all relevant information when searching for the ideal tourism place. To do so, this study will analyse how Multi-staged Binary Tree and BPNN will decrease the search time and produce valuable information to the users in a timelier manner. Lastly, this study will extend the new method to more and general applications.
This study will further compare two other methods, Multi-staged Binary Tree and BPNN for integrating with KNN in an affective oriented recommendation system.
The researcher will also investigate another application namely, tour and vocational
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
activity selection, as an example to extend the usage of the new model to common service selection.
1.4.
Significance of this Research
Tourism Recommendation System that aims to assist users in their decision making is very widely used on the Internet throughout the world. People who are to make travel plans are always in need of a recommendation system with a highly accurate precision rate so that the most ideal decision can be reached. Researcher of this study is still highly motivated to and interested in improving the accuracy rate of the existing tourism recommendation system. Therefore, the researcher believes that a research in algorithms to generate a new approach that improves the accuracy rate is of considerable value to the existing literature and in practice. Ideally the new approach under the idea of the two-stage decision model can be utilized in not just the recommendation system for tourism but also for other applications.
1.5.
A Brief Overview of Research Methodology
In this study, the researcher will conduct a survey to collect data and use the collected sample to validate the algorithms. For each of the three algorithms technique, Decision Tree, Multi-staged Binary Tree and BPNN, the researcher of this study will examine their accuracy or precision rate when integrated with KNN for the generation of a new approach for the recommendation system. The comparison of these algorithms will be discussed in Chapter 2. The methodology of choosing the target application will be discussed in Chapter 3. The tourism industry is the chosen target as its decision making process requires the analysis of a large amount of inputs from the users. These inputs can vary significantly due to the diverse demographic background and differences in personal values of the users. Moreover, evident from the statistics
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
from the UNWTO (2013), tourism is an industry with a vast base of participants. As a result, an improvement to the accuracy of such an extremely complicated decision making process will most likely, if not certainly, creates significant value to the global economy.
The results and suggestions regarding whether Multi-staged Binary Tree or BPNN should be integrated with KNN in becoming the new and better approach for tourism recommendation system domain will be discussed in Chapter 4. The overall conclusion of this study will be made in Chapter 5.