社群商務決策支援機制之設計

(1)

國立交通大學

資訊管理研究所

博士論文

社群商務決策支援機制之設計

Designing Social Commerce Decision Support Mechanisms

研究生：李易霖

指導教授：李永銘博士

(2)

i

社群商務決策支援機制之設計

Designing Social Commerce Decision Support Mechanisms

研究生：李易霖指導教授：李永銘博士

Student: Yi-Lin Lee Advisor: Dr. Yung-Ming Li

國立交通大學資訊管理研究所

博士論文

A Dissertation

Submitted to Institute of Information Management College of Management

National Chiao Tung University in Partial Fulfillment of the Requirements

for the Degree of

Doctor of Philosophy in Information Management July 2012

Hsinchu, Taiwan, the Republic of China

(3)

ii

中文摘要

社群商務決策支援機制之設計

研究生：李易霖

指導教授：李永銘博士

國立交通大學資訊管理研究所博士班

摘要

隨著社群網站的蓬勃發展，以其為基礎的商業用途應用程式也越來越多。然而目前所知多數相關研究以及應用程式開發，其目的多在於建立品牌形象以及支援客戶互動。與前述情況相比，對於購買決策相關應用則較少論及。事實上，許多消費者在購買商品時，會聽取朋友的意見與建議，以作為選擇最終購買商品的參考依據。本研究之目的在於以消費者之線上社群網路為基礎，透過社會心理學以及消費者購買行為決策流程，建立社群商務購買決策支援機制。現實生活中，互動頻繁的朋友較可能是親密的朋友，但在線上社群網站中此種情況是否仍然如此，在進行決策機制設計前必須先加以驗證。透過蒐集本研究所使用之實驗平台上的各項互動，以及實際調查所得到之社會關係指標，利用社群網路分析之 MRQAP 法對此推論進行驗證的結果，證實了此一關係的存在。此一關聯性被確認後，本研究接著針對三種常見的消費者購買決策情境，設計了不同的決策支援機制。消費者進行購買決策時，通常會處於以下三種狀況其中之一。第一，消費者已經找到數種符合需求的商品，需要在其中挑選一項作為最終購買商品。第二，消費者已經列出了某些評選商品的考量因素，但卻不知道有哪些商品符合所列條件。第三，消費者僅僅知道要購買某種商品，但卻不知道從何處開始著手。本研究針對以上情境，分別設計了相對應的決策支援機制。在第一種情境中，本研究設計了決策支援小組的篩選機制，以找出適當的參考團體。而改良過後的投票機制則被用來選出最終的建議購買產品。而在第二種情境中，決策支援小組以 QOC 表達方式，針對消費者所在乎之考量因素給予權重，而後形成最後建議。在形成最後建議的過程中，決策小組成員間彼此相互影響的程度也被納入考量。第三種情境裡，考量朋友之間的友誼會因時間產生變化，因此甄選決策小組的條件增加了時間因素。此外，決策小組發表的各項意見與建議，透過文字處理篩選出評選商品的考量因素，經由人工智慧的工具，做出最後的建議。除此之外，本研究也進行相關實驗以確認各機制之可行性。實驗結果確認本研究所提之機制，與其他決策方法比較後，能提供給消費者較佳的決策支援訊息。 關鍵字：社群網路，決策支援，消費者決策行為，電子商務

(4)

iii

ABSTRACT

Designing Social Commerce Decision Support Mechanisms

Student: Yi-Lin Lee

Advisor: Dr. Yung-Ming Li

Institute of Information Management

National Chiao Tung University

Abstract

With the vigorous development of the social networking sites, many application systems have been developing for the purpose of branding and consumer service. In contrast, researches on consumer purchase decision making is relatively rare. In fact, many consumers collect advices and suggestions from friends as a reference for final decision. In this study, purchase decision support mechanisms were designed to support the operation of social commerce for different scenarios. In the first scenario, the consumer has found several products that meet requirements. For the second scenario, the consumer knows only selection criteria about the item required. In the third scenario, the consumer just wants to buy something, but has no idea about how and what to buy.

A screening mechanism was designed for first scenario to identify appropriate friends as support group, and an improved majority voting mechanism was proposed. For the second scenario, a personalized and socialized recommendation tool was designed. During the consensus-making process, the degree of mutual influence among the members of the decision group was also taken into account. For third scenario, the time factor was included in the decision group screening mechanism. By using part-of-speech processing technique the possible selection criteria were identified, and artificial intelligence methods were used to propose product reference list. In addition, the experimental results confirmed that the proposed mechanisms can provide better support when compared with other benchmark methods.

Keywords: Social Network, Social Commerce, Decision Support, Consumer Behavior

(5)

iv

致謝

鳳凰花開的季節，空氣裡滿是回憶的味道，重拾書本的博士班生活，終於畫下了句點。六年時間彷彿昨日，此刻的我沒有太多情緒起伏，只覺得完成了一個人生中重要的里程碑。這段不算輕鬆的過程，首先要感謝的是指導教授李永銘博士。桃李不言，下自成蹊。老師的言教，讓我知道人師之表；老師的身教，讓我看到經師之典。看著您的身影，如果有一天能執教鞭，我想我已經知道如何成為學生心目中的好老師。博士論文的完成，除了李永銘教授的指導外，也感謝論文指導委員會的清華大學服務科學研究所林福仁教授、中央大學資訊管理學系陳彥良教授、中正大學資訊管理學系古政元教授、交通大學資訊管理研究所羅濟群教授以及劉敦仁教授的提點。因為有您們的指導，讓本篇論文更嚴謹且完整。也謝謝您們的鼓勵，希望有一天我也能追隨您們的腳步，續窺學術殿堂的奧義。博士班的生活中，那些曾經筆硯相親的夥伴，謝謝你們帶給我的一切。人生如渡一重汪洋，我何其幸運，與你們相聚在同一條狹船上。勇劭學長以及治華學長，謝謝你們的協助，讓我的論文能夠順利發表，那些星期三晚上一起度過的日子，都將會是未來每次相聚時聊不完的話題。建邦、嘉豪、敬文，還記得第一年暑假，在空蕩蕩的管二館裡報論文的日子嗎？乃儁，涵文以及子鳳，你們熬夜趕論文那蒼白的臉，我會永遠記得。宗穎以及雅玲，你們兩個勞來與哈台，即使在畢業後還是繼續帶給我歡樂。與我一起畢業的蕙如、宣銘、文翔、郁卉、佳林與詩雯，因為有你們，讓我在最後一年的學生生活，多了很多的回憶。蕙如，看妳順利畢業，我也鬆了一口氣，未來的路希望妳能靠自己的努力走下去。實驗室的清潔股長、勤務兵、設備組、補給官、五百里加急快遞員宣銘，感謝你默默的付出，讓研究室的所有人沒有後顧之憂。文翔提供的各式飲料，適時的緩解了遇到瓶頸時的壓力。沒有財務大臣郁卉，我們可能得喝西北風，妳帶來了故鄉的美味，也捎來鄉愁的滋味。因為佳林妳的祈禱，上帝賜給了我人生中最好的禮物，神會賜福予妳。入學時滑壘成功的詩雯，畢業時也維持妳一貫的風格撲回本壘，回首這段日子，妳跟宣銘帶給我們的笑料，相信會是大家永遠的歡樂。致緯、竣智和潘強三位小朋友，希望我們這群瘋狂的學長姐，沒有嚇著你們。跟我相處時間最久的革命戰友無尾熊政揚，那些年我們一起喝過的酒、聊過的事，留待未來相聚再繼續。還在奮鬥的博士班學弟妹們，繼續堅持下去就會成功。在我背後默默支持我的母親、弟弟一家人以及太太曉菱，沒有你們的支持，我無法在工作與學業兩頭燒的情況下走完這段路程。最後，謹以此篇論文獻給在天上的父親，我做到了!

李易霖

2010/07/01

(6)

v

LIST OF FIGURES

Figure 1.1 A typical consumer purchase behavior ... 2

Figure 1.2 Scenarios of this study and corresponding support mechanisms ... 4

Figure 2.1 Theoretical foundation ... 6

Figure 2.2 Mapping of Simon’s and Kotler’s decision process ... 7

Figure 2.3 QOC representation schema for design rationale ... 12

Figure 3.1 Scenario of social support mechanism ... 14

Figure 3.2 Social support mechanism system framework ... 17

Figure 3.3 Combination of social similarity ... 19

Figure 3.4 Experiment process for social support mechanism ... 25

Figure 3.5 Social network before group formation ... 27

Figure 3.6 The result of decision maker locating ... 28

Figure 3.7 Example of group centrality ... 29

Figure 3.8 Average usefulness level about product ranking ... 30

Figure 3.9 Average usefulness level about store ranking... 30

Figure 4.1 Scenario of social recommendation mechanism ... 36

Figure 4.2 Social recommendation mechanism system framework ... 37

Figure 4.3 Alternatives selection process ... 41

Figure 4.4 Experiment process for social recommendation mechanism ... 44

Figure 4.5 Average stay time for different groups and methods ... 47

Figure 4.6 Average usefulness level for different groups and methods ... 47

Figure 5.1 Scenario of social intelligence mechanism ... 51

Figure 5.2 Social intelligence mechanism system framework ... 52

Figure 5.3 The components of social source credibility ... 54

Figure 5.4 A sample social network for social similarity calculation ... 54

Figure 5.5 Example of the hyponym taxonomy in WordNet ... 58

Figure 5.6 The detail process of product selection criteria discovering module ... 59

Figure 5.7 Experiment process for social intelligence mechanism... 62

Figure 5.8 Precision and similarity calculation process of selection criteria ... 66

Figure 5.9 Similarity comparison of Epinions keyword and extracted criteria ... 67

Figure 5.10 Average stay time for different groups and methods ... 68

(9)

viii

LIST OF TABLES

Table 3.1 MRQAP analysis for friendship and social interactions ... 16

Table 3.2 Symbols used in social support mechanism ... 18

Table 3.3 Example of proposed voting mechanism ... 24

Table 3.4 Facebook personal profile data for analyzing social similarity ... 25

Table 3.5 Characteristics of the three networks ... 26

Table 3.6 Experiment settings of social support mechanism ... 27

Table 3.7 Tests of between-subjects effects for average product usefulness level 31 Table 3.8 Tests of between-subjects effects for average store usefulness level ... 31

Table 3.9 Multiple comparisons of product voting usefulness level ... 32

Table 3.10 Multiple comparisons of store voting usefulness level ... 32

Table 3.11 Example of Kendall’s Tau value calculating ... 33

Table 3.12 Kendall’s



value of system and user ranking similarity ... 34

Table 4.1 Symbols used in social recommendation mechanism ... 38

Table 4.2 Translation of QOC schema ... 41

Table 4.3 Product candidates table ... 42

Table 4.4 Product candidates table after recommendation conflict resolution ... 43

Table 4.5 Characteristics of the three networks ... 44

Table 4.6 Experiment settings of social recommendation mechanism ... 45

Table 4.7 Tests of between-subjects effects for average stay time ... 48

Table 4.8 Tests of between-subjects effects for average usefulness level... 48

Table 4.9 Multiple comparisons of average stay time ... 49

Table 4.10 Multiple comparisons of average usefulness level ... 49

Table 5.1 Symbols used in social intelligence mechanism ... 53

Table 5.2 Characteristics of the three networks ... 62

Table 5.3 Experiment setting of social intelligence mechanism ... 64

Table 5.4 The coefficients of regression models ... 65

Table 5.5 Tests of between-subjects effects for average stay time ... 69

Table 5.6 Tests of between-subjects effects for average usefulness level... 69

Table 5.7 Multiple comparisons of stay time ... 70

Table 5.8 Multiple comparisons of usefulness ... 70

(10)

-1-

CHAPTER 1 INTRODUCTION

INTRODUCTION

1.1 Background

Social networks are the grouping of individuals, and online social network platforms are now one of the most popular online communities. Most online social network services are used for sharing what you’ve done or what you’re doing, but this may not be the only thing they can do. Companies have been devoting their efforts to explore the opportunities of social network over the past years. The increased popularity of social network has opened opportunities for electronic commerce, often referred to as social commerce. Social network not only provides a new platform for pioneers to innovate, but also raise a variety of new research problems for electronic commerce researchers.

There’s more evidence that online social network can be a conduit to social commerce [24]. Social network provide an open platform for social commerce consumers and vendors to search, share and advertise product information. From a survey conducted by Gartner [7], 40% of consumers regularly search products information on social media, 34% are more likely to share product information on social media with their friends than in e-commerce sites, and 77% of online shoppers use reviews. The survey also showed that 75% trust personal recommendations, and 75% are more likely to purchase if a friend endorses. This open up the gate to provide product information meeting consumer’s personal preference based on social relation. At the same time, 81% of consumers receive advice from friends. The result implies that a social support mechanism for product selection would be helpful to consumers. In this study, the phenomena were addressed based on consumer purchase behavior. According to OTX’s purchase intention survey [65], 70% of consumers visit social media websites to collect information on a product. According to the Nielsen Global Online Consumer Survey of over 25,000 Internet consumers from 50 countries in 2009, 90% of consumers trust the opinions of personal acquaintances [18]. IBM’s survey in 2011 found that 50% of 16-64 year olds who use online social networking sites such as Twitter and Facebook admit to using these online social networks to assist with

(11)

-2-

shopping decisions while 35% stated they use online social networks to rank products and services. These respondents believe it is important to be able to use online social networks to assist with buying decisions [75].

1.2 Research Problem

Behind the visible action of making a purchase in social commerce lies a decision process that must be investigated. Consumer behavior involves study of how and why they buy. It blends the elements from psychology, sociology and economics. It also tries to assess the influence on the consumer from groups such as family, friends, reference groups and society in general. Consumer purchase decision is the processes undertaken by consumer in regard to a transaction during the purchase. A typical consumer purchase decision making process is depicted in Figure 1.1.

I need a digital camera, which model should I choose?

After survey, several products are found

Product candidates are identified by comparing various criteria and listening to others’ opinions

Select final decision from product candidates

Share user experience with others

Figure 1.1 A typical consumer purchase behavior

Social commerce is a subset of electronic commerce (EC) that uses social network to supports social interaction, to assist in the online buying and selling of products and services. It includes tools that enable consumers to get advice from trusted individuals, find goods and services and then purchase them. Social commerce helps consumers make smart and savvy purchase, and consumers now are looking for ways to leverage each other’s expertise, understand what they are purchasing, and make more informed and accurate purchase decisions.That is, they are increasingly influenced by online social networks when it comes to purchase decision making. Despite its growing interests, however, there are relatively few studies on social commerce support mechanism. For the purpose of helping consumer with making purchasing decision, it is desired to have proper social commerce support mechanisms based on online social networks. Moreover, as research suggests that customers value and respect personal sources more than other sources [29, 57], it would be ideal to construct decision support groups from their online social networks.

(12)

-3-

In real life, we are constantly influenced by other factors than just information, such as friends, social classes and psychological needs when making purchase decision. A consumer can obtain information from several sources:

 Personal experience: past purchase history, experience of similar products etc.

 Personal sources: family, friends, colleagues etc.

 Commercial sources: advertising, company websites, and salespeople.

In social commerce context, consumers also collect information from these sources. As social relation is the core of social commerce, in this study it was used to be the infrastructure of social commerce support mechanisms.

Consider three common scenarios of product purchasing (see Figure 1.2). Suppose that customer A wants to buy a digital camera,

Scenario 1: He/she has searched for various products, and at this point he is interested in several models. However, he/she is unable to make up his/her mind, so his/her friends or family are consulted to rank the products for him/her.

Scenario 2: He/she has identified price, megapixels and LCD size as selection criteria and needs someone to recommend products based on them.

Scenario 3: He/she has no idea about digital camera and just asks his/her friends or family to tell him/her what factors should be considered and what to buy.

Naturally, the following research problems arise when designing social commerce support mechanisms:

 For scenario 1, how to find adequate group with similar taste or preference so that

consumer can get advice from the group and the group can rank the products for consumer.

 For scenario 2, how to design the recommendation mechanism so as to utilize

friend network to recommend items based on consumer’s selection criteria, that is, how to design a personalized while socialized recommendation results.

 For scenario 3, how to build up functionalities so that consumers can discover

product information based on personal and/or commercial sources.

To address these scenarios and meet the requirements of social commerce, in this study the corresponding mechanisms were designed. For more vivid picture of the study, Figure 1.2 serves as the research paradigm.

(13)

-4- I need a digital camera, which

model should I choose?

Survey products based on selection criteria

After survey, several product candidates are found

Select final decision from product candidates

Share user experience with others

Problem Recognition Information Search Evaluation of Alternatives Purchase Decision Post-purchase Behavior

I need a digital camera, which model should I choose?

Share user experience with others Scena rio 2 Scena rio 2

Social Recommendation Mechanism

Scena

rio

3

Scena

rio 3

Social Intelligence Mechanism

Scena

rio 1

Scena

rio 1

Social Support Mechanism

Selection Criteria Product Candidate List Product Ranking List

Figure 1.2 Scenarios of this study and corresponding support mechanisms

1.3 Research Contributions

While there are on-going researches on social network and its effects on business, there is relatively little solid research on social commerce support. The contributions of this study are listed as follow.

 Social Support Mechanism for scenario 1:

This work designed a mechanism to find the most fit group for consumer, and an adaptive majority voting method was used to rank items.

1. Regarding support group selection, other than selecting group members based

on some predefined measures, in this study the agglomerative hierarchical clustering method was used to find the proper home group for consumer.

2. Concerning improved voting method, social power was used to weight each

voting and suggest the ranking of items.

 Social Recommendation Mechanism for scenario 2:

This study proposed a system framework for personalized (personal preference) recommendation results based on socialized information sources (friend networks)

(14)

-5-

1. For personalized recommendation, an assistant tool based on QOC schema was

designed to recommend proper items based on consumer’s preference. And a recommendation conflict resolution was also proposed to solve the recommendation inconsistency on certain product.

2. As for socialized recommendation, by introducing social impact theory of

social psychology into social network analysis process, a new decision group recruiting method was designed to select friends with higher impact power.

 Social Intelligence Mechanism for scenario 3:

This research suggested a set of functions to collect information from personal or commercial channel based on trustworthy sources.

1. With regard to sources selecting function, source credibility including

friendship, social similarity, network centrality and expertise was used to recruit proper members. A PageRank-like index based on post-reply was proposed to measure expertise on products.

2. As to information collecting function, artificial intelligence techniques were

used to reduce human’s intervention.

1.4 Outline of the Study

The remaining part of this paper is organized as follows. In chapter 2, existing literatures related to this study were reviewed. The corresponding support mechanisms were demonstrated in chapter 3, 4 and 5 respectively. The system framework, experiment and discussions are also included in each chapter. Finally, chapter 6 concludes research contributions and presents future research directions.

1.5 Chapter Summary

In this chapter, the applications of social network in electronic commerce environment were introduced and pointed out the imperious demands of support mechanisms. In addition, the research questions this study tried to address were also highlighted, and the important contributions were also spotlighted in this chapter.

(15)

-6-

CHAPTER 2 LITERATURE REVIEW

LITERATURE REVIEW

In this chapter, a consumer purchase decision support system framework was built based on the following theories. First, consumer purchase decision-making process was studied to understand the decision-making stages. Second, social psychology was investigated to understand the characteristics a friend should have so as to be selected as a reference group member. Third, in order to identify the decision reference group social network analysis was used to analyse the members within social network. The complete theoretical foundation related to this research is shown in Figure 2.1.

Social Media Social Commerce Informatio n T e chnol ogy Soci al Psychol ogy Con sumer Beh avior

Figure 2.1 Theoretical foundation

2.1 Consumer Purchase Decision Making

Human decision making process has been characterized as relatively sequential, and it becomes more complex with distributed source of information and the quantity of information available through networked sources. The way people make decisions varies considerably. Early research has focused on the way people are observed to make decisions and the way in which people should theoretically make decisions. Depending on their methodological foundation, these models can be classified as: descriptive, prescriptive or normative. A simple way of distinguishing between these modes of decision making is [26]:

(16)

-7-

 Descriptive: What people actually do;

 Prescriptive: What people should and can do;

 Normative: What people should do.

From a psychological perspective, it is necessary to examine individual decisions in the context of needs, preferences and values. From a normative perspective, the analysis of individual decisions is concerned with the logic of decision making and rationality. The rationality is ensured if the process of decision making is carried out systematically. As purchasing decisions are often influenced by people who the consumer knows [44], this study focused on what consumers actually do when making purchasing decisions, that is, the descriptive mode was discussed. In consumer decision-making models, Utility theory proposes that consumers make decisions based on the expected outcomes of their decisions. However, in this model consumers are viewed as rational actors who were able to estimate the probabilistic outcomes [83]. As one might expect, consumers are typically not completely rational [69]. In contrast with this view, Simon was interested in the mechanics of the decision-making process [74], in that he considered how a decision maker evaluates all the consequences and compares them with each other. He proposed three principal phases:

 Intelligence: think of the problem and find out what the alternatives to the given

problem;

 Design: determine all the possible consequences of these alternatives;

 Choice: evaluate all the possible consequences.

In the consumer purchase decision-making process proposed by Kotler [45], the consumer passes through five stages: problem recognition, information search, evaluation and selection of alternatives, decision implementation, and post-purchase evaluation. This process is an extension of Simon’s model as three stages are included in Kotler’s model (see Figure 2.2).

Problem Recognition Information Search Evaluation of Alternatives Purchase Decision Post-purchase Behavior

Intelligence Design Choice

Kotler Simon

(17)

-8-

Once consumers perceive a need, they begin to search for information needed to make a purchase decision. The initial search effort often consists of an attempt to recall past experiences. If the internal search does not collect enough information, the external sources are consulted. Empirical study found that consumers relied more heavily on personal sources of information for decisions [29, 57]. After acquiring information during the information search stage, the consumer proceeds to alternative evaluation. All the identified alternatives must be evaluated against some established criteria. These criteria might base on past experiences or the comments of friends. At purchase decision stage, the consumer stops searching for and evaluating information, and make purchase decision. From a consumer-behavior perspective, the products that consumers select can be influenced by their reference groups [5, 17]. Reference groups are people to whom an individual looks as a basis for self-appraisal or as a source of personal standards, and they have important influence on the purchase behavior. As dual process theory suggests, reference groups can be divided into normative and informational [25]. The former one is based on the desire to conform to the expectations of others, and the later one is based on the acceptance of information from others [40]. Essentially, the personal source is individual’s online social network because it is constructed based on friends. Besides, online social network can be normative as well as reference group as friends can not only provide information but also influence each other.

2.2 Social Influence, Social Impact and Social Choice

In social network, social psychology, communication and information technology are essential in building meaningful relationships and influencing behavior. Today, the area of social commerce has been expanded to include the range of social network tools. Examples of these tools include consumer ratings and reviews, user recommendations and referrals, forums and communities, social network optimization and social applications. As the fast development of internet, together with the booming of online social network, it is much easier to collect information from personal sources. Many consumers are getting used to make decisions based on comments collected from their own online social networks. While conventional decision support system has been extensively investigated, little specific mechanism on social commerce is developed. For the purpose of helping consumer with making purchasing decision, it is desired to have proper social commerce support mechanism based on online social networks. Moreover, as research suggests that consumers value and respect personal sources

(18)

-9-

more than other sources [29, 57], it would be ideal to construct decision support groups from their online social networks.

In real-world decision-making process, human can experience emotional intensity and information overload that may affect their choices. Better decision support system should address these issues and assist human decision making by developing systems that integrate capabilities from human and computational intelligence. Social influence is the process by which individuals make real changes to their feelings and behavior as a result of interaction with others who are perceived to be similar, desirable, or expert [51, 68]. Social influence does not necessarily require face-to-face interaction, but is based on information about other people [70]. Social impact theory is widely cited in the research literature in social psychology, it provides a useful framework for understanding how a person is affected by social environment [61].

Social impact theory states that social influence is proportional to a multiplicative function of the strength, immediacy and number of sources [49]:

 Strength: the importance of the reference group to the individual.

 Immediacy: the closeness of the influencing group to the individual (in space and

time) at the time of the influence attempt.

 Number: how many people there are in the reference group.

Research on social inﬂuence demonstrates that one’s attitude and judgment tend to conform to those held by the majority of others [59]. Conformity can be due to either social pressure or one’s belief that the majority is likely to be correct [25]. When a large portion of a reference group holds a particular attitude, it is likely that the individual will adopt it as well [68]. Social choice theory is concerned with relationships between individuals' preferences and social choice [28, 73], and decision making and social choice theory are strong connected [4, 15]. The method of majority decisions has been widely discussed in the context of social choice theory. Voting-based procedures are entirely natural for some kinds of social choice problems [72]. Research on consumer decision involving multi-attribute options provides empirical evidence for use of the majority rule [71, 86]. A weighted voting system is one in which the preferences of some voters carry more weight than the preferences of other voters. However, in most of the social choice literature, all voters are treated equally. In fact, some voters are more important than others.

(19)

-10-

2.3 Social Network Analysis

An online social network is a social structure made of people who are tied by one or more specific types of interdependency. Research on online social network has captured the effect of social influence on consumers’ purchase decisions across a variety of context [6, 38, 55]. Online social network analysis (SNA) refers to techniques used to analyse online social networks. Online social network can be analysed in node level and dyadic properties. The most popular metrics used are degree, betweenness and closeness centrality [31]. Degree centrality can be used to see if someone in an online social network is involved in large number of interactions. Betweenness centrality is a metric to verify if an individual is an important node who lies on a high proportion of paths between others. A user with higher betweenness centrality is often considered as an opinion leader [31], and a higher closeness centrality indicates that a user is highly related to all others [64]. At the dyadic level the two properties are dyadic cohesion and equivalence [9, 10]. Dyadic cohesion describes to the social closeness of a pair of nodes. Equivalence refers to the extent to which pairs of nodes is similar.

Social impact theory suggests that social status, power and credibility can impact on decision [50]. Social status can be the proxy to estimate strength [62]. In-degree centrality, betweenness centrality [30, 31], and Bonacich power centrality can be used to measure social status [8, 30, 62]. Moreover, a member with high cognitive centrality would acquire pivotal power in a group and exert more influence on decision making [39]. In social impact theory, immediacy is used to describe group structure. Group structure can be treated as a pattern of immediacies between group members, and immediacies is the distances between individuals [61]. Furthermore, closeness may increase the power of social influence by making a source of influence more immediate [49, 61], hence closeness centrality can be used as the proxy of immediacy.

The studies of social network have examined a diverse set of properties, and these properties are classified as relational properties and structural properties [76]. Relational properties focus on the content of the relationship between network members and on the form of these relationships, while structural properties describe the way members fit together to form social networks. Human relationships are maintained, renewed, or deteriorate over time [77], but time factor is missing from the above properties.

(20)

-11-

2.4 Multiple Regression Quadratic Assignment Procedure

Some data sets contain observations corresponding to pairs of entities (e.g., friends), and these data are not independent. The multiple regression quadratic assignment procedure (MRQAP) is commonly used in social network analysis. MRQAP is a nonparametric statistical algorithm regressing a dependent matrix on one or several independent matrixes. It is a standard technique to analyse social network data and to discover behavioral characteristics of friendship [85]. Therefore network regression measures are the most appropriate statistical method for testing them. However, these data are not independent and do not satisfy the assumptions of ordinary least squares regression, therefore requiring the use of the multiple regression quadratic assignment procedure (MRQAP) to test social network data [9].

MRQAP has been widely used in social network related research [27, 37, 48, 84]. However, in the development of social network applications, to my knowledge little effort has been devoted to test if the data collected from online social networks can be used to maintain online relationship. For example, some social network-based recommendation systems used interaction data such as comment, share, interests in common to measure online relationship [52, 53], but they are not empirically examined. To make this research more solid, this method was introduced to test if the interaction data on online social network can really reflection social relation.

2.5 Design Rationale and Representation Schema

Due to the complexity of decision problem and communication process between decision-making group members, there is a strong need in formatting the solution design process to help members record, access and assess design rationale. Design rationale is used to provide information about why certain decisions were made. A clear design rationale provides a medium for communicating between decision group members. A design rationale is an important tool in arriving at the initial decision alternatives in the first place, and a representation is needed for capturing design rationale. A good representation schema is vital to enabling effective design and discuss. A representation schema explicitly documents the reasoning and argumentation occurring in design. It determines the methods used to capture and retrieve the design rationale.

(21)

-12-

One design rationale representation schema known as Questions, Options and Criteria (QOC) developed by McLean et al. [54]. It focuses on three basic concepts indicated in its name. QOC represents the design space using three components:

 Questions(Requirements): identify key issues for structuring the space of

alternatives

 Options(Alternatives): provide possible answers to the questions

 Criteria: provide the bases for evaluating and choosing from among the options.

A design rationale presented by QOC schema is depicted in Figure 2.3.

Option Question Criteria Option Criteria Criteria Criteria Option Question Option Criteria Question Option

Link between Question and responding Option Link between Option and consequent Question Positive Assessment

Positive Assessment

Figure 2.3 QOC representation schema for design rationale

2.6 Fuzzy Analytic Hierarchy Process and Fuzzy Delphi

The Delphi method is a group decision making technique. Murry et al. integrated the concept of traditional Delphi method and fuzzy theory to improve the vagueness of the Delphi method [58]. Fuzzy Delphi is a good method for group decision to solve the fuzziness of common understanding of experts’ comments [60]. Analytic hierarchy process (AHP) is a structured, multi-criteria decision-making approach, and it is widely used for dealing with quantifiable and intangible criteria that can be applied to numerous areas such as decision theory [82]. Hsu and Chen [34] proposed a fuzzy similarity aggregation method, in which similarities between experts were collated and fuzzy numbers were assigned directly to each expert to determine the agreement degree between them. Fuzzy AHP (FAHP) translates the viewpoints of experts from definite values into fuzzy numbers and membership functions, and presents triangular fuzzy numbers in paired comparison of matrices to develop FAHP. Consequently, the

(22)

-13-

comments of experts approach human thinking model, so as to achieve more reasonable evaluation criteria.

2.7 Chapter Summary

The primary objectives of this chapter were to provide a brief overview of related theories used in this study. By investigating Simon’s human decision process and Kotler’s consumer purchase decision-making process, their relationship was mapped. Furthermore, the key role of reference group in purchasing behavior was also pinpointed. Social impact, social influence and social choice drawn from social psychology were reviewed to understand how people influence each other within social context. For the purpose of representing the relation between selection criteria and possible candidates, design rationale representation scheme was also covered in literature review. Last but not the least, information technologies used in this study were also discussed- namely, social network analysis, Fuzzy Delphi and Fuzzy AHP.

(23)

-14-

CHAPTER 3 SOCIAL SUPPORT MECHANISM

SOCIAL SUPPORT MECHANISM

In our daily life, we usually make purchasing decisions. Some can be easily made because we are familiar with the items we need, while others may be much more complex. In this chapter, a support mechanism is presented to help consumers with ranking desired products.

3.1 Scenario of Social Support Mechanism

For the consumer who is experienced in the product he/she wants to buy, there may have been a candidate list in mind after surveying available items. However, consumer may hesitate about which one to buy. Under the circumstances, advice from friends can be an important reference. In this chapter, a social support mechanism was designed to deal with this requirement. An illustration of this mechanism is depicted in Figure 3.1. Output Product Ranking Process Social Support Input

Product Candidate List

1 2

3 4

5 6

Figure 3.1 Scenario of social support mechanism

3.2 Empirical Analysis of Online Social Network Interaction

Basically, two types of data can be collected in online social network sites: social network structure characteristics and social interaction. Online social network analysis (SNA) is the measuring of relationships and flows between people. The nodes in the network are the people while the links show relationships or information flows. In summary, SNA provides a mathematical analysis of human relationships objectively.

(24)

-15-

In contrast, social interaction is more subjective. In famous social network sites such as Facebook, we can see mutual interactions like: comment, like, photo tag, share and join activity. Before proceeding further, it would be ideal to make sure these subjective data collected from online social network sites can be used to identify proper reference groups for the purpose of decision support. More precisely, can these data be used to be the proxies of identifying influential friends?

One of the issues facing researchers who analyse online social networks is that standard statistical tests may be inappropriate [42]. As social network interaction data do not satisfy assumptions of statistical inference in classical regression because the observations are not independent. Consequently, multiple regression quadratic assignment procedure (MRQAP) was used to run the multiple regressions [9, 46]. MRQAP tests are permutation tests for data organized in square matrices of relationships. Such a data structure is typical in social network studies, where variables indicate some type of relation between a given set of friends [23].

The social relationships index (SRI) was developed as a self-report version of the social support interview [11, 12, 66], and this scale has demonstrated good test–retest reliability and internal consistency [79]. It was designed to examine positivity and negativity in social relationships. Besides, the SRI can be used to assess specific individuals within one's social network and provides a summary within relationship categories. In this research, SRI was used to assess if the friends are supportive. Meanwhile, MRQAP was used to test the relationship between interactions and the usefulness of friends.

For the SRI, participants were instructed to rate how helpful they feel their friends are in a decision support context (i.e., when they need advice, understanding, or suggestion; 1 = not at all, 10 = very much). Thus, three items were used to measure friendship

positivity [12]. At the end of this process, a n n SRI matrix was constructed as the

output. Next, the social interaction data including activity, comment, share, photo tag

and like was collected, and five n n matrices were constructed. Then the MRQAP

was executed to test the relations between the friendship and social interactions. Five models were tested during MRQAP, and the result is shown in Table 3.1.

(25)

-16-

Table 3.1 MRQAP analysis for friendship and social interactions

MODEL 1 MODEL 2 MODEL 3 MODEL 4 MODEL 5

STANDARDIZED COEFFICIENT Comment 0.797967* 0.374767* 0.370484* 0.360939* 0.356697* Like 0.601515* 0.593344* 0.580499* 0.574102* Share 0.048184* 0.046932* 0.046455* Photo tag 0.065572* 0.064567* Activity 0.043365* R2 0.637 0.819 0.822 0.826 0.827 Adjusted R2 0.637 0.819 0.822 0.825 0.827 N=11881 *p<0.001

Table 3.1 shows that these social interactions are positively and significantly related to social relation, so in the system they were all included in the calculation of interaction.

3.3 System Framework

Based on consumer purchase decision-making process, the proposed system supports a consumer with necessary functions in purchase decision stage. The requirements for this system were governed by the objective of designing a system to support product purchasing decision processes on online social network. For more vivid picture of the study, Figure 3.2 serves as the research paradigm, and the symbols used in the proposed mechanism are listed in Table 3.2. In the social network analysis module, the centralities of each individual are calculated, and the similarity is measured in social similarity analysis. These data are processed in social influence analysis module to discover the influential friends, and the reference group (hereinafter referred to as “decision group”) is constructed. And then the final selections are the output of alternatives selecting module. In the following, the important system modules are described in detail.

(26)

-17-

(27)

-18-

Table 3.2 Symbols used in social support mechanism

SYMBOL DESCRIPTION

( , )

a i j Adjacency of user i and j, a i j( , ) 1 if they are connected

( , )

D

SS i j Directly connected friend similarity between user i and j

( , )

ID

SS i j Indirectly connected friend similarity between user i and j

( , )

P

SS i j Personal profile similarity between user i and j

( , )

SA i j Social interaction between user i and j

( , )

SS i j Social similarity between user i andj

( )

PI i Number of profile items used to describe the characteristics of i

( , )

DI i G Degree of interaction between user i and group G ( , )

p

N i j Number of post between user i andj

( , ) R

N i j Number of reply between user i andj

( , ) T

N i j Number of photo tag between user i andj

N Number of users in a specific social network or group

( , )

d i j Social distance between user i and j

( )

v i Voting of user i

( )

w i Voting power of user i

GC Group centrality

3.3.1 Home Group Locating

3.3.1.1 Group Formation

In this module, the social network members were divided into groups, and the degree of activity and similarity were used to decide which group the consumer (decision maker) is belonged to. The primary goal of cluster analysis is to classify objects into categories, and the resulting clusters should show high internal homogeneity and external heterogeneity. In social network, friends in a clique are likely to share some characteristics and interests. So in this study, agglomerative hierarchical clustering was used to divide social network members into different groups. Centroid method, average linkage complete linkage, single linkage and Ward method are commonly used in hierarchical clustering. In this module, Ward method was used to perform

(28)

-19-

clustering, and Sum of Square Error (SSE) was used to measure the effectiveness of clustering. The SSE is formulated as:

2 1 1 ( ) C n i j j i j SSE z x   



 (3.1)

, where C is the number of groups after clustering process, and C is the members i

in a group, x is a certain member in a group, z is the jth attribute/characteristic of i_j

centroid member in a group, and x_j is the jth attribute/characteristic of x.

3.3.1.2 Isolated Group Members Detection and Deletion

To avoid the problem of being influenced by the isolated members in a group, some detection and deletion methods would be necessary. As a group is supposed to be homogeneous after the clustering process, so the similarity was used to detect the isolated members within certain group. The correlation between social similarity and inﬂuence has been studied and confirmed [20]. The more similar an online friend is in personal characteristics, the stronger is the social tie. In this research, an index was defined to measure social similarity, as shown in Figure 3.3.

Social Similarity

Friends in common

Characteristics in common

Directly connected friends

Indirectly connected friends

Personal profiles

Mutual interactions

Figure 3.3 Combination of social similarity

The similarity index is composed of four parts; the first one is: directly connected friends in common. It is reasonable to say that two individuals have something in common if they share many of the same friends. In this research, the number of directly connected friends in common was used to measure social similarity. Jaccard index was used here: ( ) ( ) ( , ) , ( ) ( ) D F i F j SS i j F i F j    (3.2)

(29)

-20-

where F i is the friends of i. For the second part: indirectly connected friends, ( ) i is

similar to j if ihas a friend k that is similar to j . Thus, the indirect connection

friend similarity is:

( , ) ( , ) ( , ),

ID ID

k

SS i j 



a i k SS k j (3.3)

wherea i k( , )is an element of the adjacency matrix of the social network. a i k( , )1 if

there is a direct connection between i and j, 0 otherwise.

The third part of social similarity is characteristics in common. Personal profile similarity was defined by the number of profile data items shared. That is,

0 if ( ) or ( ) 0, ( , ) ( ) ( ) otherwise, ( ( ), ( )) P P i P j SS i j P i P j min P i P j       (3.4)

where P i( )is the number of profile items used to describe the characteristics of i, and

( ) ( )

P i P j is the number of items both i and j have the same profile value. The

fourth part is social interaction. The social interaction is defined as:

1 1 0 if ( , ) or ( , ) is 0, ( , ) _{( , )} otherwise, ( , ) n k n k I i k I i j SA i j _{I i j} I i k          



(3.5)

where I i j( , ) is the total number of interactions between i and j. By combining all

the above similarities together, the similarity index is defined as:

( , ) _D( , ) _ID( , ) _P( , ) ( , ).

SS i j SS i j SS i j SS i j SA i j (3.6)

Those who have low similarity were excluded in the group.

3.3.1.3 Decision Maker Locating

To find which group a consumer belongs to, it is required to identify his main group. In this study, degree of interaction was defined to figure out the level of interaction

(30)

-21-

for each user. If a consumer is more active in a certain group, it is likely to say that he should be identified as the member of that group. Degree of interaction is defined as:

1 ( , ) ( , ) ( , ) ( , ) , N P R T j N i j N i j N i j DI i G N    



(3.7)

where G is a specific group and N is the number of group members. N i j ,P( , )

( , )

R

N i j , andN i j are the number of post written by decision maker, reply on _T( , ) posts and tags on photo respectively. At the end of this process, the decision maker

will be located in a certain group called home group (HG).

3.3.2 Social Power Calculating

3.3.2.1 Social Network Analysis

The purpose of this module was to collect data related to strength in social impact theory. People's brains are more responsive to friends than to strangers, even if the stranger has more in common [47]. There are psychological and evolutionary arguments for the idea that the social factors of ‘similarity’ and ‘closeness’ could get privileged treatment in the brain. However, a study suggests that social closeness is the primary factor, rather than social similarity, as previously assumed [47]. Measuring the network position is finding the centrality of an individual. These measures give us insight into the various roles and groupings within an online social network. Since SNA was introduced to analyse complex networks [16], in the proposed model three commonly used centrality metrics, i.e., closeness, betweenness and degree centrality were chose to be decision group selection factors.

Closeness is used to measure the immediacy in social impact [68]. It is defined as the total distance of a user from all other users, and can be formulated as [30]:

1 1 ( ) , ( , ) C N j C i d i j  



(3.8)

where N is the number of users and d i j( , )is the distance between decision maker i

(31)

-22-

greater power in the network [43]. Betweenness centrality tracks the number of geodesic paths through the entire social network, and it is an approximation of influence [16]. Besides, betweenness centrality best measures which members, in a set of members, are viewed most frequently as a leader, than other social network analysis measures [5]. The betweenness centrality is defined as [30]:

( , , ) ( ) , , ( , ) B j k g j i k C i i j k g j k  



  (3.9)

where g j k( , )is the number of geodesic paths from j to k, and g j i k( , , )is the number

of these geodesics that pass through node i.

Degree refers to the attribute that can present an initiative action from a user. The higher the number of degree, the more motivation a user has to interact with others. When a target user posts comments or sends links to others, they make links of this type. Degree centrality is defined as [30]:

1 ( ) ( , ), N D j C i a i j  



(3.10)

where a i j( , )1 if and only if i and j are connected. Otherwise, a i j( , )0.

3.3.2.2 Social Influence Analysis

As social impact theory states, social influence is a function of strength, immediacy and number of influencing source. In this research, follow the similar idea, the power of social influence for individuals within the decision group was formulated as:

,

I C

 

S (3.11)

where I [I_i₁]is a n1 matrix describing the value of social influence, S [SS i j( , )]

is a n n matrix for describing similarity between group members. The centralities

(betweenness, closeness and degree) of all the users in a social network can be

represented by centrality C , where C is a n3 matrix of the above three centralities.

(32)

-23-

Both



and  can be predefined or calculated from social network. To demonstrate

both scenarios, in this research



is predefined as:

1 3 1 3 , 1 3             (3.12)

and  is derived from social network.  [ ( , )] i j is defined as:

0 if there is no path from to ,

( , ) 1 otherwise. ( , ) i j i j d i j       (3.13)

After the social influence is calculated, it will be used in the product candidates choosing module as the weight of voting.

3.3.3 Product Candidates Choosing

3.3.3.1 Social Influence Voting

In a conventional majority voting system, people are treated equally. For example, a home group with N members votes on whether to recommend a certain product p.

Assume the sum of total voting is 1, and every member can vote v i( ) (agree/disagree)

with(1, 0) . Therefore, the product will be recommended if:

1 ( ) 0.5. N i v i N  



(3.14)

In this study, the voting mechanism was improved by introducing social influence as weight of voting.

The consensus weight of the ith member depends on the influence of the member’s strength relative to other members of the group. The stronger that member’s position is to other members’ positions, the more weight that member is given in defining the group consensus. Thus, the normalized voting power of the ith member is defined by:

(33)

-24- 1 1 1 ( ) i , N j j I w i I  



(3.15)

so the product will be recommended if:

1 ( ) ( ) 0.5. N i v i w i  



(3.16)

Take Table 3.3 for example. In the conventional majority voting, P1 will be recommended and P2 will not. By introducing social influence into voting mechanism, decision group won’t recommend P1, and P2 will be their choice.

Table 3.3 Example of proposed voting mechanism

M1 M2 M3 M4 M5 RESULT RECOMMEND?

Voting on product P1( ( ))v i 0 1 0 1 1 3 Y

Majority voting 0.2 0.2 0.2 0.2 0.2 0.6 Y

Social influence voting( ( ))w i 0.3 0.2 0.2 0.2 0.1 0.5 N

Voting on product P2( ( ))v i 1 0 0 0 1 2 N

Majority voting 0.2 0.2 0.2 0.2 0.2 0.4 N

Social influence voting( ( ))w i 0.4 0.1 0.1 0.2 0.2 0.6 Y

3.4 Experiment

3.4.1 Experiment Process

To further prove the feasibility of this design, an empirical study alone with system development was conducted. The procedures of experiment are described as Figure 3.4. To implement this system, one of the most popular social network sites Facebook was selected to be experiment platform to collect required data. To register a Facebook account, a user must provide the profile information (see Table 3.4). In this experiment, these data were collected to be master database for personal profile. A snowball sampling procedure with S stages K names is defined as follows. A random sample of individuals is drawn from a given population. Each one in the sample is asked to name

K different persons, where K is a predefined number. For example, each person is asked

to name K best friends. The persons who were not in the random sample but were named by individuals in it form the first stage. Each of the individuals in the first stage

(34)

-25-

is then asked to name K different persons. This procedure repeats S times to complete the sampling process.

Start

Snowball sampling to invite participants Select decision makers

from participants

Retrieve personal profile data from Facebook

Perform group clustering and find proper group for decision maker (Home Group Locating) Analyze social power of

decision group members (Social Power Calculating) Vote on decision problem

(Product Candidates Choosing)

Decision makers evaluation End

Figure 3.4 Experiment process for social support mechanism

In the initial stage, by using snowball sampling 18 Facebook users were drawn randomly and divided them into groups by their lifestyle. After filtering out those users who were not willing to join the experiment, three groups were identified: student group, office worker and random member (those who cannot be classified into student or office worker) groups. By using 3 (S) stages 3 (K) names snowball sampling 120 participants were included in each group. That is, a specific network was formed by a continuous node expending process until a predefined maximum distance of connections (i.e. 3 hops in the experiment) was reached. After filtering the people not interested in this experiment, finally each group had 60 unique participants. Of all the 180 participants, 50 users were randomly selected to collect data required for SRI survey. The average year of Facebook usage was 2.1 years and the average number of

Table 3.4 Facebook personal profile data for analyzing social similarity

PROFILE CATEGORY DATA

Basic Sex/Home town/Country

Personal Activities/Interests

Education College/High school

Work Employer

(35)

-26-

friends was 216. The characteristics of these social networks are summarized in Table 3.5.

Table 3.5 Characteristics of the three networks

ATTRIBUTES SOCIAL NETWORKS

STUDENT OFFICE WORKER RANDOM MEMBER

Number of participants 36 36 36 Age 20~32 26~45 22~35 Gender Male: 56% Female: 44% Male: 47% Female: 53% Male: 64% Female: 36%

Average betweenness centrality 27.901 37.103 36.221

Average closeness centrality 31.919 35.956 39.440

Average degree 1.781 3.136 2.742

In the experiment, 18 randomly selected users (6 from each group) were invited to be decision makers. They can issue a decision problem and evaluate the effectiveness of decision criteria and alternatives. The decision makers were asked to issue 2 decision problems during the experiment, what to buy and where to buy. A desire product together with possible alternatives (what to buy) and shopping stores was listed (where to buy), and these problems were delivered to the decision group members selected by this system.

When the decision problem was presented to decision group, members were asked to vote on the candidate products together with stores to buy them. Follow the method proposed in this research, the suggested alternatives were collected and presented to decision makers. After reviewing the suggested alternatives, the decision makers were asked to rate how much they were satisfied. A 5-point Likert Scale was used for each alternatives presented to decision makers: Very Useful, Useful, Neither Useful nor Useless, Useless and Very Useless by a rating score of 5, 4, 3, 2 and 1. The experiments were conducted for each of the three group 3 times, and each experiment lasted for one week. The related settings of this experiment are listed in Table 3.6, and the result of group formation process is depicted in Figure 3.5 and Figure 3.6.

(36)

-27-

Table 3.6 Experiment settings of social support mechanism

ITEM SETTING

Type of support Product candidate list ranking

Participant sampling Snowball sampling

No. of participants Student: 60 Office worker:60 Random member: 60 No. of requestors Student: 6 (out of 60) Office worker:6 (out of 60) Random member: 6 (out of 60)

No. of candidate list 2 for each support requestor

Digital camera/mobile phone/notebook

Benchmark method

Random: rank products in candidate list randomly

Social network analysis: select support group members by SNA Group centrality: select support group by group centrality

Evaluation method

Clickstream: browsing time on the product pages provided

Perceived helpfulness: questionnaire survey by 5-point Likert scale

Ranking comparison (Kendall's  ): order of requestors and system

(37)

-28-

Figure 3.6 The result of decision maker locating

3.4.2 Benchmark Methods

To compare the proposed mechanism with others, three methods were selected as benchmark.

 Random: this method was used as baseline benchmark. All the experiment process

was the same as proposed mechanism except the alternatives and stores were randomly selected from the desire list of decision makers.

 Social network analysis (SNA): the decision group members were selected by

considering centrality only. That is, I 



C.

 Group centrality method (GCM): to further compare the influence of different

decision support group, a group centrality measure was designed to be another method for selecting decision group from clustering results. Suppose DG is a

decision group, and V is the complete social network. Assume N DG is the ( )

number of members who are not in DG but connect with members in DG , and

社群商務決策支援機制之設計

國 立 交 通 大 學

資訊管理研究所

博 士 論 文

社 群 商 務 決 策 支 援 機 制 之 設 計

Designing Social Commerce Decision Support Mechanisms

研 究 生： 李易霖

指導教授： 李永銘 博士

社 群 商 務 決 策 支 援 機 制 之 設 計

Designing Social Commerce Decision Support Mechanisms

中文摘要

社群商務決策支援機制之設計

研 究 生： 李易霖

指導教授： 李永銘博士

國立交通大學資訊管理研究所博士班

摘要

ABSTRACT

Designing Social Commerce Decision Support Mechanisms

Student: Yi-Lin Lee

Advisor: Dr. Yung-Ming Li

Institute of Information Management

National Chiao Tung University

Abstract

致謝

李易霖

TABLE OF CONTENTS

LIST OF FIGURES

LIST OF TABLES



CHAPTER 1

INTRODUCTION

INTRODUCTION

1.1 Background

1.2 Research Problem

1.3 Research Contributions

1.4 Outline of the Study

1.5 Chapter Summary

CHAPTER 2

LITERATURE REVIEW

LITERATURE REVIEW

2.1 Consumer Purchase Decision Making

2.2 Social Influence, Social Impact and Social Choice

2.3 Social Network Analysis

2.4 Multiple Regression Quadratic Assignment Procedure

2.5 Design Rationale and Representation Schema

2.6 Fuzzy Analytic Hierarchy Process and Fuzzy Delphi

2.7 Chapter Summary

CHAPTER 3

SOCIAL SUPPORT MECHANISM

SOCIAL SUPPORT MECHANISM

3.1 Scenario of Social Support Mechanism

3.2 Empirical Analysis of Online Social Network Interaction

3.3 System Framework

















 











3.4 Experiment



國立交通大學

博士論文

社群商務決策支援機制之設計

研究生：李易霖

指導教授：李永銘博士

社群商務決策支援機制之設計

研究生：李易霖

指導教授：李永銘博士