CHAPTER 4 EXPERIMENTS
4.2 Data Collection and preprocessing
Data collection includes two parts: group discussion messages collection and social information collection.
In the part of group discussion message collection, our experiments have three scenarios mentioned earlier. Then system will suggest some option criteria to support discussion. In the food scenario, group members will get three restaurant options to discuss, such as McDonald's, KFC and Pizza Hut. In the travel scenario, group members need to discuss with
39
three option about sight. In the shopping scenario, group members discuss what kind of group product they want to purchasing. In the experiments, we collect 37 Facebook Club and there are 184 Facebook Club participants expressing comments on the options. The data was gathered from 2014/03/30 to 2014/04/15.
In social information collection part, we collected the social information of each group member, such as their common friends, common Facebook Club, and their liked fans pages.
In the real world, some people care about information security, so they locked their information if you are not their friend. Some of social information data can not completely be collected and we eliminate the incomplete data. After data cleanness, there are 33 groups and 166 participants’ data we can use. The dataset summary before data cleanness is shown in Table 6. The dataset summary after data cleanness is shown in Table 7.
Table 6. The Dataset Summary before Data Cleaning
Title Value
Duration of Experiment 2014/03/30 to 2014/04/15 The Number of Participants 184 participants
The Number of Groups 37 groups
Table 7. The Dataset Summary after Data Cleaning
Title Value
Duration of Experiment 2014/03/30 to 2014/04/15 The Number of Participants 166 participants
The Number of Groups 33 groups
40
In this research experiment, we further analyze 166 participant’s information. Their gender distribution and age distribution are shown in Figure 12.
Figure 12. Profile of Participants Social Interaction) need to compute. In social similarity, we get participants mutual friends and groups to calculate their group each person’s social similarity. In social interaction, we get participants’ interactions on Facebook between the group members. After normalizing both social influence scores, we aggregate these two scores and normalize it again to gain final social influence score.
Participant Expertise Computing: we analyze the information that members clicked
“like” button of the fans pages with the same tools (Facebook API and Facebook Query Language). We find fans pages about eating, purchasing, travel; each commerce category has 50 fans pages. Then we calculate participant expertise score with respect to every decision category and normalize it to gain final participant expertise score.
60%
41
Group Discussion Message Computing: according to the adjectives included the discussion messages, we can calculate the options scores by each discussion group members.
So the same option might receive different scores by different persons. Finally we can get the options scores with respect to different members.
4.4 Weight Generation
In this part, we need to decide the factor weight in different scenarios after computing scores steps. Adapting ANP model can find every factor weight, so in order to generate weight scores, we build a pairwise comparison matrix model by using questionnaires which can let us to set correlation importance between each criteria. Table 8 is weight setting questionnaire between each factor in our mechanism.
Table 8. Factor Weight Setting Questionnaire
Question When you make a purchasing decision with a group which factor is more important?
Very Important Important Equal Important Very Important
Expertise members. The AHP comparison table shown in Figure 13.
42
Figure 13. APH Comparison Table
4.4.1 Factor Weighting Determination
In order to get the factors weighting scores, we use AHP method to generate it. We let each participants determine the weight of each factor in different scenario by asking them the ratio between each two factors and preform pairwise comparison. Finally we calculate the average weight of each component and show the result in Table 9.
Table 9. Each Factor Weight in Different Scenario
Friends Opinion Expertise Opinion Discussion Information
Food 0.687 0.17 0.143
Shopping 0.322 0.313 0.365
Travel 0.258 0.32 0.422
43
CHAPTER 5 RESULTS AND EVALUATIONS
In this section, we have two methods to evaluate and discuss the experiment performance of the proposed mechanism. First, we evaluate the group members who will buy the products recommended by our mechanism. Second we ask group members’ evaluation on the satisfaction of the recommended ranking list.
5.1 Hit Ratio
In the experiment, we evaluate the group member who will buy the products recommended by our mechanism. If the group discussion members feels satisfied with and the social support mechanism also recommends purchasing it. That is to say, we will evaluate our mechanism performance by comparing whether the decision made by the group members matches the first recommending option created by our proposed mechanism. A hit ratio means correct social decision is made.
# '
Where #ofOptionRecommendToUser stands for the set of products recommended for purchasing. #ofOptionThatHitTheUser sSelection' stands for the set of satisfactory products group member purchased.
5.2 Factor Weighting Determination
In order to determine the weighting approach that brings better performance to the recommendation, we evaluate the weight of each factor by two different approaches: (1) equally weighting approach, (2) group weighting approach. Equally weighting approach assigns the weight equally as 33% for each factor, group weighting approach assigns the weight based on average weight of the each group member. Figure 14 is performance of
44
different weighting approaches, and Table 10 is statistical verification results of weighting approaches.
Figure 14. Performance of Different Weighting Approaches Table 10. Statistical Verification Results of Weighting Approaches
Paired Group Mean Std Dev t-Value Sig(2-tailed) Group
Weight
Equal Weight
0.0384 0.01648 7.71 0
As shown in the figure, because the average weight decided by the group members, so the performance of group weighting approach is better than equal weighting approach. So we utilize group weight approach to decide each factor weighting by the scores that we calculate in chapter 4.
5.3 Performance of Recommendation Factors
We compare three factors, social influence, and participant expertise and discussion message with different combinations in different scenarios (food, travel and shopping).
Figure 15 is the average of accuracy including all scenarios. As shown in the figure, we can
food travel shopping
45
find our proposed mechanism is higher than other six recommend approaches.
Figure 15. The Average of Accuracy Including All Scenarios
Figure 16 is accuracy of food scenario. As shown in the figure, the model considering social influence will perform better than the other model. And our proposed mechanism have better performance than others.
Figure 16. Accuracy of Food Scenario
Figure 17 is accuracy of travel scenario. As shown in the figure, the model considering group discussion message will perform better than the other model. And our proposed mechanism have better performance than others.
0.25
SI PP DM SI+PP DM+PP SI+DM SI+PP+DM
Accuracy
SI PE DM SI+PE DM+PE SI+DM SI+PE+DM
Food Scenario
46
Figure 17. Accuracy of Travel Scenario
Figure 18 is accuracy of shopping scenario. As shown in the figure, the model considering participant expertise will perform better than the other model. And our proposed mechanism have better performance than others.
Figure 18. Accuracy of Shopping Scenario
Furthermore, we use a statistic method- the paired sample t-test in 95% significant level, the all the pair test is significant under 0.05. In other words, our method is the best compared with others. The Table 11, 12, 13 is shown as statistical verification of the similarity.
0.25
SI PE DM SI+PE DM+PE SI+DM SI+PE+DM
Travel Scenario
SI PE DM SI+PE DM+PE SI+DM SI+PE+DM
Shopping Scenario
47
Table 11. Statistical Verification of the Accuracy in Food Scenario Paired Group Mean Std Dev t-Value Sig(2-tailed)
SI+PP+DM SI+DM 0.17785 0.15159 15.1116 0
DM+PP 0.22129 0.18509 15.404 0
SI+PP 0.16466 0.16355 12.972 0
DM 0.09877 0.19282 6.6 0
PP 0.10721 0.19806 6.974 0
SI 0.13354 0.18072 9.521 0
Table 12. Statistical Verification of the Accuracy in Travel Scenario Paired Group Mean Std Dev t-Value Sig(2-tailed)
SI+PP+DM SI+DM 0.25665 0.17145 19.286 0
DM+PP 0.22684 0.17012 17.180 0
SI+PP 0.26622 0.18084 18.967 0
DM 0.16049 0.17753 11.648 0
PP 0.15293 0.17872 11.025 0
SI 0.08943 0.15851 7.270 0
Table 13. Statistical Verification of the Accuracy in Shopping Scenario Paired Group Mean Std Dev t-Value Sig(2-tailed)
SI+PP+DM SI+DM 0.17602 0.18458 12.287 0
DM+PP 0.14261 0.19111 9.615 0
SI+PP 0.11914 0.18610 8.248 0
DM 0.06544 0.17862 4.720 0
PP 0.04485 0.17069 3.385 0
SI 0.06275 0.17752 4.554 0
48
5.4 Participant’s satisfaction rate
The figure 19 is the average scores of group participant’s satisfaction rating, and figure 20, 21 and 22 is the satisfaction rating in food, travel and shopping scenario. As shown in the figures the average rating of our proposed ranking list was better than others in different scenario.
Figure 19. The Average of Satisfaction
Figure 20. The Satisfaction in Food Scenario
2.5
49
Figure 21. The Satisfaction in Travel Scenario
Figure 22. The Satisfaction in Shopping Scenario
Furthermore, we use a statistic method- the paired sample t-test in 95% significant level, the all the pair test is significant under 0.05. In other words, our method is the best compared with others. The Table 14, 15 and 16 is shown as statistical verification of the satisfaction.
2.5
50
Table 14. Statistical Verification of the Satisfaction in Food Scenario Paired Group Mean Std Dev t-Value Sig(2-tailed) SI+PP+DM Traditional Group
Decision System
1.01205 0.09955 10.167 0
Random 2.42547 0.15281 15.872 0
Table 15. Statistical Verification of the Satisfaction in Travel Scenario Paired Group Mean Std Dev t-Value Sig(2-tailed) SI+PP+DM Traditional Group
Decision System
1.14254 0.09633 12.432 0
Random 2.67231 0.18431 16.177 0
Table 16. Statistical Verification of the Satisfaction in Shopping Scenario Paired Group Mean Std Dev t-Value Sig(2-tailed) SI+PP+DM Traditional Group
Decision System
0.09775 0.09555 11.577 0
Random 2.54343 0.17382 15.663 0
51
CHAPTER 6 DISSSUSION AND CONCLUSION
With the development of social media, the electronic commerce has evolved to a new paradigm of social network driven commerce or social commerce. For example, Facebook provides fans page for users to share and exchange goods information or user’s experiment.
Recently, with group commerce development, most people organize a group for collective purchasing some suitable products or services. While many recommender systems are developed to support the group commerce vendor to promote their products or services. The group decision systems for supporting group commerce customers are still little. In this study, we proposed a social decision support mechanism for group purchasing, which utilizes three components: social influence, personal preference and discussion context. The proposed mechanism can recommend the fittest option set for group members, quantify the evaluations of group members and use social influence adjusted voting mechanism to recommend a list of ranked options according to this discussion information over social media. The results of the experiment show that the proposed mechanism has the better performance than other benchmark methods.
6.1 Research Contribution
This study makes some significant contributions described as follows.
Firstly, from the practical aspect, most of decision support system mainly use past data and expert opinions to determine the best option or strategy. None of these systems consider that group decision should integrate social influence between group members, participant expertise, and group discussion message information. The three types of information can provide more suitable option ranking list to group members. Moreover based on the dynamic discussion message analysis, the proposed system can extract and recommend the fittest options for group, then support them to reach common consensus fast.
52
Secondly, from the methodological aspect, this study integrate the techniques of mining and social network analysis, and MCDM techniques to identify important criteria from discussion context and discover influenced person who is opinion leader or close friends, and determine criteria weights to consolidate the group decision processes under social media environment.
Thirdly, from the empirical aspect, we discover that personal preference is a more important factor than two others in eating scenario, discussion message is a more important factor than two others in travel scenario, and personal preference is a more important factor than two others in purchasing scenario. According to the result of the experiment, the similarity will be significantly improved when system considers more factors.
6.2 Research Limitations
There are some limitations to this research.
Firstly, the mechanism analyzes personal preference based on the information whether the user clicks the “like” button of fans pages on the social network platform. But some fans pages is not popular on Facebook. Their fans pages click like button number nearly rare. We have to look for the fans pages which are representative. Secondly, in the discussion process, there are a lot of not meaningful conversations during group participants’ discussion. So we have to extract meaningful part to analyze. Thirdly, there is the security issue in the system when we want to collect group participant’s social network information. Some people lock their Facebook personal information such as total friend number or mutual friends. So some people’s social network information is incomplete. For correctly evaluating mechanism performance we have to eliminate these data. Lastly, the proposed mechanism has the problem of cold start. The mechanism requires enough numbers of users in the database and maintain users’ behavior and interaction on social media to provide more suitable option ranking list.
53
6.3 Future Studies
There are several related issues which could be further studied. Firstly, in this research, we mainly use Facebook fans pages whether the user click the Like button to find the opinion leader in the discussion group. In the future, the factor of user’s activity or online behavior can be added into the system to help determine user’s social preference and then find the group opinion leader in different scenarios. Secondly, with rapid development of mobile device and techniques and people’s opinion may be influenced and changed under different contexts (e.g. location or time), so our system can combine mobile techniques to get group discussion messages in real-time. The data collected will closely reflect their current needs.
Thirdly, in our mechanism, we consider social influence data such as mutual friend, mutual Facebook Club or comment to increase group satisfaction. However, there are still several other social data which is possible to compute the social influence between two person, such as tags, pokes or frequency of messages sent. Fourthly, with rapid development of group commerce, we can implement proposed mechanism in group commerce website such as Groupon. Lastly, in the real world people will ask their friends when they make a purchasing discussion, so our mechanism can consider evaluation from their friends who are not in discussion processing.
54
REFERENCES
1. Anagnostopoulos. Sellers in e-marketplaces: A Fuzzy Logic based decision support system. Information Sciences, Volume 278, 10 September 2014, Pages 267-284 2. A. Agarwal and P. Bhattacharya, Augmenting Wordnet with polarity information on
adjectives. in Proceedings of the 3rd International Wordnet Conference, Jeju Island, Korea, 2006.
3. Abramovici, M. and A. Lindner, Providing product use knowledge for the design of improved product generations. CIRP Annals-Manufacturing Technology, 2011.
4. Alireza Abbasi, Rolf T. Wigand and Liaquat Hossain. Measuring social capital through network analysis and its influence on individual performance. Library & Information Science Research, Volume 36, Issue 1, January 2014, Pages 66-73
5. Andreas Klein and Parimal Bhagat, We-Commerce: Evidence On A New Virtual Commerce Platform, Global Journal Of Business Research, Vol 4, No. 4,2010.
6. Andrew T. Stephen and Olivier Toubia*,Deriving Value from Social Commerce Networks, Journal of Marketing Research, Forthcoming, January 9, 2009
7. Bearden, W. O. and Etzel, M. J. Reference Group Influence on Product and Brand Purchase Decisions. Journal of Consumer Research, 9, 2(1982), 183-194.
8. Byers,J.W.,Mitzenmacher,M., Zervas,G.,The Groupon Effect on Yelp Ratings: A Root Cause Analysis, ACM Journal Name, Vol. X, No. X, Article X,,February 2012.
9. Childers, T. L. and Rao, A. R. The Influence of Familial and Peer-Based Reference Groups on Consumer Decisions. Journal of Consumer Research, 19, 2(1992), 198-211.
10. Ching-Yung Lin etc. Social Network Analysis in Enterprise, Proceedings of the IEEE (Volume:100 , Issue: 9 ), Sept. 2012, pages 2759 – 2776
11. C.J. Hinde, R.S. Patching, and S.A. McCoy, Semantic transfer and contradictory evidence in intuitionistic fuzzy sets, in: Proceedings of the 2008 IEEE International Conference on Fuzzy Systems, Hong Kong, 2008, 2095-2102.
12. Dan J. Kim, Donald L. Ferrin and H. Raghav Rao. A trust-based
consumer decision-making model in electronic commerce: The role of trust, perceived risk, and their antecedents. Decision Support Systems, Volume 44, Issue 2, January 2008, Pages 544-564
13. Ecommercetaiwanapp,URL:[http://ecommercetaiwanapp.blogspot.tw/2013/12/2013-o2o
55
_8196.html], Accessed:12/15/2013
14. EZprice,URL:[http://news.ezprice.com.tw/3133/#.U3LkUvmSySo],Accessed:03/30/201 4
15. Fodness, D. and Murray, B. A Model of Tourist Information Search Behavior. Journal of Travel Research, 37, 3(1999), 220-230
16. GROUPON, URL: [http://www.groupon.com/deals/cafepress-youngstown], Accessed: 04/11/2011
17. Hsu, H.-M. and Chen, C.-T. Aggregation of Fuzzy Opinions under Group Decision Making. Fuzzy Sets and Systems, 79, 3(1996), 279-285.
18. J. Kamps, M. Marx, R.J. Mokken, and M.D. Rijke, Using Wordnet to measure semantic orientation of adjectives, in: Proceedings of the LREC, 4th International Conference on Language Resources and Evaluation, 1115-1118, 2004.
19. Kim, Y. A. and Srivastava, J. Impact of Social Influence in E-Commerce Decision Making. In Proceedings of the 9th International Conference on Electronic Commerce, Minneapolis, Minnesota, USA: ACM, 2007, pp.293-302.
20. Kotler, P. Marketing Management: Analysis, Planning, Implementation, Control. 9th ed:
Prentice Hall, 1997.
21. K.T. Atanassov, Intuitionistic fuzzy sets, Fuzzy Sets and Systems 20 (1) (1986), 87-96.
22. LIAO S H, et al. Mining customer knowledge for exploring online group buying behavior. Expert Systems with Applications, 2012, 39(3): 3708-3716.
23. Liu Bao-zheng and Wang Ding-Wei, Research on the realization and the risk of group-buying pattern, Control and Decision Conference (CCDC), 2013 25th Chinese, date 25-27 May 2013, pages 1248 – 1251
24. M. Kilduff, The friendship network as a decision-making resource:
Dispositionalmoderators of social influences on organizational choice, Journal of Personality and Social Psychology 62 (1) (1992), 168-180.
25. Moutusy Maity and Mayukh Dass, Consumer decision-making across modern and
56
traditional channels: E-commerce, m-commerce, in-store. Decision Support Systems, Volume 61, May 2014, Pages 34-46
26. Murray, K. B. A Test of Services Marketing Theory: Consumer Information Acquisition Activities. The Journal of Marketing, 55, 1(1991), 10-25.
27. Nascimento, V etc. Enhancing Ontology Alignment Recommendation by Exploring Emergent Social Networks, Web Intelligence and Intelligent Agent Technology (WI-IAT), 2012 IEEE/WIC/ACM International Conferences on (Volume:2 )
28. OTX, The Impact of Social Media on Purchasing Behavior, 2008.
29. Paez, A., Scott, D. M., & Volz, E. (2008). A discrete-choice approach to modeling social influence on individual decision making. Environment and planning. B, Planning &
design, 35(6), 1055.
30. P.D. Turney and M.L. Littman, Measuring praise and criticism: Inference of semantic orientation from association, ACM Transactions on Information Systems 21 (4) (2003), 315-346.
31. Peter, J.P. & Olson (1990). Consumer Behavior and Marketing Strategy. Richard D.
Irwin Inc. (pp.5-6)
32. Peter Leitner and Thomas Grechenig, Collaborative Shopping Networks: Sharing the Wisdom of Crowds in E-Commerce Environments, 21st Bled eConference eCollaboration: Overcoming Boundaries through Multi-Channel Interaction ,June 15 - 18, 2008; Bled, Slovenia
33. Pew Internet (c), 28% of American adults use mobile and social location-based services, Available at: [http://pewinternet.org/Reports/2011/Location.aspx], Accessed:
19/05/2013.
34. Piller, F.T. and D. Walcher, Toolkits for idea competitions: a novel method to integrate users in new product development. r&D management, 2006. 36(3): p. 307-318.
35. Rashotte, L., Social Influence, in The Blackwell Encyclopedia of Social Psychology, Ritzer, G., Editor. 2007, Blackwell Publishing: Malden, Massachusetts, USA. p.
4426-4429.
57
36. Sanghyun Kim & Hyunsun Park(2013), Effects of various characteristics of social commerce (s-commerce) on consumers’s trust and trust performance, International Journal of Information Management, Volume 33, Issue 2, April 2013, Pages 318–332 37. S. C. Alexander, J. L. Peterson, and A. B.Hollingshead. Help is at your keyboard:
Support groups and the internet. In L. R. Frey, editor, Group Communication in Context:
Study of Bona Fide Groups, 2nd ed, pages 309–334. Lawrence Erlbaum Associates, 2003.
38. Sherry M.B. B2B e-commerce adoption decisions in Taiwan: The interaction of cultural and other institutional factors, Electronic Commerce Research and Applications, Volume 5, Issue 2, Summer 2006, Pages 92-104
39. Simon, H. A. A Behavioral Model of Rational Choice. The Quarterly Journal of Economics, 69, 1(1955), 99-118.
40. Statistic Brain, URL: [http://www.statisticbrain.com/facebook-statistics/], Accessed:01/01/2014
41. S.Y. Chou, Y.H. Chang, and C.Y. Shen, A fuzzy simple additive weighting system under group decision-making for facility location selection with objective/subjective attributes, European Journal of Operational Research 189 (1) (2008), 132-145.
42. Taiwantrade,URL:[http://www.taiwantrade.com.tw/CH/bizsearchdetail/42526/I], Accessed: 03/29/ 2012
43. techcrunch,URL:[http://techcrunch.com/2010/12/01/groupon/],Accessed:01/10/2010 44. Tse-Ming Tsai,Ping-Che Yang, Wen-Nan Wang, Pilot Study toward Realizing Social
Effect in O2O Commerce Services. Social Informatics Lecture Notes in Computer Science Volume 8238, pp 268-273, 2013.
45. Vaidya, O. S. and Kumar, S. Analytic Hierarchy Process: An Overview of Applications.
European Journal of Operational Research, 169, 1(2006), 1-29.
46. Von Neumann, J.; Morgenstern, O.; Rubinstein, A. and Kuhn, H. W. Theory of Games