xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 2200385Ka 4500
controlfield tag 001 002029654
007 cr mnu|||uuuuu
008 090918s2009 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002976
h [electronic resource] /
by Rebecca Anderson.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 47 pages.
Thesis (M.S.)--University of South Florida, 2009.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
ABSTRACT: Social shopping is one of the latest trends on the Internet. Websites dedicated to social networking with a focus on shopping have been emerging on the web for a few years. The basic idea is that consumers are looking for product information on the Internet and social shopping sites provide a place for consumers to find this information from other consumers. These sites provide a place for their users to engage in socialization and shopping simultaneously, sometimes following recommendations of premier users, who are labeled from other users. However, purchases aren't made through these sites. So, there may still be something missing from the experience. For these sites, social pricing mechanisms may be implemented to provide revenue. Major ecommerce websites have begun focusing on increasing social features throughout the transaction process. For example, more websites are including ratings, reviews and recommendations of products and services by other consumers.However, pure ecommerce websites do not provide functionality that allows consumers to communicate in real time. Hence, there are some features missing from the social experience. Also, the social functionality included in pure e-commerce websites, tends to be utilized for the benefit of the Web site, as opposed to the consumers. Both social shopping sites and ecommerce sites have seen independently successful though few sites have been able to truly integrate these together at this point. It may be more beneficial to the end user if these sites could work in unison. This thesis is an exploratory study of the emerging social shopping phenomenon.The contributions of this work include analysis of the social shopping phenomenon and identifying metrics and Web sites that incorporate social shopping, a survey of academic literature related to social shopping and social pricing and a review of current recommender system algorithms with a discussion on how to incorporate social networking data into the algorithms to improve recommendations. Improvement suggestions include incorporating customer purchase history with social networking information. Potential future research ideas are included.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Advisor: Balaji Padmanabhan, Ph.D.
x Information Systems and Decision Sciences
t USF Electronic Theses and Dissertations.
Social Shopping by Rebecca Anderson A thesis submitted in partial fulfillment of requirements for the degree of Master of Science Department of Information Systems and Decisions Sciences College of Business Administration University of South Florida Major Professor: Balaji Padmanabhan, Ph.D. Min-Dong Paul Lee, Ph.D. Rahul Tripathi, Ph.D. Date of Approval: April 27, 2009 Keywords: Social Pricing, Recommender Systems, Social Ne tworks, E-Commerce, Trust Copyright 2009, Rebecca Anderson
i Table of Contents List of Tables ii List of Figures iii Abstract iv Chapter 1: Social Shopping on the Internet 1 1.1 What is Social Shopping? 1 1.2 Matrix of Social Shopping Websites 3 1.3 Sites in the Matrix 8 1.4 What is Social Shopping on the Internet? 19 1.5 Scoring/Social Index 20 Chapter 2: Survey of Research in Social Shopping 22 2.1 Online Purchase Decisions 22 2.2 Impact of User Reviews 24 2.3 Impact of Recommender Systems 26 2.4 Impact of Social Networks 28 Chapter 3: Pricing Mechanisms and Social Shopping: A Case for Demand Aggregation 31 Chapter 4: A Social Recommendation Algorithm 34 4.1 Examining Collaborative Filtering 34 4.2 Examining Amazon.com 36 4.3 Examining SNACK 38 4.4 Utilizing the Social Network 40 Chapter 5: Conclusion 43 References Cited 44
ii List of Tables Table 1-A Matrix of Social Shopp ing Websites 4 Table 1-B Matrix of Social Shopp ing Websites 5 Table 2 Social Index 20 Table 3 Weighted Social Index 21 Table 4 Collaborative Filtering 34 Table 5 Item-to-Item Collaborative Filter ing .36 Table 6 Customer Interest Level in Movies by Genre 41
iii List of Figures Figure 1. Woot 11 Figure 2. Kaboodle 11 Figure 3. Zebo 12 Figure 4. ThisNext 12 Figure 5. StyleHive 13 Figure 6. StyleFeeder 13 Figure 7. ShopStyle 14 Figure 8. Amazon 14 Figure 9. Walmart 15 Figure 10. Target 15 Figure 11. JC Penney 16 Figure 12. QVC 16 Figure 13. Sears 17 Figure 14. Overstock 17 Figure 15. Kohls 18 Figure 16. MacyÂ’s 18 Figure 17. Network Classification Mode l 29 Figure 18. SNACK Network Nodes 39
iv Social Shopping Rebecca Anderson ABSTRACT Social shopping is one of the latest tre nds on the Internet. Websites dedicated to social networking with a focus on shopping have been emerging on the web for a few years. The basic idea is that consumer s are looking for product information on the Internet and social shopping sites provide a place for consumers to find this information from other consumers. These sites provi de a place for their users to engage in socialization and shopping simultaneously, sometimes following recommendations of premier users, who are labeled from other users. However, purchases arenÂ’t made through these sites. So, there may still be something missing from the experience. For these sites, social pricing mechanisms may be implemented to provide revenue. Major ecommerce websites have begun focusing on in creasing social feat ures throughout the transaction process. For example, more websites are includi ng ratings, reviews and recommendations of products and services by other consumers. However, pure ecommerce websites do not provide functionality that allows consumers to communicate in real time. Hence, there are some featur es missing from the social experience. Also, the social functionality included in pure e-co mmerce websites, tends to be utilized for the benefit of the Web site, as opposed to the c onsumers. Both social shopping sites and ecommerce sites have seen independently suc cessful though few sites have been able to truly integrate these together at this point. It may be more benefici al to the end user if these sites could work in unison. This thes is is an explorator y study of the emerging social shopping phenomenon. The contributions of this work include analysis of the social shopping phenomenon and identifying me trics and Web sites that incorporate social shopping, a survey of academic literatu re related to social shopping and social pricing and a review of cu rrent recommender system algorithms with a discussion on
v how to incorporate social networking data into the algorithms to improve recommendations. Improvement suggestions include incorporating customer purchase history with social networking information. Potential future research ideas are included.
1 Chapter 1: Social Shopping on the Internet The popularity of social networking sites ha s steadily increased over the past few years. This is important for businesses becau se it means that larg e groups of people are congregating and communicating in the same pl ace online on a consistent basis. Ideally businesses should harness the buying power of th e social networks, but this is not always easy. Gaining the trust of many social network users is a difficult task. Social websites are places where people go to interact with fr iends and meet new people. This is often not the atmosphere in which users want to in teract with businesses or be bombarded with advertising. A possible soluti on is social shopping websites. Essentially, social shopping encompasse s various types of viral marketing. Businesses are increasingly paying attention to this. People who participate on these websites have already identified themselves as buyers and have indicated which products they are interested in. Further, users read ily give up information about themselves, such as location and hobbies and they have a group of friends with similar interests. Given the extent of explicitly revealed information these sites are potential gold mines for companies in terms of learning about customer s. With social shopping sites, businesses can more easily find and target future customers. 1.1 What is Social Shopping? There are many definitions for social shoppi ng. Wikipedia defines it as Â“a method of e-commerce and of traditional shopping in wh ich consumers shop in a social networking environment similar to MySpace.Â” Entrep eneur.com defines it as Â“the intriguing offspring of social networking and online s hopping.Â” The New York Times calls it Â“a new category of e-commerce that tries to co mbine two favorite onlin e activities: shopping and social networking.Â” About.c om describes it as Â“the comb ination of social media and e-commerce. In essence, it is taking all th e key aspects of the social web Â– friends,
2 groups, voting, comments, discussions Â– a nd focusing them on the worldÂ’s favorite activity: shopping.Â” GetElast ic describes is as Â“a mashup that resembles social bookmarking, social networking and comparison shopping in a blender.Â” Inc.com calls it Â“services [that] combine the networking power of MySpace with the data-crunching power of Google and in the process bring a little more humanity to the act of shopping online.Â” The LATimes.com wrote that it is Â“combine[s] two of the WebÂ’s most prominent activities: engaging in commerce and chatting with like-minded folk.Â” Hitwise finds social shopping to be Â“a gr oup of websitesÂ… that center around the users creating customized wish lists to share with friends or people with similar tastes, rather than aggregating content around the product or retailer.Â” Te chCrunch calls it Â“a strange grab-bag of sites all trying to crack the nut of how to monetize so cial networking around shopping, which is most social when it is real-world, not virtual.Â” These definitions are limiting as they fo rce the definition around e-commerce, as opposed to all forms of commerce. Only TechCrunch and Answers.com reference the idea that social shopping could extend beyond the web. What is social shopping really? In this thesis we define social shopping as shopping in which a customerÂ’s purchasing process is in part affected by communicati on between the customer and others who are not affiliated with a product. For exam ple, when two friends are shopping at a department store, one friend may make a co mment about an item and the other friend makes a purchase based, at least in part, on that comment. This concept includes a wide spectrum of communication and shopping scenarios. Some environments create more sociability than others. For instance, shopping malls tend to provide a more social environment than stand alone stores. This is because shopping malls include activities other than shopping that prom ote social activities, such as dining, movie theaters and arcades. Als o, shopping malls play host to a variety of retail stores which allows shoppers to comple te multiple tasks in one outing and compare prices and selections between similar items. All these characteristics combine to make shopping malls highly social in na ture. Stand alone stores do not provide the same social environment that shopping malls do. It is less likely to see dini ng or other activities offered in these stores. Also, the shopper is unable to complete as many tasks and is
3 unable to comparison shop. Sta nd alone stores do not prohi bit social behavior; they simply do not create a social environment. The goal of the stand alone store is product driven, while the goal of the shoppi ng mall is socially driven. Social shopping can also occur through ot her forms of commerce, such as shopping using the television. This is most comm only done with companies HSN and QVC that have their own cable television networks. Shopping in this form occurs when viewers purchase an item seen on television by calli ng a phone number during the program. This form of shopping is not social in nature, but can become social when two or more people are communicating about a product as the sa le is happening. For instance, two people could be together in a room, talking over th e phone, chatting on the Internet, etc. while watching the television broadcast. Auctions are a form of shopping that can be social in nature. Shopping in this form occurs when many people bid up the price of an item and the person with the highest bid gets the item. As large groups of people fo rm to bid on products, conversations between individuals are likely and re sult in social shopping. S hopping over the Internet, or ecommerce, is also popular and can be soci al in nature. E-commerce began as an individual activity, but with the advent of social networking sites, has increasingly become a social activity. These can be split su ch as brick and mortar stores into concepts of shopping malls and stand alone stores. There are social networking sites themed around shopping, such a Kaboodle.com, which co rrelates to the shopping mall. There are product driven sites with social elements that translate to th e stand alone store, such as Amazon.com, which most people are accustomed to. E-commerce also takes place in the form of auctions through sites such as eBay.com 1.2 Matrix of Social Shopping Websites Below is a matrix of the top eight soci al shopping websites in the United States according to Alexa.com, and the top ten e-co mmerce websites (excluding eBay Express, since that has since shut down) according to HitWise. There are a total of fourteen prominent social features, displaye d below in two separate tables.
4 Table 1-A Matrix of So cial Shopping Websites Rating Review Poll Profile Comment Wall Forum Email Woot X Kaboodle X X X X X Zebo X X X X X ThisNext X X X X X StyleHive X X X X X X StyleFeeder X X X X ShopStyle X X X X Amazon X X Walmart X X Target X X JC Penney X X X QVC X X X Sears X X Overstock X X X X X Kohls X MacyÂ’s X X X
5 Table 1-B Matrix of Social Shopping Sites Blog Games/ Quizzes Share on Other Sites Social Network Support Social Pricing Recs Social Recs Woot X X Kaboodle X X Zebo X X ThisNext X X X X N StyleHive X X X X StyleFeeder ShopStyle X Amazon X X Y Walmart X N Target JC Penney X Y QVC X X Y Sears Overstock X X Y Kohls MacyÂ’s X N The categories in the columns were chosen because they allow for communication between at least two people. When customers rate products, as seen in figure 16, they quickly communicate about the value, usefulness or other measur ement to other potential customers. Reviews, as seen in figure 3 and figure 9, are similar except that customers have the ability to offer more information a bout a product than a rating. Ratings and reviews are common across both socially dr iven and product driven websites.
6 Polls, as seen in figure 2, allow peop le to ask others for their choice among specific products. So, if someone is inte rested in buying a pair of jeans and is looking at different brands or styles, that person can pos t a poll requesting others to vote for the best pair. Profiles, as seen in figure 7, give peop le the opportunity to, not only broadcast information about themselves, but search th rough profiles to find others like them. Profiles allow people with similar tastes a nd interests to find and connect with one another. Comment walls, as seen in figure 6, allo w people to write personalized messages to others. Forums, as seen in figure 1, provide a sp ace for people to discuss products, brands or anything else as a group. Many sites provide an email feature, as seen in figures 11 and 15, which allows someone to send information about a specific product to another person. Blogs, as seen in figures 5 and 14, are sim ilar to reviews in th at they allow people to write, in as much detail as they wis h, about any aspect of a product or brand. Some people post games and quizzes, as seen in figure 5, which allows other people to interact with different products. Another common feature is th e ability to shar e product information from one site on any other site, as seen in figure 8. This allows one person to communicate with multiple people in various places from one medium. Social network support, as seen in figure 4, lets users use exis ting social networks formed on popular social networking s ites, such as Facebook, explicitly. Social pricing occurs when users get fi nancial incentives or different product prices based on the degree of their so cial involvement in the site. Recommendations are system genera ted and often based on proprietary algorithms. Social recommendations ar e also system generated based on algorithms that generate recommendations specifically for a user but based on other usersÂ’ data.
7 There are few conclusive studies that s how which of these particular features influences a purchasing decision and it is also possible that the effects of these features can vary based on the site. However, all the li sted features have th e potential to influence decisions as they give people the opport unity to interact with the product and communicate with others in the shopping process. As the matrix shows, recommendations are the most popular of the social features, while social pricing in rare. The matrix can also be viewed as a brief summary of the features that the Internet shop ping population is looking for today. Online shoppers want an easy way to sort through a ll the available choices and they prefer recommendations from people like them selves, or social recommendations. Recommendations are also an easy way to a ggregate opinions. With technologies like instant messaging a user can only contact other users one at a time. With recommendations, users can quickly see many opinions at once. The advantage that instant messaging has is that the conversations occur in real time and with someone that the user is already familiar with. Familiarity is helpful because there is already a trust in the taste and opinion of the person giving th e recommendation. Amazon, attempts to alleviate the concern in trusting a recomme ndation by allowing users to review the recommendations. In other words, customer s that have utilized a recommendation in a purchase decision can then give a rating a bout the helpfulness of the recommendation. Amazon sorts recommendations by rating, so new customers first see what other customers found to be helpful. It would be interesting for Amazon to allow users to create taste based profiles and append this information to the recommendations. There is clearly potential for new social technologies to emerge in this space as well. Real time video messaging and voice co mmunications would be natural since they are technologies already being utilized on th e web. These kinds of technologies would increase direct communication between two pe ople already known to one another. Since people prefer recommendations fr om others that they trust, these technologies have the potential to increase purchases. However, they decrease communi cation to many people at once, which is what social shopping is founded on. Perhaps the next generation in
8 social shopping will come full circle back to the people already known to us, as opposed to strangers on the Internet. In terms of comparing specific e-comm erce sites today, based on the matrix Overstock offers the most social features of the e-commerce sites and Woot offers the fewest social features of the social shopping si tes. This shows that there is a wide range of sociability on many different types of retail websites. It is difficult to know which features are the most useful at various points in the decision pro cess. It is also difficult to know whether it is more beneficial for product sites to integrate more social features or for social sites to sell an actual product. 1.3 Sites in the Matrix Figure 1 through figure 16 illustrates how specifi c social features are usually provided in e-commerce sites. Below are brief descriptions of the sites in the matrix. Woot.com sells one product a day and every day is therefore Â“a different productÂ”. There is a forum on the site where members discuss the daily product, post any reviews they have found on it and what it retails for on other sites. Also, Woot.com shows the community real time statistics about how many items have been sold, where the purchases are coming from across the count ry, the percentage of people who bought multiple items and more. Kaboodle.com does not actually sell anyt hing. This is primarily a social networking site with a focus on shopping. Us ers create profiles a nd shopping lists and interact with other users with similar shopping tastes. The items in the shopping lists are from third party e-commerce sites and not so ld directly by Kaboodle. Currently Kaboodle is one of the most popular social shopping websites. Zebo.com is also a social networking site geared toward shopping, but also allows users to buy and sell items through the site. This makes sellers buyers and vice versa all in the same place. Also, this site is a little more competitive in that users are ranked based on how much they own, or which brands they own. Revenue comes from people selling items on the site and from paid advertisements.
9 ThisNext.com has all the social networking elements of a social shopping site. It allows users to make widgets to showcase produ cts they like on other s ites. It also has mavens, which are users that have reached cel ebrity status on the site by many other users following their recommendations and purchases. The site further categorizes mavens by the regions and product categories they are most popular in. Mavens al so play a role in the siteÂ’s marketing efforts. Revenue come s from referrals based on reviews on the site. StyleHive.com is a social shopping site th at focuses on fashion. The site allows users to tag trends found across the Internet and also posts information on shopping deals and discounts. Users of the site then pr omote high end fashion, so it is targeted specifically to a certain segment. Further, the site allows retailers and designers to introduce their products through various featur es, such as the blogs or communities. StyleFeeder.com is a social bookmarking site. Users bookmark various products across the Internet that interests them. StyleFeeder creates a kind of personalized shopping engine for users based on the items that theyÂ’ve bookmarked. StyleFeeder has also created widgets that allow users to pos t their personal stylefeed on other sites across the web. ShopStyle.com is a social networking site focused completely on fashion and accessories. Users can browse through multiple retailers through the si te and create their own look based on products found on the site. Users can also sign up to have sales on their favorite brands emailed to them from the site. The brands feat ured on this site tend to be high end and high priced. Amazon.com is one of the oldest e-commer ce websites. This site carries almost anything a user could want to purchase on line from highly recognizable brands and products to some of the most obscure. Th e site features a reputed recommendation system that matches users to potential products. It is also widely used for getting detailed and very useful customer reviews for products. Walmart.com is the online presence of one of the largest retailers in the world. The site features close out prices on all products, just as the brick and mortar stores do. It offers products across multiple categories and highlights coupons and savings on the site.
10 Target.com carries products across multiple categories. Most products carried are also available in brick and mortar stores except a bridal fashion line only available online. It also offers an online deal of the week. JC Penney.com carries the same products as the brick and mortar department store, with a few additional items only found online. Users can rate and review online items, which are popular social features. QVC.com offers the same products as on te levision, but they are always available, until they are sold out, on the site. Users can also watch the televised item in real time while on the site and can check program gui des for future televised item sales. Sears.com offers the same product lines as th e brick and mortar store. They try to market the site through email newsletters and incent users to sign up for them by offering $10 in coupons. Users can also find outlet items on the site. Overstock.com is an e-commerce site that offers low prices on products across multiple categories. It also has auction, auto and real estate sections. Overstock offers users recommendations based on items theyÂ’v e looked at or purch ased in the past. Kohls.com offers the same product line that is found in the brick and mortar store. It appears to have the fewest social features of all the site s listed based on the metrics listed here. Macys.com features the same product line as the brick and mortar stores, though occasionally offers discounts only available online. It offers an application to be downloaded to the desktop that shows products offers personalized to the user. It also has few social features compar ed to others discussed here.
F F igure 1 Wo o igure 2 Kab o t oodle 11
F F igure 3 Zeb o igure 4 Thi s o s Next 12
F F igure 5 Styl e igure 6 Styl e e Hive e Feede r 13
F F igure 7 Sho p igure 8 Am a p Style a zon 14
F F igure 9 Wal m igure 10 Ta r m art r get 15
F F igure 11 JC igure 12 Q V Penney V C 16
F F igure 13 Se a igure 14 O v a rs v erstock 17
F F igure 15 K o igure 16 M a o lhs a cyÂ’s 18
19 1.4 What is Social Shopping on the Internet? Sites like Amazon.com were in existence ev en prior to the dot com boom in the 20002001 periods. What is becoming increasingl y popular is combining e-commerce with social features to create so cial shopping. Currently, soci al shopping sites are product driven and the kinds of products that us ually show up on these sites are personal products, such as clothing, home dcor, music and music accessories and gadgets. Further, the newest and trendi est items on the market are ge nerally quick to be found and discussed in these networks. Often, the site s do not actually sell anything. To purchase any of the products seen, the user must return to the product providerÂ’ s site. These kinds of sites aggregate data by allowing users to tag products on other sites using a browser plug-in or by uploading informa tion about a product straight to the site. The users that generally frequent the social shopping sites today are predominantly younger and technology savvy. A feature of some of the most social sites is the Â“premier shopperÂ”. Essentially, site users can reach celebrity status by being Â“followedÂ” by other site users. The recommendations of the premier users carry more weight than those of other users. They become premier users through their reputation of promoting products that are well liked by other users. Reputation systems are employed in order for users to rate one another. Also, the websites promote their premier users, further increasing the celebrity status. When shopping in a mall, there are a finite number of items from which to choose. A potential customer can see, touch and interact with the item. Potential customers can also compare similar items and see other people interacting with and purchasing items of interest. On the Internet, these things are not as easy to do and the processes, such as in comparing similar items, can be overwhelmi ng. Also, on the Internet the number of possible items from which to choose is often very large. Social shopping sites offer a solution to this problem by providing a pl ace online for customers to review and recommend products and for potential cust omers to easily find and utilize this information. The sites also allow users to add other users with similar interests and tastes to their networks. This allows users to have connections with one person who shares
20 their taste in music, another w ho shares their taste in interior design, a third with a shared taste in clothing and so on. Social shopping sites give online shoppers th e ability to create a profile, specifying their tastes in various products and to networ k with others with similar tastes. Users provide the content for the si te and users set the product tr ends, often leading to viral marketing for various products. The future of these sites as standalone destinations will depend on how these business models thrive. Ho wever the social features incorporated on these sites and the technologies used to en able user interactio ns may gradually be incorporated into most e-commerce sites. 1.5 Scoring/Social Index The sites in the matrix above were scor ed based on which soci al features were present. A simple rating score comprised on the number of social feat ures present is used to rank the sites in Table 2. Table 2 Social Index Rating Site 10 StyleHive 8 ThisNext 7 Kaboodle 7 Zebo 7 Overstock 5 ShopStyle 5 StyleFeeder 5 QVC 4 Amazon 4 JC Penney 3 MacyÂ’s 3 Woot 2 Sears 2 Target 2 Walmart 1 Kohls
21 An interesting observation is that Overstoc k actually has more so cial features than some of the social sites, whereas Woot is le ss social than some of the e-commerce sites. It might be more useful to weight some of th e social features as more important than the others. Social features that promote a communication to a single person may be more effective than communication directed at a gr oup of people. Such features are comment walls, polls, social recommendations and email. A weighting that rates these twice as heavily as others is used in Table 2. The ch anges to the table are minor, with effects to only two sites, Kaboodle and Kohls. Table 3 Weighted Social Index Rating Site 12 StyleHive 10 Kaboodle 10 ThisNext 9 Zebo 9 Overstock 6 ShopStyle 6 StyleFeeder 6 QVC 6 JC Penney 5 Amazon 4 MacyÂ’s 3 Sears 3 Woot 2 Kohls 2 Target 2 Walmart
22 Chapter 2: Survey of Research in Social Shopping Given that the social shopping phenomenon is recent there is lit tle work that has studied this area directly. However there are bodies of work in related topics such as factors influencing online purchase decisions, the impact of user reviews in e-commerce, the impact of online recommender systems and the interaction between social networks and e-commerce. In this chapter we su rvey work in these related areas. 2.1 Online Purchase Decisions Multiple factors influence a customerÂ’s online purchase decision. Two major factors are trust and reputation and these are greatly influenced through site design, security and social factors (Yoon 2002). Social influence affects tr ust and reputation as users recommendations tend to increase these fa ctors when the site is unknown to users. Also, users tend to perceive a site as being valuable when other people have used it or made purchases from it in the past. Non-soci al factors are also important in increasing trust and reputation. A site should have the capability to secu re usersÂ’ personal information. Further, when a site is poor ly designed it increases frustration and dissatisfaction, which is linked to a decrease in reputation. There are two main trust definitions: Relia bility and Decision (Jsang et al 2006). Reliability Trust is reliability in somebody or something. Decision Trust is the extent to which one party is willing to depend on someone or something in a given situation with a feeling of relative security, even though there could be nega tive consequences. In other words, reliability trust equates to having fa ith in someone or something. Decision trust equates to having faith in someone in a give n situation with some level of risk. An e-commerce site must have certain feat ures in place to establish trust. These include security and privacy. Consumers will not consider any other aspects of the site unless security and privacy have been esta blished (Chen and Barnes 2007). These are
24 what other people have already done. When a new user on an e-commerce site can see that other people have used the site and found it useful, that new userÂ’s trust in the site is increased. Online there are many ways of providing information that will influence reputation. Some types of reput ation influencers are: rating, reviews, sales volumes and recommendations. Ratings are shown (Chen 2007) to have a positive impact on reputation as are sales volumes. These are both indicators that other people, not only find the site to be useful, but have enough trust in the site to ma ke a purchase. These tend to be displayed graphically. Also, these tend to be greatly us eful to users of the website because it is simple and quick to leave a rating. Experts and consumers both leave produc t reviews, but which opinion matters more? Consumers tend to trus t the reviews of other consum ers more than experts (Chen 2007). There may be a level of skepticism when accepting the recommendation of an expert since an expert could have been incented to speak on behalf of the product. However, a group of consumers are not likely to have the same incentive motivation to recommend a product. This is also true of recommendations from a recommender system as opposed to recommendations from a webs ite owner. Consumers tend to trust the recommender system over the website owner (Chen 2007). 2.2 Impact of User Reviews Online user reviews positively affect purchase decisions on e-commerce websites (Duan et al. 2008). However, reviews are more influential in terms of creating awareness, than in actually persuading a user to purchase a product. Further, as mentioned in the previous section, reviews ar e successful in establishing trust in an ecommerce site for new users. The type of us er making the review is also important as consumer reviews tend to carry more weight than expert reviews. Online reviews influence consumers in two wa ys. First, reviews are influential in perception of product quality, ge nerally through a detailed review (Duan et al 2008). Second, reviews are influentia l in increasing product aw areness, generally when
25 dispersed through online communities (Duan et al 2008). What appears to happen is that users are influenced by awareness more than by persuasion. For example, when a new iPhone is released user reviews abound thr oughout the Internet on product sites, blogs, social networks, etc. The more information there is about the product, the more aware a new user is about the product. Greater aw areness has a larger influence on purchase decisions than persuasion. This is probably because people are not willing to take the time to read many reviews. However, if th ere are many reviews, then many people must have an interest in the produc t. The more reviews are disp ersed throughout the Internet, the greater affect it will have (Duan et al 2008) on purchase decisions. Other studies suggest that consumer re views do have an effect on persuasion. Reviews on a website also have an effect on herd behavior, which is the concept that people will do what other peopl e have already done. In f act, reviews can actually be more effective than ratings in this sense. When new users saw reviews by consumers that contradicted ratings they change their choi ces (Chen 2007). It could be the case that reviews can change opinions at the time of purchase, which would mean the decision to buy has already been made. In this case, inte nt becomes important. Also, the reviews are located on an e-commerce site at influentia l points during the tr ansaction process. However, if the purchase decision has not been made, reviews may not persuade a consumer to purchase, but awareness of the pr oduct may aid in coming to the decision. The herd behavior concept also applies in terms of who has left the review, as mentioned in the previous s ection. Consumers tend to infl uence other consumers more than experts do (Chen 2007). Consumers also tend not to be influenced by reviews of website owners. This is because the owner has something to gain, in terms of revenue, by leaving a review or making a recommendation. A review like this can seem more like a sales pitch. Consumers leaving reviews ha ve nothing to gain, as they have already purchased and are utilizing the product or service. Consum ers simply find value in the product and wish to inform others about it.
26 2.3 Impact of Recommender Systems Recommender systems, or system ge nerated recommendations, are becoming ubiquitous across e-commerce websites mostly because they are viewed as more influential than user reviews in the online purchase decision. How recommender systems are implemented varies by website. Ther e are four distinct types of systems: collaborative filtering, content based, recomm endation support system and social data mining (Terveen and Hill 2001). In collabora tive filtering systems recommendations for a user are collaborative in that they are ba sed on purchases made by Â“similarÂ” users. The more user purchase data the better the sy stem is at giving product recommendations. Content based systems are based on the cont ent of the item viewed. For instance, a content based system may recommend other comedies to users who viewed a comedy film. Support systems do not actually make r ecommendations. Instead they support users in making recommendations for and finding reco mmendations from other users. Social data mining systems mine preferences from soci al interactions with other users. These systems require no input directly from the user in making recommendations. Some sites use one specific system while others combine tools from various systems to guide user purchase decisions. Recommendations provided through recomm ender systems are more influential than recommendations provided by human expe rts (Senecal and Nantel 2004). Further, recommendations provided through recommen der systems are actually judged more valuable than recommendations provided by friends (Sinha and Swearingen 2001). A part of the reason for this is that recomme nder systems offer a level of personalization because recommendations are based off previ ous purchases and other customers similar to the user. Human experts are merely gi ving a review of a pr oduct, not necessarily specified for any individual user. Friends, most likely, have a better idea of a userÂ’s tastes, but may not have a complete idea of purchase history in all domains. Further, users do not mind spending some time rating item s, if this provides them with useful recommendations from the system. The only request the user has is to provide enough information about the product for him to ma ke a purchase decision. The problem is defining how much information is enough, si nce this will gene rally vary by user.
27 A newer area of study in terms of reco mmender systems involves incorporating social information data along with purchase data in providing recommendations (Kim and Srivastava 2007). This information coul d come from the number of nodes between people in a social network, information ga rnered from products between people in a network, for instance, using the Tell A Friend email feature, data mining social information communicated through profiles, comments and blogs, etc. SNACK (Lam 2004) is an example of this kind of recomm ender system. SNACK provides a weight to recommendations based on closeness between people in a social network (Lam 2004). Kim and Srivastava (2007) suggest an approach base d on encouraging users to recommend a product to a friend and capturi ng the data between the two people (also, capture purchase and product from data from online communities as well as reviews). With enough data a fairly substantial repres entation of user taste can be built out which will enable valuable recommendations. However, transparency in data collection would have to be present for this model to work. Otherwise users may feel their privacy has been invaded. Incorporating context into recommenda tions could add value to recommender systems. For example, in the service i ndustry recommendations for a business lunch would be different than recommendations for a dinner date (Bonhard and Sasse 2006). So, incorporating event or s ituational context co uld improve recommendations made by recommender systems. The question is how to get this information and when to present it. Bonard and Sasse (2006) f ound that two situations had to be met for a user to trust a contextual recommendation. First, the advice seeker knows they have taste overlap with the recommender. Second, the advice seeker and recommender know each other well enough and have enough mutual taste that the recommender has a high likelihood of providing a valuable recommenda tion. To apply this to a recommender system, it would make sense to incorporate taste in terms of rated or selected products when making recommendations. To do this the system should provide informa tion regarding profile similarity and overlap in ra tings and reviews of items complementary to the product being recommended (Bonhard et al 2006). Inte restingly, familiarity with the profile does
28 not have an effect on trust in the recommendation. This sy stem would work only in an online environment in which sociability is promoted. 2.4 Impact of Social Networks Word of mouth has a positive impact on online purchase decisions (Dellarocas 2003), as does profile similarity between advi ce seekers and recommenders. Most word of mouth actually takes place implicitly. In other words, users endorse products by using them as opposed to talking about them. Th ere is explicit communication, which occurs when a user discusses satisfac tion in a product. This often comes in the form of user reviews, which can be found on an e-commer ce website or a weblog (blog). Further, there are programs to target influence ma ximizers within a network, sometimes called network targeting (Kempe et al. 2003). Influence maximi zers are users in a social network who have the ability to impact many other users within their own networks. By targeting only the maximizers the message will be sent more efficiently than by targeting the whole network. Network targeting is becoming more prev alent as online communities continue to flourish on the web. There are many people s earching for content and network targeting seeks to get the right content in front of the right people. Ac cording to Hill et al (2006), one model of network targeti ng is the network classificati on model. This model uses knowledge of the links between entities in a network to estimate quantity of interest. Typically, users are most influenced by other users closest to them in a network. For instance consider figure 17 below.
29 Figure 17 Network Classification Model Using this model, Person 1 would be most influenced by Person 2 and Person 3 because they are closest to him in the network. Pe rson 1 would be less influenced by Person 4, Person 5 and Person 6 because they are furt hest away from him in the network. However, taste becomes important as well. It is possible for people far away in terms of network nodes to share similar taste in Pr oduct X than people in the closest nodes. Closeness does not necessarily imply taste similarity. Intent of the online community becomes rele vant here. The intentions of users in a community like Myspace or Facebook is mostly sociability, the intention is not to make a purchase. So nodes, in terms of closeness, in a network like this may not have similar purchase histories or similar product tastes. The intentions of user s in a community like Kaboodle or ThisNext is to find and purchase pr oducts. For example, it is like going to a shopping mall with a group of friends. So nodes, in terms of closeness, in a network like this would be more likely to have similarity in purchase histories and product tastes. Essentially, the community from which this information is gathered is important. One model of getting information and r ecommendations into a network is the Push-Poll recommender system. This approach seeds a item into a social network (Push), then queries adjacent users about whether the item should be recommended for the current user (Poll) (Webster and Vassileva 2007). The process would begin with analysis and classification of content found in the netw ork and user rating of content. Then an Person 1 Person 2 Person 4 Person 5 Person 3 Person 6
30 item determined to be relevant to th e network would be pushed through. Some advantages to this model are utilizing soci al networks to create recommendations as opposed to setting rules (Webster and Va ssileva 2007). Also, new items can be introduced with minimum analysis (Webster and Vassileva 2007), since the network is actually making the recommendation of the item. This may increase efficiency in terms of computation of the algorithm.
31 Chapter 3: Pricing Mechanisms and Social Shopping: The Case of Demand Aggregation Can online users efficiently form (electronic) communities that might be able to receive or negotiate better pr ices? If an online mechanism ex isted for easy participation within a group purchase would buyers avai l themselves of this opportunity? What products or services might this work for? When and why would retailers Â– who are profit maximizers and often averse to discounting wa nt to participate in such a mechanism? Indeed these questions are important and timely. Some early firms, such as Mercata and Mobshop, who were in the bus iness of facilitating retail group-buying online, have failed and shut down operati ons. There are some insights from recent academic work that has studied this (relatively new) pricing mechanism. So why did some early group-buying attemp ts fail? One proposed factor is the purchase uncertainly (Tan et al 2007) that might come from a group-buying consortium. For instance, in some ways in which this is implemented, prospective buyers may not know how long they might have to wait to r eceive the product (Cook 2001), or what they might eventually have to pay for the product (T an et al. 2007). It has also been proposed that the concept may be Â“too difficultÂ” (C ook 2001) for retail buyers. This type of purchasing works best when placed at the right place in the supply, though it may be difficult to know what the Â“right placeÂ” is. Fo r retailers, this type of purchasing may not always make sense because they are last in the supply chain, meaning they would have to sacrifice some of thei r profits to make it work. Howe ver, products at the beginning or end of their lifecycles might fit we ll into a group-buying mechanism. To an extent, these issues can be addr essed by better system transparency, where buyers can see all the other orders, and pric e-conditional orders, in which a buyer places an order at a maximum price at which it can be, executed (Tan et al. 2007). Indeed in such a mechanism it has been shown (Che n 2002) that the optimal bid price for a prospective buyer is influenced mainly by he r reservation prices and not by when the
32 buyer comes into the market or what other pros pective buyers do. Still, to date, it has not been shown if making these changes result ed in any successful retail group-buying market. However, they stand as independe nt design suggestions worthy of greater exploration. There has also been case-study research that looked at a variety of firms (Kauffman and Wang 2002). In this work some additional conjectures as to why early online retail group-buying did not work include (1) The critical mass effect Â– perhaps early group buying implementations online did not each have enough transaction volume to better compete with larger retail wholes alers who do better with bulk discounting (2) The product variety issue Â– perhaps there was too much product variety in online retail, leading to order fragmentation, making it ha rder to Â“groupÂ” similar orders (3) The product lifetime effect perhap s group buying may work for Â“newerÂ” products that attract many customers, suggesting that this may be a pricing mechanism that may be used once early in the lifetime of a product and (4) The se arch cost effects perhaps current search (engine) options favor fixed price mechanisms since there is no way of determining the final price for an ongoing group purchase o ffer until the price discovery process is completed. Many of the above are also ec hoed in Tang (2008) where many failed groupbuying schemes in the US are compared to an emerging group-buying scheme in China called Â“tuangouÂ”. So when exactly can this be superior to the traditional posted fixed-price offers in the retail space? There have been two atte mpts to address this, coming from different perspectives. Anand and Aron (2003) as well as Chen (2004) study this from a sellerÂ’s perspective. Under various stylized assu mptions both these papers study analytical models that compare group buying to the trad itional fixed price mechanism. Anand and Aron (2003) argue that group buying in retail may be superior to fixed price mechanisms when one of two different factors come in. Fi rst, if there is sufficient demand uncertainty then sellers may benefit from using a gr oup buying mechanism. Demand uncertainty is usually rare for a common produc t that is well-unde rstood, hence this again suggests that there may be specific markets where the mechanism might better work in. Second, a
33 combination of economies of scale and produc tion postponement matter. Specifically, if the seller realized scale ec onomies and can wait until demand aggregation occurs before production then group buying can be superior Chen (2004) extend these insights and argue that the sellerÂ’s pencha nt for risk matters as well, with higher risk favoring the group buying outcome. In their work, Chen (2004) characterizes higher risk as a greater preference for selling higher volumes. What if, from a sellerÂ’s perspective, demand uncertainty and the ability to postpone production do not exist? Can there st ill be conditions under which the seller might benefit from a group buying mechanism on line? This is an important question that needs to be examined in future research. A totally different take on this has been from the buyerÂ’s side, where research has looked at how buyers can be efficiently grouped to make group-buying work. Hyodo (2003) and Li (2004) study mechanism design for coalition formation Â– i.e. how can groups be formed efficiently? Using a game theoretic perspective Li (2004) show the difficulty in designing efficient mechanis ms, arguing that an incentive compatible mechanism that induces buyers to reveal the truth may be elus ive in some situations. On the other hand, using a very different com putational approach, Hyodo (2003) show how Genetic Algorithm-based simulations can be us ed to optimally group buyers into clusters for specific product categories. To an extent th is shows that research into the dynamics of social behavior can play a role in determin ing when and how this pricing mechanism will work online.
34 Chapter 4: A Social Recommendation Algorithm This chapter examines recommendation algorithms and discusses how to enhance recommender systems by utilizing social data. 4.1 Examining Collaborative Filtering Collaborative filtering systems are one of the most popular types of recommender systems. It works by representing a customer as an item vector (Linden et al 2003). This system then finds customers most similar to a given user to make item recommendations. See Table 4 as an example. Table 4 Collaborative Filtering Item 1 Item 2 Item 3 Item 4 Item 5 Item 6 Item 7 Customer 1 0 1 0 1 1 0 1 Customer 2 1 1 1 0 1 0 0 Customer 3 1 0 0 0 0 1 0 Customer 4 0 0 1 0 1 0 0 Customer 5 1 1 1 1 0 1 0 Customer 6 0 0 0 0 1 1 1 Customer 7 1 0 1 1 0 1 0 The recommendations are made based on similarity between customers, where 1 represents a purchase. The system uses the cosine measure (Linden et al 2003) to find a similarity coefficient between 0 and 1 wh ere zero equates to no similarity and one equates to 100% similarity. These similarity numbers ma y then be used to identify similar customers. As an example, based on the Table 4, data similarity between Customer 1 and the other Customers are listed below.
35 Customer 1Â’s Similarity to: Customer 2 Similarity = 0.5 Customer 3 Similarity = 0 Customer 4 Similarity = 0.4 Customer 5 Similarity = 0.4 Customer 6 Similarity = 0.6 Customer 7 Similarity = 0.3 Therefore, the order of relevance for Customer 1 is: . Similarly, Customer 6Â’s similarities to others may be calculated as: Customer 6Â’s similarity to: Customer 1 Similarity = 0.6 Customer 2 Similarity = 0.3 Customer 3 Similarity = 0.4 Customer 4 Similarity = 0.4 Customer 5 Similarity = 0.2 Customer 7 Similarity = 0.3 Therefore the order of rele vance for Customer 6 is: Collaborative filtering is commonly used, but technically cumbersome (Linden et al. 2003). Similarity between customers tends to be sparse, so the system churns through data looking for customers to compare. This lengthens the time to provide a recommendation. Also, customers have to rate or purchase an item before a recommendation can be made, since compar isons are between customers. Further, customers have to be Â“signed inÂ” before any recommendations can be made.
36 4.2 Examining Amazon.com A method used by Amazon.com (Linden et al. 2003) takes the transpose of the collaborative filtering matrix, previously discussed. This means that Amazon represents items as a dimensional vector of customers (Linden et al 2003). In other words, the system finds items that are most similar to items being viewed by a user to make item recommendations. See Table 5 as an example. Table 5 Item-to-Item Collaborative Filtering Customer 1 Customer 2 Customer 3 Custom er 4 Customer 5 Customer 6 Customer 7 Item 1 0 0 1 0 1 1 0 Item 2 0 1 0 0 1 0 0 Item 3 1 1 0 1 1 0 0 Item 4 0 0 1 0 0 1 0 Item 5 1 0 1 1 0 0 1 Item 6 1 0 0 1 0 0 0 Item 7 0 0 0 1 0 1 0 The recommender system is looking for similar items, where 1 represents a purchase, instead of similar customers. By flipping the vector the algorithm can quickly identify which items have been purchased t ogether, correlating these as similar items (Linden et al 2003). The reason Amazon does this is to increase speed in making a recommendation. The data in a collaborative filtering system can be slow to search through because it is based on customers as opposed to items. Customer purchases tend to be sparse when compared with items. However, search ing by items that have been purchased is less tenuous since most items have been purchased at least once. Most rows
37 in the inverted matrix are likely to have a si gnificantly larger number of ones than in the original matrix. In this system each item receives a sim ilarity coefficient between 0 and 1, zero being no similarity at all and one being 100% similar using the same cosine metric (Linden et al. 2003). As an example, item sim ilarity for items 1 and 3 are listed below: Item 1Â’s similarities to: Item 2: Similarity = 0.4 Item 3: Similarity = 0.3 Item 4: Similarity = 0.8 Item 5: Similarity = 0.3 Item 6: Similarity = 0 Item 7: Similarity = 0.3 Therefore the order of re levance for Item 1 is:
- Item 3Â’s similarities to: Item 1: Similarity = 0.3 Item 2: Similarity = 0.7 Item 4: Similarity = 0 Item 5: Similarity = 0.2 Item 6: Similarity = 0.7 Item 7: Similarity = 0.4 Therefore the order of re levance for Item 3 is:
- An item that a user has purchased, rated or is currently viewing is matched to similar items in the table to provide recomme ndations to the customer. Consider a couple examples.
38 First, Customer 5 visits Amazon.com. The system has a record of this customerÂ’s purchase history. Since Custom er 5 has already purchased it em 1 and item 3, but has not purchased item 4 or item 6, the system recommends item 4 and item 6 as soon as the customer enters the website. Item 4 and It em 6 are recommended because they are the closest in similarity to the previ ous purchases made by customer 5. Second, an anonymous user comes to Amazon.com. No recommendations can be provided until that user view s a product, since there is no available purchase history. However, if this user views item 1, the recommender system will recommend item 4 to the user. If this user views item 3, the sy stem will recommend item 6. Even though there is no purchase history, viewing an item show s a level of interest. Since Amazon has already rated item similarity from purchases of previous users, the system is able to make recommendations. Amazon.comÂ’s recommender system is valuable to all users on the website. However, a critical mass is needed for th is recommender algorithm to succeed. This means that the site has to have a large number or products and a large number of customers. If the website has few product and/or few customers, the recommendations will most likely not be valuable. Is there a way to utilize a recommender system without having a critical mass? A site could pull an XML feed from a produc t database, like that of Best Buy to gain a large number of products If the site could gather a large number of products, then customers could be pulled from s ites that have a critic al mass of users. Perhaps, information from a social network could be utilized in this situation. 4.3 Examining SNACK The SNACK (Social Network in Automated Collaborative-filtering of Knowledge) begins with the collaborative fi ltering algorithm (refer to table 4), then incorporates information from the advice se ekerÂ’s social network, utilizing multiple nodes beyond the first ones (refe r to figure 17). SNACK give s a weight to the similar customers provided by the collaborative filter ing system, based on closeness in nodes to the user (Lam 2004). Let us examine the co llaborative filtering example provided above.
39 Customer 1Â’s closest customers: Customer 6Â’s closest customers: Now, let us include social network information. Figure 18 SNACK Network Nodes Since there are five levels within the networ k, let us provide a weight of 0.4 to the first level, 0.3 to the second level, 0.2 to the third level, 0.1 to the fourth level and 0 to the fifth level. That means Customer 1Â’s customer similarity profile will change in the following manner: Customer 1Â‘s modified similarities to: Customer 2 Similarity = 0.5 + 0.4 = 0.9 Customer 3 Similarity = 0 + 0.3 = 0.3 Customer 4 Similarity = 0.4 + 0.3 = 0.7 Customer 5 Similarity = 0.4 + 0.4 = 0.8 Customer 6 Similarity = 0.6 + 0.3 = 0.9 Customer 1 Customer 2 Customer 3 Customer 4 Customer 5 Customer 6 Customer 7
40 Customer 7 Similarity = 0.3 + 0.2 = 0.5 Therefore the order of relevan ce for Customer 1 changes to: Customer 1Â’s closest customers: In a similar manner, Customer 6Â’s cust omer similarity profile will change: Customer 6Â’s changed similarities to: Customer 1 Similarity = 0.6 + 0.3 = 0.9 Customer 2 Similarity = 0.3 + 0.2 = 0.5 Customer 3 Similarity = 0.4 + 0.1 = 0.5 Customer 4 Similarity = 0.4 + 0.1 = 0.5 Customer 5 Similarity = 0.2 + 0.4 = 0.6 Customer 7 Similarity = 0.3 + 0.4 = 0.7 Therefore the order of relevan ce for Customer 6 changes to: Customer 6Â’s closest customers: Customer 1 had slight variations when the social weight was added. Customer 6 had more drastic variations. In this model, Cu stomers 3 and 4 flipped places with Customers 7 and 5 from the collaborative filtering model. 4.4 Utilizing the Social Network The Amazon recommender system finds similarity between products and the SNACK recommender system weights customer similarity based on nodes in a social network. Is this the best ther e is? There is such an abunda nce of information all over the Internet, that it seems recommender syst ems are missing potentially important data points. The first thing to do is identify a community w hose intention is focused on
41 making a purchase decision. This will provid e a network of people with similar taste profiles and a plethora of data mining opportuni ties to look match produ cts with users. In any given social network a user cr eates a profile that generally includes information on taste, such as favorite books, mo vies, music, etc. There is also textual information left in comment walls, blogs and reviews that can be mined. Also, there are connections between individual people that can be utilized and connections between individuals and a user group that can be utilized. Any algorithm for a social recommender sy stem should categorize a userÂ’s tastes. So, given the information available for collect ion, the system should categorize available products, such as Horror Films and Romantic Co medies. Then assign each user a weight, based on taste, for each category. An example is given in table 6. Table 6 Customer Interest Level in Movies by Genre Customer 1 Customer 2 Customer 3 Custom er 4 Customer 5 Customer 6 Customer 7 Horror 0.1 0.2 0.9 0.2 0.3 0.1 0.7 Action 0.5 0.1 0.3 0.1 0.5 0.6 0.2 Drama 0.3 0.7 0.6 0.7 0 0.4 0.5 Comedy 0.8 0.3 0 0.3 0.9 0.8 1 Romantic Comedy 0 0.5 0.2 0.6 0.4 0.7 0.8 Sci-Fi 0.7 1 0.5 0.8 0.7 0.3 0.3 Children 0.4 0.6 0.1 0.5 0.6 1 0.4 Based on this information the system knows th at Customer 2 and Customer 4 have very similar tastes in terms of movies. Howeve r, when the two customers are compared by taste in clothing, there is little similarity. So, using taste cat egorization allows the system to make very specific recommendation. Closeness in user network nodes and user group when making a recommendation should also be considered. The SNACK al gorithm does a good job of weighting users in terms of closeness (refer to figure 18). However, the nodes should not be considered unless the users match in taste categories. So, the system already knows that Customer 2 and Customer 4 are similar in te rms of movie taste. If they are also close in terms of
42 network nodes or they are the same user group in a community, the system should give them weight when providing a recommendation. Essentially, the system needs to a viab le social network in order to provide valuable recommendations. Then the system needs to incorporate categorized taste recommendations that are weighted if users ar e close in the network or are both in a taste based user group.
43 Chapter 5: Conclusion This thesis is an exploratory study on online social shopping, which is an emerging phenomenon in e-commerce. We discussed cu rrent industry trends, social features, academic research, social pricing and recomm ender algorithms. There are a myriad of questions that need to be answered related to this phenomenon in future work. Will sites that have high social capital, but do not generate revenue outside of advertising be able to succeed long term? Could sites with brick and mo rtar stores benefit from a dding social features and if so how exactly? Which social features influence pu rchasing decisions and by how much? At what point in the purchase decision does the user utilize social features? Does a communal focus create purchase c onversions better than an individual focus? How easy is it for users to find other users with similar taste profiles? Should the system aid in connecti ng people with similar profiles? Under what circumstances could demand aggregation succeed? What factors need to be included to successfully implement social recommendation algorithms? In conclusion, the popularity of social shopping websites is clearly increasing. In industry, at least, it seems that there is a belief that soci al communication has a positive effect on purchase decisions. More research ne eds to be done to validate this hypothesis and to determine long term viability of the so cial shopping sites that exist today. Also, more research needs to be done to determine how to utilize social information to improve recommendations made by recommender systems.
44 References Cited Anand, K. and Aron, R., 2003. Group Buying on the Web: A Comparison of PriceDiscovery Mechanisms, Mana gement Science, v.49 n.11, p.1546-1562. Bergstein, B. 2008. Making Shopping a Virtual Social Experience. Retrieved December 7, 2008, from LATimes.com Web site: http://articles.latimes.com/2008 /jan/25/business/fi-socialshop25 Bonhard, P., Harries, C., McCarthy, J. and Sa sse A. 2006. Accounting for taste: Using profile similarity to improve recomme nder systems. CHI 2006 Proceedings, Social Computing 2. Bonhard, P. and Sasse, M.A. 2006. Â‘Knowi ng me, knowing youÂ’ Using profiles and social networking to improve recomme nder systems. BT Technology Journal Vol 24 No 3. Bustos, L. 2007. Social Shopping Roundup for Online Retailers. Retrieved December 7, 2008, from GetElastic.com Web site: http ://www.getelastic.com/social-shopping/ Butcher, M. 2008. Skimbit Tries to Crack the Social Shopping Model. Retrieved December 7, 2008 from TechCrunch.com Web site: http://www.techcrunch.com/2008/09/17/sk imbit-tries-to-crack-the-socialshopping-model/ Chen, J., Chen, X. and Song, X., 2002. Bidder's strategy under group-buying auction on the Internet. IEEE Transactions on Syst ems, Man and Cybernetics. Part A. Systems and Humans. v32. 680-690. Chen, J., Chen, X., and Song, X. 2007. Comp arison of the group-buying auction and the fixed pricing mechanism. Decision Support Systems 43, 2 (Mar. 2007), 445-459. Chen, Y. 2007. Herd behavior in purcha sing books online. Computers in Human Behavior 24 (2008) 1977Â–1992. Chen, Y. and Barnes, S. 2007. Initial tr ust and online buyer be havior. Industrial Management & Data Systems 2007 Vol.107 (No.1/2). Cook, 2001. Venture Capital: Where Mercata led, consumers were unwilling to follow. Seattle Post, January 12, 2001.
45 Dellarocas, C. 2003. The Digitizat ion of Word of Mouth: Pr omise and Challenges of Online Feedback Mechanisms. Management Science, 49, 10, 1407-1424. Dougherty, H. 2007. Social Shopping Still Small, but Usage Increasing. Retrieved January 5, 2009 from Hitwise.com Web site: http://weblogs.hitwise.com/heatherdougherty/2007/12/social_shoppi ng_still_small_bu_1.html Duan, W., Gu, B. and Whinston, A. 2008. Do online reviews really matter? Â– An empirical investigation of panel data Decision Support Systems 45 (2008) 1007Â– 1016. Egger, F. 2000. Â“Trust me, IÂ’m an onlin e vendor: Towards a model of trust for commerce site design. Conference on Hu man Factors in Computing Systems. Gordon, K. 2007. The Power of Social Shopping Networks. Retrieved December 7, 2008, from Entrepeneur.com Web site: http://www.entrepreneur.com/marke ting/onlinemarketing/article174746.html Hill, S., Provost, F. and Volinsky, C. 2006. Network-based marketing: Identifying likely adopters via consumer networks. Stat istical Science, 2006, Vol. 21, No. 2, 256Â– 276. Hyodo, M., Matsuo, T., Ito, T., 2003. An optim al coalition formation algorithm for electronic group buying. In SICE 2003 Annual Conference vol.3, no., pp. 34023407. Jsang, A., Ismail, R. and Boyd, C. 2006. A survey of trust and online reputation systems for online service provision. Decision Support Systems 43 (2007), 618644. Kauffman, R.J., Wang, B., 2002. Bid Together, Buy Together: On the Efficacy of GroupBuying Business Models in Internet-based Selling. In: Lowry, P.B., Cherrington, J.O., Watson, P.R. (eds.) Handbook of El ectronic Commerce in Business Society, CRC Press, Boca Raton, FL (2002). Kempe, D., Kleinberg, J., and Tardos, . 2003. Maximizing the spread of influence through a social network. In Pr oceedings of the Ninth ACM SIGKDD international Conference on Knowle dge Discovery and Data Mining (Washington, D.C., August 24 27, 2003), ACM, New York, NY, 137-146. Kim, Y. and Srivastava, J. 2007. Impact of social influence in e-commerce decision making. ACM International Confer ence Proceeding Series; Vol. 258.
46 Lam, C. 2004. SNACK: Incorporating So cial Network Information in Automated Collaborative Filtering. In Proceedings of the 5th ACM Conference on Electronic Commerce. ACM, Ne w York, NY, 254-255. Li, C., Shuchi Chawla, Uday Rajan, Katia Sycara, 2004. Mechanism design for coalition formation and cost sharing in groupbuying markets. Electronic Commerce Research and Applications, Vo lume 3, Winter 2004, Pages 341-354. Linden, G., Smith, B. and York, J. 2003. Am azon.com Recommendations: Item-to-Item Collaborative Filtering. Internet Computing, IEEE Vol 7 No. 1 pp. 76-80. McCarthy, R. 2007. The Power of Suggestion. Retrieved December 7, 2008, from Inc.com Web site: http://www.inc.com/magazine/20070201/technologyecommerce.html Nations, D. 2007. What is Social Shopping? Retrieved January 5, 2009, from About.com Web site: http://webtrends.about.com/od/web20/a/social-shopping.htm Senecal, S. and Nantel, J. 2004. The influence of online product recommendations on consumersÂ’ online choices. J ournal of Retailing 80 (2004) 159Â–169. Social Shopping. 2007. Retrieved October 11, 2008, from Wikipedia.org Web site: http://en.wikipedia.o rg/wiki/Social_shopping Sinha, R. and Swearingen, K. 2001. Comparing recommendations made by online systems and friends. In Proceedings of the DELOS-NSF Workshop on Personalization and Recommender Sy stems in Digital Libraries. Tan, C. Khim-Yong Goh1 and Hock-Hai Teo, 2007. An Investigation of Online Group-Buying Institution and Buyer Beha vior. Lecture Notes in Computer Science, Springer Berlin / Heidelberg Volume 4553/2007 Tang, C., 2008. United We May Stand. Sl oan Management Review, May 12, 2008. Tedeschi, B. 2006. Like Shopping? Social Networking? Try Social Shopping. Retrieved December 7, 2008, from NYTimes.com Web site: http://www.nytimes.com/2006/09/11/technology/11ecom.html?n=Top/News/Busi ness/Companies/MySpace.com Terveen, L. and Hill, W. 2001. Beyond recomme nder systems: Helping people help each other. In Jack Carroll (Ed.), HCI in the New Millennium. Addison Â– Wesley.
47 Webster, A. and Vassileva, J. 2007. Pushpoll recommender system: Supporting word of mouth. Lecture Notes In Artif icial Intelligence; Vol. 4511. Yoon, S. J. 2002. The antecedents and consequences of trust in online-purchase decisions. Journal of Interact ive Marketing 16 (2), 47-63.