Accepted Papers - Day 2 (Thursday)

Plenary Track - User Generated Content (10.45-12.15)

Who are the Most Influential Users in a Recommender System?

Mohammad Amin Morid, Mehdi Shajari, and Alireza Hashemi Golpayegani

Abstract: Collaborative filtering (CF) is a popular method for personalizing product recommendations for e-commerce applications. In order to recommend a product to a user and predict her preference, CF utilizes product evaluation ratings of the like-minded users. This process of finding the like-minded users causes a social network to be formed among all users. In this social network, each link between a couple of users presents an implicit connection between them. Here, there are some users who have more connections with others and are called the most influential users. This paper attempts to model and analyze the behavior of these users by employing data mining techniques. First, the most important features which present a user’s influence were selected with a linear regression method, and then, the modeling was performed by a decision tree. Based on our results, the most influential users are users who show more interest to rate more than average number of items with low frequency. Moreover, other most influentials are users who rate in moderation items which have been seen in moderation. In addition, these items are rated with good degree of agreement with other users’ rates on the items. We achieved a high accuracy with this model.

A Multi-Agent Prediction Market based on Partially Observable Stochastic Game

Janyl Jumadinova and Prithviraj Dasgupta

Abstract: We present a novel, game theoretic representation called POSGI (partially observable stochastic game with information) for distributed information aggregation using a multi- agent based prediction market model. We then describe a correlated equilibrium (CE)-based solution strategy for this game which enables each agent to dynamically calculate the prices at which it should trade a security in the prediction market. We have extended our results to risk averse traders and shown that a Pareto optimal correlated equilibrium strategy can be used to incentively truthful revelations from risk averse agents. Simulation results comparing our CE strategy with five other strategies commonly used in similar markets, with both risk neutral and risk averse agents, show that the CE strategy improves price predictions and provides higher utilities to the agents as compared to other existing strategies.

Mining Millions of Reviews: A Technique to Rank Products Based on Importance of Reviews

Kunpeng Zhang, Yu Cheng, Wei-keng Liao, and Alok Choudhary

Abstract: As online shopping becomes increasingly more popu- lar, many shopping web sites encourage existing cus- tomers to add reviews of products purchased. These re- views make an impact on the purchasing decisions of potential customers. At Amazon.com for instance, some products receive hundreds of reviews. It is overwhelm- ing and time restrictive for most customers to read, com- prehend and make decisions based on all of these re- views. Customers most likely end up reading only a small fraction of the reviews usually in the order which they are presented on the product page. Incorporating various product review factors, such as: content related to product quality, time of the review, content related to product durability and historically older positive cus- tomer reviews will have different impacts on the prod- ucts rankings. Thus, the automated mining of product reviews and opinions to produce a re-calculated prod- uct ranking score is a valuable tool which would allow potential customers to make more informed decisions. In this paper, we present a product ranking model that applies weights to product review factors to calculate a products ranking score. Our experiments use the cus- tomer reviews from Amazon.com as input to our prod- uct ranking model which produces product ranking re- sults that closely relate to the products sales ranking as reported by the retailer.

Track A - Information Uncertainty in Online Markets (14.00-15.30)

Online information search and utilization of electronic word-of-mouth

Essi Pöyry, Petri Parvinen, Jari Salo, Hedon Blakaj, and Olli Tiainen

Abstract: Research on online consumer information search behavior has typically concentrated on search-type information instead of experience information. This article focuses on electronic word- of-mouth (eWOM) as a source of experience information, and we study the relationships between the antecedents, amount and utilization of eWOM searched. Using survey data from 1660 customers of two travel agencies, we find that 1) the search for eWOM differs distinctively from the search for marketer- generated online content, and 2) the more eWOM is being searched, the less it is being utilized in the final purchase decision.

Demographic Factors in Assessing Perceived Risk in Online Shopping

Anthony Griffin, Dennis Viehland

Abstract: Research in online shopping was the focus of many studies as the age of electronic commerce began in the late 1990’s. More recently, research in this area has declined, even as shopping on the Internet continues to increase and now dominates some product categories. This research offers a timely update on this literature by investigating online shopping from a perceived risk perspective. The results find that overall perceived risk is low with only some consumer concerns in psychological, time and performance risk. Analysis of perceived risk across six product categories and four demographic factors finds a significant level of perceived risk for lower income individuals when purchasing consumer electronics, but not in any other construct examined in this research. Overall, this study provides empirical evidence to substantiate the common perception that perceived risk in online shopping is declining and does not differ greatly across product category or demographic factor.

An Empirical Study on Quality Uncertainty of Products and Social Commerce

Kyunghee Lee, and Byungtae Lee

Abstract: With the advance of Social Network Service (SNS) like Facebook, Social Commerce (SC) such as Groupon now prospers, which provides daily deals at a highly discounted price by gathering buying power of consumers through SNS. From the perspective of quality-uncertainty, it is unusual to sell experience and credence goods/services on the internet as Groupon does. In traditional E- commerce (EC) purchasing decisions rely on information provided after the actual use of products by other consumers, while in Groupon it heavily depends on opinion even before purchasing. For example, traditional sites use a third-party recommendation including feedback mechanism, while Groupon encourages consumers to post and share their preference on goods/services over SNS. Given this difference, focusing on the effect of SNS, we collect and analyze changes of sales for deals Groupon provided, using an econometric model that reflects our understanding of consumer behavior in the presence of different degrees of quality-uncertainty. The information from SNS is captured by using a function called ―Facebook Like‖ that is a recommendation system in which suggestions are brought by one’s friends, and it is a module that can be installed in any website. In this study, we demonstrate that the information from SNS positively affects sales for deals, which implies that SNS provides recommendation and encourages consumers to purchase by reducing encountered uncertainty. In addition, we also find that the effect of SNS is enlarged as the extent of the quality- uncertainty increases. This result means that under the presence of high degree of uncertainty, the information from SNS gives consumers a stronger belief in quality than information from a third-party. Besides, as many other studies proved, we also confirm that the internet turns experience goods into search goods by substituting in-store visits with virtual encounters.

Track B - Market Design & Analysis (14.00-15.30)

A Model of Peer-to-Peer (P2P) Social Lending in the Presence of Identification Bias

Frederick J. Riggins, and David M. Weber

Abstract: The Internet has created new opportunities for peer-to-peer (P2P) social lending platforms to emerge which have the potential to transform the way microfinance institutions (MFIs) raise and allocate funds used for poverty reduction. Depending upon where decision making rights are allocated, there is the potential for identification bias whereby lenders may be motivated to give to specific projects with which they have a personal interest or affinity without regard to whether or not it represents a particularly sound financial investment. In this paper, we present an analytical model where an individual lender can use a P2P social lending network to provide funds to entrepreneurs seeking funding in developing nations. We show that in the presence of identification bias, the P2P social lending network can be used to increase overall contributions for poverty reduction despite the fact that such a network may result in inefficient allocation of funds. Even so, in the presence of strong identification bias this inefficient mechanism can result in improved poverty reduction through the provisioning of financial services in the microfinance industry.

Stakeholder Interaction and Internet Auction Outcomes: Analyzing Active Disclosure

Ananth Srinivasan, and Fangxing Liu

Abstract: Better understanding of information asymmetry in internet auctions by researchers has led to improved online auction designs, increased market efficiency, and therefore better outcomes for all stakeholders. In this paper we focus on a specific aspect of internet auctions that has not received much attention in the literature: the presence of interaction between buyers and sellers while an auction is in progress – so called “live interaction”. Examples of such interaction include characteristics of questions from buyers, answers and other disclosures from sellers, conversation threads, etc. We believe that such interaction differentiates active disclosure from seller volunteered information (passive disclosure). The facilitation of live interaction is a feature of the auction site TradeMe which represents 60% of all internet traffic in New Zealand. We collected data from 532 auctions of used cars over a three week period. In addition to data about the auction itself, we collected data about the number of questions asked and answered, the average length of the questions and answers, number of conversation threads and the use of specific textual triggers in the questions that encompass sentiments such as intent, politeness, and courtesy words. We modeled the problem using logistic regression to isolate live interaction based determinants of auction outcomes. We studied the effects of live interaction variables on outcomes in two ways: by themselves; and embedded in a larger model encompassing more traditional auction characteristics. The results show which specific aspects of live interaction between buyers and sellers are significant in determining auction outcomes. We propose that such interaction is greatly facilitated by the use of mobile devices and building it in as a necessary design feature can produce superior outcomes.

Automated Analysis of Weighted Voting Games

Shaheen Fatima, Michael Wooldridge, and Nicholas Jennings

Abstract: Weighted voting games (WVGs) are an important mechanism for modeling scenarios where a group of agents must reach agreement on some issue over which they have different preferences. How- ever, for such games to be effective, they must be well designed. Thus, a key concern for a mechanism designer is to structure games so that they have certain desirable properties. In this context, two such properties are PROPER and STRONG. A game is PROPER if for every coalition that is winning, its complement is not. A game is STRONG if for every coalition that is losing, its complement is not. In most cases, a mechanism designer wants games that are both PROPER and STRONG. To this end, we first show that the problem of determining whether a game is PROPER or STRONG is, in general, NP-hard. Then we determine those conditions (that can be evaluated in polynomial time) under which a given WVG is PROPER and those under which it is STRONG. Finally, for the general NP-hard case, we discuss two different approaches for over- coming the complexity: a deterministic approximation scheme and a randomized approximation method.