Online reputation mechanisms are emerging as a promising alternative to more traditional trust building mechanisms, such as branding and formal contracting, in settings where the latter may be ineffective or prohibitively expensive; a lot of electronic trading communities fall under these categories. Although a number of commercial websites already employ various forms of reputation mechanisms, rigorous research into their properties is still in its infancy. This fledgling field can benefit from past results in economics and game theory. Moreover, in order to translate the stylized results of game theory into concrete managerial guidance for implementing and participating in effective reputation mechanisms further advances are needed in a number of important areas: First, the design space of such mechanisms needs to be scoped and the effects of different design choices on performance need to be better understood. Second, the economic efficiency of various classes of reputation mechanisms needs to be quantified and compared to that of alternative mechanisms for building trust. Third, the robustness of those mechanisms against boundedly rational players, noisy ratings and strategic manipulation needs to be studied and improved. This paper surveys past results that have been derived in a variety of contexts, but which are relevant as a basis for building online reputation systems, presents two analytical models that illustrate the role of such systems in electronic markets and identifies opportunities for further MS/OR research in this fascinating area.

Recommender systems have become valuable resources for users seeking intelligent ways to search through the enormous volume of information available to them. One crucial unsolved problem for recommender systems is how best to learn about a new user. In this paper we study six techniques that collaborative filtering recommender systems can use to learn about new users. These techniques select a sequence of items for the collaborative filtering system to present to each new user for rating. The techniques include the use of information theory to select the items that will give the most value to the recommender system, aggregate statistics to select the items the user is most likely to have an opinion about, balanced techniques that seek to maximize the expected number of bits learned per presented item, and personalized techniques that predict which items a user will have an opinion about. We study the techniques thru offline experiments with a large preexisting user data set, and thru a live experiment with over 300 users. We show that the choice of learning technique significantly affects the user experience, in both the user effort and the accuracy of the resulting predictions.

This paper introduces a model for analyzing marketplaces, such as eBay, which rely on binary reputation mechanisms for quality signaling and quality control. In our model sellers keep their actual quality private and choose what quality to advertise. The reputation mechanism is primarily used to determine whether sellers advertise truthfully. Buyers may exercise some leniency when rating sellers, which needs to be compensated by corresponding strictness when judging sellers' feedback profiles. It is shown that, the more lenient buyers are when rating sellers, the more likely it is that sellers will find it optimal to settle down to steady-state quality levels, as opposed to oscillating between good quality and bad quality. Furthermore, the fairness of the market outcome is determined by the relationship between rating leniency and strictness when assessing a seller's feedback profile. If buyers judge sellers too strictly (relative to how leniently they rate) then, at steady state, sellers will be forced to understate their true quality. On the other hand, if buyers judge too leniently then sellers can get away with consistently overstating their true quality. An optimal judgment rule, which results in outcomes where at steady state buyers accurately estimate the true quality of sellers, is analytically derived. However, it is argued that this optimal rule depends on several system parameters, which are difficult to estimate from the information that marketplaces, such as eBay, currently make available to their members. It is therefore questionable to what extent unsophisticated buyers are capable of deriving and applying it correctly in actual settings.

Recently, Artificial Intelligence techniques have proved useful in helping users to handle the large amount of information on the Internet. The idea of personalized search engines, intelligent software agents, and recommender systems has been widely accepted among users who require assistance in searching, sorting, classifying, filtering and sharing this vast quantity of information. In this paper, we present a state-of-the-art taxonomy of intelligent recommender agents on the Internet. We have analyzed 37 different systems and their references and have sorted them into a list of 8 basic dimensions. These dimensions are then used to establish a taxonomy under which the systems analyzed are classified. Finally, we conclude this paper with a cross-dimensional analysis with the aim of providing a starting point for researchers to construct their own recommender system.

We consider the problem of user agents selecting processor agents to processor tasks. We assume that processor agents are drawn from two populations: high and low-performing processors with di#erent averages but similar variance in performance. For selecting a processor, a user agent queries other user agents for their high/low rating of di#erent processors. We assume that a known percentage of "liar" users, who give inverse estimates of processors. We develop a trust mechanism that determines the number of users to query given a target guarantee threshold likelihood of choosing high-performance processors in the face of such "noisy" reputation mechanisms. We evaluate the robustness of this reputation-based trusting mechanism over varying environmental parameters like percentage of liars, performance difference and variances for high and low-performing agents, learning rates, etc.

Online retailers have access to large amounts of transactional data but current recommender systems tend to be short-sighted in nature and usually focus on the narrow problem of pushing a set of closely related products that try to satisfy the user's current need. Most ecommerce recommender systems analyze a large amount of transactional data without actually having any idea of what the items in the transactions mean or what they say about the customers who purchased or browsed those items. In this paper, we present a case study of a system that recommends items based on a custom-built knowledge base that consists of products and associated semantic attributes. Our system fn'st extracts semantic features that characterize the domain of interest, apparel products in our case, using text learning techniques and populates a knowledge base with these products and features. The recommender system analyzes descriptions of products that the user browses or buys and automatically infers these semantic attributes to build a model of the user. This abstraction allows us to not only recommend other items in the same class of products that "match" the user model but also gives us the ability to understand the customer's "tastes" and recommend items across categories for which traditional collaborative filtering and contentbased systems are unsuitable. Our approach also allows us to "explain" the recommendations in terms of qualitative features which, we believe, enhances the user experience and helps build the user's confidence in the recommendations.

The complexity of product assortments offered by online selling platforms makes the selection of appropriate items a challenging task. Customers can differ significantly in their expertise and level of knowledge regarding such product assortments. Consequently, intelligent recommender systems are required which provide personalized dialogues supporting the customer in the product selection process. In this paper we present the domainindependent, knowledge-based recommender environment CWAdvisor which assists users by guaranteeing the consistency and appropriateness of solutions, by identifying additional selling opportunities, and by providing explanations for solutions. Using examples from different application domains, we show how model-based diagnosis, personalization, and intuitive knowledge acquisition techniques support the effective implementation of customer-oriented sales dialogues. In this context, we report our experiences gained in industrial projects and present an evaluation of successfully deployed recommender applications.

Recommender systems have become paramount to customize information access and reduce information overload. They serve multiple uses, ranging from suggesting products and artifacts (to consumers), to bringing people together by the connections induced by (similar) reactions to products and services. This thesis presents a graph-theoretic model that casts recommendation as a process of ‘jumping connections’ in a graph. In addition to emphasizing the social network aspect, this viewpoint provides a novel evaluation criterion for recommender systems. Algorithms for recommender systems are distinguished not in terms of predicted ratings of services/artifacts, but in terms of the combinations of people and artifacts that they bring together. We present an algorithmic framework drawn from random graph theory and outline an analysis for one particular form of jump called a ‘hammock.’ Experimental results on two datasets collected over the Internet demonstrate the validity of this approach.

Abstract not found

We describe our work on extracting attribute and value pairs from textual product descriptions. The goal is to augment databases of products by representing each product as a set of attribute-value pairs. Such a representation is beneficial for tasks where treating the product as a set of attribute-value pairs is more useful than as an atomic entity. Examples of such applications include demand forecasting, assortment optimization, product recommendations, and assortment comparison across retailers and manufacturers. We deal with both implicit and explicit attributes and formulate both kinds of extractions as classification problems. Using single-view and multi-view semi-supervised learning algorithms, we are able to exploit large amounts of unlabeled data present in this domain while reducing the need for initial labeled data that is expensive to obtain. We present promising results on apparel and sporting goods products and show that our system can accurately extract attribute-value pairs from product descriptions. We describe a variety of application that are built on top of the results obtained by the attribute extraction system. 1.