Skip to main content
main-content

Über dieses Buch

This book describes research performed in the context of trust/distrust propagation and aggregation, and their use in recommender systems. This is a hot research topic with important implications for various application areas. The main innovative contributions of the work are: -new bilattice-based model for trust and distrust, allowing for ignorance and inconsistency -proposals for various propagation and aggregation operators, including the analysis of mathematical properties -Evaluation of these operators on real data, including a discussion on the data sets and their characteristics. -A novel approach for identifying controversial items in a recommender system -An analysis on the utility of including distrust in recommender systems -Various approaches for trust based recommendations (a.o. base on collaborative filtering), an in depth experimental analysis, and proposal for a hybrid approach -Analysis of various user types in recommender systems to optimize bootstrapping of cold start users.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Introduction

Abstract
Although the saying above is an old one, it is surprisingly applicable to the Information Age we are living in now. We are flooded with social networking sites on which we can manage our friends, relatives, or business relations. Some of them are merely used to keep track of our acquaintances, but others can be quite convenient for other purposes too, think e.g. of the business oriented social networking tool LinkedIn or the image hosting website and online community Flickr. Many other useful applications will follow in the next sections.
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 2. Trust Models

Abstract
Multi-agent systems consist of a large number of intelligent, interactive and (partially) autonomous agents that must cooperate to complete a certain task, often too difficult to solve for an individual agent. Such systems are used in a wide range of applications, ranging from mobile environments [73], over the creation of crowd-related effects for movies1, to online trading [57]. Multi-agent systems can often benefit from a trust system, especially when the circumstances do not allow for perfect information about the interaction partners’ behavior and intentions [117]. They may for example incorporate a trust network to monitor and control the behavior of the agents that participate in a process, think e.g. of an online market place such as eBay. Another nice illustration can be found in [66], in which a trust network is used to alleviate the problem of corrupt sources in peer-to-peer file-sharing networks by keeping track of the peers’ trustworthiness. With the advent of the Semantic Web [12], even more applications and systems will need solid trust mechanisms. The Semantic Web is an extension of the current web where content is annotated (see RDF2 and OWL3) such that machines and computers are able to understand its meaning and reason with it. Hence, since more and more intelligent agents will take over human tasks in the future, they also require an automated way of inferring trust in each other, see for instance [123].
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 3. Trust Propagation

Abstract
How often do you not hear someone exclaiming “it’s a small world”, astonished at bumping into somebody he/she thought almost impossible to meet, or at discovering they have a mutual acquaintance. This ‘small world’ idea dates back ages, and has ever since found its way into our daily lives and even our popular culture; think for example of the movies1 and the song lyrics2. And it is not only an established expression in the English language: the Spanish compare the world to a handkerchief3, and the Namibians have an old, even more poetic proverb saying that it is only the mountains that never meet4. Despite their old age, these sayings are still remarkably up-to-date and widely applicable; not only in our everyday life, but also online.
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 4. Trust Aggregation

Abstract
In the previous chapter, we thoroughly discussed the trust propagation problem. However, besides propagation, a trust metric must also include an aggregation strategy. After all, in large networks it will often be the case that not one, but several paths lead to the user for whom we want to obtain a trust estimate. When this is the case, the trust estimates that are generated through the different propagation paths must be combined into one aggregated estimation; see for instance the situation depicted in Fig. 4.1.
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 5. Social Recommender Systems

Abstract
The wealth of information available on the web has made it increasingly difficult to find what one is really looking for. This is particularly true for exploratory queries where one is searching for opinions and views. Think e.g. of the many information channels you can try to find out whether you will love or hate the first Harry Potter movie: you may read the user opinions on Epinions.com or Amazon.com, investigate the Internet Movie Database1, check the opinion of your favorite reviewers on Rotten Tomatoes1, read the discussions on a Science Fiction & Fantasy forum2, and you can probably add some more possibilities to the list yourself. Although today it has become very easy to look up information, at the same time we experience more and more difficulties coping with this information overload. Hence, it comes as no surprise that personalization applications to guide the search process are gaining tremendous importance. One particular interesting set of applications that address this problem are online recommender sytems [2, 15, 121, 125, 138].
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 6. Trust and Distrust-Based Recommendations

Abstract
When a web application with a built-in recommender offers a social networking component which enables its users to form a trust network, it can generate more personalized recommendations by combining content from the user profiles (ratings) with direct and/or propagated and aggregated information from the trust network. These are the so-called trust-enhanced recommendation systems. As we will explain later on, to be able to provide the users with enough accurate recommendations, the system requires a trust network that consists of a large number of users: the more connections a user has in the trust network, the more recommendations can be generated. Furthermore, more trust connections create more opportunity for qualitative or accurate recommendations. Hence, it is important to trust as many users as possible. However, at the same time, the trust connections you make should reflect your real opinion, otherwise the recommendations will become less accurate. In other words, on the one hand it is advisable to make many trust connections, but on the other hand you need to pay enough attention to which people you really want to trust; in some cases, even distrust can be beneficial for the quality of the recommendations you receive. Consequently, every user needs to find the right balance to get the best out of a trust-based recommendation system.
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 7. Connection Guidance for Cold Start Users

Abstract
In Chap. 5, we briefly discussed the most common limitations of recommender systems. In this chapter, we go more deeply into one of their main challenges, namely the user cold start problem. Due to lack of detailed user profiles and social preference data, recommenders often face extreme difficulties difficult to generate in generating adequately personalized recommendations for new users. Some systems therefore actively encourage users to rate more items. The interface of the online DVD rental service Netflix for example explicitly hides two movie recommendations, and promises to reveal these after the user rates his most recent rentals. Since it is very important for e-commerce applications to satisfy their new users (who might be on their way to become regular customers), it does not come as a surprise that the user cold start problem receives a lot of attention from the recommender system community.
Patricia Victor, Chris Cornelis, Martine de Cock

Chapter 8. Conclusions

Abstract
With the advent of e-commerce applications and the ever growing popularity of social networking tools, a novel kind of recommender systems has been born; the so-called trustenhanced recommenders which infer trust information from the social network between their users, and incorporate this knowledge into the recommendation process to obtain more personalized recommendations. Since the pioneering work of Jennifer Golbeck and Paolo Massa, research on trust-based recommendations is thriving and attracts and inspires an increasing number of scientists around the world. In this book, we contributed to some of the most recent and exciting developments in this still nascent domain, namely the potential of distrust, recommendations for controversial items, and connection guidance for cold start users.
Patricia Victor, Chris Cornelis, Martine de Cock

Backmatter

Weitere Informationen

Premium Partner

    Bildnachweise