nach oben

2013 | Buch

Kapitel lesen Erstes Kapitel lesen

Social Web Artifacts for Boosting Recommenders

Theory and Implementation

verfasst von: Cai-Nicolas Ziegler

Verlag: Springer International Publishing

Buchreihe : Studies in Computational Intelligence

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten

Über dieses Buch

Recommender systems, software programs that learn from human behavior and make predictions of what products we are expected to appreciate and purchase, have become an integral part of our everyday life. They proliferate across electronic commerce around the globe and exist for virtually all sorts of consumable goods, such as books, movies, music, or clothes.

At the same time, a new evolution on the Web has started to take shape, commonly known as the “Web 2.0” or the “Social Web”: Consumer-generated media has become rife, social networks have emerged and are pulling significant shares of Web traffic. In line with these developments, novel information and knowledge artifacts have become readily available on the Web, created by the collective effort of millions of people.

This textbook presents approaches to exploit the new Social Web fountain of knowledge, zeroing in first and foremost on two of those information artifacts, namely classification taxonomies and trust networks. These two are used to improve the performance of product-focused recommender systems: While classification taxonomies are appropriate means to fight the sparsity problem prevalent in many productive recommender systems, interpersonal trust ties – when used as proxies for interest similarity – are able to mitigate the recommenders' scalability problem.

Inhaltsverzeichnis

Frontmatter

Laying Foundations

Frontmatter

Introduction

Abstract

We are living in an era abounding in data and information. And we human beings are ever-hungry for acquiring knowledge. However, the sheer masses of information surrounding us are making the task of dissecting noise from signal, or relevant from irrelevant information, virtually impossible.

Cai-Nicolas Ziegler

On Recommender Systems

Abstract

Recommender systems [Resnick and Varian, 1997] have gained wide-spread acceptance and attracted increased public interest during the last decade, levelling the ground for new sales opportunities in e-commerce [Schafer et al, 1999; Sarwar et al, 2000a]. For instance, online retailers like Amazon.com (http://www.amazon.com) successfully employ an extensive range of different types of recommender systems.

Cai-Nicolas Ziegler

Use of Taxonomic Knowledge

Frontmatter

Taxonomy-Driven Filtering

Abstract

One of the primary issues that recommender systems are facing is rating sparsity, resulting in a decrease of the recommendations’ accuracy.Hence, high-quality product suggestions are only feasible when information density is high, i.e., large numbers of users voting for small numbers of items and issuing large numbers of explicit ratings each. Smaller-sized, decentralized and open communities are typical for the Web 2.0. Here, ratings are mainly derived implicitly from user behavior and interaction patterns. However, these communities poorly qualify for blessings provided by recommender systems.

Cai-Nicolas Ziegler

Topic Diversification Revisited

Abstract

Chapter 3 has introduced topic diversification as an efficient means to avoid topic overfitting in our taxonomy-driven filtering approach. However, the topic diversification method can be applied to any recommender system that generates ordered top-N lists of recommendations, as long as taxonomic domain knowledge is available for the recommendation domain in question.

Cai-Nicolas Ziegler

Taxonomies for Calculating Semantic Proximity

Abstract

The two preceding chapters have demonstrated that classification taxonomies can be put to use in improving recommender systems in terms of the quality of their recommendations. The taxonomy we resorted to for all the empirical evaluations was the one from Amazon.com. Now we want to give an example how Web 2.0 taxonomies, having been crafted by collective efforts of several thousands of volunteering editors, can likewise be used to these ends.

Cai-Nicolas Ziegler

Recommending Technology Synergies

Abstract

Chapter 5 has laid out an approach to identify the semantic proximity between two named entities like brand and product names, locations, and so forth. The underlying chapter will now work on this model and put it to use for the identification and recommending of technology synergies. That is, given a set of technologies associated with different business units, we want to know which are the pairs of technologies that exhibit the largest synergies.

Cai-Nicolas Ziegler

Social Ties and Trust

Frontmatter

Trust Propagation Models

Abstract

Part II has discussed making use of taxonomies for improving recommender systems, in particular with regard to the quality of recommendations. Part III will now shift the focus from taxonomies to interpersonal trust, which abundantly manifests via the rife social networks and platforms on the Web 2.0. In contrast to taxonomies, trust is a means to address the scalability and cold-start problem of recommenders.

Cai-Nicolas Ziegler

Interpersonal Trust and Similarity

Abstract

Recently, the integration of computational trust models [Marsh, 1994b; Mui et al, 2002; McKnight and Chervany, 1996] into recommender systems has started gaining momentum [Montaner et al, 2002; Kinateder and Rothermel, 2003; Guha, 2003; Massa and Bhattacharjee, 2004], synthesizing recommendations based upon opinions from most trusted peers rather than most similar ones. Likewise, for social filtering within a spread out and decentralized recommender framework, we cannot rely upon conventional collaborative filtering methods only, owing to the neighborhood computation scheme’s poor scalability. Some more natural and, most important, scalable neighborhood selection process schemes become indispensable, e.g., based on trust networks.

Cai-Nicolas Ziegler

Amalgamating Taxonomies and Trust

Frontmatter

Decentralized Recommender Systems

Abstract

Preceding chapters, particularly Chapter 3 and Chapter 7, have presented methods and techniques based on Web 2.0 information structures that are, among other things, able to address specific issues of decentralized recommender systems.Moreover, Chapter 8 has shown that, to a certain extent, trust implies similarity and thus becomes eligible as a tool for CF neighborhood formation, which is generally performed by applying some rating-based or attribute-based similarity measure (see Section 2.3.2).

Cai-Nicolas Ziegler

Conclusion

Abstract

Undoubtedly, recommender systems are becoming increasingly popular, owing to their versatility and their ability to reduce complexity for the human user.

Cai-Nicolas Ziegler

Backmatter

Titel: Social Web Artifacts for Boosting Recommenders
verfasst von: Cai-Nicolas Ziegler
Verlag: Springer International Publishing
Electronic ISBN: 978-3-319-00527-0
Print ISBN: 978-3-319-00526-3
DOI: https://doi.org/10.1007/978-3-319-00527-0