Book contents
- Frontmatter
- Contents
- Foreword
- Preface
- 1 Introduction
- PART I INTRODUCTION TO BASIC CONCEPTS
- 2 Collaborative recommendation
- 3 Content-based recommendation
- 4 Knowledge-based recommendation
- 5 Hybrid recommendation approaches
- 6 Explanations in recommender systems
- 7 Evaluating recommender systems
- 8 Case study: Personalized game recommendations on the mobile Internet
- PART II RECENT DEVELOPMENTS
- Bibliography
- Index
4 - Knowledge-based recommendation
from PART I - INTRODUCTION TO BASIC CONCEPTS
Published online by Cambridge University Press: 05 August 2012
- Frontmatter
- Contents
- Foreword
- Preface
- 1 Introduction
- PART I INTRODUCTION TO BASIC CONCEPTS
- 2 Collaborative recommendation
- 3 Content-based recommendation
- 4 Knowledge-based recommendation
- 5 Hybrid recommendation approaches
- 6 Explanations in recommender systems
- 7 Evaluating recommender systems
- 8 Case study: Personalized game recommendations on the mobile Internet
- PART II RECENT DEVELOPMENTS
- Bibliography
- Index
Summary
Introduction
Most commercial recommender systems in practice are based on collaborative filtering (CF) techniques, as described in Chapter 2. CF systems rely solely on the user ratings (and sometimes on demographic information) as the only knowledge sources for generating item proposals for their users. Thus, no additional knowledge – such as information about the available movies and their characteristics – has to be entered and maintained in the system.
Content-based recommendation techniques, as described in Chapter 3, use different knowledge sources to make predictions whether a user will like an item. The major knowledge sources exploited by content-based systems in-clude category and genre information, as well as keywords that can often be automatically extracted from textual item descriptions. Similar to CF, a major advantage of content-based recommendation methods is the comparably low cost for knowledge acquisition and maintenance.
Both collaborative and content-based recommender algorithms have their advantages and strengths. However, there are many situations for which these approaches are not the best choice. Typically, we do not buy a house, a car, or a computer very frequently. In such a scenario, a pure CF system will not perform well because of the low number of available ratings (Burke 2000). Furthermore, time spans play an important role. For example, five-year-old ratings for computers might be rather inappropriate for content-based recommendation. The same is true for items such as cars or houses, as user preferences evolve over time because of, for example, changes in lifestyles or family situations.
- Type
- Chapter
- Information
- Recommender SystemsAn Introduction, pp. 81 - 123Publisher: Cambridge University PressPrint publication year: 2010
- 4
- Cited by