<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2805">
  <Title>Learning to Recognize Blogs: A Preliminary Exploration</Title>
  <Section position="2" start_page="0" end_page="24" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> In recent years, weblogs (online journals in which the owner posts entries on a regular basis) have not only rapidly become popular as a new and easily accessible publishing tool for the masses, but its content is becoming ever more valuable as a &amp;quot;window to the world,&amp;quot; an extensive medium brimming with subjective content that can be mined and analysed to discover what people are talking about and why. In recent years the volume of blogs is estimated to have doubled approximately every six months. Technorati  report that about 11% of internet users are blog readers and that about 70 thousand new blogs are created daily. Popular blogosphere (the complete collection of all blogs) analysis tools estimate the blogosphere to contain anywhere between 20  and 24 million  blogs at time of writing. Given this growing popularity and size, research on blogs and the blogosphere is also increasing. A large amount of this research is being done on the content provided by the blogosphere and the nature of this content, like for example (Mishne and de Rijke, 2005), or the structure of the blogosphere (Adar et al., 2004).</Paragraph>
    <Paragraph position="1"> In this paper, however, we address the task of binary blog classification: given a (web) document, is this a blog or not? Our aim is to base this classification mostly on blog characteristics rather than content. We will by no means ignore content but it should not become a crucial part of the classification process.</Paragraph>
    <Paragraph position="2"> Reliable blog classification is an important task in the blogosphere as it allows researchers, ping feeds (used to broadcast blog updates), trend analysis tools and many others to separate real blog content from blog-like content such as bulletin boards, newsgroups or trade markets. It is a task that so far has proved difficult as can be witnessed by checking any of the major blog update feeds such as weblogs.com  or blo.gs.</Paragraph>
    <Paragraph position="3">  Both will at any given time list content that clearly is not a blog. In this paper we will explore blog classification using machine learning to improve blog detection and experiment with several methods to try and further improve the percentage of instances classified correctly. The main research question we address in this paper is exploratory in nature: - How hard is binary blog classification? Put more specifically,  - What is the performance of basic off-the-shelf machine learning algorithms on this task? and - Can the performance of these methods be improved using resampling methods such as bootstrapping and co-training? An important complicating factor is the lack of labeled data. It is widely accepted that given a sufficient amount of training data, most machine learning algorithms will achieve similar performance levels. For our experiments, we will have a very limited amount of training material available. Therefore, we expect to see substantial differences between algorithms.</Paragraph>
    <Paragraph position="4"> In this paper we will first discuss related work in the following section, before describing the experiments in detail and reporting on the results. Finally, we will draw conclusions based on the experiments and the results.</Paragraph>
  </Section>
class="xml-element"></Paper>