Discovering Important Bloggers based on Analyzing Blog Threads Shinsuke Nakajima Junichi Tatemura Yoichiro Hino National Institute of NEC Laboratories America,Inc Dept. of Social Informatics, Information and 10080 North Wolfe Road, Kyoto University Communications Technology Suite SW3-350, Cupertino, CA Yoshida Honmachi, Sakyo-ku 3-5 Hikari-dai Seika-cho 95014, USA Kyoto, Japan Soraku-gun Kyoto, Japan [email protected] [email protected] [email protected] labs.com u.ac.jp Yoshinori Hara Katsumi Tanaka Internet Systems Research Dept. of Social Informatics, Laboratories, NEC Kyoto University Corporation Yoshida Honmachi, Sakyo-ku 8916-47, Takayama-cho, Kyoto, Japan Ikoma, Nara, Japan [email protected] [email protected] u.ac.jp ABSTRACT 1. INTRODUCTION A blog (weblog) lets people promptly publish content (such The broadband infrastructure and ubiquitous computing as comments) relating to other blogs through hyperlinks. have created an environment in which people are continu- This type of web content can be considered as a conversa- ally online on the World Wide Web (WWW). Given this tion rather than a collection of archived document. To cap- environment, people are increasingly publishing their reac- ture ’hot’ conversation topics from blogs and deliver them tions (e.g., comments and opinions) to current events (e.g., to users in a timely manner, we propose a method of dis- news). Users may state their opinion of a current news ar- covering bloggers who take an important role in conversa- ticle, followed by other users who react to their opinions tions. We characterize bloggers based on their roles in pre- by stating a different opinion. In this sense, the web can vious blog threads (a set of blog entries comprises a con- be seen as a place for conversation rather than for archived versation),. We provide several definitions of bloggers’ roles documents. Triggered by an event, a hot conversation may including (1) agitators who stimulate discussion, and (2) quickly propagate from one site to another through the web. summarizers who provide summaries of the discussion. We A weblog, or blog for short, is a tool or web site that consider that these bloggers are likely to be useful in iden- enables people to publish content promptly. The “Glossary tifying hot conversations. In this paper, we discuss models of Internet Terms [1]” says that of blogs and blog thread data, and methods of extracting blog threads, discovering important bloggers, and acquiring “A blog is basically a journal that is available on the web. important content from their entries. The activity of updating a blog is “blogging” and someone who keeps a blog is a “blogger.” Blogs are typically updated Categories and Subject Descriptors daily using software that allows people with little or no tech- nical background to update and maintain the blog. Postings H.4.3 [Information Systems]: Information Systems Ap- on a blog are almost always arranged in chronological order plicationsCommunications Applications; H.2.8 [Information with the most recent additions featured most prominently.” Systems]: Database ManagementDatabase Applications; J.4 [Computer Applications]: Social and behavioral sci- A blog entry, a primitive entity of blog content, typically ences has links to web pages or other blog entries, creating a con- versational web through multiple blog sites. General Terms Since conventional search engines treat the web as a snap- shot of hyperlinked documents, they are not very effective Human Factors for capturing conversational web content such as blogs. A new approach is required for timely delivery of hot conver- Keywords sations over multiple blogs on the web. blog thread, important blogger, agitator To capture potentially hot conversations on the web, we Copyright is held by the author/owner(s). propose a method for discovering bloggers who take impor- WWW2005, May 10–14, 2005, Chiba, Japan. tant roles in these conversations. This information on blog- . RSS URL micro behavior of blogspace. In particular, they focused on not only the explicit link structure but also the implicit Blog site Blog site routes of transmission for finding blogs that are sources of PermaLink information. Blog entry A news site However, their purpose was not to acquire important web Blog entry Blog entry content. There are numerous reports of studies on topic detection 䋺 PermaLink A Web site 䋺 and tracking[5]. In particular, there have been several stud- Blog entry ies on first story detection (FSD). The goal of FSD is to Blog site recognize when a new topic appears that has not been pub- Reply Link PermaLink Blog entry lished previously. In this paper, we adopt a similar technique Blog entry to discriminate agitators (aspect 3 (5.2)). Allan et al. stud- Blog entry ied first story detection[6]. According to this study, when a 䋺 䋺 䋺 䋺 new story arrives, its feature set is compared to those of all 䋺 䋺 past stories. If it is sufficiently different, the story is marked as a first story; otherwise, it is not. Figure 1: Typical blog site. Though these previous studies of FSD are relevant to us, we cannot adopt these methods directly since blog threads include relations between entries based on replylinks, which ger characteristics can then be used to acquire important are different from a simple news stream. hot conversations Figure 1 shows a schematic of a typical blog site. A blog 3. BLOGS AND BLOG THREAD DATAMODEL site is usually created by a single owner/blogger and con- sists of his or her blog entries, each of which usually has There are several definitions of blogs other than the one a permalink (URL: uniform resource locator) to enable di- given in Section 1. Some are shown below. rect access to the entry. Blog readers can discover bloggers’ • From ”Web log:” A blog is basically a journal that characteristics (e.g., their interests, role in the community, is available on the web. The activity of updating a etc.) by browsing their past blog entries. If readers know blog is ”blogging” and someone who keeps a blog is a the characteristics of a particular blog, they can expect sim- ”blogger.” [7] ilar characteristics to appear in future entries in that blog. • Our goal is to develop a method of capturing hot conver- An easily updated personal website, generally updated sations by automating readers’ processes for characterizing daily and expressing opinions. [8] and monitoring blogs. • Web pages that are constantly updated with new com- In our method, an important blogger is defined on the ba- mentary and links relating to a particular topic. Often sis of his or her role in a blog thread, i.e., a set of blog entries very personal. [9] comprising a conversation on a specific topic. We think it is likely that bloggers take various roles in a thread, including That is, a blog is a website that anybody can easily update acting as (1) an agitator who stimulates discussion, or (2) a and use to express his/her own opinions in a public space. summarizer who provides summaries of the discussion. We To put it another way, blogs are a storehouse of information believe that these three types are important bloggers who that reflects public opinion. Although there is a lot of trivial are useful for identifying hot conversations. information in blogspace, there is also a lot of important We first describe related work and describe blogs and a information. model of blog thread data, and then the extraction of blog Before defining our model of blog thread data , let us threads. This is followed by a discussion of how important discuss these definitions of blog sites and blog entries. Ex- bloggers are discovered, and examples of applications of our amples are shown in Figures 2 and 3. method. We end with a summary and outline our plans for + + future work. site =(siteURL, RSS, blogger , siteName, entry ) entry =(permaLink, blogger, time, title?, description, comment∗) 2. RELATED WORK comment =(blogger, permaLink, content, time) In related work on analyzing blogspace, Kumar et al. studied the burstiness of blogspace[2]. They examined 25,000 A blog site has a site URL, RSS (really simple syndica- blog sites and 750,000 links to the sites. They focused on tion), site name, and entries, and is managed by one or clusters of blogs connected via hyperlinks named blogspaces more bloggers. A blog entry has a permalink for access, a and investigated the extraction of blog communities and the publication time, title, and entry description. A comment evolution of the communities. includes the content of the comment and the time when it Gruhl et al. studied the diffusion of information through was written. A blogger posts a blog to an entry identified blogspace[3]. They examined 11,000 blog sites and 400,000 by a permalink. links in the sites, and tried to characterize macro topic- diffusion patterns in blogspaces and micro topic-diffusion replyLink =(ei,ej ), (ei → ej ) patterns between blog entries. They also tried to model trackbackLink =(ei,ej ), (ei → ej ) topic diffusion by means of criteria called Chatter and Spikes. sourceLink =(ei,wi), (ei → wi) Adar et al. studied the implicit structure and dynamics where ei,ej ∈ E , E is a set of blog entries, and wi ∈ W , of blogspace[4]. They also examined both the macro and W is a set of Web pages except blog entries. rootEntry replyLink SiteName blogger replyLink sourceLink Blog site Blog entry Blog site Blog site RSS SiteURL Blog entry Blog entry Blog site Web site Blog entry Blog site Blog site Blog site Blog entry Blog entry Blog entry rootEntry Blog entry 䋺 Figure 4: Example of blog thread 䋺 䋺 䋺 rected connected graph and is defined as follows.