Related article search provides an effective browsing tool for PubMed users, allowing them to navigate the document collection without explicitly issuing queries. Any given MEDLINE citation is connected to a number of related articles, which are in turn connected to even more related articles, and so on. Thus, any single citation represents a node in a vast related document network defined by content-similarity links. We explore the hypothesis that these networks can be exploited for document retrieval, in the same manner as hyperlink graphs in the Web environment.
The PubMed search engine provides the context for this work. Whenever the user examines an abstract in PubMed, the right panel of the browser is automatically populated with titles of articles that may also be of interest, as determined by a probabilistic content-similarity algorithm [3]; see Figure 1 for an example. In other words, each abstract view automatically triggers a related article search: the top five results are integrated into a "Related Articles" panel in the display. Note that although MEDLINE records contain only abstract text, it is not inaccurate to speak of searching for articles since PubMed provides access to the full text when available. We use "document" and "article" interchangeably in this article.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment