Citation-Based Retrieval for Scholarly Publications

queries against a database of stored, indexed documents. However, many of these search engines have proved ineffective for searching scholarly publications accurately. Researchers have developed autonomous citation indexing agents, such as CiteSeer,4 to search computer-science-related literature online. These agents extract citation information from the literature and store it in a database. CiteSeer can convert PostScript and PDF documents to text using the pstotext program from the Digital Virtual Paper project (see www.research.digital.com/SRC/virtualpaper/home. html and www.research.digital.com/SRC/virtualpaper/ pstotext.html). The citation indices created are similar to the Science Citation Index.5 As such, you can generate a citation database to store citation information. When new publications become available, the citation indexing agent can detect them and store them in the database. Citation databases contain rich information that you can mine to retrieve publications. Our intelligentretrieval technique for scholarly publications stored in a citation database uses Kohonen’s self-organizing map (KSOM).6 The technique consists of a training process that generates cluster information and a retrieval process that ranks publications’ relevance on the basis of user queries.