olds matching, except that (1) it uses only the titles, authors
and years from each citation, and (2) it creates a clustering
that is non-overlapping. Our final method is to perform
the complete hierarchical agglomerative clustering with the
string edit distance metric. This represents a very expensive
baseline, but one that should perform accurately. All methods
are implemented in Perl, except for the existing Cora
algorithm, which is implemented in C. Experiments are run
on a 300 MHz Pentium-II with enough memory that there
is no paging activity