What have you tried so far? I'd simply parse out the information with a regular expression (untested):

use strict;

for my $item (<DATA>) {
    chomp $item;
    if ($item !~ /^\\bibitem\[.*?\]   # skip the command and optional 
+argument
                   \{([A-Z\d]+)\]     # extract the bibliography item,
+ all caps letters
                   \s*(.*?)           # extract the author list
                   \((\d{4})\)        # extract the year
                   (.*?)              # extract the publication title
                   \{\\it\s(.*?)\},   # extract the category
                   \s*\{\\bf\s(\d+)\},# extract the volume
                   \s*(\d+-\d+)       # extract the page numbers
                   \.                 # a dot at the end
                 /x) {
        print "%% Ignoring weird bibliography item: >$item<\n";
        next;
    };
    printf <<'TEX',$1,$2,$3,...
@article{%s,
    ...
TEX
};
__DATA__
\bibitem[Ao {\it et~al}., 2004]{AGKM04} Ao, W. {\it et~al.} (2001) Som
+e long title 1, {\it Science}, {\bf 305}, 1743-1746.
\bibitem[Bailey and Elkan, 1995]{BE95} Bailey, T. Elkan, C. (1995) Som
+e long title 2 {\it Machine Learning}, {\bf 21}, 51-80.
[download]