Mathematical Document Classification via Symbol Frequency Analysis

summary:Earlier work has examined the frequency of symbol and expression use in mathematical documents for various purposes including mathematical handwriting recognition and forming the most natural output from computer algebra systems. This work has found, unsurprisingly, that the particulars of symbol and expression vary from area to area and, in particular, between different top-level subjects of the 2000 Mathematical Subject Classification. If the area of mathematics is known in advance, then an area-specific information can be used for the recognition or output problem. What is more interesting is that although the specifics of which symbols are ranked as most frequent vary from area to area, the shape of the relative frequency curve remains the same. The present work examines the inverse problem: Given the relative frequencies of symbols in a document, is it possible to classify the document and determine the most likely area of mathematics of the work? We examine the symbol frequency “fingerprints” for the different areas of the Mathematical Subject Classification