The Effects of Tabular-Based Content Extraction on Patent Document Clustering

Data can be represented in many different ways within a particular document or set of documents. Hence, attempts to automatically process the relationships between documents or determine the relevance of certain document objects can be problematic. In this study, we have developed software to automatically catalog objects contained in HTML files for patents granted by the United States Patent and Trademark Office (USPTO). Once these objects are recognized, the software creates metadata that assigns a data type to each document object. Such metadata can be easily processed and analyzed for subsequent text mining tasks. Specifically, document similarity and clustering techniques were applied to a subset of the USPTO document collection. Although our preliminary results demonstrate that tables and numerical data do not provide quantifiable value to a document’s content, the stage for future work in measuring the importance of document objects within a large corpus has been set.