Automatic Text Summarization of Patent Documents

Abstract: This thesis investigates Automatic Text Summarization and how it can be used to generate summaries of patent documents. The purpose of the generated summary was to convey the patent document’s subject so that the reader could decide whether the patent document is relevant and should be read in full or could be discarded. This, in order to reduce the time patent attorneys need to spend reading full patent documents in their daily work. A summarizing tool using Extraction-based summarization, and a Graph-based ranking method was implemented and tested. Summaries were generated from ten patent descriptions using the implemented tool. Each summary was also evaluated using human evaluation. The method showed very promising results in summarizing patent descriptions. In addition, the results highlighted some areas of improvement for future work.