Abstract
A patent specification is a document with legal protection. It records the information of technologies and its execution method. With the effective analysis and usage of patent documents, it not only can stimulate the development of industries and progress of science, but also can catch on the technical level and the key technologies of the industry and the science, and has the newest market information in hand. However, due to the specific terms and the specific writing formats of patent document, it is hard to understand for human beings within limited time. And another problem is that, the size of the patent documents is growing rapidly. If we still rely on human to conduct the analysis of the patent documents, it will be very costly in time and human power. In this thesis, we provide an approach to automatically extract the claim structure from patent documents and show the results in visualization to help human beings to understand much more easily and effectively without wasting time and energy. There are two major goals in our research. One is the element extraction and the other is the triple extraction. In element extraction section, we use statistical method to count the frequency of NGram word and use the combination of part-of-speech to extract the element. And In triple extraction section, we develop a wrapper environment for user to wrap the training data and automatically generate its corresponding patterns with minimal user efforts, and provide an induction algorithm that is an adaptation of string edit distance to induce our training results into abstracted extraction patterns, and use the result patterns to extract the claim structure.