| dc.description.abstract | A patent provides inventors/owners with a set of exclusive rights granted by a government to inventors to utilize it in the industry or commerce. Identifying objects in a patent can detect the problem in a XML-based patent, which improves the consistency and persuasiveness of a patent. This also leads to a better exploitation and protection of a patent. In filing patent application process, owners are required to put a number next to one object to help readers can distinguish it among other objects and interact with the drawings. Unfortunately, existing works cannot sufficiently identify all the significant objects in a patent since it is manually written by the owners with complex structures. Patents are written in specific fields with abundance of technical words, it is hard to read and may take long amount of time to completely understand.
In this thesis, we propose an object-identification approach that logically uses XML parser, Natural Language Processing (NLP) techniques in order to not only recognize objects but also detect semantic relationships among them, which contributes to enhancing the specification of patents and searching process.
Our approach is threefold. First, we suggest new method to identify all objects related to a patent to copes with two semantic problems causing inconsistent issues, which are number-concept conflicts and omitting numbers when describing concepts. Second, we also comes up with an approach to detect and extract part-of relationship among objects based on linguistic features, which are preposition-based patterns and verb-based patterns. The final results of the approach are identified objects with their coordinating numbers, an edited detail description and part-of relationships among those objects, which gives a high-level specification to the patent | en_US |
