Contribuţii la aplicarea tehnicilor de data mining în inventica asistată de calculator

Abstract

The research summarized in this thesis started from a set of practical needs encountered in the process of inventions patenting. In this patenting process a series of steps are followed, including one of the most important, which is the search for information and references available to date, relevant to the applicant's invention. Currently there are many search engines in inventions or scientific literature databases, some of them even belonging to patent offices, other belonging to private companies. Few are those that offer the possibility of ordering the search results by relevance, using one of the existing specific data mining algorithms. Myself, as a patent applicant, I was put in a position to make such searches and have been forced to look through dozens of databases and hundreds of results where the vast majority proved irrelevant and I felt from my own experience the need for a more efficient tool to assist me in the search stage [1], [2], [3], [4]. Therefore, the inventors are put in a position to search for useful information through lists of hundreds of results simply ordered only by one of the existing fields in the database, usually by application date. Hence, there is a real need to find and list the relevant patent search results and a solution to this need is proposed in the present paper. Another problem faced by those involved in patenting is the right selection of the class in which the new patent will be classified. Currently there are several types of classifications used for patents, but because by definition an invention brings something new, never seen before, there will always be a chance that new classes or even new industries to appears and which sometimes these new classes or industries are "forced" classified into one of the existing classes. The standard classifications are updated periodically, but it is quite difficult to say when a new class or a new industry emerged. An independent grouping of existing classification could say if a new class appeared. The present thesis aims to help in solving this problem by using data mining techniques such as clustering and by improving and adapting them to the particular case of patents.