Method And System Implementation Of Patent Index Mining Based On TRIZ

As the first driving force to lead development,innovating can be used as an important measure to evaluate the core competitiveness and market capacity of enterprises.Patent is the main carrier of innovation output.The research on patent is conducive to the targeted choice of technological innovation strategy and the rational allocation of resources for enterprises.Therefore,the establishment of innovation index system to mine the technological innovation of patent text is an important issue of the times.In the current research,the definition of patent index is lack of unified specification,and the construction of patent index system is mostly completed manually.These problems lead to different definitions of indexes by different experts in different fields,which will ignore the information of the patent text itself,and the manual method will consume a lot of human resources.Based on this,this thesis defines the indicators in combination with TRIZ(Theoriya Resheniyva Izobretatel’skikh Zadatch)theory,so that the index can reflect the final problem to be solved in the patent,and proposes two methods of automatic mining of patent indicators based on machine learning.The specific research includes the following contents:(1)Combined with the analysis of patent invention based on TRIZ theory,this thesis studies the characteristics of unstructured patent text,studies and summarizes the general process of patent invention,and defines the index based on TRIZ for the first time.(2)This thesis presents two patent index mining methods based on convolution neural network and graph neural network.Among them,the former uses dependency syntax analysis to generate a dependency tree of syntactic relationships and establish domain word dependencies and modification relationships according to the characteristics of patent text index words and the interaction between technologies,and then mines nouns,verbs and adjectives which are dependent on the subject and object of the patent text as features,and uses convolution neural network to mine indicators.The latter uses the agent and patient with subject and object as features,and uses the graph neural network training model to mine indexes according to the modification relationship of context and the impact on word meaning.This thesis uses Python crawler technology to crawl a large amount of patent data,and divides it into different fields.Two methods are used to experiment respectively,and achieved good results for each field.(3)According to the above experimental results,a better method is selected.And a set of patent index mining system is designed and developed by applying system development technology based on this method.The system can help users effectively mine patent indicators,and further verify the effectiveness of the method.