Research And Implementation Of Evaluation System Of The Value Of Patents Based On Improved DBNs Algorithm

In the environment that many lessons for innovation and invention blow out,in the era that dominated by the knowledge economy,intangible assets like patent has become the dominant trading theme in business activities.However,in the base of patent application and authorization of growing environment,domestic technology conversion rate has not been significantly improved.The influence factors are including patent holding and buyer's personal subjective factors,national policy and industrial maturity,etc.Among them,the patent value evaluation technology and market maturity is not high is a limiting factor,Existing methods mainly include income method,cost method and market method,other methods like fuzzy evaluation and option method are still in the paper,but because of the correspondence between real value and cost of the patent is not high,the instability of long-term returns,and technology replacement cycles of the existence of uncertainty factors,often makes it hard for patent value evaluation in the process of transforming deal to assess the fair value of all parties are satisfied.In this study,we use an interdisciplinary research method,and combine computer science with econometrics,using the deep learning algorithm to realize the patent value evaluation,and implement it,to reduce the subjectivity in traditional evaluation methods,increase the accuracy.The main content of the research is as follows:(1)Through studying the literature of patent value evaluation in our country and abroad,analyzing the normal patent evaluation system and research interview,find the influence factors of patent value evaluation,then refer to the references and expert's advice,finally,we will design a reasonable patent value evaluation index system.(2)Quantify the indicator system we have just established in order to process the indicators easily by the algorithm.At the same time,we take a sample data set,design and realize preprocessing algorithm including data standardization,data reduction and deviated data cleansing.(3)Realize three kinds of algorithms in python which have a good generalization ability,they are backpropagation neural network(BPNN),support vector machine(SVM)and deep belief networks(DBNs).Then we analyze the advantages and disadvantages of three algorithms,while select a small sample standard database to compare,and use mean square error(mse),Pearson coefficient to compare them through testing them using UCI sample data sets.(4)According to the results of comparison,DBNs is the best algorithm fitting with patent value evaluation.Then we optimize the best algorithm from multiple perspectives according to its characteristics,and compare to the original DBNs network.Thus,we design the algorithm model which has a better performance in the patent value evaluation.(5)Use crawlers to obtain basic information about patents,visiting experts in various fields to evaluate and quantify the patents according to the design patent value assessment system,and establish a patent value training database.Finally complete the system based on SOA architecture,including data crawler module,algorithm iterative update module,user's module.This article finally completed the patent value evaluation system based on DBNs that we optimized,and we achieve the expected functions,use ten patents which have been sold to test them,and the test passes.