Research On Patent Analysis With Internet Data

Network makes the amount of data growth rapidly,and a vast amount of patent data are pouring into people's life.In order to make more accurate development strategies,companies need to understand the related patent information.But some hidden patent literature information have not been fully utilized,artificial statistics method ignores their existence and there is only some manual statistical analysis results of patents in patent reports.Therefore,the research focus on organizing patent analysis contents,we calculated related technology life cycle parameters with traditional method,extracted patent topics from text data set,achieved the automatic classification of patent documents.In order to make up for the lack of the traditional patent analysis report and the automatic writing,this study also aims to enrich the content of the patent analysis report,and to realize the automatic writing system of the report.The first problem should be solved is patent data acquisition,in order to get more relevant patent data and improve the performance of patent retrieval,invested the effect of query expansion on the results.Query expansion based on dictionary and Baidu platform,although the results were comprehensive but not accurate enough,and contrary to the results of relevance feedback.In consideration of advantages and disadvantages of each method,we combined dictionary and relevance feedback.The improved method improved all precision and recall.In order to optimize the predict of technology maturity,we introduced the concept of degree of novelty.It introduced the calculation of text similarity analysis,the result of novelty degree was between 0.4 and 0.6 from different angles of divided categories.The side shows that there is no major breakthrough technology in recent years.In order to explore the change trend of annual patent applications,use time series prediction algorithm to deal with data sequence,exponential smoothing and ARMA achieved good results,and verified technology life factors do affect the data sequence.When the patent data is obtained based on the crawler technology,in order to optimize the predict of technology maturity,we introduced the concept of degree of novelty.It is added to the calculation of text similarity analysis,and the data set from different angles of classification to calculate the degree of technological innovation.In order to explore the change trend of annual patent applications,use time series prediction algorithm to deal with data sequence,exponential smoothing and ARMA achieved good results,and verified technology life factors do affect the data sequence.Patent IPC is not the only way to get the topics,through topic extraction algorithms applying to the collected patent documents,we can get more targeted and more detailed technical topics.In this paper,we researched three algorithms:Text Rank,LDA and TFIDF.In order to measure the results of them,we use the variable of novelty degree.Text Rank algorithm achieved 0.63,the result of LDA is0.55,although Text Rank is higher than LDA,it depends on single document too much.By adjusting the initial number of topics selected by LDA,we found that the confusion degree was the smallest when the number of topics was 4.For the automatic classification of patent documents,the experimental results on the large categories are less than or equal to 0.7,the experimental results in small categories was significantly improve,the lowest value was also close to 0.7,and the R value of kNN reached 0.88.Based on the existing research results,this paper discussed the realization of patent analysis system,and the writing of patent analysis report.