Studies On Patent Evaluation And Prediction Based On Associations

With the development of economic globalization, knowledge and useful information have seen an explosive growth, and the ratio of knowledge-based industries in the domestic economy is continuously increasing. Intellectual property has become the competitive focus between countries and enterprises. As one kind of the most important intellectual properties, patent is the carriers of scientific and technological knowledge and the largest technical information source in the world. According to related statistical data from the World Intellectual Property Organization (WIPO), it shows that 90%to 95%of the inventions and research results in the world could be found in the patents each year. If the patent information is effectively used, not only the research level and the starting point of research projects can be improved, but also 60%of the research time and 40% of the research funds can be saved.Due to the importance of patents, enterprises and researchers usually search and analyze patents manually, so as to access knowledge and information. However, with the improvement of people's awareness on intellectual property protection in recent years, the number of patent applications gets a rapid growth year by year. As a result, artificial methods for patent anlaysis has been unable to meet the needs of users. Based on these reasons and the important role of patents in intellectual properties, automatic technologies for patent analysis have become one of the significant themes. Some important international conferences and organizations, such as SIGIR, ACL and NTCIR, have set up a series of of workshops for patent analysis, and a number of patent systems, such as Google Patent, SooPAT and PatFT, also provide patent search and analysis services.Though many studies have been done in patent analysis, and fruitful research results are achieved on patent classification, competitor discovery, patent trend analysis, cooperation recommendation and so on, there still exist a lot of disvantages. For example, current methods generally take patents as independent individuals, and a variety of associations in patents, such as similarity, citations and cooperations between patent applicants are always ignored.To solve the problem, this paper proposed association-based patent evaluation and forecasting techniques, in order to provide decision support for users effectively. This paper mainly includes the following four works:(1)A Latent-Citation-Network-Based Patent Value Evaluation MethodExisting patent value evaluation methods are mainly based on training or citation analysis. However, the methods based on training depend too much on the experimental parameters, which results in a weak credibility. On the other hand, the methods based on citation analysis only consider direct citations during the evaluation leaving indirect citations and published time of patent neglected. For these reasons, this paper presents a latent-citation-network-based patent value evaluation method and implements a basic algorithm to evaluate the value of each patent effectively. Further, an improved algorithm is proposed to solve the problem of inefficiency of the basic algorithm. Finally, to handle the value variations caused by the arrival of a new patent, a dynamic patent value evaluation algorithm is designed to efficiently update the value of the original patents.(2)Finding Novel Patents Based on Patent AssociationExisting methods for patent novelty evaluation generally take patents as independent individuals, whereas the associations between patents, which would influence the novelty of patents, are always ignored. To solve the problem, in this paper, we provide a novel patent finding method that utilizes associations between patents. Firstly, we construct a novelty-association network according to the similarities and citations between patents. Further, a new concept of patent novelty is defined on the network and a rank algorithm is proposed to rank the patents in the patent dataset. Finally, to handle the rank variations caused by the arrival of new patents, an update algorithm is designed to efficiently update the patent rank.(3) Predicting Competitors' Future Research Topics on Heterogeneous NetworkExisting methods for topic prediction generally predict future development trend of the whole domain through the analysis of the techniques based statistics. Although we can use these methods to analyze the development trend of a specific competitor, it would not get a good result, as the number of compeitors' published patents may be very small. In addition, some associations that would influence a company's inclination on selecting future research direction are ignored. Based on these reasons, we proposed a new method to predict the future research of competitors on heterogeneous network. In our method, we first model companies, techniques and the relationship between them as a heterogeneous network. Then, we extract topological features that could influence a company's inclination on selecting future research direction. Finally, a prediction model is constructed based on the training set and applied to calculate and rank the probability that a given company will perform research on each topic.(4)Constructing Patent Search and Analysis Prototype SystemBased on the proposed patent evaluation and prediction methods, this paper adds the functions of patent valuation evaluation, novel patents finding and future research topic prediction of competitors into the existing prototype system. By implementing the above functions, the prototype system can provide decision support in patent transactions, help users to find advanced technical information, and forecast research topics of competitors, thereby avoid competitive risk from others effectively.