Scientometrics
Faramarz soheili; shiva shirdavani
Abstract
Background and Objectives: Citation is an important element in scientific writing and has a prominent role in the production and dissemination of information. Citation analysis is one of the applications of citation that examines the relationship between citing and cited document and studies the rules ...
Read More
Background and Objectives: Citation is an important element in scientific writing and has a prominent role in the production and dissemination of information. Citation analysis is one of the applications of citation that examines the relationship between citing and cited document and studies the rules governing this relation. Citation analysis of patents is nowadays frequently observed in studies and is an important tool for identifying and analyzing the technical knowledge within patents. This research tries to identify determinants of patent citations by using a survival analysis.
Methodology: Research method was patent citation analysis. Research population consisted of 25,392 patents in the USPTO database in the Data Processing: Artificial Intelligence (AI) (Category: 706). There were 25392 AI patents in the USPTO database, among them, 13644 patents remained after deleting duplicates. Therefore, the new list, containing 13,644 patents, was saved in a plain text file by patents number, and these file was used to download the by Ravar Premep software. By using this software and using data mining technique, patents were saved in a separated location whit HTML format. Information needed for this analysis was: patent number, filed year, issue date, title, abstract, inventors, assignee, citations, and categories. This information was extract from patents in a comprehensive file. There were about 80 patent applications in the field at the USPTO, with the manually extracted information from the database added to this comprehensive file. To classify patent, first, patent that did not receive any citation in the field of AI were grouped into separate group and then the rest of the patents were divided into 3 groups using Bradford law based on number of citations. Group 1: Patents that have received at least 22 citations (412 patents).
Group 2: Patent applications that have received at least 9 and at most 21 citations (1123 patents). Group 3: Patents that have received at least 1 and up to 8 citations (4975 patents). Group 4: Patent Letters That Did Not Receive Any Citation (7134 patents). In this study 8922 inventors were retrieved as the first inventors for 13,644 patents. Since the classification of the first inventors did not fit well with any of the conventional classification methods in scientometrics, the inventors were divided into three groups according to the number of inventions. Inventors with more than 20 patents in the first group, inventors with 11 to 20 inventions in the second group and inventors with 10 inventions and fewer in the third group. In the field of AI, 2898 assignees hold 13644 patents. Bradford law was used to classify assignees in terms of frequency of inventions because of their high fit to the data. According to this method, assignees were classified into three groups in terms of their ability to produce patents. First Class (Strong): assignees who have produced at least 87 patents (19 assignees). Second Class (Medium): assignees who have produced at least 7 patents (233 assignees). Third Class (Poor): Proprietors who have produced at least 1 patent (2646 assignees). Bradford law was also used to classify assignees in terms of citation. Since in this method the minimum number is equal to one, assignees whose patents did not receive any citations were placed in a separate group (Group 4) and patent holders whose patents received at least one patent. They were classified into 3 groups according to Bradford method. Cox and Kaplan-Meier regression were used to analyze the data.
Findings: The research findings showed that of the 6749 patents that received first citation, in the first two years, the probability of citing each patent was less than 50 percent, and the probability of being cited increased over time. So that probability has risen to over 90% after 84 months. Kaplan-Meier test results showed that it takes between 38 and 40 months on average to a patent obtain first citation in the field of AI. The Kaplan-Meier test results in survival analysis showed that one year after a patent first was cited, the probability of being cited for the ninth time was 2.2%, and that probability increased over time. The findings also showed that on average, a patent may be cited for the ninth time between 69 and 74 months after receiving its first citation. Kaplan-Meier test showed that in the first 4 months after receiving the ninth citation, the probability of receiving the 22th citation for patent applications was zero, and then the probability increased. Also the finding showed that of the 6749 patents reviewed citation for the first time, in the first 2 years, the probability of citing each patent was less than 50%, and the probability of being cited increased over time. So the probability has risen to over 90% after 84 months.
Discussion: The results of the Cox test showed that, at the significant level of 0.05, inventors and assignees were influenced the productivity and receiving citation of patents. The results showed that the frequency distribution of citations received by patents based on the first inventor and proprietor complies with Bradford’s scattering law. The results showed that the percentage of cited patents increased logarithmically over time. In other words, after a few years, the chances of citing a patent are reduced to a fixed amount. The results of the relationship between the power of first inventors both in terms of productivity and receiving citation showed that, the more the first inventors had a patent or citation, the shorter length of time need to be cited for the first time. The results of Cox test showed that at the significant level of 0.05 percent, inventors and assignees were influential in productivity and receiving citation. In other words, the chance of getting the first citation of assignees whose first inventor is in the first group (strong group) is 1.926 times higher than the patent of their first inventor in the third group. The results also showed that the probability of receiving the first citation for patents whose first inventors were in the first group was 1.925 times higher than that of the third group. Also those patent that the inventors where in second groups 1.44 times higher than of the third group. The results also showed that assignees were influenced by both the time they produced the invention and the citation received. The more assignees have had more or patents or citations, the less time it takes for them to be cited for the first time. The results also showed that the strongest assignees groups were more likely to receive their first citation at the same time interval than the weaker assignees. It may be argued that the role of countries in citing to patent applications is not very influential, but other factors are influential, one of which being the relevance of other patents that the inventor may cited or the evaluator of that role. And the type of citations in patent differs from those based on articles and other scientific documents.
