Beyond Patent Citations as Measures of Innovative Search and Success

成百上千的研究论文使用专利引用数据来研究创新搜索和影响。然而，最近的批评提出了一些实质性的问题，即专利引文测量是否准确地反映了专利的新颖性和重要性。利用大数据技术，包括LASSO内容标签、这些标签的语义距离测量，以及对专利申请人和审查员之间的交流进行计算语言学分析，本研究将补充（甚至可能取代）现有的衡量专利重要性及其相对于其他专利发明的地位的方法。专利重要性的新衡量标准是基于有多少未来的专利申请被某项专利“阻挡”，而需要验证的假设是，“阻挡专利”是有价值的技术空间。为了描述技术空间中的专利特征，本研究将使用计算机文本分析来识别给定专利中最具区别的单词，相对于1975年以来授予的所有其他专利，并测试类似“标记”的专利在技术空间中彼此接近的假设，从而代表本地搜索。这些技术和公共数据将为创新历史、社会福利和科技政策的政策评估、发明者和受让人消除歧义、现有技术搜索以及计量经济学估计的处理和控制专利的匹配开辟研究。随着大型数据库的广泛使用，利用专利数据来研究创新及其社会和经济影响已经变得越来越普遍。这种方法的优点是易于计算，并且可以跨数百万个文档和数千种技术进行一致的度量。缺点是度量（例如，通过专利引用现有技术）与实际现象（如新颖性、经济影响、创造就业机会或专利诉讼的可能性）之间提供了较差的间接相关性。本研究运用计算社会科学，运用改进的大数据技术，结合调查验证，可以改善不足，并提供给研究者，显著提高创新与技术战略文献的研究水平。除了促进更好的研究之外，这项工作还提供了更广泛的社会效益，帮助发明人、律师、审查员和投资者进行现有技术搜索，改进专利数据库的消歧，并更快地识别重要专利。

Hundreds, possibly thousands, of research papers have used patent citation data to study innovative search and impact. Recent critiques, however, have raised substantial questions about whether patent citation measures accurately capture a patent's novelty and importance. Using big data techniques including LASSO content labels, semantic distance measures with those labels, and computational linguistics analysis of communication between patent applicants and their examiners, this research will supplement (or possibly even replace) existing approaches to measuring a patent's importance and its position relative to other patented inventions. The new measure of patent importance is based on how many future patent applications are "blocked" by a particular patent, and the assumption to be validated is that "blocking patents" are valuable pieces of technology space. To characterize patents in technology space, this research will use computerized text analysis to identify the most differentiating words for a given patent, relative to all other patents granted since 1975, and test the hypothesis that similarly "tagged" patents are close to one another in technology space, and thus represent local search. The techniques and public data will open up research in the history of innovation, policy assessments of social welfare and technology and science policy, inventor and assignee disambiguation, prior art search, and matching of treated and control patents for econometric estimation. Broader impacts The use of patent data to study innovation and its social and economic impact has become increasingly common with the widespread availability of large databases. The advantages of such approaches are easily calculated and consistent measures across millions of documents and thousands of technologies. The disadvantages are that the measures (prior art references by patents, for example) provide poor and indirect correlations with actual phenomena (such as novelty, economic impact, job creation, or potential for patent litigation). Using computational social science, this research applies improved big data techniques, combined with survey validation, that could ameliorate the disadvantages, and made available to researchers, significantly improve research in the innovation and technology strategy literatures. In addition to enabling better research, this work provides wider societal benefit in helping inventors, lawyers, examiners, and investors in prior art search, improved disambiguation of the patent database, and faster identification of important patents.