Evaluating and ranking patents with multiple criteria: How many criteria are required to find the most promising patents?
Introduction
Patents contain a wealth of information about technological progress and market trends. Although patent documents have a well-defined format, they are lengthy and include many technical terms, which requires significant effort to analyze. Hence, an entire research area, called patent mining or patent analysis, aims to assist patent analysts and policy makers in finding, processing, and analyzing patents. Patent analysis can reveal market trends and novel industrial solutions that can lead to investment decisions (Campbell, 1983).
The basic tasks that a patent mining system performs are Zhang et al. (2015):
- •
Patent search and retrieval: the system searches for relevant patents from patent databases according to a keyword/phrase. Natural language processing and data mining methods have been used to improve the relevance of the returned results for a query (Konishi, 2005, Mahdabi, Keikha, Gerani, Landoni, Crestani, 2011, Sheremetyeva, 2003, Shinmori, Okumura, Marukawa, Iwayama, 2003, Stein, Hoppe, Gollub, 2012, Tannebaum, Rauber, 2012, Xu, Croft, 1996, Zhang, Liu, Li, Shen, Li, 2017a).
- •
Patent visualization: patents are presented in ways that help analysts understand key relationships and concepts. Graph theory, network analysis, and text mining methods have been applied to visualize patents that contain both structured and unstructured data (Huang, Chiang, Chen, 2003, Kim, Suh, Park, 2008, Lee, Yoon, Park, 2009, Tang, Wang, Yang, Hu, Zhao, Yan, Gao, Huang, Xu, Li, et al., 2012, Yoon, Park, 2004, Yoon, Yoon, Park, 2002, Yoon, Seo, Coh, Song, Lee, 2017).
- •
Patent evaluation: patents are evaluated according to their importance or potential. This challenging but important task can help decision makers invest in new novel industrial solutions. Natural language processing and data mining methods have been used for patent evaluation (Fujii, 2007, Hu, Huang, Xu, Li, Usadi, Zhu, 2012, Liu, Hseuh, Lawrence, Meliksetian, Perlich, Veen, 2011) (a detailed literature review is provided in Section 2). The most frequently used technique is citation analysis (Albert, Avery, Narin, McAllister, 1991, Ellis, Hepburn, Oppenhein, 1978). Although citation analysis can reveal a seminal discovery through the number of citations that a patent received, it may miss important recently issued patents.
In this paper, we propose a new methodology for patent evaluation and ranking. Our methodology takes into account several pieces of information about patents, including citation counts, pagerank, and the existence of clusters in citation graphs. The relative importance of the proposed criteria is determined through the solution of a linear optimization model that provides weights that satisfy a number of intuitive constraints amongst the proposed criteria. These weights constitute the main ingredients of a systematic patent evaluation and ranking method that utilizes the multicriteria methodology TOPSIS (Hwang and Yoon, 1981). The proposed approach represents the first multiple criteria approach to patent evaluation and ranking that does not require estimations from decision makers. Instead, we rely on common sense constraints that are easier to argue and negotiate between experts. Additionally, we investigate computationally the impact of the number of criteria used to evaluate patents. These computations reveal that at least five criteria are needed to obtain somewhat similar rankings and that none of the eight proposed criteria is redundant in the sense that their inclusion in the model leads to different rankings.
The remainder of this paper is organized as follows. In Section 2, we present a literature review on patent evaluation and ranking techniques. Section 3 details the proposed methodology. Section 4 describes a computational implementation of the proposed approach in a web-based decision support system. Section 5 presents computational experiments, including finding the most important patents for the production processes of twenty-two chemicals. Conclusions from this research are presented in Section 6.
Section snippets
Related work
Patent evaluation techniques aim to support decision makers by assessing the quality of patents. Most existing works have relied on citations to rank patents. This includes forward citations, i.e., citations received by a patent from patents granted at a subsequent point in time, and backward citations, i.e., citations given by a patent to patents granted at an earlier time. Previous studies Trajtenberg (1990), Harhoff et al. (1999) and Hall, Jaffe, Trajtenberg, 2005, Hall, Jaffe, Trajtenberg,
Patent search and retrieval
The aim of the first step of our methodology is to search and retrieve patents relevant to a given query. The Quid software (Quid Inc.) is used for this task. Quid is a business intelligence platform that assists decision making through visualization of complex and unstructured information. Among other features, Quid can find relevant patents according to a user query. Quid analyzes interactions among collected patents and represents patent relationships as a network map. The software uses
Decision support system
In order to facilitate application and validation of the proposed methodology, we have implemented a web-based decision support system (DSS). The DSS has been implemented using PHP, MySQL, Ajax, and jQuery. Fig. 2 presents the decision-making process that a decision maker can follow in order to retrieve and rank patents according to a specific query. Initially, the decision maker submits a query to find relevant patents using Quid. Then, the decision maker reviews the collected patents and
Computational experiments
We illustrate the proposed methodology to find the most novel patents that describe new processes to produce three well-known chemicals:
- 1.
ammonia process synthesis
- 2.
olefin synthesis
- 3.
polyethylene synthesis.
After exporting patent data from Quid, we upload each case to our DSS. Three different sets of weights, denoted here as sets A, B and C, were used to rank the patents for each case. Sets A and B correspond to the two solution points of the linear optimization formulation presented in Section 3.4.
Conclusions
We presented a methodology to rank patents based on multiple criteria and an intuitive linear optimization formulation that reveals how to weigh different criteria. We also implemented a web-based decision support system to automate the proposed methodology. This system was used to validate the methodology in the context of finding the most important patents that describe new processes for three well-known chemicals: (i) ammonia process synthesis, (ii) olefin synthesis, and (iii) direct
References (43)
- et al.
Direct validation of citation counts as indicators of industrially important patents
Res. Policy
(1991) - et al.
A state-of the-art survey of TOPSIS applications
Expert Syst. Appl.
(2012) - et al.
A multi-expert system for ranking patents: an approach based on fuzzy pay-off distributions and a TOPSIS–AHP framework
Expert Syst. Appl.
(2013) - et al.
Studies on patent citation networks
J. Documentation
(1978) - et al.
Valuing patents utilizing consensual dynamics in a real options framework
IIE Annual Conference. Proceedings, Institute of Industrial Engineers-Publisher
(2014) - et al.
The optimality of non-additive approaches for portfolio selection
Expert Syst. Appl.
(2011) - et al.
Building queries for prior-art search
- et al.
Patent evaluation based on technological trajectory revealed in relevant prior patents
Advances in Knowledge Discovery and Data mining
(2014) - et al.
On the development and application of a self–organizing feature map–based patent map
R&D Manag.
(2002) Systems selection by multiple attribute decision making
(1981)
Patsearch: an integrated framework for patentability retrieval
Knowl. Inf. Syst.
Consensus modeling in multiple criteria multi-expert real options-based valuation of patents
Intelligent systems 2014
Chemical process design
Patent trends as a technological forecasting tool
World Patent Inf.
Enhancing patent retrieval by citation analysis
Proceedings of the 30th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval
Chemical process systems design
Market value and patent citations
RAND J Econ
The NBER patent citation data file: Lessons, insights and methodological tools
Tech. rep
Citation frequency and the value of patented inventions
Rev. Econ. Stat.
Coa: finding novel patents through text analysis
Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
Finding nuggets in IP portfolios: core patent mining through textual temporal analysis
Proceedings of the 21st ACM International Conference on Information and Knowledge Management
Cited by (8)
Twenty years of US nanopatenting: Maintenance renewal scoring as an indicator of patent value
2023, World Patent InformationQualitative and quantitative patent valuation methods: A systematic literature review
2022, World Patent InformationCitation Excerpt :Wand and Hsieh [60] propose a system that targets institutions with heterogeneous patent portfolios such as universities and research institutes. Several MCDM studies align with the methodology but are different in terms of criteria [26,61–63]. Grimaldi et al. [64] suggest a framework in which value indicators and company manager interviews are used to value the firm's patent portfolio in terms of scores.
Industrial text analytics for reliability with derivative-free optimization
2020, Computers and Chemical EngineeringCitation Excerpt :Text analytics with maintenance records from chemical processes is a topic that has not been reported in the past. Text mining techniques have only recently been put into use in process systems applications (Ploskas et al., 2019; Zhang et al., 2019b; 2019a). In addition to human errors, the datasets we are interested in are characterized by imbalances, i.e., have significantly more instances of certain classes than other minority classes.