1. bookAHEAD OF PRINT
Détails du magazine
License
Format
Magazine
eISSN
2444-8656
Première parution
01 Jan 2016
Périodicité
2 fois par an
Langues
Anglais
Accès libre

Evaluation and Verification of Patent Value Based on Combination Forecasting Model

Publié en ligne: 15 Jul 2022
Volume & Edition: AHEAD OF PRINT
Pages: -
Reçu: 14 Feb 2022
Accepté: 16 Apr 2022
Détails du magazine
License
Format
Magazine
eISSN
2444-8656
Première parution
01 Jan 2016
Périodicité
2 fois par an
Langues
Anglais
Preface

According to data from the official website of the State Patent Office, China received 1.54 million patent applications, accounting for 46.4% of the global patent applications received that year in 2018 which ranks first in the world. The United States, in second place, received 597, 000 patent applications and Japan, in third place, received 314, 000, while Europe received 174, 000 applications in the same year. With the increasing popularity of domestic patent applications, there are also social demands for the evaluation of patent value. Various patent cooperation and patent transfer businesses, as well as research on patent support policies, all require the support of patent evaluation model.

In the early value evaluation, it is carried out through the regression analysis of the past transaction process which means the patent value of the actual profitability of the previous patents for the projects that have already produced patent conversions is evaluated according to the profitability model of intangible assets, or the transfer transaction price before the patent. However, the model cannot provide an effective evaluation for patents that have never had transactions and conversion cooperation. In the current patent value evaluation process from most domestic institutions, the expert subjective evaluation method is used to evaluate the corresponding value, where expert scoring method is adopted. It is too subjective and not scientific, although it can be improved by the brainstorm of a group of experts.

With the gradual promotion and application of machine learning technology and artificial intelligence technology in recent years, intelligent patent evaluation algorithms based on various artificial intelligence theories have also emerged. In particular, gray relational analysis models and rough set models as the bottom layer with the complex neural networks such as deep convolutional neural networks and multi-column neural networks for deep and learning big data analysis at the same time is used in some research, but it still comes out with a large amount of shortcomings. As mentioned above, the reliability of the analysis results is in doubt when evaluating patents with unclear value added.

This paper intends to use the state-space method to construct a patent value evaluation model, which can evaluate the value of patents that have been converted or transferred, or that have not been converted or transferred.

1. It takes utility model patents in the field of clothing design as the research object in the paper. We’ve have adopted 100 patents that have been transformed into results as the data source where 70 patents is randomly selected for state space model construction on the actual conversion income and the remaining 30 patents is used for model validation.

For the patents mentioned above, AHP characteristic factor decomposition can be performed according to Figure 1:

Figure 1

Diagram of Patent Characteristic Factor Decomposition (AHP)

In Figure 1, the audience gender (N1) can be divided into male (1) female (2) neutral (3) for assignment. The audience age (N2) should conform to the law of normal distribution, and the 20–40 years old with the strongest purchasing power should be set as the highest point, followed by expansion along the normal relationship and the integer distribution on the interval 1–5. In terms of application location, the product of the value of the accessory and the amount of the accessory is used as the state space impact factor (N3) of the application location, that is, the total output value of the application location to which the utility model patent is applied as a state space standard. The total annual output value of the conversion organization and the related output value of the accessory (the output value of the final clothing product applicable to the accessory) are defined as N4 and N5 respectively.

The conditional polynomial expansion of N2 can be obtained as follows: N2={5,Age(0,5](20,40)4,Age(5,18)3,Age(18,20][80,)2,Age(60,80)1,Age(40,60] {N_2} = \left\{{\matrix{{5,Age \in (0,5] \cup (20,40)} \hfill \cr {4,Age \in (5,18)} \hfill \cr {3,Age \in (18,20] \cup [80,\infty)} \hfill \cr {2,Age \in (60,80)} \hfill \cr {1,Age \in (40,60]} \hfill \cr}} \right.

Where: Age means the corresponding age range of N2 that comes from social questionnaire survey results and e-commerce sales platform reports;

The actual value of the patent is embodied as the entropy value of the relevant output value N5 in the total annual output N4 of the enterprise, and final value entropy S is: S=N5/N4N3/N5=N5N3N4 S = {{{N_5}/{N_4}} \over {{N_3}/{N_5}}} = {{{N_5}} \over {{N_3} \cdot {N_4}}}

Where: the higher the S value is, the higher the added value of the patent will be. On the contrary, the added value will be lower. When the entropy value is greater than 1.000, it is considered that the patent has a positive meaning for the market competitiveness of related products. Otherwise, it is negative.

However, it should be statistically correlated with the above 5 N values when the evaluation value of a patent is set as B. Then the actual evaluation model is: B=F[f1(N1),f2(N2),f3(N3),f4(N4),f5(N5)] B = F[{f_1}\left({{N_1}} \right),{f_2}({N_2}),{f_3}({N_3}),{f_4}({N_4}),{f_5}({N_5})]

Where: f1(N1), f2(N2), f3(N3), f4(N4), f5(N5) are five regression models; F() is the state-space integration model; the dimension of B is RMB Yuan;

Among the 70 data used to build the state space model, the actual value analysis model adopts the Hejun evaluation model. The final assessment value R comes from the product of the average patent added value in the past 1 year (12 monthly cycles) and 30, namely: R=30(112n=112Bn) R = 30 \cdot \left({{1 \over {12}} \cdot \sum\limits_{n = 1}^{12} {{B_n}}} \right)

Where: Bn is the total added value of the patent generated when the NTH monthly cycle is pushed forward in the current month; Dimension of R is RMB Yuan;

A preliminary statistical analysis of the data of the selected 70 patents revealed that the data analysis rules are shown in Table 1.

Basic Distribution Law of State Space Model Building Data

0 0, 000 RMB) 0, 000 RMB)
X. 7 84 34
2 572 .25
G 4 6 68
R 2 3 4

In Table 1, it is found that the highest production value entropy of related patents can reach 4.197 and the lowest can reach 0.562. Among the 70 patent data used to construct the state space, some patents have a relatively positive effect on the added value of product patents, while some patents have a relatively negative effect. Therefore, the actual appraised value of some patents turns out positive, and some negative in the process of confirming B and R. According to statistics, it has 46 positive B and R, accounting for 65.7%, and 24 negative of them, accounting for 34.3% among the 70 patents

Construction and Demonstration of State Space Model
Data Verification Calculation during Model Construction

The state space model is constructed on the actual data of the above 70 patents to, form S, B, and R for the state space curve of N1, N2, N3, N4, N5 according to the situation of both the inventor and the patent transferor during the patent evaluation process to obtain the evaluation result of the future transformation effect of the patent. In other words, the actual value of the model is not simply evaluated, but also the possible impact of patent transformation enterprises or other patent transformation institutions on the patent transformation process can be comprehensively evaluated under this model. It believes that for the same patent, different patent conversion parties may bring different actual values to the patent.

By substituting the remaining 30 known data into the above model, the results in Table 2 can be obtained.

Basic Distribution Law of State Space Model Construction Data

N=30 S B (10, 000 RMB) R (10, 000 RMB)
X 3 52 45
MIN 0.277 −36.743 −529.63
AVG 1.985 −3.471 −19.68
VAR 0.417 2.283 36.73
R2 0.996 0.997 0.995

In Table 2, their R2 are all over 0.995, which means that the 30 verification values and the 70 modeling values are considered to fit well, although the 30 verification data have a slightly lower impact point than the 70 modeling data. According to the general laws of statistics, the state space model supported by 70 modeling data can be effectively verified by 30 verification values unexpectedly.

State Space Data Visualization

Take the N2 variable, the influence variable of age adaptability of patent-related products, as an example. The actual R value distribution is shown in Figure 2.

Figure 2

Graph of Statistical Relationship between Patent Value R and Variables

In Figure 2, taking the logarithm of N2 and the R respectively, you will find that the final data distribution results show a more significant linear relationship, and R2=0.995, which is statistically significant. The relationship between R and the N2 accords with the inverse power ratio distribution. That is, there is a normal power rate distribution relationship between the reciprocal of N2 and the R. As N2 increases, its patent value increases sharply.

The value of patents is concentrated in the high N2 value range with strong purchasing power and desire.

In addition to N2 among the five control values such as N1, N2, N3, N4, N5, there is also a significant statistical relationship among other control values, and R2 is not less than 0.995 which means their statistical relationship is also significant. The distribution of the model between 70 modeling data and 30 verification data conforms to the statistical law with a significant relationship.

Artificial Intelligence Application Scenarios for Patent Value Evaluation
The Construction of Artificial Intelligence Architecture

After the data relationship constructs the state space model, its greatest application value comes from the application in the field of artificial intelligence, which can realize the patent value evaluation based on artificial intelligence machine learning based on data logic. The actual patent value estimation results after the cooperation between the patent and the institution can be obtained by inputting the patent data of the patent publication platform, and performing data integration analysis with the data of institutions that expect to transform the patent into the market. What's more, it can realize the data function through a single neural network module as shown in Figure 3.

Figure 3

Diagram of Artificial Intelligence Application Scenarios of the Patent Value Model

In Figure 3, if the estimation of patent added value is carried out by a patent conversion agency, it would search for related patents on the patent platform, extract its data, and integrate the factors to form the model required. With 5 control values such as N1, N2, N3, N4, N5, the neural network would evaluate the added value of each optional patent. If it is performed by the patent holder, it would search for the information of the patent conversion agency and extract relevant factors on the corporate information disclosure platform, such as the chamber of commerce platform or other third-party corporate information display platform. And then integrate data with its own patents to form 5 control values such as N1, N2, N3, N4, N5 required by the model for the evaluation of the added value of each optional patent through a neural network.

For this neural network, it has 5 input values, all of which can be managed as separate variables in Double format. The statistical significance of the neural network aims to form a combined prediction model for patent value evaluation based on the state space model that can accommodate the previous analysis in the artificial intelligence big data space, and use the simulation projection method to fully evaluate the patent value based on this model. After the variable is input, it undergoes hidden layer functions such as logarithmic projection, polynomial function normalization, power rate statistics, and de-fuzzy process, and finally realizes that the neural network directly outputs the patented value-added estimation data with RMB Yuan.

Where, the logarithmic projection process simulates the data pre-management process and the normalization process of polynomial function simulates the state space construction process in the model mentioned above in the model mentioned above; the power rate statistical process closely approximates the nature of the power rate function to the state space; the de-fuzzy process is used to dimensionally integrate the fuzzy data output by the neural network. The neural network architecture is shown in Table 3.

Neural Network Structure Design

Hidden layer level Node number Function Node function
1 5 Logarithmi c projection
2 11 Y=(AlogeXi+B) Y = \sum {(A \cdot {{\log}_e}{X_i} + B)}
3 17 Polynomial normalization
4 29 Y=j=05AjXij Y = \sum {\sum\limits_{j = 0}^5 {{A_j}X_i^j}}
5 37 Power rate statistics
6 29 Y=(AeXi+B) Y = \sum {\left({A \cdot {e^{{X_i}}} + B} \right)}
7 13 De-fuzzy data output
8 3 Y=(AXi+B) Y = \sum {\left({A \cdot {X_i} + B} \right)}

In Table 3, Xi is the input variable of the node, Y is the output variable of the node, and A and B are the variables to be regressed; j is the polynomial order; i is the index number; e is the natural constant, which is approximately 2.7182818; After the data processing of the neural network, you can directly extract the basic information of enterprises and institutions that can be used for patent conversion collected on the Internet for comparison by patent holders, or extract the patent information collected on the Internet for comparison by patent conversion agencies.

Empirical analysis of artificial intelligence architecture

In this paper, the above 70 modeling data are taken as the training data of the neural network, and the 30 data are taken as the verification data. By comparing the evaluation results directly using the model in Chapter 2 with the evaluation results automatically interpreted by artificial intelligence, Table 4 can be obtained.

Value R Accuracy Comparison Table of Empirical Results of Artificial Intelligence Interpretation

[] [] []
[]N=30 []Manual calculation results []Machine learning results
MAX 507.45 508.13
MIN −529.63 −527.27
AVG −19.68 −19.14
VAR 36.73 33.42
R2 0.995 0.999

In Table 4, the final interpretation results of the two interpretation modes for the R are basically the same, but the R2 of the machine learning result reaches 0.999, which is higher than the manual calculation result of 0.995, that is, the machine learning result is more consistent with the model. In view of the fact that the machine learning process does not construct a substantial state space, but only uses a neural network system that can accommodate the state space which means machine learning has better control over data than the calculation mode of manually constructing a state space model.

Summary

In the studies, the grey relational model and rough set model are used to evaluate the patent value in machine learning. However, these models all have shortcomings in the patent evaluation that has not gone through the early transformation process as discussed above. In this paper, a state space model is constructed through known patent value data, and a high-dimensional (5 control variables) data projection space is maintained. After extracting data from patent-related information and patent conversion agency information, the state-space model is projected to obtain the corresponding patent value data, especially the patent value-added income data that the patent may generate in the institution. The basic model of patent value evaluation in this article is 30 times of the average performance of patent value added in the past 12 months, that is, the commercial value evaluation model proposed by Hejun Consulting. The significant statistical results with R2 higher than 0.995 were obtained by manually calculating the model or building a neural network machine learning architecture based on the model, which means it is suitable for patent value evaluation scenarios.

Figure 1

Diagram of Patent Characteristic Factor Decomposition (AHP)
Diagram of Patent Characteristic Factor Decomposition (AHP)

Figure 2

Graph of Statistical Relationship between Patent Value R and Variables
Graph of Statistical Relationship between Patent Value R and Variables

Figure 3

Diagram of Artificial Intelligence Application Scenarios of the Patent Value Model
Diagram of Artificial Intelligence Application Scenarios of the Patent Value Model

Value R Accuracy Comparison Table of Empirical Results of Artificial Intelligence Interpretation

[] [] []
[]N=30 []Manual calculation results []Machine learning results
MAX 507.45 508.13
MIN −529.63 −527.27
AVG −19.68 −19.14
VAR 36.73 33.42
R2 0.995 0.999

Basic Distribution Law of State Space Model Building Data

0 0, 000 RMB) 0, 000 RMB)
X. 7 84 34
2 572 .25
G 4 6 68
R 2 3 4

Basic Distribution Law of State Space Model Construction Data

N=30 S B (10, 000 RMB) R (10, 000 RMB)
X 3 52 45
MIN 0.277 −36.743 −529.63
AVG 1.985 −3.471 −19.68
VAR 0.417 2.283 36.73
R2 0.996 0.997 0.995

Neural Network Structure Design

Hidden layer level Node number Function Node function
1 5 Logarithmi c projection
2 11 Y=(AlogeXi+B) Y = \sum {(A \cdot {{\log}_e}{X_i} + B)}
3 17 Polynomial normalization
4 29 Y=j=05AjXij Y = \sum {\sum\limits_{j = 0}^5 {{A_j}X_i^j}}
5 37 Power rate statistics
6 29 Y=(AeXi+B) Y = \sum {\left({A \cdot {e^{{X_i}}} + B} \right)}
7 13 De-fuzzy data output
8 3 Y=(AXi+B) Y = \sum {\left({A \cdot {X_i} + B} \right)}

Wei Jiaqi, Bai Bo. A Method for Evaluating the Value of Patent Groups Based on Grey Relational Analysis: A Case Study of Beijing T Security Technology Company [J]. Cultural Studies, 2020(07): 6–11. JiaqiWei BoBai A Method for Evaluating the Value of Patent Groups Based on Grey Relational Analysis: A Case Study of Beijing T Security Technology Company [J] Cultural Studies 2020 07 6 11 Search in Google Scholar

Guo Ling, Xie Kaicheng. R&D Enterprise Patent Value Evaluation and Grading System Construction Method[J]. China Invention and Patent, 2020, 17(07): 89–95. LingGuo KaichengXie R&D Enterprise Patent Value Evaluation and Grading System Construction Method[J] China Invention and Patent 2020 17 07 89 95 Search in Google Scholar

NTT Data Inc.; Patent Issued for Providing Quantitative Evaluations of Friction within a Customer Experience to Reduce Abandonment and Improve Conversion of Transactions (USPTO 10, 699, 287)[J]. Computer Technology Journal, 2020. NTT Data Inc. Patent Issued for Providing Quantitative Evaluations of Friction within a Customer Experience to Reduce Abandonment and Improve Conversion of Transactions (USPTO 10, 699, 287)[J] Computer Technology Journal 2020 Search in Google Scholar

Wang Youhong. Research on the Intangible Asset Value Evaluation System[J]. Value Engineering, 2020, 39(21): 11–13. YouhongWang Research on the Intangible Asset Value Evaluation System[J] Value Engineering 2020 39 21 11 13 Search in Google Scholar

Liu Wutang, Wang Xiaoran, Xiao Linzhi, Chen Cheng, Zhao Ningru. Research on Patent Value Evaluation Based on Big Data[J]. China Assets Appraisal, 2020(07): 4–11+63. WutangLiu XiaoranWang LinzhiXiao ChengChen NingruZhao Research on Patent Value Evaluation Based on Big Data[J] China Assets Appraisal 2020 07 4 11+63 Search in Google Scholar

Xie Wenjing, Bao Xinzhong, Zhang Nan. Patent Value Evaluation Based on Rough Set Theory and its Empirical Research [J/OL]. Journal of Information: 1–6 [2020-08-09]. http://kns.cnki.net/kcms/detail/61.1167.G3.20200708.1729.020.html. WenjingXie XinzhongBao NanZhang Patent Value Evaluation Based on Rough Set Theory and its Empirical Research [J/OL] Journal of Information 1 6 [2020-08-09]. http://kns.cnki.net/kcms/detail/61.1167.G3.20200708.1729.020.html. Search in Google Scholar

Liu Linlin. Patent Evaluation Model and Analysis under Litigation Risk[J]. Science and Technology Innovation and Application, 2019(05): 30–32. LinlinLiu Patent Evaluation Model and Analysis under Litigation Risk[J] Science and Technology Innovation and Application 2019 05 30 32 Search in Google Scholar

Cao Zanhua. The Evolution of Patent Evaluation Models[N]. Chinese Journal of Science, 2018-12-12(008). ZanhuaCao The Evolution of Patent Evaluation Models[N] Chinese Journal of Science 2018-12-12(008). Search in Google Scholar

Zhang Yinying. Research on the Evaluation of Patent Pledge Value[D]. Xi'an University of Architecture and Technology, 2016. YinyingZhang Research on the Evaluation of Patent Pledge Value[D] Xi'an University of Architecture and Technology 2016 Search in Google Scholar

Li Bing. Research on Patent Value Evaluation of Patent Securitization of High-tech Enterprises [D]. Beijing Jiaotong University, 2015. BingLi Research on Patent Value Evaluation of Patent Securitization of High-tech Enterprises [D] Beijing Jiaotong University 2015 Search in Google Scholar

Articles recommandés par Trend MD

Planifiez votre conférence à distance avec Sciendo