Patent tradeability prediction method and system based on machine learning

Abstract

The invention discloses a patent tradeability prediction method and system based on machine learning, and belongs to the field of asset assessment, and the method comprises the steps: obtaining a feature variable of a to-be-tested patent; the patent tradeability prediction model is input into the trained patent tradeability prediction model, and a patent tradeability prediction result is output; wherein the training of the patent tradeability prediction model comprises the following steps: constructing a characteristic variable data set; random Forest, GBM, XGBoost, LightGBM, Catboost and LR are adopted as prediction models, hyper-parameters of the models are preferentially selected by using a grid search method, and a prediction result is obtained; comparing evaluation results of the model evaluation indexes to obtain an optimal prediction model; and adjusting a weight value of a loss function in each model through a weight balance method, optimizing the six machine learning models, and obtaining a final patent tradeability prediction model based on an evaluation result. According to the invention, the highest prediction accuracy is realized with the least variable input, and the patent tradeability is predicted by using the existing data.

Description

Technical Field

[0001] The present invention belongs to the field of asset evaluation, and specifically relates to a patent tradability prediction method and system based on machine learning. Background Art

[0002] As an important intellectual property, the transfer of patents has a profound impact on technological development and business strategies. The transfer of patents can not only bring economic returns to inventors, but also promote the dissemination and application of technology, thereby accelerating the innovation process. However, the patent transfer market is complex and changeable, and is affected by many factors such as technological value, market demand, and legal environment. In this context, the transferability of patents has become a key indicator for assessing their market potential. It is well known that the uncertainty and unpredictability of patent transactions pose a huge challenge to both holders and potential buyers. The economic losses caused by inaccurate patent valuations are estimated to account for a significant proportion of global intellectual property transactions each year. Given the importance of patent transactions and the extensiveness of related economic activities, it is particularly important to develop a reliable and accurate prediction method.

[0003] The tradability of patents is affected by many factors, including market demand, technology maturity, legal environment, etc. These factors interact with each other, increasing the complexity and uncertainty of the prediction system. For example, rapid changes in the market may affect the demand for a certain technology, while changes in the legal environment may affect the scope of protection of the patent and its commercial value. In addition, various data directly related to patents, such as the number of citations, the number of claims, the number of patent family members, etc., can be regarded as important factors in the prediction model. The complex correlation and dynamic changes between these data have brought significant challenges to the accurate prediction of patent transferability.

[0004] In current patent evaluation practice, the prediction of a patent's tradability usually relies on the subjective judgment of experts. This method requires experts to evaluate the value of patents based on their detailed understanding of the corresponding technical field, market trends, and legal environment. Although the rich experience of experts plays a key role in this process, the strong subjectivity of this method is a disadvantage that cannot be ignored. Because the differences in opinions and understandings among experts may lead to inconsistent evaluation results. In addition, expert evaluation is inefficient when processing large amounts of patent data, especially when facing a huge patent library. This traditional evaluation method is difficult to meet the needs of fast and large-scale analysis.

[0005] Given the shortcomings of existing methods for predicting patent tradability, there is an urgent need to develop new data processing and modeling technologies to more accurately and reliably predict the possibility of patent transfer. Machine learning technology can automatically extract logic from large amounts of data and use this logic to predict or classify new data. Machine learning is particularly good at handling complex relationships between multiple variables and can continuously improve prediction accuracy as the amount of data increases, showing self-evolutionary characteristics that traditional methods do not have. Therefore, it is particularly important to develop a method that uses machine learning to process and model patent-related data to predict whether it is tradable.

[0006] Therefore, it is necessary to design and develop a patent tradability prediction method and system based on machine learning to process and model patent-related data and predict whether it is tradable in response to the above problems. Summary of the invention

[0007] The purpose of the present invention is to provide a patent tradability prediction method based on machine learning to address the problems existing in the prior art. Through hyperparameter optimization and evaluation index evaluation model, it is possible to achieve the prediction of patent tradability using existing data without relying on a large number of expert reviews, thereby greatly saving time costs.

[0008] According to one aspect of the present invention, a method for predicting patent tradability based on machine learning is provided, comprising:

[0009] Obtain characteristic variables of the patent to be tested;

[0010] Input the acquired characteristic variables of the patent to be tested into the trained patent tradability prediction model, and output the patent tradability prediction result; wherein the training of the patent tradability prediction model includes:

[0011] Identify the characteristic variables that affect patent tradability and construct a characteristic variable dataset;

[0012] Select characteristic variables that affect patent tradability from internal and external environmental factors that affect patent tradability, and construct a characteristic variable data set;

[0013] Based on the feature variable data set, the Random Forest model (RF), Gradient Boosting model (GBM), Extreme Gradient Boosting Tree model (XGBoost), Light Gradient Boosting model (LightGBM), Category Boosting model (Catboost) and Logistic Regression model (LR) were used for training, and the grid search method was used to select the best hyperparameters of each model to ensure that each model achieved the best prediction effect;

[0014] Based on the models after obtaining the optimal hyperparameters, the weight value of the loss function in each model is adjusted through the weight balancing method, the models are optimized, and the results of the evaluation indicators of each model after optimization are compared to obtain the optimal prediction model as the patent tradability prediction model.

[0015] Furthermore, the grid search method selects the best hyperparameters of each model, including:

[0016] Based on a parameter grid defining each hyperparameter and its possible values, the possible values ​​of the hyperparameters of each model are combined with the possible values ​​of other hyperparameters in each model;

[0017] For each set of hyperparameter combinations, the cross-validation method is used to evaluate the performance of each model and obtain the evaluation results;

[0018] Based on the evaluation results, the best hyperparameter combination is selected as the hyperparameters of the best prediction model.

[0019] Furthermore, the calculation formula of the weight balancing method is as follows:

[0020]

[0021] in, represents the actual category of the sample, is the probability that the model predicts the positive class, and are the weights of the positive and negative classes respectively; higher weights are given to positive samples , giving lower weights to negative samples .

[0022] Furthermore, the evaluation indicators include recall rate, F1 and AUC indicators, wherein the recall rate calculation formula is as follows:

[0023]

[0024] Among them, TP is the number of samples that are actually positive and are correctly classified as positive, FN is the number of samples that are actually positive and are misclassified as negative, and FP is the number of samples that are actually negative and are misclassified as positive.

[0025] F1 comprehensively evaluates the classification performance of the model through the weighted harmonic mean of Precision and Recall. The calculation formula is as follows:

[0026]

[0027] Among them, Precision is the precision rate, Recall is the recall rate;

[0028] The AUC metric describes the model's ability to distinguish between positive and negative samples. The calculation formula is as follows:

[0029]

[0030] Among them, FPR is the proportion of incorrect samples that the classification model misclassifies as correct samples under a specific threshold, and TPR is the proportion of correct samples that the model correctly predicts and distinguishes under a specific threshold. is the number of thresholds, is the current index.

[0031] Furthermore, the characteristic variables that affect the tradability of patents include the following characteristic variables:

[0032] Nationality of applicant, number of cohort countries, type of applicant, number of cohort countries, time to enter substantive examination, examination time, number of applicants, number of citations, priority countries, number of inventors, number of IPC classification numbers, number of words in the first claim, grant time, number of dependent claims, number of pages of documents, number of independent claims, number of priority numbers, number of licenses, number of lawsuits and number of pledges.

[0033] According to one aspect of the present invention, a patent tradability prediction system based on machine learning is provided, comprising:

[0034] A data acquisition module is used to obtain characteristic variables of the patent to be tested;

[0035] The patent tradability prediction module is used to input the acquired characteristic variables of the patent to be tested into the trained patent tradability prediction model and output the patent tradability prediction result; wherein the training of the patent tradability prediction model includes:

[0036] Select characteristic variables that affect patent tradability from internal and external environmental factors that affect patent tradability, and construct a characteristic variable data set;

[0037] Based on the feature variable data set, the random forest model, gradient boosting model, extreme gradient boosting tree model, lightweight gradient boosting model and category boosting model were used for training respectively, and the grid search method was used to select the best hyperparameters of each model to ensure that each model achieved the best prediction effect;

[0038] Using evaluation indicators to evaluate each model, comparing the evaluation results, and obtaining the best prediction model;

[0039] The weight value of the loss function in the optimal prediction model is adjusted through the weight balancing method, the above six machine learning models are optimized, and the results of the evaluation indicators of each model after optimization are compared to obtain the optimal prediction model as the patent tradability prediction model.

[0040] According to one aspect of the present specification, there is provided an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method for predicting patent tradability based on machine learning when executing the computer program.

[0041] According to one aspect of the present specification, there is provided a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the patent tradability prediction method based on machine learning.

[0042] According to one aspect of the present invention, a computer program product comprising instructions is provided, which, when executed on a computer, enables the computer to execute the steps of the patent tradability prediction method based on machine learning.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention proposes a patent tradability prediction method based on machine learning, which adopts six machine learning models for training, optimizes hyperparameters through grid search method, and evaluates the model using evaluation indicators to achieve the best prediction model with fewer feature variable inputs.

[0045] 2. The present invention proposes a patent tradability prediction method based on machine learning and a weight balancing method, and finally obtains a patent tradability prediction model. It does not need to rely on a large number of expert reviews, and can use existing data to predict patent tradability, greatly saving time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A flowchart of a patent tradability prediction method based on machine learning provided in an embodiment of the present invention;

[0048] Figure 2 A flowchart of the patent tradability prediction model training provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The embodiment of the present invention provides a patent tradability prediction method based on machine learning, such as Figure 1-2 As shown, the method comprises the following steps: obtaining characteristic variables of the patent to be tested; inputting the obtained characteristic variables of the patent to be tested into the trained patent tradability prediction model, and outputting the patent tradability prediction result; wherein, the training of the patent tradability prediction model comprises: selecting characteristic variables affecting patent tradability from internal and external environmental factors affecting patent tradability, and constructing a characteristic variable data set; based on the characteristic variable data set, respectively adopting a random forest model, a gradient boosting model, an extreme gradient boosting tree model, a lightweight gradient boosting model, a category boosting model and a logistic regression model for training, and using a grid search method to select the best hyperparameters of each model to ensure that each model achieves the best prediction effect; using an evaluation index to evaluate each model, and comparing the evaluation results to obtain the best prediction model; based on each model after obtaining the optimal hyperparameters, adjusting the weight value of the loss function in each model by a weight balancing method, optimizing each model, and comparing the evaluation results of the evaluation index of each model after optimization, and obtaining the best prediction model as the patent tradability prediction model.

[0051] Specifically, the feature variable data set is divided into a feature variable training set and a feature variable test set in a ratio of 7:3.

[0052] Specifically, according to the internal and external environmental factors that affect the tradability of patents, 20 characteristic variables are selected as the initial data set, which includes the nationality of the applicant, the number of cognate countries, the type of applicant (enterprise, individual, university or research institute), the number of cognate countries, the time to enter substantive examination, the examination time, the number of applicants, the number of citations, the priority country, the number of inventors, the number of IPC classification numbers, the number of words in the first claim, the authorization time (the number of days from application to authorization), the number of dependent claims, the number of pages of literature, the number of independent claims, the number of priority numbers, the number of licenses, the number of lawsuits and the number of pledges.

[0053] Specifically, the grid search method used is to conduct a comprehensive search for all possible parameter combinations by setting a series of candidate values ​​for hyperparameters. This process involves training and evaluating the model for each set of parameter configurations, and finally selecting the parameter combination with the best performance as the final hyperparameter of the model. The grid search method is based on a well-defined parameter grid, which lists each hyperparameter and its potential range of values. Each possible value of these hyperparameters is combined with the possible values ​​of other hyperparameters. For each hyperparameter combination, the model performance is evaluated through cross-validation or other validation methods. Finally, the optimal hyperparameter combination is selected based on performance to determine the best prediction model.

[0054] Specifically, the random forest model is built on decision trees and uses a tree-like structure algorithm to analyze the data set. When building each tree, a feature is first selected as the root node. The basis for this selection is that the feature can minimize the Gini coefficient when splitting the data, thereby optimizing the classification efficiency of the model. The Gini coefficient is a standard for measuring data purity, and its calculation formula is:

[0055]

[0056] in, represents the Gini coefficient of the data set D, is the sum of the categories, Indicates the first The lower the Gini coefficient is, the more samples in the node tend to be in the same category.

[0057] Specifically, the decision trees in the random forest model branch in order of the Gini coefficient from small to large, starting from the root node (i.e. the selected feature) and gradually differentiating downward to the leaf nodes (the end of the tree), which represent the final decision of the model (i.e. the target value). Random forest trains a single decision tree by randomly selecting n samples with replacement from the original training sample set to form a new training set. This process is repeated m times to generate m trees. After these trees are gathered into a forest, the classification result of the new data is determined by the voting results of each tree, and the category with the highest score is the final classification result. As a reinforcement of the decision tree, random forest effectively enhances the generalization ability of the model and reduces the risk of overfitting by integrating multiple trees, each of which is independently trained on randomly selected samples.

[0058] Specifically, the gradient boosting model iteratively optimizes a series of decision trees to gradually reduce the prediction error, thereby improving the fitting accuracy. Its model function is established based on the principle of gradient descent of the loss function, as follows:

[0059]

[0060] in, It is the decision tree In the iterations, the samples predictions, Representative The prediction results of the tree, represents the set and sum of possible trees. In addition, the gradient boosting model reduces the prediction residual left by the previous model by adding a new decision tree in each iteration. The selection of each new tree depends on the effectiveness of the tree in reducing the overall loss function, that is, selecting the tree that can maximize the reduction of the gradient of the loss function under the current model.

[0061] Specifically, the extreme gradient boosting tree model iteratively optimizes a series of decision trees to reduce prediction errors and improve fitting accuracy. Its model function is established based on the optimization principle of the regularized objective function, as follows:

[0062]

[0063] in, It is In the iterations, the samples predictions, Indicates The newly added decision tree in the iteration, represents the learning rate. The extreme gradient boosting tree model emphasizes that the selection of each new tree is based on reducing the regularized loss function. In each iteration, the new tree The choice of depends on the contribution of the tree to maximizing the overall model performance, which also includes reducing the risk of overfitting the model. This strategy, which combines loss reduction with model complexity control, ensures that the extreme gradient boosting tree model improves the prediction accuracy while also enhancing the generalization ability and stability of the model.

[0064] Specifically, the lightweight gradient boosting model iteratively optimizes a series of decision trees to reduce prediction errors and improve fitting accuracy. Its model function is established based on the gradient descent principle of the loss function and a unique histogram optimization method, as follows:

[0065]

[0066] in, It is In the iterations, the samples predictions, It is The newly added decision tree in the iteration, is the learning rate. The lightweight gradient boosting machine model optimizes the implementation of the traditional gradient boosting machine by using a histogram-based decision tree splitting algorithm to reduce memory consumption and increase computing speed. In addition, the model supports efficient processing of high-dimensional data and large-scale data sets, while selecting the most informative features for splitting through the gradient distribution of feature splitting points, thereby reducing the risk of overfitting when the model processes sparse data.

[0067] Specifically, the category boosting model iteratively optimizes a series of decision trees to reduce prediction errors and improve fitting accuracy. Its model function is established based on the gradient descent principle of the loss function, as follows:

[0068]

[0069] in, It is In the iterations, the samples predictions, Indicates The newly added decision tree in the iteration, is the learning rate. The category boosting model reduces the risk of overfitting during model training and improves the efficiency of processing category data through its innovative sequential boosting and combined category feature processing techniques. In addition, the category boosting model also introduces a statistical transformation of the target variable, in this way, each tree can focus more accurately on difficult-to-predict samples.

[0070] Specifically, the logistic regression model is a widely used statistical learning method, mainly used to solve binary classification problems. The probability of a sample belonging to a certain category is determined by mapping the output of the linear regression model to a probability value between 0 and 1. The logistic regression function is:

[0071]

[0072] Among them, z is the output of the logistic regression model, that is, z=wx+b, w is the weight vector, x is the feature vector, and b is the bias term.

[0073] Specifically, the selected patent feature variable data set is used as the input variable, and whether the patent has been traded is used as the prediction result. The prediction results of each model are evaluated based on three evaluation indicators (recall rate, F1 value, AUC) to obtain the best prediction model; among them, the calculation formulas of the two indicators of recall rate and F1 value are as follows:

[0074]

[0075]

[0076] Among them, TP (True Positive) is the number of samples that are actually positive and correctly classified as positive, FN (False Negative) is the number of samples that are actually positive and incorrectly classified as negative, and FP (False Positive) is the number of samples that are actually negative and incorrectly classified as positive. Recall measures the model's ability to identify positive samples, and F1-score comprehensively evaluates the model's classification performance through the weighted harmonic mean of Precision and Recall.

[0077] The calculation formula of the AUC indicator is as follows:

[0078]

[0079] Among them, FPR is the proportion of incorrect samples that the classification model misclassifies as correct samples under a specific threshold, and TPR is the proportion of correct samples that the model correctly predicts and distinguishes under a specific threshold. represents the number of thresholds, Represents the current index, which is used to traverse a series of thresholds. The AUC indicator describes the model's ability to distinguish between positive and negative samples.

[0080] Specifically, the calculation formula for the weight balancing method used to optimize the machine learning model is as follows:

[0081]

[0082] in, represents the actual category of the sample, is the probability that the model predicts the positive class, and are the weights of the positive and negative classes respectively. Through this strategy, the model gives higher weights to minority class samples during training. , while giving lower weights to majority class samples .

[0083] The embodiment of the present invention also provides evaluation results of each model, as shown in Table 1, each model is evaluated by three evaluation indicators: recall, F1 value, and AUC, among which the best model is the lightweight gradient boosting model.

[0084] Table 1 Evaluation results of each model

[0085] Model Recall F1-Score AUC LR 0.500 0.472 0.633 RF 0.500 0.471 0.696 XGBoost 0.530 0.530 0.740 GBM 0.532 0.533 0.738 LightGBM 0.532 0.533 0.741 Catboost 0.528 0.525 0.744

[0086] The weight value of the loss function in the lightweight gradient boosting model is adjusted by the weight balancing method, and the above six machine learning models are optimized. Finally, the evaluation results of each optimized model are shown in Table 2.

[0087] Table 2 Evaluation results of each model after optimization

[0088] Model Recall F1-Score AUC LR 0.613 0.492 0.642 RF 0.649 0.539 0.705 XGBoost 0.664 0.560 0.744 GBM 0.671 0.570 0.744 LightGBM 0.672 0.570 0.749 Catboost 0.673 0.579 0.749

[0089] Combining the model evaluation results in Table 1 and Table 2, it can be found that after the present invention adopts the weight balance method for optimization, the prediction accuracy of each model is significantly improved.

[0090] The implementation basis of each embodiment of the present invention is to implement programmed processing through a device with a processor function. Therefore, in engineering practice, the technical solutions and functions of each embodiment of the present invention are encapsulated into various modules. Based on this reality, on the basis of the above embodiments, an embodiment of the present invention provides a patent tradability prediction system based on machine learning, which is used to execute a patent tradability prediction method based on machine learning in the above method embodiment.

[0091] The embodiment of the present invention provides a patent tradability prediction system based on machine learning, and the specific implementation process thereof includes: obtaining characteristic variables of a patent to be tested through a data acquisition module; inputting the acquired characteristic variables of the patent to be tested into a trained patent tradability prediction model through a patent tradability prediction module, and outputting a patent tradability prediction result; wherein the training of the patent tradability prediction model includes selecting characteristic variables affecting patent tradability from internal and external environmental factors affecting patent tradability, and constructing a characteristic variable data set; based on the characteristic variable data set, respectively using a random forest model, a gradient boosting model, an extreme gradient boosting tree model, a lightweight gradient boosting model and a category boosting model for training, and using a grid search method to select the best hyperparameters of each model to ensure that each model achieves the best prediction effect; using an evaluation index to evaluate each model, comparing the evaluation results, and obtaining the best prediction model; adjusting the weight value of the loss function in the best prediction model through a weight balancing method, optimizing the above six machine learning models, and comparing the evaluation results of the evaluation indexes of each model after optimization, obtaining the best prediction model as the patent tradability prediction model, and realizing a more accurate prediction of the tradability of the patent.

[0092] Based on the same inventive concept as the aforementioned embodiment, an embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions, so as to implement the patent tradability prediction method based on machine learning as proposed in the aforementioned embodiment.

[0093] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it is used to achieve the highest prediction accuracy with the least variable input, and predict the tradability of the patent through the patent tradability prediction model finally obtained. The storage medium can be any non-volatile storage device such as a hard disk, a solid-state hard disk, a flash drive, an optical disk, etc., which is used to store computer program codes and necessary data files. The stored computer program includes: a data acquisition module, a patent tradability prediction module; wherein the patent tradability prediction module includes: a data set construction unit, a model building unit and a patent tradability prediction unit.

[0094] The embodiment of the present invention also provides a computer program product containing instructions, which, when executed on a computer, generates in whole or in part the patent tradability prediction method based on machine learning proposed in the above embodiment. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0095] Finally, it should be pointed out that the above specific embodiments are only representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments, and there are many variations. Any simple modification, equivalent changes and modifications made to the above specific embodiments based on the technical essence of the present invention should be considered to belong to the protection scope of the present invention.

Claims

1. A patent tradability prediction method based on machine learning, characterized in that: include: Obtain characteristic variables of the patent to be tested; Input the acquired characteristic variables of the patent to be tested into the trained patent tradability prediction model, and output the patent tradability prediction result; wherein the training of the patent tradability prediction model includes: Select characteristic variables that affect patent tradability from internal and external environmental factors that affect patent tradability, and construct a characteristic variable data set; Based on the feature variable data set, the random forest model, gradient boosting model, extreme gradient boosting tree model, lightweight gradient boosting model, category boosting model and logistic regression model were used for training respectively, and the grid search method was used to select the best hyperparameters of each model to ensure that each model achieved the best prediction effect; Based on the models after obtaining the optimal hyperparameters, the weight value of the loss function in each model is adjusted through the weight balancing method, the models are optimized, and the results of the evaluation indicators of each model after optimization are compared to obtain the optimal prediction model as the patent tradability prediction model.

2. The patent tradability prediction method based on machine learning according to claim 1, characterized in that: The grid search method selects the best hyperparameters of each model, including: Based on a parameter grid defining each hyperparameter and its possible values, the possible values ​​of the hyperparameters of each model are combined with the possible values ​​of other hyperparameters in each model; For each set of hyperparameter combinations, the cross-validation method is used to evaluate the performance of each model and obtain the evaluation results; Based on the evaluation results, the best hyperparameter combination is selected as the hyperparameters of the best prediction model.

3. The patent tradability prediction method based on machine learning according to claim 1, characterized in that: The calculation formula of the weight balancing method is as follows: , in, represents the actual category of the sample, is the probability that the model predicts the positive class, and are the weights of the positive and negative classes respectively; higher weights are given to positive samples , giving lower weights to negative samples .

4. The patent tradability prediction method based on machine learning according to claim 1, characterized in that: The evaluation indicators include recall rate, F1 and AUC indicators, where the recall rate calculation formula is as follows: , Among them, TP is the number of samples that are actually positive and are correctly classified as positive, FN is the number of samples that are actually positive and are misclassified as negative, and FP is the number of samples that are actually negative and are misclassified as positive. F1 comprehensively evaluates the classification performance of the model through the weighted harmonic mean of Precision and Recall. The calculation formula is as follows: , Among them, Precision is the precision rate, Recall is the recall rate; The AUC metric describes the model's ability to distinguish between positive and negative samples. The calculation formula is as follows: , Among them, FPR is the proportion of incorrect samples that the classification model misclassifies as correct samples under a specific threshold, and TPR is the proportion of correct samples that the model correctly predicts and distinguishes under a specific threshold. is the number of thresholds, is the current index.

5. The patent tradability prediction method based on machine learning according to claim 1, characterized in that: The characteristic variables that affect the tradability of patents include the following characteristic variables: Nationality of applicant, number of cohort countries, type of applicant, number of cohort countries, time to enter substantive examination, examination time, number of applicants, number of citations, priority countries, number of inventors, number of IPC classification numbers, number of words in the first claim, grant time, number of dependent claims, number of pages of documents, number of independent claims, number of priority numbers, number of licenses, number of lawsuits and number of pledges.

6. A patent tradability prediction system based on machine learning, characterized in that: include: A data acquisition module is used to obtain characteristic variables of the patent to be tested; The patent tradability prediction module is used to input the acquired characteristic variables of the patent to be tested into the trained patent tradability prediction model and output the patent tradability prediction result; wherein the training of the patent tradability prediction model includes: Select characteristic variables that affect patent tradability from internal and external environmental factors that affect patent tradability, and construct a characteristic variable data set; Based on the feature variable data set, the random forest model, gradient boosting model, extreme gradient boosting tree model, lightweight gradient boosting model, category boosting model and logistic regression model were used for training respectively, and the grid search method was used to select the best hyperparameters of each model to ensure that each model achieved the best prediction effect; Based on the models after obtaining the optimal hyperparameters, the weight value of the loss function in each model is adjusted through the weight balancing method, the models are optimized, and the results of the evaluation indicators of each model after optimization are compared to obtain the optimal prediction model as the patent tradability prediction model.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the patent tradability prediction method based on machine learning described in any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the patent tradability prediction method based on machine learning described in any one of claims 1 to 6 are implemented.

9. A computer program product comprising instructions, characterized in that When it is run on a computer, it enables the computer to execute the steps of the patent tradability prediction method based on machine learning as described in any one of claims 1 to 6.

Images