Intellectual property operation management system based on big data

Abstract

The invention belongs to the technical field of intelligence, and discloses an intellectual property operation management system based on big data, comprising: acquiring behavior data and basic data of intellectual property; carrying out missing value and abnormal value processing on the behavior data of the intellectual property to obtain a preprocessed data set; predicting value index data of the intellectual property by using an SVR model based on the preprocessed data set; an ETL tool is used for extraction, conversion and loading, a TF-IDF technology is used for processing the application name, a keyword group B is extracted, and the keyword group B serves as a new field to be stored in a target database; based on a text input by a user, obtaining a keyword A by using a TF-IDF technology; matching with the keyword group B to obtain comprehensive matching degree data; based on the comprehensive matching degree data, a threshold value is set, and whether recommendation is carried out or not is judged; and the intellectual property operation management efficiency and accuracy are improved.

Description

[0001] The present invention relates to the field of intelligent technology, and in particular to an intellectual property operation and management system based on big data. Background Art

[0002] The existing big data intellectual property operation and management system has exposed many problems: first, it fails to adopt intelligent, data-driven prediction models, but relies on simple rules and statistical data, which cannot accurately assess the potential commercial value, technological innovation or market influence of intellectual property; Most existing systems only match keywords based on text data, ignoring the value indicators of intellectual property rights. This means that even if the intellectual property rights have a high degree of textual matching, it does not necessarily mean that they have actual value to user needs. Existing systems usually use static keyword matching rules and fixed matching thresholds, which means that once the matching standards are set, the system cannot be adaptively adjusted according to changes in data or user needs. Fixed thresholds cannot be dynamically adjusted according to subtle differences between intellectual property rights, which may cause the match to be too loose or too strict, affecting the accuracy of the recommendation.

[0003] In view of this, the present invention proposes an intellectual property operation and management system based on big data to solve the above problems. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an intellectual property operation and management system based on big data, comprising: Data acquisition module: obtains behavioral data and basic data of intellectual property rights; Data processing module: Process missing values ​​and outliers for the behavior data of intellectual property rights to obtain a preprocessed data set; Prediction module: Based on the preprocessed data set, the SVR model is used to predict the value indicator data of intellectual property rights; Data storage module: Use ETL tools to extract, transform and load the basic data, pre-processed data sets and value indicator data of intellectual property rights, and use TF-IDF technology to process the application name, extract keyword group B, and store keyword group B as a new field in the target database; Keyword matching module: Based on the user input text, use TF-IDF technology to obtain keyword A; and match it with keyword group B to obtain comprehensive matching data; Recommendation module: Based on the comprehensive matching data, set the recommendation threshold and determine whether to make a recommendation.

[0005] Furthermore, the behavior data of the intellectual property rights include features of click count, collection count, download count, and citation count; Basic data includes application name, application number, application date, validity period and applicant name.

[0006] Furthermore, the method for obtaining the preprocessed data set includes: For the intellectual property behavior data, the stacking method is used to form a data set, and the forward filling method and backward filling method are used to process the missing values ​​of the data set to obtain the missing value processing data set. Then, the isolation forest model is used to identify and process the outliers, and the processed data is standardized to obtain the preprocessed data set. The isnull technology is used to traverse the data set composed of click count features, favorite count features, download count features and citation count features, and identify missing values. For missing values, a forward filling method is used to obtain the forward data value of the missing value, and a backward filling method is used to obtain the backward data value. The average of the forward data value and the backward data value is calculated, and the missing value is filled with the average value.

[0007] Furthermore, the method of using the isolation forest model to identify outliers and process them includes: Initialize hyperparameters: Initialize the hyperparameters of the isolation forest, including the number of trees SL, the maximum depth of each tree SD, and the number of samples per tree mm; Select training subset: set SL trees, set the training subset size of each tree to mm, and select a subset from the missing value processing data set for training each time; Feature selection: Use multiple evaluation metrics to select split features; Split point selection: For the selected split feature, a split point is randomly selected to divide the data into two subsets; Recursive splitting: For each subset, repeat feature selection and split point selection, and continue to split the data into smaller subsets until the maximum depth SD of the tree is reached; Repeat the above process: build M_L trees, each tree is trained on a different subset of the data; Calculate the path length of each data point: In each tree, calculate the path length of the data point , that is, the number of split nodes from the root node to the data point, where Represents the path length of the i-th data point in the j-th tree; Calculate the average path length over all trees: For each data point , calculate the average path length of the data point in all trees ,in, Represents data points The path length in the jth tree, Indicates the number of trees; Calculate outlier score: Evaluate the outlier score of each data point by calculating its average path length ,in, is a constant calculated based on the number of samples m, representing the expected path length, and the formula is: ,in is a constant; Stop condition: stop iteration when the maximum depth of the tree is reached; Output: Setting the outlier threshold ,when When , it indicates that the data point is normal; when When , it means that the data point is abnormal. The backward filling method is used to replace and fill the abnormal data points to obtain the pre-processed data set.

[0008] Furthermore, the specific method of using multiple evaluation indicators to select split features includes: By calculating the information gain, Gini index and variance of each feature and standardizing them, the entropy weight method is used to calculate the information gain, Gini index and variance to set the weight, and the comprehensive score of each feature is calculated ,in, , and Respectively represent The standardized information gain, Gini index and variance of the eigenvalues, , and Represent the weights of information gain, Gini index and variance respectively, and select the comprehensive score The highest feature is used as the selected feature; The entropy weight method calculates the weight by calculating the information entropy value of each evaluation indicator. ,in, Indicates The entropy of the evaluation index, Indicates The probability distribution of the tz-th eigenvalue of the evaluation indicator, Indicates the total number of eigenvalues; According to information entropy Calculating redundancy ; Calculate the weight of each evaluation indicator according to the redundancy , where 3 represents the total number of evaluation indicators.

[0009] Furthermore, the setting of the outlier threshold The specific methods include: Using neighborhood distance to set outlier threshold , calculate the mean value of the missing value processing data set and standard deviation , and use the coefficient of variation to calculate , and set the K value to ,in, represents the number of samples in the dataset, represents the adjustment coefficient, , represents the smoothing factor; Determine the K value. For each data point, use the Euclidean distance to calculate the distance between the data point and its K nearest neighbors, and perform an average calculation to obtain the average distance of the K nearest neighbors. According to the rule of thumb, set the distance threshold to 95%. When the average distance of the K nearest neighbors of a data point is greater than the distance threshold of 95%, the data point is an outlier.

[0010] Furthermore, the specific method of using the SVR model to predict the value indicator data of intellectual property rights based on the preprocessed data set includes: The sample set includes J_o groups of samples, each of which includes a preprocessed data set and corresponding intellectual property value indicator data; Initialize hyperparameters: Select radial basis function as kernel function, determine the penalty coefficient, epsilon parameter, gamma parameter hyperparameter combination, and use grid search method to determine the optimal hyperparameter combination; Calculate the initial loss function: Calculate the initial loss function value of the model , as the starting point of optimization, where represents regularization, represents the weight vector of the input features, represents the bias term, represents the error term, is the penalty coefficient, n2 represents the total number of training samples, o represents the sample index, represents the error of the oth sample on the upper deviation, represents the error of the o-th sample on the lower deviation; Iterative optimization: Use gradient descent to update model parameters in each iteration and , calculate the new loss function during the update process; Check convergence conditions: After each round of iteration, compare the change in the current loss function value with that of the previous round. When the loss change between two adjacent iterations is less than the preset error threshold, stop the iteration.

[0011] Furthermore, the specific method of using the ETL tool to extract, transform and load the basic data, pre-processed data set and value indicator data of the intellectual property rights includes: Extraction: Through the real-time data streaming technology Apache Kafka, the basic IP data is obtained from L_L data sources in real time and directly transferred to the ETL tool for processing. At the same time, the value indicator data predicted based on the IP behavior data and the processed data set are imported into the ETL tool. The focus of this stage is to ensure that all relevant data from different sources can be completely and accurately extracted for subsequent conversion and loading; Conversion: For the basic intellectual property data, value indicator data and processed data sets imported into the ETL tool, duplicate removal, missing value processing, date format unification, field name adjustment are performed to convert the original data into the standard format of the target database; Loading: Import the converted data into the target database and design the table structure, field types and indexes of the target database; Furthermore, the specific method of using TF-IDF technology to extract keyword group B from the application name includes: Application name preprocessing: Use the HanLP tool to segment the application name and obtain subwords; remove stop words and convert the uppercase letters of the text to lowercase to obtain preprocessed text; Count the frequency of each subword in the preprocessed text ,in, represents the frequency of occurrence of word d, Indicates the number of times word d appears in the preprocessed text, Indicates the total number of words in the preprocessed text; Calculate the inverse text frequency to calculate the attractiveness of a subword in the corpus. The larger the IDF, the more recognizable the word is to a specific document. The formula is as follows: ,in, Indicates the total number of texts, represents the amount of text containing subword d, represents the inverse text frequency of subword d; Calculate the TF-IDF value. Multiply the occurrence frequency TF of each subword by the inverse text frequency to get the TF-IDF value. This value reflects the importance of word t in document d. The higher the TF-IDF value of a subword, the more important it is as a keyword. Sort by the TF-IDF value of each subword, set a threshold A_Q, and when the TF-IDF value is greater than the threshold, the subword is used as a keyword to obtain keyword group B; The method of obtaining keyword A based on the user input text using TF-IDF technology and matching keyword group B to obtain comprehensive matching data includes: Preprocess the user input text and extract keyword group A using TF-IDF technology; Retrieve keyword group B from the database and match it with keyword group A input by the user. Get the vectors of keyword group A and keyword group B, and construct the vectors based on the TF-IDF value of each subword in the keyword group; Use cosine similarity to calculate the similarity between keyword group A and keyword group B: ,in, A vector representing the keyword group A, represents the vector of keyword group B, represents the inner product, and Represents the Euclidean norm; the calculated cosine similarity value is between 0 and 1, and the larger the value, the more similar the two keyword groups are; Calculate the Jaccard similarity, the formula is: ,in, Indicates the size of the intersection of keyword group A and keyword group B. Indicates the size of the union of keyword group A and keyword group B. The closer the value is to 1, the more similar the two keyword groups are, and vice versa. By calculating the cosine similarity and Jaccard similarity, combined with the intellectual property value index, the comprehensive matching degree is calculated: ,in, , and Respectively represent the weights of cosine similarity, Jaccard similarity and intellectual property value index, Represents the intellectual property value indicator.

[0012] Furthermore, the specific method of setting the recommendation threshold based on the comprehensive matching data and determining whether to make a recommendation includes: Use the sliding window method to set the threshold and set a window of fixed length C_K; Calculate the mean and standard deviation of the comprehensive matching degree within the window and set the initial recommendation threshold: ,in, Represents the mean of the comprehensive matching degree in the window, Represents the standard deviation of the comprehensive matching degree in the window, represents the adjustment factor, ; When a new comprehensive match is received, the window slides forward, removes the oldest comprehensive match, adds the new match, obtains an updated window, and recalculates the mean and standard deviation to obtain the recommended threshold. ;in, represents the mean of the update window, represents the standard deviation of the update window; when When the IP is selected, the IP is recommended, and the IP has a high matching degree and meets the needs of the user; when , the intellectual property is not recommended, the matching degree of the intellectual property is low and does not meet the user's needs.

[0013] The technical effects and advantages of the intellectual property operation and management system based on big data of the present invention are as follows: The present invention integrates the behavioral data, basic data and predicted value indicators of intellectual property through big data technology, and provides a data-driven intellectual property operation and management solution. It uses the SVR model to predict the value of intellectual property, the TF-IDF technology to extract keywords, and calculates the comprehensive matching degree through cosine similarity and Jaccard similarity to provide users with accurate intellectual property recommendations. The sliding window method dynamically adjusts the recommendation threshold to ensure that the system can adapt to changing data and user needs in real time, thereby improving the accuracy of matching and the effectiveness of recommendations. The use of intelligent prediction models can accurately predict the potential commercial value and technological innovation of intellectual property, greatly improving the accuracy of the evaluation; at the same time, keyword matching not only relies on text content, but also combines the value indicators of intellectual property to improve the relevance of matching results. The dynamic threshold adjustment mechanism ensures that the system can flexibly adjust the recommendation standards according to actual data changes, avoiding matching deviations caused by static rules, thereby optimizing user experience and recommendation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a schematic diagram of an intellectual property operation and management system based on big data of the present invention; Figure 2 This is a schematic diagram of an intellectual property operation management method based on big data of the present invention. DETAILED DESCRIPTION

[0015] 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.

[0016] Example 1 See also Figure 1 As shown, the intellectual property operation and management system based on big data described in this embodiment includes: Data acquisition module: acquires the behavior data and basic data of intellectual property rights; the behavior data of intellectual property rights include the characteristics of click count, collection count, download count and citation count; Basic data includes application name, application number, application date, validity period and applicant name; The behavior data of intellectual property rights include the number of clicks, number of favorites, number of downloads and number of citations, which are obtained by regularly crawling intellectual property websites using crawler technology; The application name, application number, application date, validity period and applicant name can be obtained through the public query interface or API interface provided by the intellectual property database; Data processing module: Process missing values ​​and outliers for the behavior data of intellectual property rights to obtain a preprocessed data set; Prediction module: Based on the preprocessed data set, the SVR model is used to predict the value indicator data of intellectual property rights; Data storage module: Use ETL tools to extract, transform and load the basic data, pre-processed data sets and value indicator data of intellectual property rights, and use TF-IDF technology to process the application name, extract keyword group B, and store keyword group B as a new field in the target database; Keyword matching module: Based on the user input text, use TF-IDF technology to obtain keyword A; and match it with keyword group B to obtain comprehensive matching data; Recommendation module: Based on the comprehensive matching data, set the recommendation threshold and determine whether to make a recommendation.

[0017] The methods for obtaining the preprocessed data set include: For the intellectual property behavior data, the stacking method is used to form a data set, and the forward filling method and backward filling method are used to process the missing values ​​of the data set to obtain the missing value processing data set. Then, the isolation forest model is used to identify and process the outliers, and the processed data is standardized to obtain the preprocessed data set. The isnull technology is used to traverse the data set composed of click count features, favorite count features, download count features and citation count features, and identify missing values. For missing values, a forward filling method is used to obtain the forward data value of the missing value, and a backward filling method is used to obtain the backward data value. The average of the forward data value and the backward data value is calculated, and the missing value is filled with the average value.

[0018] Ways to use the Isolation Forest Model to identify outliers and handle them include: Initialize hyperparameters: Initialize the hyperparameters of the isolation forest, including the number of trees SL, the maximum depth of each tree SD, and the number of samples per tree mm; Select training subset: set SL trees, set the training subset size of each tree to mm, and select a subset from the missing value processing data set for training each time; Feature selection: Use multiple evaluation metrics to select split features; Split point selection: For the selected split feature, a split point is randomly selected to divide the data into two subsets; Recursive splitting: For each subset, repeat feature selection and split point selection, and continue to split the data into smaller subsets until the maximum depth SD of the tree is reached; Repeat the above process: build M_L trees, each tree is trained on a different subset of the data; Calculate the path length of each data point: In each tree, calculate the path length of the data point , that is, the number of split nodes from the root node to the data point, where Represents the path length of the i-th data point in the j-th tree; Calculate the average path length over all trees: For each data point , calculate the average path length of the data point in all trees ,in, Represents data points The path length on the jth tree, Indicates the number of trees; Calculate outlier score: Evaluate the outlier score of each data point by calculating its average path length ,in, is a constant calculated based on the number of samples m, representing the expected path length, and the formula is: ,in is a constant; Stop condition: when the maximum depth of the tree feature iteration is reached; Output: Setting the outlier threshold ,when When , it indicates that the data point is normal; when When , it means that the data point is abnormal. The backward filling method is used to replace and fill the abnormal data points to obtain the pre-processed data set.

[0019] The specific method of using multiple evaluation indicators to select split features includes: By calculating the information gain, Gini index and variance of each feature and standardizing them, the entropy weight method is used to calculate the information gain, Gini index and variance to set the weight, and the comprehensive score of each feature is calculated ,in, , and Respectively represent The standardized information gain, Gini index and variance of the eigenvalues, , and Represent the weights of information gain, Gini index and variance respectively, and select the comprehensive score The highest feature is used as the selected feature; Here This is because the smaller the Gini index, the better, while the larger the information gain, the better, so the directions of the two indicators are opposite, and the Gini index needs to be negated so that they both become "the larger the better" indicators; The entropy weight method calculates the weight by calculating the information entropy value of each evaluation indicator. ,in, Indicates The entropy of the evaluation index, Indicates The probability distribution of the tzth eigenvalue of the evaluation indicator, Indicates the total number of eigenvalues; According to information entropy Calculating redundancy ; Calculate the weight of each evaluation indicator according to the redundancy , where 3 represents the total number of evaluation indicators.

[0020] Setting the outlier threshold The specific methods include: Using neighborhood distance to set outlier threshold , calculate the mean value of the missing value processing data set and standard deviation , and use the coefficient of variation to calculate , and set the K value to ,in, represents the number of samples in the dataset, represents the adjustment coefficient, , represents the smoothing factor; Determine the K value. For each data point, use the Euclidean distance to calculate the distance between the data point and its K nearest neighbors, and perform an average calculation to obtain the average distance of the K nearest neighbors. According to the rule of thumb, set the distance threshold to 95%. When the average distance of the K nearest neighbors of a data point is greater than the distance threshold of 95%, the data point is an outlier.

[0021] Based on the preprocessed data set, the specific methods of using the SVR model to predict the value indicator data of intellectual property rights include: The sample set includes J_o groups of samples, each of which includes a preprocessed data set and corresponding intellectual property value indicator data; Initialize hyperparameters: Select radial basis function as kernel function, determine the penalty coefficient, epsilon parameter, gamma parameter hyperparameter combination, and use grid search method to determine the optimal hyperparameter combination; Calculate the initial loss function: Calculate the initial loss function value of the model , as the starting point of optimization, where represents regularization, represents the weight vector of the input features, represents the bias term, represents the error term, is the penalty coefficient, n2 represents the total number of training samples, o represents the sample index, Represents the error and It represents the error of the o-th sample on the lower deviation (the predicted value is smaller than the true value); Iterative optimization: Use gradient descent to update model parameters in each iteration and , calculate the new loss function during the update process; Check convergence conditions: After each round of iteration, compare the change of the current loss function value with that of the previous round. When the loss change between two adjacent iterations is less than the preset error threshold, stop the iteration. In the SVR model, the convergence error threshold is set by an empirical rule and is set to a fixed value , when the loss function change between adjacent conjunction iterations is less than this value, the model has converged.

[0022] The specific methods of using ETL tools to extract, transform and load basic data, pre-processed data sets and value indicator data of intellectual property include: Extraction: Through the real-time data streaming technology Apache Kafka, the basic IP data is obtained from L_L data sources in real time and directly transferred to the ETL tool for processing. At the same time, the value indicator data predicted based on the IP behavior data and the processed data set are imported into the ETL tool. The focus of this stage is to ensure that all relevant data from different sources can be completely and accurately extracted for subsequent conversion and loading; Conversion: For the basic intellectual property data, value indicator data and processed data sets imported into the ETL tool, duplicate removal, missing value processing, date format unification, field name adjustment are performed to convert the original data into the standard format of the target database; Loading: Import the converted data into the target database and design the table structure, field types and indexes of the target database; Data sources include intellectual property databases, scientific research paper databases, business analysis platforms, user behavior data, social media, and internal enterprise systems.

[0023] The specific methods of using TF-IDF technology to extract keyword group B from the application name include: Application name preprocessing: Use the HanLP tool to segment the application name and obtain subwords; remove stop words and convert the uppercase letters of the text to lowercase to obtain preprocessed text; Count the frequency of each subword in the preprocessed text ,in, represents the frequency of occurrence of word d, Indicates the number of times word d appears in the preprocessed text, Indicates the total number of words in the preprocessed text; Calculate the inverse text frequency to calculate the attractiveness of a subword in the corpus. The larger the IDF, the more recognizable the word is to a specific document. The formula is as follows: ,in, Indicates the total number of texts, represents the amount of text containing subword d, represents the inverse text frequency of subword d; Calculate the TF-IDF value. Multiply the occurrence frequency TF of each subword by the inverse text frequency to get the TF-IDF value. This value reflects the importance of word t in document d. The higher the TF-IDF value of a subword, the more important it is as a keyword. Sort by the TF-IDF value of each subword, set a threshold A_Q, and when the TF-IDF value is greater than the threshold, the subword is used as a keyword to obtain keyword group B; The threshold A_Q is set according to the empirical method and is 70%; HanLP is an efficient, open source natural language processing toolkit that focuses on the processing of Chinese text. It provides rich functions, including word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, sentiment analysis, etc. It can handle complex language characteristics and is suitable for text mining, search engines, information extraction and other application scenarios. HanLP has high accuracy and flexibility, supports multiple languages ​​and customized training, and is an important tool in the field of Chinese processing.

[0024] Based on the user input text, the TF-IDF technology is used to obtain keyword A; and it is matched with keyword group B. The methods for obtaining comprehensive matching data include: By extracting keywords and calculating similarity through TF-IDF, we ensure accurate understanding of user input; cosine similarity and Jaccard similarity evaluate keyword similarity from different angles, enhancing the comprehensiveness of matching; at the same time, combined with the value index of intellectual property rights, we can further screen out intellectual property rights with actual value based on text similarity, thereby improving the accuracy and practicality of recommendations; Set a user input text box, where users can input according to their needs, obtain the user input text, perform preprocessing based on the user input text, and use TF-IDF technology to extract keyword group A; Retrieve keyword group B from the database and match it with keyword group A input by the user, obtain the vectors of keyword group A and keyword group B, and construct the vector based on the TF-IDF value of each subword in the keyword group; Use cosine similarity to calculate the similarity between keyword group A and keyword group B: ,in, A vector representing the keyword group A, represents the vector of keyword group B, represents the inner product, and Represents the Euclidean norm; the calculated cosine similarity value is between 0 and 1, and the larger the value, the more similar the two keyword groups are; Calculate the Jaccard similarity, the formula is: ,in, Indicates the size of the intersection of keyword group A and keyword group B. Indicates the size of the union of keyword group A and keyword group B. The closer the value is to 1, the more similar the two keyword groups are, and vice versa. By calculating the cosine similarity and Jaccard similarity, combined with the intellectual property value index, the comprehensive matching degree is calculated: ,in, , and Respectively represent the weights of cosine similarity, Jaccard similarity and intellectual property value index, Represents the intellectual property value indicator.

[0025] Based on the comprehensive matching data, the recommendation threshold is set and the specific methods for determining whether to make a recommendation include: Use the sliding window method to set the threshold and set a window of fixed length C_K; Calculate the mean and standard deviation of the comprehensive matching degree within the window and set the initial recommendation threshold: ,in, Represents the mean of the comprehensive matching degree in the window, Represents the standard deviation of the comprehensive matching degree in the window, represents the adjustment factor, ; When a new comprehensive match is received, the window slides forward, removes the oldest comprehensive match, adds the new match, obtains an updated window, and recalculates the mean and standard deviation to obtain the recommended threshold. ;in, represents the mean of the update window, represents the standard deviation of the update window; when When the user's IP is selected, the IP is recommended. The IP has a high matching degree, meets the user's needs, and is valuable to the user. when , the intellectual property is not recommended, the matching degree of the intellectual property is low, it does not meet the user's needs, and it has no value to the user.

[0026] This embodiment integrates the behavior data, basic data and predicted value indicators of intellectual property rights through big data technology, provides a data-driven intellectual property operation management solution, uses the SVR model to predict the value of intellectual property rights, uses the TF-IDF technology to extract keywords, and calculates the comprehensive matching degree through cosine similarity and Jaccard similarity to provide users with accurate intellectual property recommendations. The recommendation threshold is dynamically adjusted through the sliding window method to ensure that the system can adapt to changing data and user needs in real time, thereby improving the accuracy of matching and the effectiveness of recommendations. In addition, the use of intelligent prediction models can accurately predict the potential commercial value and technological innovation of intellectual property, greatly improving the accuracy of the evaluation; at the same time, keyword matching not only relies on text content, but also combines the value indicators of intellectual property, which improves the relevance of matching results. The dynamic threshold adjustment mechanism ensures that the system can flexibly adjust the recommendation standards according to actual data changes, avoiding the matching deviation caused by static rules, thereby optimizing user experience and recommendation efficiency.

[0027] Example 2 See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an intellectual property operation management method based on big data, including: Step S1, collecting behavioral data and basic data of intellectual property rights; Step S2, processing missing values ​​and outliers for the behavior data of intellectual property rights to obtain a preprocessed data set; Step S3, inputting the preprocessed data set into the SVR model to predict the value index data of the intellectual property rights; Step S4: Use ETL tools to extract, transform and load the intellectual property basic data, the processed data set and the value indicator data, and use TF-IDF technology to extract the keyword group B from the application name, and store the keyword group B as a new field in the target database; Step S5: Based on the user input text, use TF-IDF technology to obtain keyword A; and match it with keyword group B to obtain comprehensive matching data; Step S6: Based on the comprehensive matching data, a recommendation threshold is set and a decision is made as to whether to make a recommendation.

[0028] Example 3 This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the intellectual property operation and management system based on big data provided above is implemented.

[0029] Since the electronic device introduced in this embodiment is an electronic device used to implement an intellectual property operation and management system based on big data in the embodiment of this application, based on the intellectual property operation and management system based on big data introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the intellectual property operation and management system based on big data in the embodiment of this application, it belongs to the scope of protection of this application.

[0030] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0031] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A big data-based intellectual property operation and management system, characterized in that: include: Data acquisition module: obtains behavioral data and basic data of intellectual property rights; Data processing module: Process missing values ​​and outliers for the behavior data of intellectual property rights to obtain a preprocessed data set; Prediction module: Based on the preprocessed data set, the SVR model is used to predict the value indicator data of intellectual property rights; Data storage module: Use ETL tools to extract, transform and load the basic data, pre-processed data sets and value indicator data of intellectual property rights, and use TF-IDF technology to process the application name, extract keyword group B, and store keyword group B as a new field in the target database; Keyword matching module: Based on the user input text, use TF-IDF technology to obtain keyword A; And match it with keyword group B to obtain comprehensive matching data; Recommendation module: Based on the comprehensive matching data, set the recommendation threshold and determine whether to make a recommendation.

2. According to the big data-based intellectual property operation and management system of claim 1, it is characterized in that: The behavioral data of the intellectual property rights include characteristics of click count, collection count, download count, and citation count; Basic data includes application name, application number, application date, validity period and applicant name.

3. According to the big data-based intellectual property operation and management system of claim 2, it is characterized in that: The method for obtaining the preprocessed data set includes: For the intellectual property behavior data, the stacking method is used to form a data set, and the forward filling method and backward filling method are used to process the missing values ​​of the data set to obtain the missing value processing data set. Then, the isolation forest model is used to identify and process the outliers, and the processed data is standardized to obtain the preprocessed data set. The isnull technology is used to traverse the data set composed of click count features, favorite count features, download count features and citation count features, and identify missing values. For missing values, a forward filling method is used to obtain the forward data value of the missing value, and a backward filling method is used to obtain the backward data value. The average of the forward data value and the backward data value is calculated, and the missing value is filled with the average value.

4. According to the big data-based intellectual property operation and management system of claim 3, it is characterized in that: The method of using the isolation forest model to identify outliers and process them includes: Initialize hyperparameters: Initialize the hyperparameters of isolation forest; Select training subset: set SL trees, set the training subset size of each tree to mm, and select a subset from the missing value processing data set for training each time; Feature selection: Use multiple evaluation metrics to select split features; Split point selection: For the selected split feature, a split point is randomly selected to divide the data into two subsets; Recursive splitting: For each subset, repeat feature selection and split point selection, and continue to split the data into smaller subsets until the maximum depth SD of the tree is reached; Repeat the above process: build M_L trees, each tree is trained on a different subset of the data; Calculate the path length of each data point: In each tree, calculate the path length of the data point , that is, the number of split nodes from the root node to the data point, where Represents the path length of the i-th data point in the j-th tree; Calculate the average path length over all trees: For each data point , calculate the average path length of the data point in all trees ,in, Represents data points The path length on the jth tree, Indicates the number of trees; Calculate outlier score: Evaluate the outlier score of each data point by calculating its average path length ,in, is a constant calculated based on the number of samples m, representing the expected path length, and the formula is: ,in is a constant; Stop condition: stop iteration when the maximum depth of the tree is reached; Output: Setting the outlier threshold ,when When , it indicates that the data point is normal; when When , it means that the data point is abnormal. The backward filling method is used to replace and fill the abnormal data points to obtain the pre-processed data set.

5. According to the big data-based intellectual property operation and management system of claim 4, it is characterized in that: The specific method of using multiple evaluation indicators to select split features includes: By calculating the information gain, Gini index and variance of each feature and standardizing them, the entropy weight method is used to calculate the information gain, Gini index and variance to set the weight, and the comprehensive score of each feature is calculated , select Comprehensive Rating The highest feature is used as the selected feature; in, , and Respectively represent The standardized information gain, Gini index and variance of the eigenvalues, , and Represent the weights of information gain, Gini index and variance respectively; The entropy weight method calculates the weight by calculating the information entropy value of each evaluation indicator. ,in, Indicates The entropy of the evaluation index, Indicates The probability distribution of the tzth eigenvalue of the evaluation indicator, Indicates the total number of eigenvalues; According to information entropy Calculating redundancy ; Calculate the weight of each evaluation indicator according to the redundancy , where 3 represents the total number of evaluation indicators.

6. According to the big data-based intellectual property operation and management system of claim 5, it is characterized in that: Setting the outlier threshold The specific methods include: Using neighborhood distance to set outlier threshold , calculate the mean value of the missing value processing data set and standard deviation , and calculate the coefficient of variation , and set the K value to ,in, represents the number of samples in the dataset, represents the adjustment coefficient, , represents the smoothing factor; Determine the K value. For each data point, use the Euclidean distance to calculate the distance between the data point and its K nearest neighbors, and perform an average calculation to obtain the average distance of the K nearest neighbors. According to the rule of thumb, set the distance threshold to; when the average distance of the K nearest neighbors of a data point is greater than the distance threshold, the data point is an outlier.

7. The intellectual property operation and management system based on big data according to claim 6 is characterized in that: The specific method of using the SVR model to predict the value indicator data of intellectual property rights based on the preprocessed data set includes: The sample set includes J_o groups of samples, each of which includes a preprocessed data set and corresponding intellectual property value indicator data; Initialize hyperparameters: Select radial basis function as kernel function, determine the penalty coefficient, epsilon parameter, gamma parameter hyperparameter combination, and use grid search method to determine the optimal hyperparameter combination; Calculate the initial loss function: Calculate the initial loss function value of the model , as the starting point of optimization, where represents regularization, represents the weight vector of the input features, represents the bias term, represents the error term, is the penalty coefficient, n2 represents the total number of training samples, o represents the sample index, represents the error of the oth sample on the upper deviation, represents the error of the o-th sample on the lower deviation; Iterative optimization: Use gradient descent to update model parameters in each iteration and , calculate the new loss function during the update process; Check convergence conditions: After each round of iteration, compare the change in the current loss function value with that of the previous round. When the loss change between two adjacent iterations is less than the preset error threshold, stop the iteration.

8. The big data-based intellectual property operation and management system according to claim 7 is characterized in that: The specific method of using the TF-IDF technology to process the application name and extract the keyword group B includes: Application name preprocessing: Use the HanLP tool to segment the application name and obtain subwords; remove stop words and convert the uppercase letters of the text to lowercase to obtain preprocessed text; Count the frequency of each subword in the preprocessed text ,in, represents the frequency of occurrence of word d, Indicates the number of times word d appears in the preprocessed text, Indicates the total number of words in the preprocessed text; Calculate the inverse text frequency, the formula is as follows: ,in, Indicates the total number of texts, represents the amount of text containing subword d, represents the inverse text frequency of subword d; Calculate the TF-IDF value by multiplying the occurrence frequency TF of each subword by the inverse text frequency to get the TF-IDF value; Sort by the TF-IDF value of each subword, set a threshold A_Q, and when the TF-IDF value is greater than the threshold, the subword is used as a keyword to obtain keyword group B.

9. The intellectual property operation and management system based on big data according to claim 8 is characterized in that: Based on the user input text, the keyword A is obtained using TF-IDF technology; And match it with keyword group B. The specific methods of obtaining comprehensive matching data include: Use TF-IDF technology to extract keyword group A from user input text; Get the vectors of keyword group A and keyword group B, and use cosine similarity to calculate the similarity between keyword group A and keyword group B: ,in, A vector representing the keyword group A, represents the vector of keyword group B, represents the inner product, and represents the Euclidean norm; Calculate the Jaccard similarity, the formula is: ,in, Indicates the size of the intersection of keyword group A and keyword group B. Indicates the union size of keyword group A and keyword group B; By calculating the cosine similarity and Jaccard similarity, combined with the intellectual property value index, the comprehensive matching degree is calculated: ,in, , and Respectively represent the weights of cosine similarity, Jaccard similarity and intellectual property value index, Represents the intellectual property value indicator.

10. The intellectual property operation and management system based on big data according to claim 9, characterized in that: The specific method of setting the recommendation threshold based on the comprehensive matching data and determining whether to make a recommendation includes: Use the sliding window method to set the threshold and set a window of fixed length C_K; Calculate the mean and standard deviation of the comprehensive matching degree within the window and set the initial recommendation threshold: ,in, Represents the mean of the comprehensive matching degree in the window, Represents the standard deviation of the comprehensive matching degree in the window, represents the adjustment factor, ; When a new comprehensive match is received, the window slides forward, removes the oldest comprehensive match, adds the new match, obtains an updated window, and recalculates the mean and standard deviation to obtain the recommended threshold. ;in, represents the mean of the update window, represents the standard deviation of the update window; when When the IP is selected, the IP is recommended, and the IP has a high matching degree and meets the needs of the user; when , the intellectual property is not recommended, the matching degree of the intellectual property is low and does not meet the user's needs.

Images