Intellectual property management system and management method

Abstract

The invention belongs to the technical field of data management, and discloses an intellectual property management system and management method.The intellectual property management method comprises the steps that firstly, intellectual property data are collected and preprocessed, and a high-quality data set is obtained; secondly, key parameters are extracted from the high-quality data set, the data quality is preliminarily evaluated, a preliminary data quality index is obtained, and a preliminary data state signal is generated; then, according to the initial data state signal, extracting features from a high-quality data set, and generating a low-dimensional feature matrix; and finally, constructing a data quality evaluation model, taking the low-dimensional feature matrix as input, calculating the data quality of the intellectual property data, obtaining a data quality index, and generating a data quality report, thereby not only improving the efficiency of data processing and quality evaluation, but also providing technical support for continuous optimization of the data quality, and improving the data quality evaluation efficiency. The method is suitable for intellectual property management requirements of various enterprises and institutions.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and more specifically, to an intellectual property management system and management method. Background Art

[0002] With the development of science and technology and economy, intellectual property (IP) has become an important part of the core competitiveness of enterprises. However, traditional IP management relies on manual processing of large amounts of document information, which is inefficient and prone to errors. In addition, the lack of a systematic data quality control mechanism leads to problems in data accuracy, completeness and consistency, which affects data analysis and decision support.

[0003] In recent years, advances in big data and artificial intelligence technologies have provided new ways to improve the level of intellectual property management. However, data processing in existing intellectual property management systems is still not intelligent enough and data quality is difficult to ensure. How to effectively extract valuable information from complex and diverse intellectual property data and ensure the quality of this information remains a challenge.

[0004] In view of this, the present invention proposes an intellectual property management system and management method to solve the above problems. Summary of the invention

[0005] 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 management system, comprising:

[0006] Data processing module: used to collect and pre-process intellectual property data to obtain high-quality data sets;

[0007] Key parameter extraction module: used to extract key parameters from high-quality data sets, preliminarily evaluate data quality, obtain preliminary data quality index, and generate preliminary data status signals;

[0008] Feature extraction module: used to extract features from high-quality data sets based on preliminary data state signals and generate low-dimensional feature matrices;

[0009] Quality assessment module: used to build a data quality assessment model, take the low-dimensional feature matrix as input, calculate the data quality of intellectual property data, obtain the data quality index, and generate a data quality report.

[0010] Furthermore, the intellectual property data includes basic information, legal status, technical details, financial information and administrative and judicial information of the intellectual property item;

[0011] Basic information includes application number, applicant, name and date. Date includes application date, publication date and grant date. Technical details include abstract, claims, description and drawings.

[0012] Legal status includes current status, ongoing status, and licensing history, and administrative judicial information includes administrative decisions and litigation records;

[0013] Financial information including fee payment records and valuations;

[0014] The preprocessing method includes data cleaning, missing value processing, outlier detection, data conversion and data standardization;

[0015] Integrate the pre-processed intellectual property data, using intellectual property entries as rows and corresponding intellectual property data as columns to construct a high-quality data set with a table structure;

[0016] Furthermore, the method of calculating the preliminary data quality index of the high-quality data set and generating the preliminary data status signal includes:

[0017] Calculate the accuracy score, completeness score, and consistency score of high-quality data sets, and average them to obtain a preliminary data quality index;

[0018] A preliminary data quality threshold is preset. If the preliminary data quality index is less than the preliminary data quality threshold, the data quality of the high-quality data set is determined to be low, and a preliminary data quality unqualified signal is generated. Otherwise, a preliminary data quality qualified signal is generated.

[0019] Furthermore, the method of calculating the accuracy score, completeness score and consistency score of the high-quality data set includes:

[0020] The accuracy score is calculated as:

[0021] Divide the high-quality data set into S_a data subsets by type, and sort each data subset in ascending order;

[0022] The quartile method was used to calculate the upper quartile P1, median P2, and lower quartile P3 of each data subset;

[0023] Calculate the difference between P1 and P2 to obtain the quartile difference PX;

[0024] The sum of P1 and 1.5 times PX is defined as the maximum value of the normal range of the data subset;

[0025] The difference between P3 and 1.5 times PX is defined as the minimum value of the normal range of the data subset;

[0026] The values ​​in the data subset that are beyond the normal range are defined as outliers, and a threshold value of the number of outliers is preset. If the number of outliers in any data subset is greater than the threshold value of the number of outliers, the accuracy of the data subset is determined to be low;

[0027] If the accuracy of any data subset is low, the accuracy of the high-quality data set is judged to be low;

[0028] If the accuracy of the high-quality dataset is low, the accuracy score is recorded as 0, otherwise it is 1;

[0029] The completeness score is calculated as:

[0030] Mark the required and optional fields in the high-quality data set, traverse the high-quality data set, identify whether each field has a value and assign a field score. If there is a value, the field score is recorded as 1, otherwise, it is recorded as 0;

[0031] The field scores of the required fields and the field scores of the optional fields are averaged to obtain the comprehensive integrity scores of the required fields and the comprehensive integrity scores of the optional fields. The two scores are added together to obtain the data integrity score of the high-quality data set.

[0032] A completeness threshold is preset. If the data completeness score is less than the completeness threshold, the high-quality data set is judged to have low completeness.

[0033] If the completeness of the high-quality dataset is low, the completeness score is recorded as 0, otherwise it is 1;

[0034] The consistency score is calculated as:

[0035] Read dates from a high-quality dataset and define a logical order for each IP entry, including:

[0036] Application date < publication date < grant date;

[0037] If there are dates that are not in logical order, mark them;

[0038] Read and identify the uniqueness of the application number, identify whether the applicants under the same application number are consistent, and mark if the application number is not unique or the applicants are inconsistent;

[0039] Read and identify the current status, check the authorized intellectual property items, and mark if the authorization date is empty or the fee payment record is empty;

[0040] If any marker exists in the high-quality dataset, the consistency of the high-quality dataset is judged to be low;

[0041] If the consistency of the high-quality dataset is low, the consistency score is recorded as 0, otherwise it is recorded as 1.

[0042] Furthermore, the method of extracting features from the high-quality data set according to the preliminary data state signal to generate a low-dimensional feature matrix includes:

[0043] Step S51: Real-time monitoring of preliminary data status signals. When a signal of data quality failure appears, the high-quality data set is encoded, all data is converted into features in numerical form, and the original matrix is ​​constructed with the intellectual property entries as rows and the corresponding features as columns;

[0044] Step S52: preset an e value, and find e nearest neighbors for each data point in the original matrix;

[0045] Among them, a data point is an IP entry, and the nearest neighbor is the other data point that is closest to the data point;

[0046] Step S53: Calculate the reconstruction weight of each data point in the nearest neighbor and construct a weight matrix WL;

[0047] Step S54: construct a Laplace matrix ML=(S-WL) according to the weight matrix T (S-WL);

[0048] Among them, S is a square matrix with the same size as the weight matrix ML. The values ​​on the main diagonal position are all 1, and the rest of the values ​​are all 0;

[0049] Step S55: Use linear algebra methods to process the Laplace matrix to obtain the eigenvalues ​​and eigenvectors of the Laplace matrix, set a d value, select the eigenvectors corresponding to the smallest d non-zero eigenvalues, and construct a low-dimensional feature matrix.

[0050] Furthermore, the reconstruction weight of each data point in its nearest neighbor is calculated to construct the weight matrix WL in the following manner:

[0051] For each data point x i , with each nearest neighbor x j With x i The difference is a column, constructing a local reconstruction matrix ZL i ;

[0052] Where i is the index of the data point in the original matrix, i = 1, 2, ..., N, N is the number of data points, and j is the index of the nearest neighbor, j = 1, 2, ..., e;

[0053] The local reconstruction matrix ZL is calculated by the weight vector formula i Calculate and get the data point x i The reconstruction weight w ij , the reconstruction weights of all data points are combined into a weight matrix WL. For data points that are not nearest neighbors, the weight w ij Set to 0;

[0054] The calculation method of the weight vector formula is:

[0055]

[0056] in, yes The inverse matrix of , l is the index of the number of columns of the inverse matrix, and m is the index of the number of rows of the inverse matrix.

[0057] Furthermore, the method of constructing the data quality assessment model is:

[0058] Use elastic network regression to build a data quality assessment model, with a low-dimensional feature matrix as input and a data quality index for each data point as output;

[0059] The training process of the data quality assessment model is:

[0060] Collect historical data sets containing historical intellectual property data and their corresponding label data, and construct training data sets (Y, y) through the historical data sets;

[0061] Among them, Y is a low-dimensional feature matrix, y is the corresponding label data, and the label data is an N-dimensional vector, which represents the data quality index of each data point;

[0062] The training data set is divided into two data groups according to a preset ratio, which are used as training set and validation set respectively;

[0063] Select the best hyperparameters α and r, input the training set into the data quality assessment model, use the gradient descent method to minimize the loss function of the data quality assessment model, and obtain the optimal parameter β0 value and β value;

[0064] The validation set is input into the data quality assessment model to obtain the predicted data quality index for each data point. 2 The score is used as an indicator to evaluate the model performance and calculate the R of the predicted data quality index. 2 Score, if R 2 If the score is greater than or equal to the preset evaluation index threshold, the model performance meets the standard;

[0065] The loss function is:

[0066]

[0067] Among them, yc i is the true value of the data quality index of the ith data point, u is a vector of all 1s with a size of N × 1, and β0 and β are the parameters of the data quality assessment model.

[0068] Furthermore, the method of selecting the best hyperparameters α and r is:

[0069] Step S81: define the value range of α and r, preset a series of values ​​of α and r, list all possible values ​​of each hyperparameter, and form a hyperparameter network;

[0070] Step S82: For each pair of [α, r] combinations, use the current hyperparameter combination to train the data quality assessment model on the training set;

[0071] Step S83: define mean square error as the evaluation index of the hyperparameter, set a k value, use k-fold cross validation method to evaluate the performance of the model, and calculate and record the evaluation index of each hyperparameter combination;

[0072] Step S84: taking the average of the k evaluation indicators of each hyperparameter combination to obtain the average evaluation indicator of the hyperparameter combination;

[0073] Step S85: Select the hyperparameter combination with the highest average evaluation index as the optimal hyperparameters α and r.

[0074] Furthermore, the method of calculating the data quality of the intellectual property data, obtaining the data quality index, and generating the data quality report includes:

[0075] Use the trained data quality assessment model to predict the low-dimensional feature matrix, obtain the data quality index of each data point, organize it, generate a data quality report and analyze it;

[0076] The data quality report includes the quality index of each data point, overall data quality statistical indicators and visual charts;

[0077] Ways to generate data quality reports include:

[0078] Number and organize the data quality index and data points of each data point to create a visual data frame;

[0079] In the visualization data frame, the basic statistical indicators and distribution statistical indicators of each data quality index are calculated and integrated into a table form, which is recorded as the overall data quality statistical indicators;

[0080] The basic statistical indicators include the mean, standard deviation, minimum, maximum and median of the quality index; the distribution statistical indicators are the percentiles of the quality index;

[0081] Draw a histogram and box plot of the quality index, and draw a scatter plot of the quality index and each feature in the corresponding data point to integrate into a visual chart.

[0082] Furthermore, the method of generating a data quality report and performing analysis includes:

[0083] Set the difference thresholds for mean and median, standard deviation threshold, normal range values ​​for minimum and maximum values, and percentile gap thresholds;

[0084] Analyze the data quality report. If the difference between the mean and the median is greater than the set difference threshold, it is determined that the intellectual property data is skewed or outliers exist;

[0085] If the standard deviation is greater than the standard deviation threshold, it is determined that the difference in the intellectual property data is too large;

[0086] If the minimum or maximum value is greater than the normal range, it is determined that there is an input error in the intellectual property data;

[0087] If the difference between each percentile is greater than the gap threshold, it is determined that there is an abnormal distribution problem in the intellectual property data.

[0088] The technical effects and advantages of an intellectual property management system and management method of the present invention are as follows:

[0089] The invention provides an efficient, comprehensive and intelligent intellectual property management system; the system realizes data cleaning, missing value processing, outlier detection, data conversion and standardization through an automated data processing module, thereby improving the efficiency and accuracy of data processing; a key parameter extraction module and a feature extraction module evaluate the accuracy, completeness and consistency of data from multiple dimensions, generate a preliminary data quality index, and lay a foundation for in-depth evaluation; a data quality evaluation model constructed using elastic network regression, combined with a low-dimensional feature matrix, can effectively process complex data environments and improve the accuracy and reliability of evaluation; the best hyperparameter combination is selected through a k-fold cross-validation method to ensure the generalization ability and stability of the model; the generated data quality report includes detailed data quality indexes, statistical indicators and visual charts to help users intuitively understand the data quality status; a diagnostic improvement module monitors data quality in real time, automatically identifies and diagnoses problems, formulates and implements improvement measures, and continuously improves data quality; the system also automatically identifies problems such as data skew, outliers, excessive differences, input errors and abnormal distribution by setting thresholds, and performs targeted optimization in combination with visual charts to ensure continuous improvement of data quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A schematic diagram of an intellectual property management system of the present invention;

[0091] Figure 2 A schematic diagram of an intellectual property management method of the present invention. DETAILED DESCRIPTION

[0092] 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.

[0093] Example 1

[0094] See also Figure 1 As shown, the intellectual property management system described in this embodiment includes:

[0095] Data processing module: used to collect and pre-process intellectual property data to obtain high-quality data sets;

[0096] Key parameter extraction module: used to extract key parameters from high-quality data sets, preliminarily evaluate data quality, and generate preliminary data quality index;

[0097] Feature extraction module: used to extract features from high-quality data sets based on preliminary data quality indexes and generate low-dimensional feature matrices;

[0098] Quality assessment module: used to take the low-dimensional feature matrix as input, build a data quality assessment model, calculate the data quality of intellectual property data, obtain the data quality index, and generate a data quality report;

[0099] Diagnosis and improvement module: used to diagnose problems and make optimizations based on data quality reports;

[0100] Furthermore, the intellectual property data includes basic information, legal status, technical details, financial information and administrative and judicial information;

[0101] Basic information includes application number, applicant, name and date. Date includes application date, publication date and grant date. Technical details include abstract, claims, description and drawings. They are obtained through internal databases and API interfaces of authoritative data sources.

[0102] Legal status includes current status, ongoing status, and licensing history, and administrative judicial information includes administrative decisions and litigation records, obtained through court websites and legal databases;

[0103] Financial information includes fee payment records and valuations, obtained through internal financial systems;

[0104] The preprocessing method includes data cleaning, missing value processing, outlier detection, data conversion and data standardization;

[0105] Data cleaning includes deduplication (removing duplicate records to ensure data uniqueness), formatting (converting data into a unified format to ensure data consistency), and verification (verifying data integrity and accuracy, such as checking whether required fields are filled in and whether the date format is correct);

[0106] Data standardization is the conversion of data into a unified unit and format (e.g., unified date format, currency unit);

[0107] Missing value processing is to fill or delete missing data to ensure data integrity;

[0108] Outlier detection is to identify and process abnormal data (e.g. values ​​outside a reasonable range);

[0109] Data conversion includes text conversion and encoding (cleaning and standardizing text data and converting it into numerical form), image encoding (flattening the attached image into a one-dimensional vector), categorical variable encoding (converting categorical variables into numerical form), data normalization (normalizing numerical data to a specific range), data discretization (dividing continuous numerical data into discrete intervals), and data aggregation (merging multiple related data fields into one field);

[0110] Integrate the pre-processed intellectual property data, using intellectual property entries as rows and corresponding intellectual property data as columns to construct a high-quality data set with a table structure;

[0111] It should be noted that an IP entry is an independent IP record, and each entry contains a series of attributes describing the IP, i.e., IP data (for example, if there is a dataset containing 4 patents, and the IP data of each patent has 5 data types, such as [application number, applicant, application date, valuation, number of licenses], then in this example, each row in the high-quality dataset represents a patent entry, row 1: Patent 1, row 2: Patent 2..., and each column represents a data type, then column 1: application number, column 2: applicant...);

[0112] Furthermore, the method of calculating the preliminary data quality index of the high-quality data set and generating the preliminary data status signal includes:

[0113] Calculate the accuracy score, completeness score, and consistency score of high-quality data sets, and average them to obtain a preliminary data quality index;

[0114] A preliminary data quality threshold is preset. If the preliminary data quality index is less than the preliminary data quality threshold, the data quality of the high-quality data set is determined to be low, and a preliminary data quality failure signal is generated;

[0115] Otherwise, a preliminary data quality qualified signal is generated;

[0116] Furthermore, the accuracy score is calculated as follows:

[0117] The high-quality data set is divided into S_a data subsets by type, each data subset is sorted in ascending order, and the upper quartile P1 (the value at the 75% position after the data is sorted), median P2 (the value at the 50% position after the data is sorted), and lower quartile P3 (the value at the 25% position after the data is sorted) of each data subset are calculated using the quartile method;

[0118] Calculate the difference between the upper quartile and the lower quartile to obtain the quartile difference PX, that is, P1-P2=PX;

[0119] The sum of the upper quartile and 1.5 times the quartile difference is defined as the maximum value of the normal range of the data subset, that is, P1+1.5×PX, and the difference between the lower quartile and 1.5 times the quartile difference is defined as the minimum value of the normal range of the data subset, that is, P3-1.5×PX;

[0120] The values ​​in the data subset that are beyond the normal range are defined as outliers, and a threshold value for the number of outliers is preset. If the number of outliers in any data subset is greater than the threshold value for the number of outliers, the accuracy of the data subset is determined to be low. If the accuracy of any data subset is low, the accuracy of the high-quality data set is determined to be low.

[0121] If the accuracy of the high-quality dataset is low, the accuracy score is recorded as 0, otherwise it is 1;

[0122] The completeness score is calculated as:

[0123] Mark the required and optional fields in the high-quality data set, traverse the high-quality data set, identify whether each field has a value (that is, whether each field is blank, if the field is not blank, it has a value, if the field is blank, it has no value) and assign a field score. If there is a value, the field score is recorded as 1, otherwise, it is recorded as 0;

[0124] It should be noted that in data processing and quality assessment, fields are usually divided into required fields and optional fields. These classifications help ensure the integrity and consistency of the data;

[0125] Mandatory fields are fields that must be filled in the data record. These fields are critical to the integrity of the data and subsequent analysis. Data entries that lack mandatory fields are generally considered incomplete and may result in unusable data or inaccurate analysis results.

[0126] For example, in a high-quality dataset, mandatory fields might include: application number, applicant, filing date, publication date, grant date, current status;

[0127] Optional fields are fields that can be left blank in a data record. Although these fields are not mandatory, they can provide additional information that helps to more fully understand the data. Although optional fields are not mandatory, they can enhance the richness of the data and the depth of the analysis.

[0128] For example, in a high-quality data set, optional fields might include: valuation, number of permits, number of administrative decisions, number of lawsuits;

[0129] Use the mean calculation method to process the field scores of all required fields and all optional fields in the high-quality data set respectively, obtain the comprehensive integrity scores of the required fields and the comprehensive integrity scores of the optional fields, and add them together to obtain the data integrity score;

[0130] A completeness threshold is preset. If the data completeness score is less than the completeness threshold, the high-quality data set is judged to have low completeness.

[0131] If the completeness of the high-quality dataset is low, the completeness score is recorded as 0, otherwise it is 1;

[0132] The consistency score is calculated as:

[0133] Read dates from a high-quality dataset and define a logical order for each IP entry, including:

[0134] Application date < publication date, publication date < grant date;

[0135] If there are dates that are not in logical order, mark them;

[0136] Read and identify the uniqueness of the application number, and identify whether the applicant name under the same application number is consistent. If the application number is not unique or the applicant name is inconsistent, mark it;

[0137] Read and identify the current status of the intellectual property, check the authorized intellectual property, and mark if there is an empty authorization date or fee payment record;

[0138] If any marker exists in the high-quality dataset, the consistency of the high-quality dataset is judged to be low;

[0139] If the consistency of the high-quality dataset is low, the consistency score is recorded as 0, otherwise it is recorded as 1;

[0140] It should be noted that the preliminary data quality index is used to detect whether the data has obvious quality problems or anomalies;

[0141] Furthermore, the method of extracting features from the high-quality data set according to the preliminary data state signal to generate a low-dimensional feature matrix includes:

[0142] Step S51: Real-time monitoring of preliminary data status signals. When a data quality failure signal appears, the high-quality data set is encoded (including using one-hot encoding for category data, word embedding encoding for text data, encoding date and time data as timestamps, and performing standardization on numerical data). All data in the high-quality data set (including text, category, value, and time) are converted into features in numerical form, and the original matrix is ​​constructed with intellectual property entries as rows and corresponding features as columns.

[0143] Wherein, each row represents an intellectual property entry, each column represents a feature, and the size of the original matrix is ​​N×D, where N is the number of data points and D is the dimension of the feature (for example, assuming there is a high-quality dataset containing [entry ID, application date, inventor, technical field, applicant], where the content of the first row is: entry ID: 001, application date: 2020-01-0, inventor: Zhang San, technical field: electronics, applicant: Company A. After encoding, a feature matrix in pure digital form can be obtained: entry ID: 001, application date: 1577836800, inventor: 1, technical field: 1, applicant: 1. In this example, the application date is converted to a timestamp, and the inventor, technical field, and applicant are encoded in numerical form);

[0144] Step S52: preset an e value, and for each data point in the original matrix, find the e nearest neighbor data points;

[0145] Among them, a data point in the original matrix is ​​an IP entry, which is used to represent the feature values ​​of different IP entries under the same feature, and the nearest neighbor is the other data point closest to the data point;

[0146] Step S53: Calculate the reconstruction weight of each data point in its nearest neighbor and construct a weight matrix WL;

[0147] Step S54: constructing a Laplace matrix according to the weight matrix;

[0148] Among them, the Laplace matrix M is defined as: M = (S-WL) T (S-WL), S is a square matrix with the same size as the weight matrix ML, the values ​​on its main diagonal are all 1, and the rest are all 0;

[0149] Step S55: Use linear algebra methods to process the Laplace matrix to obtain eigenvalues ​​and eigenvectors, that is, use linear algebra methods to solve the characteristic equation: M v =λ v, where v is the eigenvector and λ is the eigenvalue (for example, there is a 3×3 Laplacian matrix, assuming that the result of the eigenvalue solution is the eigenvalues: λ1=0, λ2=0.4, λ3=0.8, eigenvectors: the eigenvector corresponding to λ1=0 is v1=[0.5,0.5,0.5], the eigenvector corresponding to λ2=0.4 is v2=[0.5,-0.5,0.5], the eigenvector corresponding to λ3=0.8 is v3=[0.5,0.5,-0.5]);

[0150] Set a d value, select the eigenvectors corresponding to the smallest d non-zero eigenvalues, and construct a low-dimensional feature matrix (for example, assuming d = 2, select the eigenvectors corresponding to the smallest two non-zero eigenvalues, eigenvalues ​​λ2 = 0.4, λ3 = 0.8, the eigenvector v2 = [0.5, -0.5, 0.5] corresponding to λ2 = 0.4, the eigenvector v3 = [0.5, 0.5, -0.5] corresponding to λ3 = 0.8, and combine these eigenvectors into a low-dimensional feature matrix [0.5, 0.5, -0.5, 0.5, 0.5, -0.5]);

[0151] It should be noted that the Laplacian matrix is ​​a symmetric matrix whose eigenvalues ​​and eigenvectors have special properties;

[0152] Among them, the eigenvalues ​​of the Laplacian matrix are non-negative real numbers. These eigenvalues ​​usually represent the local structural information between data points. Smaller eigenvalues ​​correspond to the global structure in the data set, and larger eigenvalues ​​correspond to local details.

[0153] Each eigenvalue of the Laplacian matrix corresponds to an eigenvector, which is a vector of unit length and is associated with the corresponding eigenvalue. The eigenvector represents the coordinates of the data point in the new low-dimensional space;

[0154] The eigenvectors corresponding to the smallest d non-zero eigenvalues ​​are usually associated with the intrinsic structure of the data, which can help extract the main features of the data. The constructed low-dimensional feature matrix can retain the main structural information of the data, which is helpful for subsequent data analysis and processing;

[0155] Furthermore, the reconstruction weight of each data point in its nearest neighbor is calculated to construct the weight matrix WL as follows:

[0156] For each data point x i , construct an e×D local reconstruction matrix Z i , where each column is x i A nearest neighbor x j With x i Difference: Z i =[x i4 -x i ,xi2 -x i ,…,x ie -x i ](For example, suppose there is an original matrix containing 3 data points with a feature dimension of 2, that is, a 3×2 original matrix, which is flattened row-first to [1,2,3,4,5,6]. Each data point has 2 nearest neighbors, that is, e=2. The nearest neighbors of the data point x1 are x2[3,4], x3[5,6], the nearest neighbors of x2 are x1, x3, and the nearest neighbors of x3 are x1, x2. For the data point x1[1,2], construct the local reconstruction matrix Z1=[(3-1)(5-1)(4-2)(6-2)]);

[0157] The local reconstruction matrix ZL is calculated by the weight vector formula i Calculate and get the data point x i The reconstruction weight w ij (The reconstruction weight indicates how to reconstruct the current data point through the nearest neighbor data point, that is, for each data point x i , find a set of weights such that x i can be represented by its nearest neighbor x j Linear reconstruction, at the same time, the sum of weights is 1: ∑ e∈xj w ij =1), the reconstruction weights of all data points are combined into an N×N weight matrix W. In the weight matrix, each position represents an element, and the element w ij Represents data point x i At data point x j The reconstruction weight on the data point is not the nearest neighbor, and the weight w ij Set to 0 (for example, assuming the number of data points is 3, the weight matrix W is 3×3, where w12, w13 are the reconstruction weights of x1 on x2, x3, w21, w23 are the reconstruction weights of x2 on x1, x3, w31, w32 are the reconstruction weights of x3 on x1, x2, and the final weight matrix W is [0, w12, w13, w21, 0, w23, w31, w32, 0], where w11, w22, w33 are not the nearest neighbor data points, and the corresponding weights are set to 0);

[0158] Among them, the calculation method of the weight vector formula is:

[0159]

[0160] in, yes The inverse matrix of , l is the index of the number of columns of the inverse matrix, m is the index of the number of rows of the inverse matrix, Represents the inverse matrix The element in the jth row and lth column of represents the sum of the elements in the lth column of the inverse matrix;

[0161] Furthermore, the method of constructing the data quality assessment model is:

[0162] Use elastic network regression to build a data quality assessment model, with a low-dimensional feature matrix as input and a data quality index for each data point as output;

[0163] The training process of the data quality assessment model is:

[0164] Collect historical data sets containing historical intellectual property data and their corresponding label data, and construct training data sets (Y, y) through the historical data sets;

[0165] Where Y is the low-dimensional feature matrix, y is the corresponding label data, the label data is defined as an N-dimensional vector representing the data quality index of each data point, represents the data quality index of each data point, and N is the number of data points in the low-dimensional feature matrix;

[0166] The training data set is divided into two data groups according to a preset ratio, which are used as training set and validation set respectively;

[0167] Select the best hyperparameters α and r, input the training set into the data quality assessment model, use the gradient descent method to minimize the loss function of the data quality assessment model, and obtain the optimal parameter β0 value and β value;

[0168] The validation set is input into the data quality assessment model to obtain the predicted data quality index for each data point. 2 The score is used as an evaluation indicator to calculate the R of the predicted data quality index. 2 Score, if R 2 If the score is greater than or equal to the preset evaluation index threshold, the model performance meets the standard;

[0169] The loss function is:

[0170]

[0171] Among them, yc i is the true value of the data quality index of the ith data point, u is a vector of all 1s, size N×1, β0 is the parameter of the data quality assessment model, β is the vector parameter of the data quality assessment model, is the prediction error term of the data quality assessment model, which is used to measure the difference between the model prediction value and the true value, ∥yc i -β01+Yβ∥ 2 is the square of the Euclidean distance between the true value of the data quality index and the predicted value of the data quality index, that is, the residual sum of squares, is a regularization term used to prevent model overfitting, α is the regularization strength, which controls the strength of regularization, r is the proportion of L1 regularization in the total regularization, ranging from 0 to 1, ∥β∥ represents the norm of vector β, ∥β∥1 is the L1 norm of vector β, that is, the sum of the elements in vector β, representing the L1 regularization term, which is used to reduce the coefficient of the feature to zero, thereby performing feature selection, ∥β∥2 is the L2 norm of vector β, that is, the square root of the sum of the squares of the elements in vector β, It is the square of the L2 norm of the vector β, representing the L2 regularization term, which is used to smoothly reduce the coefficients of all features to deal with multicollinearity;

[0172] Furthermore, the method of selecting the best hyperparameters α and r is:

[0173] Step S81: define the value range of α and r, preset a series of values ​​of α and r, list all possible values ​​of each hyperparameter, and form a hyperparameter network;

[0174] Among them, listing all possible values ​​of each hyperparameter is to traverse all values ​​of α, and for each value of α, traverse all values ​​of r;

[0175] Step S82: For each pair of [α, r] combinations, use the current hyperparameter combination to train the data quality assessment model on the training set;

[0176] Step S83: define mean square error as the evaluation index of the hyperparameter, set a k value (e.g., 5, 6, 7), use k-fold cross validation method to evaluate the performance of the model, and calculate and record the evaluation index of each hyperparameter combination;

[0177] Among them, the method of using the k-fold cross validation method is:

[0178] Divide the training set into k subsets;

[0179] For each pair of hyperparameter combinations:

[0180] First cycle: select the first subset as the validation set, merge the remaining subsets into the training set, use the training set to train the model, use the validation set to evaluate the model, and calculate and record the evaluation indicators;

[0181] Second cycle: select the second subset as the validation set, merge the remaining subsets into the training set, use the training set to train the model, use the validation set to evaluate the model, and calculate and record the evaluation indicators;

[0182] And so on, until the kth cycle;

[0183] For each pair of hyperparameter combinations, k evaluation indicators are calculated and recorded;

[0184] Step S84: taking the average of the k evaluation indicators of each hyperparameter combination to obtain the average evaluation indicator of the hyperparameter combination;

[0185] Step S85: Select the hyperparameter combination with the highest average evaluation index as the optimal hyperparameters α and r; Further, the method of calculating the data quality of the intellectual property data, obtaining the data quality index, and generating a data quality report includes:

[0186] Use the trained data quality assessment model to predict the low-dimensional feature matrix, obtain the data quality index of each data point, organize it, generate a data quality report and analyze it;

[0187] The data quality report includes the quality index of each data point, overall data quality statistical indicators and visual charts;

[0188] Ways to generate data quality reports include:

[0189] Number and sort the data quality index and data points of each data point to create a data frame;

[0190] Calculate the basic statistical indicators and distribution statistical indicators of each data quality index and integrate them into a table form, which is recorded as the overall data quality statistical indicators;

[0191] The basic statistical indicators include the mean, standard deviation, minimum, maximum and median of the quality index; the distribution statistical indicators are the percentiles of the quality index;

[0192] Draw a histogram of the quality index (to show the distribution of the data) and a box plot (to show the distribution of the data and outliers), and draw a scatter plot of the quality index and each feature in the corresponding data point (to observe the relationship between the two), and integrate them into a visual chart;

[0193] The method of generating a data quality report and performing analysis includes:

[0194] Set the difference thresholds for mean and median, standard deviation threshold, normal range values ​​for minimum and maximum values, and percentile gap thresholds;

[0195] Analyze the data quality report. If the difference between the mean and the median is greater than the set difference threshold, it is determined that the intellectual property data is skewed or outliers exist;

[0196] If the standard deviation is greater than the standard deviation threshold, it is determined that the difference in the intellectual property data is too large;

[0197] If the minimum or maximum value is greater than the normal range, it is determined that there is an input error in the intellectual property data;

[0198] If the difference between the 25%, 50%, and 75% percentiles is greater than the gap threshold, it is determined that the intellectual property data has an abnormal distribution problem;

[0199] Based on the identified data quality issues, combined with visual charts, targeted optimization of the problems existing in intellectual property data is carried out;

[0200] Regenerate data quality reports after optimization to evaluate the effectiveness of targeted optimization until no data quality issues with the IP data are identified;

[0201] Among them, combined with visual charts, you can find out the source of the problem by checking the histogram, box plot and scatter plot, including:

[0202] By examining the histogram, we can identify whether the distribution of the IP data is as expected and whether there is obvious multi-peak or multi-modal phenomena;

[0203] By examining the box plot, identifying outliers and outliers, the central tendency and dispersion of the IP data can be analyzed;

[0204] By checking the scatter plot, analyze the relationship between each feature in the data point and the data quality index to find potential influencing factors;

[0205] In this embodiment, data cleaning, missing value processing, outlier detection, data conversion and standardization are realized through the automated data processing module, thereby improving the efficiency and accuracy of data processing; the key parameter extraction module and the feature extraction module evaluate the accuracy, completeness and consistency of the data from multiple dimensions, generate a preliminary data quality index, and lay the foundation for in-depth evaluation; the data quality evaluation model constructed using elastic network regression, combined with a low-dimensional feature matrix, can effectively handle complex data environments and improve the accuracy and reliability of the evaluation; the optimal hyperparameter combination is selected through the k-fold cross-validation method to ensure the generalization ability and stability of the model; the diagnosis and improvement module monitors data quality in real time, automatically identifies and diagnoses problems, formulates and implements improvement measures, and continuously improves data quality; the generated data quality report includes detailed data quality indexes, statistical indicators and visual charts to help users intuitively understand the data quality status; the system also automatically identifies problems such as data skew, outliers, excessive differences, input errors and abnormal distribution by setting thresholds, and performs targeted optimization in combination with visual charts to ensure continuous improvement of data quality.

[0206] Example 2

[0207] 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 management method, including:

[0208] S1. Collect and preprocess intellectual property data to obtain high-quality data sets;

[0209] S2, extract key parameters from high-quality data sets, preliminarily evaluate data quality, and generate preliminary data quality index;

[0210] S3, extract features from the high-quality data set based on the preliminary data quality index and generate a low-dimensional feature matrix;

[0211] S4. Build a data quality assessment model, take the low-dimensional feature matrix as input, calculate the data quality of the intellectual property data, obtain the data quality index, and generate a data quality report.

[0212] Example 3

[0213] 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 management method provided above is implemented.

[0214] Since the electronic device introduced in this embodiment is an electronic device used to implement an intellectual property management method in the embodiment of the present application, based on the intellectual property management method introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as a person skilled in the art implements an electronic device used in an intellectual property management method in the embodiment of the present application, it belongs to the scope of protection of this application.

[0215] 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.

[0216] 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. An intellectual property management system, characterized in that: include: Data processing module: used to collect and pre-process intellectual property data to obtain high-quality data sets; Key parameter extraction module: used to extract key parameters from high-quality data sets, preliminarily evaluate data quality, obtain preliminary data quality index, and generate preliminary data status signals; Feature extraction module: used to extract features from high-quality data sets based on preliminary data state signals and generate low-dimensional feature matrices; Quality assessment module: used to build a data quality assessment model, take the low-dimensional feature matrix as input, calculate the data quality of intellectual property data, obtain the data quality index, and generate a data quality report.

2. An intellectual property management system according to claim 1, characterized in that: The intellectual property data includes basic information, legal status, technical details, financial information and administrative and judicial information of the intellectual property items; Basic information includes application number, applicant, name and date. Date includes application date, publication date and grant date. Technical details include abstract, claims, description and drawings. Legal status includes current status, ongoing status, and licensing history, and administrative judicial information includes administrative decisions and litigation records; Financial information including fee payment records and valuations; The preprocessing method includes data cleaning, missing value processing, outlier detection, data conversion and data standardization; The preprocessed intellectual property data is integrated to construct a high-quality data set with a table structure, with intellectual property entries as rows and corresponding intellectual property data as columns.

3. An intellectual property management system according to claim 2, characterized in that: The method of calculating the preliminary data quality index of the high-quality data set and generating the preliminary data status signal includes: Calculate the accuracy score, completeness score, and consistency score of high-quality data sets, and average them to obtain a preliminary data quality index; A preliminary data quality threshold is preset. If the preliminary data quality index is less than the preliminary data quality threshold, the data quality of the high-quality data set is determined to be low, and a preliminary data quality unqualified signal is generated. Otherwise, a preliminary data quality qualified signal is generated.

4. An intellectual property management system according to claim 3, characterized in that: The method of calculating the accuracy score, completeness score and consistency score of the high-quality data set includes: The accuracy score is calculated as: Divide the high-quality data set into S_a data subsets by type, and sort each data subset in ascending order; The quartile method was used to calculate the upper quartile P1, median P2, and lower quartile P3 of each data subset; Calculate the difference between P1 and P2 to obtain the quartile difference PX; The sum of P1 and 1.5 times PX is defined as the maximum value of the normal range of the data subset; The difference between P3 and 1.5 times PX is defined as the minimum value of the normal range of the data subset; The values ​​in the data subset that are beyond the normal range are defined as outliers, and a threshold value of the number of outliers is preset. If the number of outliers in any data subset is greater than the threshold value of the number of outliers, the accuracy of the data subset is determined to be low; If the accuracy of any data subset is low, the accuracy of the high-quality data set is judged to be low; If the accuracy of the high-quality dataset is low, the accuracy score is recorded as 0, otherwise it is 1; The completeness score is calculated as: Mark the required and optional fields in the high-quality data set, traverse the high-quality data set, identify whether each field has a value and assign a field score. If there is a value, the field score is recorded as 1, otherwise, it is recorded as 0; The field scores of the required fields and the field scores of the optional fields are averaged to obtain the comprehensive integrity scores of the required fields and the comprehensive integrity scores of the optional fields. The two scores are added together to obtain the data integrity score of the high-quality data set. A completeness threshold is preset. If the data completeness score is less than the completeness threshold, the high-quality data set is judged to have low completeness. If the completeness of the high-quality dataset is low, the completeness score is recorded as 0, otherwise it is 1; The consistency score is calculated as: Read dates from a high-quality dataset and define a logical order for each IP entry, including: Application date < publication date < grant date; If there are dates that are not in logical order, mark them; Read and identify the uniqueness of the application number, identify whether the applicants under the same application number are consistent, and mark if the application number is not unique or the applicants are inconsistent; Read and identify the current status, check the authorized intellectual property items, and mark if the authorization date is empty or the fee payment record is empty; If any marker exists in the high-quality dataset, the consistency of the high-quality dataset is judged to be low; If the consistency of the high-quality dataset is low, the consistency score is recorded as 0, otherwise it is recorded as 1.

5. The intellectual property management system according to claim 1, characterized in that: The method of extracting features from a high-quality data set and generating a low-dimensional feature matrix according to the preliminary data state signal includes: Step S51: Real-time monitoring of preliminary data status signals. When a signal of data quality failure appears, the high-quality data set is encoded, all data is converted into features in numerical form, and the original matrix is ​​constructed with the intellectual property entries as rows and the corresponding features as columns; Step S52: preset an e value, and find e nearest neighbors for each data point in the original matrix; Among them, a data point is an IP entry, and the nearest neighbor is the other data point that is closest to the data point; Step S53: Calculate the reconstruction weight of each data point in the nearest neighbor and construct a weight matrix WL; Step S54: construct a Laplace matrix ML=(S-WL) according to the weight matrix T (S-WL); Among them, S is a square matrix with the same size as the weight matrix ML. The values ​​on the main diagonal position are all 1, and the rest of the values ​​are all 0; Step S55: Use linear algebra methods to process the Laplace matrix to obtain the eigenvalues ​​and eigenvectors of the Laplace matrix, set a d value, select the eigenvectors corresponding to the smallest d non-zero eigenvalues, and construct a low-dimensional feature matrix.

6. An intellectual property management system according to claim 5, characterized in that: The method of calculating the reconstruction weight of each data point in its nearest neighbor and constructing the weight matrix WL is as follows: For each data point x i , with each nearest neighbor x j With x i The difference is a column, constructing a local reconstruction matrix ZL i ; Where i is the index of the data point in the original matrix, i = 1, 2, ..., N, N is the number of data points, and j is the index of the nearest neighbor, j = 1, 2, ..., e; The local reconstruction matrix ZL is calculated by the weight vector formula i Calculate and get the data point x i The reconstruction weight w ij , the reconstruction weights of all data points are combined into a weight matrix WL. For data points that are not nearest neighbors, the weight w ij Set to 0; The calculation method of the weight vector formula is: in, yes The inverse matrix of , l is the index of the number of columns of the inverse matrix, and m is the index of the number of rows of the inverse matrix.

7. An intellectual property management system according to claim 1, characterized in that: The method of constructing the data quality assessment model is: Use elastic network regression to build a data quality assessment model, with a low-dimensional feature matrix as input and a data quality index for each data point as output; The training process of the data quality assessment model is: Collect historical data sets containing historical intellectual property data and their corresponding label data, and construct training data sets (Y, y) through the historical data sets; Among them, Y is a low-dimensional feature matrix, y is the corresponding label data, and the label data is an N-dimensional vector, which represents the data quality index of each data point; The training data set is divided into two data groups according to a preset ratio, which are used as training set and validation set respectively; Select the best hyperparameters α and r, input the training set into the data quality assessment model, use the gradient descent method to minimize the loss function of the data quality assessment model, and obtain the optimal parameter β0 value and β value; The validation set is input into the data quality assessment model to obtain the predicted data quality index for each data point. 2 The score is used as an indicator to evaluate the model performance and calculate the R of the predicted data quality index. 2 Score, if R 2 If the score is greater than or equal to the preset evaluation index threshold, the model performance meets the standard; The loss function is: Among them, yc i is the true value of the data quality index of the ith data point, u is a vector of all 1s with a size of N × 1, and β0 and β are the parameters of the data quality assessment model.

8. An intellectual property management system according to claim 7, characterized in that: The method for selecting the best hyperparameters α and r is: Step S81: define the value range of α and r, preset a series of values ​​of α and r, list all possible values ​​of each hyperparameter, and form a hyperparameter network; Step S82: For each pair of [α, r] combinations, use the current hyperparameter combination to train the data quality assessment model on the training set; Step S83: define mean square error as the evaluation index of the hyperparameter, set a k value, use k-fold cross validation method to evaluate the performance of the model, and calculate and record the evaluation index of each hyperparameter combination; Step S84: taking the average of the k evaluation indicators of each hyperparameter combination to obtain the average evaluation indicator of the hyperparameter combination; Step S85: Select the hyperparameter combination with the highest average evaluation index as the optimal hyperparameters α and r.

9. The intellectual property management system according to claim 1, characterized in that: The method of calculating the data quality of the intellectual property data, obtaining the data quality index, and generating the data quality report includes: Use the trained data quality assessment model to predict the low-dimensional feature matrix, obtain the data quality index of each data point, organize it, generate a data quality report and analyze it; The data quality report includes the quality index of each data point, overall data quality statistical indicators and visual charts; Ways to generate data quality reports include: Number and organize the data quality index and data points of each data point to create a visual data frame; In the visualization data frame, the basic statistical indicators and distribution statistical indicators of each data quality index are calculated and integrated into a table form, which is recorded as the overall data quality statistical indicators; The basic statistical indicators include the mean, standard deviation, minimum, maximum and median of the quality index; the distribution statistical indicators are the percentiles of the quality index; Draw a histogram and box plot of the quality index, and draw a scatter plot of the quality index and each feature in the corresponding data point to integrate into a visual chart.

10. An intellectual property management system according to claim 9, characterized in that: The method of generating a data quality report and performing analysis includes: Set the difference thresholds for mean and median, standard deviation threshold, normal range values ​​for minimum and maximum values, and percentile gap thresholds; Analyze the data quality report. If the difference between the mean and the median is greater than the set difference threshold, it is determined that the intellectual property data is skewed or outliers exist; If the standard deviation is greater than the standard deviation threshold, it is determined that the difference in the intellectual property data is too large; If the minimum or maximum value is greater than the normal range, it is determined that there is an input error in the intellectual property data; If the difference between each percentile is greater than the gap threshold, it is determined that there is an abnormal distribution problem in the intellectual property data.

Images