A data mining-based auxiliary system and method for technology identification

Abstract

The invention discloses an auxiliary system and method based on data mining for technology identification, and relates to the field of data matching analysis. The invention trains and tests a recurrent neural network model to construct a final recurrent neural network model, and provides a model for extracting keywords from patent document data to be analyzed later; trains and tests an SVM classification model to construct a final SVM classification model, and provides a classification model for classifying keywords from patent document data to be analyzed extracted by the final recurrent neural network model later; and simultaneously stores the queried technology identification data in separate libraries, and allocates different permissions to different users according to the number of times different users call these different types of technology identification data, thereby reducing invalid calls by users, avoiding resource occupation, and improving the efficiency of calls.

Description

Technical Field

[0001] The present invention belongs to the field of data matching analysis, and in particular, relates to an auxiliary system and method based on data mining for technology identification. Background Art

[0002] At present, the most troublesome problem for intellectual property judges in intellectual property infringement cases is the judgment and identification of professional technology. In practice, they can only rely on the technical investigator system for assistance. Due to the limited size of the technical investigator team, the accuracy and efficiency of technical identification have always been a problem. At the same time, due to the diversity of identification data, different user types of each technical identification have different usage of different types of identification data. If the data access rights of users using technical identification data cannot be controlled and optimized, not only the security and confidentiality of the identification data cannot be guaranteed, but also the utilization efficiency of the identification data is affected. Summary of the invention

[0003] In view of the problems in the related art, the present invention proposes an auxiliary system and method based on data mining for technology identification to overcome the above-mentioned technical problems existing in the existing related art.

[0004] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0005] The present invention is an auxiliary method based on data mining for technology identification, comprising the following steps:

[0006] S1. Collect multiple patent document data and corresponding keyword data; collect keyword data and corresponding technology identification data at each stage of various types of technology industries;

[0007] S2. Use the collected keyword data and the corresponding technology industry type label data to build the final SVM classification model; use the collected patent document data and keyword data to build the final recurrent neural network model;

[0008] S3, collecting patent document data to be analyzed; using the final recurrent neural network model to extract keywords from the patent document data to be analyzed to obtain a keyword set to be analyzed; using the final SVM classification model to classify the encoded keyword set to be analyzed to obtain a classification result; obtaining corresponding technical identification data according to the classification result to obtain a matched technical identification data matrix; constructing a calling interface of a database storing the matched technical identification data matrix to obtain a second database interface set;

[0009] S4. Count the number of calls of each database interface in the second database interface set;

[0010] S5. Allocate calling rights to each authentication data user according to the number of calls of each database interface;

[0011] By collecting multiple patent document data and corresponding keyword data, training data and test data are provided for the subsequent construction of a model for extracting keywords from patent document data, and training data and test data are provided for the subsequent construction of a model for determining the type of technology industry to which the patent document data belongs based on keywords; by collecting keyword data and corresponding technology identification data at each stage in various types of technology industries, a database of technology identification data corresponding to each type of patent document is constructed, which facilitates the subsequent query of corresponding technology identification data based on the type of technology industry to which the patent document data belongs; by training and testing the recurrent neural network model to construct the final recurrent neural network model, the ability of the final recurrent neural network model to extract keywords from patent document data is greatly enhanced, thereby facilitating the subsequent analysis of keywords in the patent document data. An extraction model is provided for extraction; then the SVM classification model is trained and tested to construct a final SVM classification model, so that the final SVM classification model can classify the technology industry to which the patent document data belongs according to multiple keywords of the patent document data, thereby providing a classification model for subsequent classification of keywords in the patent document data to be analyzed extracted by the final recurrent neural network model, and obtaining the technology industry type corresponding to the patent document data to be analyzed; because the needs of different users are different, the technical identification data queried according to the technology industry type corresponding to the patent document data is stored in separate databases in this solution, and different permissions are allocated to different users according to the number of times different users call these different types of technology identification data, thereby reducing invalid calls by users, avoiding resource occupation, and improving the efficiency of calls.

[0012] Preferably, the S1 comprises the following steps:

[0013] S11, collect multiple patent document data and corresponding keyword data to obtain the first patent document data set a={a1, a2, ..., a i ,...,a a′}、First keyword data matrix a i represents the i-th patent document data collected, and a′ represents the total number of patent document data collected; as follows,

[0014]

[0015] in, represents the jth keyword data in the i-th patent document data, Represents the total number of keyword data in the i-th patent document data;

[0016] S12, collecting keyword data and corresponding technology identification data of each stage in various types of technology industries, and obtaining a second keyword data matrix c and a first technology identification data matrix set d′={d1′,d2′,...,d k ′,...,d h ″},d k ′ represents the kth type of technology identification data matrix, h′ represents the total number of technology identification data types collected; c and d k ' are as follows,

[0017]

[0018] Among them, c ij The keyword dataset representing the jth stage of the i-th technology industry, d k ' ij represents the technology appraisal data set of the jth stage of the i-th technology industry in the k-th type of technology appraisal data, represents the total number of stages of the i-th technology industry collected; c ijk , Respectively represent c ij The kth keyword data, d k ' ij The k1th technical identification data in d 1ij d 2kij Respectively represent c ij The total number of keyword data in d k ' ij The total number of technical identification data in the

[0019] S13, storing the second keyword data matrix c and the first technical document data matrix d' in multiple databases according to the type of technical industry, to obtain a technical industry database set; constructing a call interface corresponding to each technical industry database in the technical industry database set, to obtain a first database interface set Represents the calling interface of the database corresponding to the i-th technology industry;

[0020] Various types of technology industries include information technology, biotechnology, energy technology, advanced manufacturing, environmental protection technology, aerospace technology, materials science and technology, artificial intelligence and machine learning, quantum computing, new energy vehicles, medical technology, agricultural technology, etc.; the stages of technology industries include research and development, prototype development, testing and verification, industrialization, growth and maturity, saturation and decline, emerging technology, etc.; by collecting keyword data and corresponding technology identification data for each stage in various types of technology industries, the richness of identification data is improved, which facilitates the subsequent query of corresponding technology identification data according to the type of technology industry in patent document data; by setting corresponding calling interfaces for the database of each type of technology identification data, the calling process is simplified, each interface corresponds to a type of technology industry, and the accuracy of the call is also improved.

[0021] Preferably, S2 comprises the following steps:

[0022] S21, setting the first keyword data matrix according to the second keyword data matrix c and the first technical document data matrix d′ The technology industry type label corresponding to each row of keyword data in the first technology industry type label dataset is obtained Represents the first keyword data matrix The technology industry type label data corresponding to the keyword data in the i-th row;

[0023] S22, constructing an initial SVM classification model; the kernel function used in the initial SVM classification model is a radial basis kernel function; using the first technology industry type label data set And the first keyword data matrix The initial SVM classification model is trained and tested; after the training and testing are completed, the final SVM classification model is obtained;

[0024] S23, constructing an initial recurrent neural network model; using the first patent document dataset a={a1, a2, ..., a i ,...,a a′} and the first keyword data matrix Train and test the initial recurrent neural network model; after the training and testing are completed, obtain the final recurrent neural network model;

[0025] By setting corresponding technology industry type labels for each row of keyword data in the first keyword data matrix, label data is provided for subsequent training and testing of the SVM classification model, making the final SVM classification model constructed more accurate.

[0026] Preferably, the S22 comprises the following steps:

[0027] S221, setting a first training data ratio; and training a first technology industry type label data set according to the first training data ratio. And the first keyword data matrix Perform data division to obtain a first technology industry type label training data set, a first technology industry type label test data set, a first keyword training data matrix, and a first keyword test data matrix;

[0028] S222, binary-encoding each keyword data in the first keyword training data matrix and the first keyword test data matrix to obtain a first keyword training encoding data matrix and a first keyword test encoding data matrix;

[0029] S223, setting a first training error threshold; inputting the first keyword training coding data matrix and the first technology industry type label training data set into the initial SVM classification model for training operation; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained SVM classification model;

[0030] The soft interval parameter of the trained SVM classification model is set to e1, recorded as the initial soft interval parameter, and the data mapping mode control parameter is set to e2, recorded as the initial data mapping mode control parameter;

[0031] S224, setting a first accuracy threshold e1′; inputting the first technology industry type label test data set and the first keyword test data matrix into the trained SVM classification model for testing; after the test is completed, obtaining a first accuracy e′2;

[0032] When e′2≥e1′, the trained SVM classification model is used as the final SVM classification model; otherwise, the initial soft interval parameter e1 and the initial data mapping mode control parameter e2 of the trained SVM classification model are adjusted and optimized to obtain the final SVM classification model;

[0033] When the test results of the trained SVM classification model do not meet the requirements, the soft interval parameters and data mapping control parameters of the trained SVM classification model are optimized so that the trained SVM classification model meets the test requirements and the classification ability of the final SVM classification model is improved.

[0034] Preferably, in S224, adjusting and optimizing the initial soft interval parameter e1 and the initial data mapping mode control parameter e2 of the trained SVM classification model to obtain the final SVM classification model includes the following steps:

[0035] S2241, build pigeon population represents the i-th pigeon in the pigeon population, Indicates the size of the pigeon population; sets the pigeon population The maximum number of iterations is f1, the current number of iterations is f2, and the search space dimension is two-dimensional;

[0036] S2242, randomly generate the initial position of each pigeon in the pigeon population according to the initial soft interval parameter e1 and the initial data mapping mode control parameter e2, and obtain the initial position set f i ′1, f i ′2 represents the one-dimensional component and the two-dimensional component of the initial position of the i-th pigeon in the pigeon population; f i ′1, f i The formula for generating ′2 is as follows:

[0037]

[0038] In the formula, rand i1 、rand i2 Respectively for f i ′1, f i ′2 is the random number generated, They represent the lower limit and upper limit of the initial soft interval parameter e1 respectively; They respectively represent the lower limit and upper limit of the value of the initial data mapping mode control parameter e2;

[0039] S2243: construct a pigeon population according to the first accuracy threshold e1′ and the first accuracy e′2 The fitness function as follows,

[0040]

[0041] In the formula, α is a positive number, indicating the correction parameter;

[0042] S2244, start iterative operation; use the fitness function in each round of iteration Calculate the fitness of each pigeon in the pigeon population to obtain a fitness value set; select the individual position corresponding to the maximum fitness in the fitness value set as the global optimal position Each round of iteration process is divided into a teacher phase and a learning phase. In the teacher phase and the learning phase, the global optimal position is The formulas for updating the position of each pigeon are as follows:

[0043]

[0044] In the formula, They respectively represent the position of the i-th pigeon in the current iteration, the position of the i-th pigeon in the next iteration, and the position of the j-th pigeon in the next iteration; g1 and g2 represent random numbers generated in the teacher stage and the learning stage respectively; They respectively represent the fitness values ​​of the positions of the i-th pigeon and the j-th pigeon in the pigeon population in the next round of iteration; β represents the learning factor; represents the average value of the positions of all pigeons in the pigeon population;

[0045] S2245. When f2≥f1, stop the iteration operation and obtain the final global optimal position g′=(g1′,g′2); g1′ and g′2 are the position components representing the soft interval parameter and the data mapping mode control parameter in the final global optimal position respectively; substitute the final global optimal position g′=(g1′,g′2) into the trained SVM classification model to obtain the optimized SVM classification model;

[0046] Input the first technology industry type label test data set and the first keyword test data matrix in S224 into the optimized SVM classification model for testing; after the test is completed, obtain the third accuracy e3′; when e3′≥e1′, use the optimized SVM classification model as the final SVM classification model; otherwise, return to S2244 and continue to iterate until e3′≥e1′, and obtain the final SVM classification model;

[0047] The principle of the pigeon flock optimization algorithm is relatively simple, and fewer parameters need to be set. Therefore, by using the pigeon optimization algorithm to perform multiple iterative optimizations on the soft interval parameters and data mapping control parameters of the trained SVM classification model, the optimization efficiency is high and the optimization process is simple; and the idea of ​​the teaching and learning algorithm is integrated into the pigeon flock optimization algorithm to improve the pigeon flock optimization algorithm, further improving the convergence speed of the pigeon flock optimization algorithm, that is, improving the search speed of the optimal soft interval parameters and data mapping control parameters.

[0048] Preferably, the S23 comprises the following steps:

[0049] S231, setting a second training data ratio; and performing a first patent document data set a={a1, a2, ..., a i ,...,a a′} and the first keyword data matrix Perform data division to obtain a first patent document training data set, a first patent document test data set, a second keyword training data matrix, and a second keyword test data matrix;

[0050] S232, setting a second training error threshold; inputting the first patent document training data set and the second keyword training data matrix into the initial recurrent neural network model for training operation; during the training process, when the training error is less than the second training error threshold, stopping the training to obtain a trained recurrent neural network model;

[0051] S233, setting a second accuracy threshold; inputting the first patent document test data set and the second keyword test data matrix into the trained recurrent neural network model for testing; after the test is completed, obtaining the second accuracy data set represents the accuracy of keyword extraction of the i-th patent document test data in the first patent document test data set by the trained recurrent neural network model, represents the total number of patent document test data in the first patent document test data set;

[0052] Calculate the second accuracy data set The average value of each accuracy data in is obtained to obtain a second accuracy mean; when the second accuracy mean is greater than or equal to the second accuracy threshold, the trained recurrent neural network model is used as the final recurrent neural network model; otherwise, return to S13 to continue training the trained recurrent neural network model until the second accuracy mean is greater than or equal to the second accuracy threshold, and the final recurrent neural network model is obtained;

[0053] After the trained recurrent neural network model is tested, the keyword data extracted by the trained recurrent neural network model is compared with the actual keyword data to obtain the accuracy of keyword extraction of the trained recurrent neural network model for the patent document data, thereby improving the accuracy of keyword extraction of the subsequent patent document data to be analyzed.

[0054] Preferably, S3 comprises the following steps:

[0055] S31, collecting patent document data to be analyzed; using the final recurrent neural network model to extract keywords from the patent document data to be analyzed to obtain a keyword set to be analyzed represents the i-th keyword extracted from the patent document data to be analyzed; represents the total number of keywords extracted from the patent document data to be analyzed;

[0056] S32: the keyword set to be analyzed Binary encode each keyword in to obtain the set of encoded keywords to be analyzed represents a coded keyword obtained by encoding the i-th keyword extracted from the patent document data to be analyzed;

[0057] S33, using the final SVM classification model to analyze the coding keyword set Perform classification operations to obtain label data of the technology industry type to be analyzed;

[0058] According to the to-be-analyzed technical industry type label data, the first database interface set The calling interface of the corresponding technology industry database is selected to call and obtain the keyword data matrix c′ to be matched; as follows,

[0059]

[0060] Among them, c i ' j represents the jth keyword data of the i-th stage in the technology industry corresponding to the technology industry type label data to be analyzed; h i represents the total number of keyword data in the i-th stage of the technology industry corresponding to the technology industry type label data to be analyzed; b represents the total number of stages in the technology industry corresponding to the technology industry type label data to be analyzed;

[0061] S34, the keyword set to be analyzed Calculate the similarity with the keyword data of each row in the keyword data matrix c′ to be matched, and obtain the matching keyword similarity data set Represents the keyword set to be analyzed The similarity with the keyword data in the i-th row in the keyword data matrix c′ to be matched;

[0062] S35, according to the matching keyword similarity data set The stage of the technology industry type corresponding to the maximum similarity value in the first technology identification data matrix set d′={d1′,d2′,...,d k ′,...,d h ″} Select the corresponding technical identification data to obtain a matching technical identification data matrix as follows,

[0063]

[0064] in, Indicates that the jth technology appraisal data in the i-th type of technology appraisal data set of the technology industry type stage is matched; l i Indicates the total number of technology appraisal data in the i-th type of technology appraisal data set that is matched to the technology industry type stage;

[0065] The matched technical identification data matrix The data are stored in multiple databases according to their types, and corresponding calling interfaces are constructed for the multiple databases storing the matched technical identification data, so as to obtain the second database interface set l′={l1′,l2′,...,l i ′,...,l h ″},l i ' represents the calling interface of the database storing the i-th type of technical appraisal data;

[0066] By adopting the final recurrent neural network model to extract the keywords in the patent document data to be analyzed, and then adopting the final SVM classification model to classify the keywords of the patent document data to be analyzed, the corresponding technology industry type label data is obtained, thereby providing a query basis for the subsequent query of the technology identification data; thereby reducing the number of comparisons and improving the data matching efficiency; after the corresponding technology identification data is queried, the queried technology identification data is classified and stored, thereby providing a technical basis for different types of users to grant different calling permissions to these different types of technology identification data.

[0067] Preferably, S4 comprises the following steps:

[0068] S41. Set user type set and the database role type set and They represent the set i-th user type and i-th database role type respectively;

[0069] Indicates the total number of set user types and database role types;

[0070] Set the user type set respectively The identification data user corresponding to each user type in the identification data user set is obtained. Indicates the identification data user corresponding to the set i-th user type;

[0071] S42, setting a statistical period; pre-assigning the database role type set Each database role type has access to all data and sets the database role type to Each database role type in the authentication data user set Each identification data user in the use;

[0072] In the second database interface set l′={l1′,l2′,...,l i ′,...,l h ″}, add a call count counter code block to each database interface; when the usage time reaches the statistical period, count the number of times each authentication data user calls the second database interface set l′={l1′,l2′,...,l i ′,...,l h The number of times each database interface is called in ″}, and the call number data matrix m is obtained; as follows,

[0073]

[0074] Among them, m ij Indicates that the i-th authentication data user calls the second database interface set l′={l1′,l2′,...,l i ′,...,l h The number of times the jth database interface is used in ″};

[0075] The number of times users call technical identification data can directly reflect the degree of demand for different types of technical identification data by these users; therefore, by collecting the number of times users call technical identification data, a data basis is provided for the subsequent allocation of user calling rights based on these call times.

[0076] Preferably, S5 comprises the following steps:

[0077] S51, set the second database interface set l'={l1',l2',...,l i ′,...,l h ″}, and obtain the call count threshold set m′={m1′,m′2,...,m i ′,...,m′ h′},m i ' represents the second database interface set l'={l1',l2',...,l i ′,...,l h The call count threshold of the i-th database interface in ″};

[0078] S52, compare each row of data in the call count data matrix m with the call count threshold set m′={m1′,m′2,...,m i ′,...,m′ h′} for comparison; when the number of calls is greater than or equal to the corresponding call number threshold, the calling permission of the corresponding database interface is granted to the database role type used by the corresponding authentication data user;

[0079] Different calling permissions are assigned to each type of user according to the number of times the user calls different types of technical identification data, making the permission assignment process more accurate and scientific.

[0080] An auxiliary system based on data mining for technology appraisal, comprising a first patent data collection module, a technology industry data collection module, an SVM classification model construction module, a recurrent neural network model construction module, a second patent data collection module, a keyword extraction module, a keyword classification module, a technology appraisal data matching module, a second call interface construction module, a call count statistics module and a permission allocation module;

[0081] The first patent data collection module is used to collect multiple patent document data and corresponding keyword data;

[0082] The technology industry data collection module is used to collect keyword data and corresponding technology identification data at each stage of various types of technology industries;

[0083] The SVM classification model building module is used to construct a final SVM classification model using the collected keyword data and the corresponding technology industry type label data;

[0084] The recurrent neural network model construction module is used to construct a final recurrent neural network model using the collected patent document data and keyword data;

[0085] The second patent data collection module is used to collect patent document data to be analyzed;

[0086] The keyword extraction module is used to extract keywords from the patent document data to be analyzed using the final recurrent neural network model to obtain a keyword set to be analyzed;

[0087] The keyword classification module is used to classify the encoded keyword set to be analyzed using the final SVM classification model to obtain a classification result;

[0088] The technology identification data matching module is used to obtain corresponding technology identification data according to the classification results to obtain a matched technology identification data matrix;

[0089] The second calling interface construction module is used to construct a calling interface of a database storing matched technical identification data matrix to obtain a second database interface set;

[0090] The call count counting module is used to count the call counts of each database interface in the second database interface set;

[0091] The authority allocation module is used to allocate calling authority to each authentication data user according to the calling times of each database interface.

[0092] The present invention has the following beneficial effects:

[0093] 1. The present invention collects keyword data and corresponding technology identification data at each stage in various types of technology industries, so as to facilitate the subsequent query of corresponding technology identification data according to the technology industry type in which the patent document data is located; by constructing the final recurrent neural network model, a model is provided for the subsequent extraction of keywords in the patent document data to be analyzed; by constructing the final SVM classification model, a classification model is provided for the subsequent classification of keywords in the patent document data to be analyzed extracted by the final recurrent neural network model; at the same time, the queried technology identification data is stored in separate libraries, and different permissions are allocated to different users according to the number of calls made by different users to these different types of technology identification data, thereby avoiding resource occupation and improving the efficiency of the call; by performing data matching, technology identification personnel can easily and accurately obtain the corresponding technology identification data, thereby improving the identification efficiency.

[0094] 2. In the present invention, the soft interval parameters and data mapping control parameters of the trained SVM classification model are iteratively optimized multiple times by using the pigeon optimization algorithm, and the optimization efficiency is high and the optimization process is simple; and the idea of ​​the teaching and learning algorithm is integrated into the pigeon flock optimization algorithm to improve the pigeon flock optimization algorithm, thereby further improving the convergence speed of the pigeon flock optimization algorithm.

[0095] 3. In the present invention, different calling rights are allocated to each type of user according to the number of times the user calls different types of technical identification data, making the process of allocating rights more accurate and scientific.

[0096] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0098] Figure 1The present invention is a flowchart of a data mining-based auxiliary system for technology identification for matching technology industry data with patent documents. DETAILED DESCRIPTION

[0099] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0100] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0101] Embodiment 1

[0102] This embodiment is an auxiliary method based on data mining applied to technology identification, comprising the following steps:

[0103] S1. Collect multiple patent document data and corresponding keyword data; collect keyword data and corresponding technology identification data at each stage of various types of technology industries;

[0104] The S1 comprises the following steps:

[0105] S11, collect multiple patent document data and corresponding keyword data to obtain the first patent document data set a={a1, a2, ..., a i ,...,a a′}、First keyword data matrix a i represents the i-th patent document data collected, and a′ represents the total number of patent document data collected; as follows,

[0106]

[0107] in, represents the jth keyword data in the i-th patent document data, Represents the total number of keyword data in the i-th patent document data;

[0108] S12, collecting keyword data and corresponding technology identification data of each stage in various types of technology industries, and obtaining a second keyword data matrix c and a first technology identification data matrix set d′={d1′,d2′,...,d k ′,...,d h ″},d k ′ represents the kth type of technology identification data matrix, h′ represents the total number of technology identification data types collected; c and d k ' are as follows,

[0109]

[0110] Among them, c ij The keyword dataset representing the jth stage of the i-th technology industry, d k ' ij represents the technology appraisal data set of the jth stage of the i-th technology industry in the k-th type of technology appraisal data, represents the total number of stages of the i-th technology industry collected; c ijk , Respectively represent c ij The kth keyword data, d k ' ij The k1th technical identification data in d 1ij d 2kij Respectively represent c ij The total number of keyword data in d k ' ij The total number of technical identification data in the

[0111] S13, storing the second keyword data matrix c and the first technical document data matrix d' in multiple databases according to the type of technical industry, to obtain a technical industry database set; constructing a call interface corresponding to each technical industry database in the technical industry database set, to obtain a first database interface set Represents the calling interface of the database corresponding to the i-th technology industry;

[0112] S2. Use the collected keyword data and the corresponding technology industry type label data to build the final SVM classification model; use the collected patent document data and keyword data to build the final recurrent neural network model;

[0113] The S2 comprises the following steps:

[0114] S21, setting the first keyword data matrix according to the second keyword data matrix c and the first technical document data matrix d′ The technology industry type label corresponding to each row of keyword data in the first technology industry type label dataset is obtained Represents the first keyword data matrix The technology industry type label data corresponding to the keyword data in the i-th row;

[0115] S22, constructing an initial SVM classification model; the kernel function used in the initial SVM classification model is a radial basis kernel function; using the first technology industry type label data set And the first keyword data matrix The initial SVM classification model is trained and tested; after the training and testing are completed, the final SVM classification model is obtained;

[0116] The S22 comprises the following steps:

[0117] S221, setting a first training data ratio; and training a first technology industry type label data set according to the first training data ratio. And the first keyword data matrix Perform data division to obtain a first technology industry type label training data set, a first technology industry type label test data set, a first keyword training data matrix, and a first keyword test data matrix;

[0118] S222, binary-encode each keyword data in the first keyword training data matrix and the first keyword test data matrix to obtain a first keyword training encoding data matrix and a first keyword test encoding data matrix;

[0119] S223, setting a first training error threshold; inputting the first keyword training coding data matrix and the first technology industry type label training data set into the initial SVM classification model for training operation; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained SVM classification model;

[0120] The soft interval parameter of the trained SVM classification model is set to e1, recorded as the initial soft interval parameter, and the data mapping mode control parameter is set to e2, recorded as the initial data mapping mode control parameter;

[0121] S224, setting a first accuracy threshold e1′; inputting the first technology industry type label test data set and the first keyword test data matrix into the trained SVM classification model for testing; after the test is completed, obtaining a first accuracy e′2;

[0122] When e′2≥e1′, the trained SVM classification model is used as the final SVM classification model; otherwise, the initial soft interval parameter e1 and the initial data mapping mode control parameter e2 of the trained SVM classification model are adjusted and optimized to obtain the final SVM classification model;

[0123] In S224, the initial soft interval parameter e1 and the initial data mapping mode control parameter e2 of the trained SVM classification model are adjusted and optimized to obtain the final SVM classification model, which includes the following steps:

[0124] S2241, build pigeon population represents the i-th pigeon in the pigeon population, e~ represents the size of the pigeon population; set the pigeon population The maximum number of iterations is f1, the current number of iterations is f2, and the search space dimension is two-dimensional;

[0125] S2242, randomly generate the initial position of each pigeon in the pigeon population according to the initial soft interval parameter e1 and the initial data mapping mode control parameter e2, and obtain the initial position set f i ′1, f i ′2 represents the one-dimensional component and the two-dimensional component of the initial position of the i-th pigeon in the pigeon population; f i ′1, f i The formula for generating ′2 is as follows:

[0126]

[0127] In the formula, rand i1 、rand i2 Respectively for f i ′1, f i ′2 is the random number generated, They represent the lower limit and upper limit of the initial soft interval parameter e1 respectively; They respectively represent the lower limit and upper limit of the value of the initial data mapping mode control parameter e2;

[0128] S2243: construct a pigeon population according to the first accuracy threshold e1′ and the first accuracy e′2 The fitness function as follows,

[0129]

[0130] In the formula, α is a positive number, indicating the correction parameter;

[0131] S2244, start iterative operation; use the fitness function in each round of iteration Calculate the fitness of each pigeon in the pigeon population to obtain a fitness value set; select the individual position corresponding to the maximum fitness in the fitness value set as the global optimal position Each round of iteration process is divided into a teacher phase and a learning phase. In the teacher phase and the learning phase, the global optimal position is The formulas for updating the position of each pigeon are as follows:

[0132]

[0133] In the formula, They respectively represent the position of the i-th pigeon in the current iteration, the position of the i-th pigeon in the next iteration, and the position of the j-th pigeon in the next iteration; g1 and g2 represent random numbers generated in the teacher stage and the learning stage respectively; They respectively represent the fitness values ​​of the positions of the i-th pigeon and the j-th pigeon in the pigeon population in the next round of iteration; β represents the learning factor; represents the average value of the positions of all pigeons in the pigeon population;

[0134] S2245. When f2≥f1, stop the iteration operation and obtain the final global optimal position g′=(g1′,g′2); g1′ and g′2 are the position components representing the soft interval parameter and the data mapping mode control parameter in the final global optimal position respectively; substitute the final global optimal position g′=(g1′,g′2) into the trained SVM classification model to obtain the optimized SVM classification model;

[0135] Input the first technology industry type label test data set and the first keyword test data matrix in S224 into the optimized SVM classification model for testing; after the test is completed, obtain the third accuracy e3′; when e3′≥e1′, use the optimized SVM classification model as the final SVM classification model; otherwise, return to S2244 and continue to iterate until e3′≥e1′, and obtain the final SVM classification model;

[0136] S23, constructing an initial recurrent neural network model; using the first patent document dataset a={a1, a2, ..., a i ,...,a a′} and the first keyword data matrix Train and test the initial recurrent neural network model; after the training and testing are completed, obtain the final recurrent neural network model;

[0137] The S23 comprises the following steps:

[0138] S231, setting a second training data ratio; and performing a first patent document data set a={a1, a2, ..., a i ,...,a a′} and the first keyword data matrix Perform data division to obtain a first patent document training data set, a first patent document test data set, a second keyword training data matrix, and a second keyword test data matrix;

[0139] S232, setting a second training error threshold; inputting the first patent document training data set and the second keyword training data matrix into the initial recurrent neural network model for training operation; during the training process, when the training error is less than the second training error threshold, stopping the training to obtain a trained recurrent neural network model;

[0140] S233, setting a second accuracy threshold; inputting the first patent document test data set and the second keyword test data matrix into the trained recurrent neural network model for testing; after the test is completed, obtaining the second accuracy data set represents the accuracy of keyword extraction of the i-th patent document test data in the first patent document test data set by the trained recurrent neural network model, represents the total number of patent document test data in the first patent document test data set;

[0141] Calculate the second accuracy data set The average value of each accuracy data in is obtained to obtain a second accuracy mean; when the second accuracy mean is greater than or equal to the second accuracy threshold, the trained recurrent neural network model is used as the final recurrent neural network model; otherwise, return to S13 to continue training the trained recurrent neural network model until the second accuracy mean is greater than or equal to the second accuracy threshold, and the final recurrent neural network model is obtained;

[0142] S3, collecting patent document data to be analyzed; using the final recurrent neural network model to extract keywords from the patent document data to be analyzed to obtain a keyword set to be analyzed; using the final SVM classification model to classify the encoded keyword set to be analyzed to obtain a classification result; obtaining corresponding technical identification data according to the classification result to obtain a matched technical identification data matrix; constructing a calling interface of a database storing the matched technical identification data matrix to obtain a second database interface set;

[0143] The S3 comprises the following steps:

[0144] S31, collecting patent document data to be analyzed; using the final recurrent neural network model to extract keywords from the patent document data to be analyzed to obtain a keyword set to be analyzed represents the i-th keyword extracted from the patent document data to be analyzed; represents the total number of keywords extracted from the patent document data to be analyzed;

[0145] S32: the keyword set to be analyzed Binary encode each keyword in to obtain the set of encoded keywords to be analyzed represents a coded keyword obtained by encoding the i-th keyword extracted from the patent document data to be analyzed;

[0146] S33, using the final SVM classification model to analyze the coding keyword set Perform classification operations to obtain label data of the technology industry type to be analyzed;

[0147] According to the to-be-analyzed technical industry type label data, the first database interface set The calling interface of the corresponding technology industry database is selected to call and obtain the keyword data matrix c′ to be matched; as follows,

[0148]

[0149] Among them, c i ' j represents the jth keyword data of the i-th stage in the technology industry corresponding to the technology industry type label data to be analyzed; h i represents the total number of keyword data in the i-th stage of the technology industry corresponding to the technology industry type label data to be analyzed; b represents the total number of stages in the technology industry corresponding to the technology industry type label data to be analyzed;

[0150] S34, the keyword set to be analyzed Calculate the similarity with the keyword data of each row in the keyword data matrix c′ to be matched, and obtain the matching keyword similarity data set Represents the keyword set to be analyzed The similarity with the keyword data in the i-th row in the keyword data matrix c′ to be matched;

[0151] S35, according to the matching keyword similarity data set The stage of the technology industry type corresponding to the maximum similarity value in the first technology identification data matrix set d′={d1′,d2′,...,d k′,...,d h ″} Select the corresponding technical identification data to obtain a matching technical identification data matrix as follows,

[0152]

[0153] in, Indicates that the jth technology appraisal data in the i-th type of technology appraisal data set of the technology industry type stage is matched; l i Indicates the total number of technology appraisal data in the i-th type of technology appraisal data set that is matched to the technology industry type stage;

[0154] The matched technical identification data matrix The data are stored in multiple databases according to their types, and corresponding calling interfaces are constructed for the multiple databases storing the matched technical identification data, so as to obtain the second database interface set l′={l1′,l2′,...,l i ′,...,l h ″},l i ' represents the calling interface of the database storing the i-th type of technical appraisal data;

[0155] S4. Count the number of calls of each database interface in the second database interface set;

[0156] The S4 comprises the following steps:

[0157] S41. Set user type set and the database role type set and They represent the set i-th user type and i-th database role type respectively;

[0158] Indicates the total number of set user types and database role types;

[0159] Set the user type set respectively The identification data user corresponding to each user type in the identification data user set is obtained. Indicates the identification data user corresponding to the set i-th user type;

[0160] S42, setting a statistical period; pre-assigning the database role type set Each database role type has access to all data and sets the database role type to Each database role type in the authentication data user set Each identification data user in the use;

[0161] In the second database interface set l′={l1′,l2′,...,l i ′,...,l h ″}, add a call count counter code block to each database interface; when the usage time reaches the statistical period, count the number of times each authentication data user calls the second database interface set l′={l1′,l2′,...,l i ′,...,l h The number of times each database interface is called in ″}, and the call number data matrix m is obtained; as follows,

[0162]

[0163] Among them, m ij Indicates that the i-th authentication data user calls the second database interface set l′={l1′,l2′,...,l i ′,...,l h The number of times the jth database interface is used in ″};

[0164] S5. Allocate calling rights to each authentication data user according to the number of calls of each database interface;

[0165] The S5 comprises the following steps:

[0166] S51, set the second database interface set l'={l1',l2',...,l i ′,...,l h ″}, and obtain the call count threshold set m′={m1′,m′2,...,m i ′,...,m′ h′},m i ' represents the second database interface set l'={l1',l2',...,l i ′,...,l h The call count threshold of the i-th database interface in ″};

[0167] S52, compare each row of data in the call count data matrix m with the call count threshold set m′={m1′,m′2,...,m i ′,...,m′ h′} for comparison; when the number of calls is greater than or equal to the corresponding call number threshold, the calling permission of the corresponding database interface is granted to the database role type used by the corresponding authentication data user.

[0168] Embodiment 2

[0169] This embodiment discloses an auxiliary system based on data mining for technology appraisal, which can implement the method of the above embodiment, including a first patent data collection module, a technology industry data collection module, an SVM classification model construction module, a recurrent neural network model construction module, a second patent data collection module, a keyword extraction module, a keyword classification module, a technology appraisal data matching module, a second call interface construction module, a call count statistics module and a permission allocation module;

[0170] The first patent data collection module is used to collect multiple patent document data and corresponding keyword data;

[0171] The technology industry data collection module is used to collect keyword data and corresponding technology identification data at each stage of various types of technology industries;

[0172] The SVM classification model building module is used to construct a final SVM classification model using the collected keyword data and the corresponding technology industry type label data;

[0173] The recurrent neural network model construction module is used to construct a final recurrent neural network model using the collected patent document data and keyword data;

[0174] The second patent data collection module is used to collect patent document data to be analyzed;

[0175] The keyword extraction module is used to extract keywords from the patent document data to be analyzed using the final recurrent neural network model to obtain a keyword set to be analyzed;

[0176] The keyword classification module is used to classify the encoded keyword set to be analyzed using the final SVM classification model to obtain a classification result;

[0177] The technology identification data matching module is used to obtain corresponding technology identification data according to the classification results to obtain a matched technology identification data matrix;

[0178] The second calling interface construction module is used to construct a calling interface of a database storing matched technical identification data matrix to obtain a second database interface set;

[0179] The call count counting module is used to count the call counts of each database interface in the second database interface set;

[0180] The authority allocation module is used to allocate calling authority to each authentication data user according to the calling times of each database interface.

[0181] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0182] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. A data mining-based auxiliary method for technology identification, characterized in that: The following steps are involved: S1. Collect multiple patent document data and corresponding keyword data; Collect keyword data and corresponding technology identification data for each stage in various types of technology industries; S2. Use the collected keyword data and the corresponding technology industry type label data to build the final SVM classification model; use the collected patent document data and keyword data to build the final recurrent neural network model; S3, collecting patent document data to be analyzed; The final recurrent neural network model is used to extract keywords from the patent document data to be analyzed, and a keyword set to be analyzed is obtained; the final SVM classification model is used to classify the encoded keyword set to be analyzed, and a classification result is obtained; Obtain corresponding technical identification data according to the classification results, and obtain a matched technical identification data matrix; Constructing a calling interface of a database storing matched technical identification data matrices to obtain a second database interface set; S4. Count the number of calls of each database interface in the second database interface set; S5. Allocate calling rights to each authentication data user according to the number of calls to each database interface.

2. The data mining-based auxiliary method for technology identification according to claim 1, characterized in that: The S1 comprises the following steps: S11, collecting multiple patent document data and corresponding keyword data to obtain a first patent document data set and a first keyword data matrix; S12, collecting keyword data and corresponding technology identification data at each stage in various types of technology industries to obtain a second keyword data matrix and a first technology identification data matrix set; S13. Store the second keyword data matrix and the first technical document data matrix in multiple databases according to the type of technology industry to obtain a technology industry database set; construct a calling interface corresponding to each technology industry database in the technology industry database set to obtain a first database interface set.

3. The data mining-based auxiliary method for technology identification according to claim 2, characterized in that: The S2 comprises the following steps: S21, according to the second keyword data matrix and the first technical document data matrix, set the technology industry type label corresponding to each row of keyword data in the first keyword data matrix to obtain a first technology industry type label data set; S22, constructing an initial SVM classification model; the kernel function used in the initial SVM classification model is a radial basis kernel function; the initial SVM classification model is trained and tested using the first technology industry type label data set and the first keyword data matrix; after the training and testing are completed, the final SVM classification model is obtained; S23, constructing an initial recurrent neural network model; using the first patent document dataset and the first keyword data matrix The initial recurrent neural network model is trained and tested; after the training and testing are completed, the final recurrent neural network model is obtained.

4. The data mining-based auxiliary method for technology identification according to claim 3 is characterized in that: The S22 comprises the following steps: S221, setting a first training data ratio; dividing the first technology industry type label data set and the first keyword data matrix according to the first training data ratio to obtain a first technology industry type label training data set, a first technology industry type label test data set, a first keyword training data matrix, and a first keyword test data matrix; S222, binary-encoding each keyword data in the first keyword training data matrix and the first keyword test data matrix to obtain a first keyword training encoding data matrix and a first keyword test encoding data matrix; S223, setting a first training error threshold; inputting the first keyword training coding data matrix and the first technology industry type label training data set into the initial SVM classification model for training operation; during the training process, when the training error is less than the first training error threshold, stopping the training to obtain a trained SVM classification model; S224, set a first accuracy threshold e1′; input the first technology industry type label test data set and the first keyword test data matrix into the trained SVM classification model for testing; after the test is completed, obtain the first accuracy e′2; when e′2≥e1′, use the trained SVM classification model as the final SVM classification model; otherwise, adjust and optimize the initial soft interval parameters and initial data mapping method control parameters of the trained SVM classification model to obtain the final SVM classification model.

5. The data mining-based auxiliary method for technology identification according to claim 4, characterized in that: In S224, the initial soft interval parameters and the initial data mapping mode control parameters of the trained SVM classification model are adjusted and optimized to obtain the final SVM classification model, which includes the following steps: S2241, constructing a pigeon population; setting the maximum number of iterations of the pigeon population to f1 and the current number of iterations to f2; S2242, randomly generating an initial position of each pigeon in the pigeon population according to the initial soft interval parameter and the initial data mapping mode control parameter to obtain an initial position set; S2243, constructing a fitness function of the pigeon population according to the first accuracy threshold e1′ and the first accuracy e′2; S2244, start iterative operation; in each round of iteration, use the fitness function to calculate the fitness of each pigeon in the pigeon population to obtain a fitness value set; select the individual position corresponding to the maximum fitness in the fitness value set as the global optimal position; divide each round of iteration into a teacher stage and a learning stage; S2245. When f2≥f1, stop the iteration operation and obtain the final global optimal position; substitute the final global optimal position into the trained SVM classification model to obtain the optimized SVM classification model; input the first technology industry type label test data set and the first keyword test data matrix in S224 into the optimized SVM classification model for testing; after the test is completed, obtain the third accuracy e3′; when e3′≥e1′, use the optimized SVM classification model as the final SVM classification model; otherwise, return to S2244 and continue to iterate until e3′≥e1′ to obtain the final SVM classification model.

6. The data mining-based auxiliary method for technology identification according to claim 5, characterized in that: The S3 comprises the following steps: S31, collecting patent document data to be analyzed; using the final recurrent neural network model to extract keywords from the patent document data to be analyzed to obtain a keyword set to be analyzed; S32, binary-encoding each keyword in the keyword set to be analyzed to obtain the coded keyword set to be analyzed; S33, using the final SVM classification model to classify the coding keyword set to be analyzed, to obtain the technology industry type label data to be analyzed; selecting a call interface of the corresponding technology industry database from the first database interface set according to the technology industry type label data to be analyzed, to obtain a keyword data matrix to be matched; S34, calculating similarity between the keyword set to be analyzed and the keyword data of each row in the keyword data matrix to be matched, to obtain a matching keyword similarity data set; S35. Select corresponding technology appraisal data from the first technology appraisal data matrix set according to the stage of technology industry type corresponding to the maximum similarity value in the matching keyword similarity data set to obtain a matched technology appraisal data matrix; store the matched technology appraisal data matrix into multiple databases according to type, and construct corresponding calling interfaces for the multiple databases storing the matched technology appraisal data to obtain a second database interface set.

7. The data mining-based auxiliary method for technology identification according to claim 6, characterized in that: The S4 comprises the following steps: S41, setting a user type set and a database role type set; respectively setting the authentication data user corresponding to each user type in the user type set to obtain an authentication data user set; S42, setting a statistical period; pre-granting each database role type in the database role type set the authority to access all data and respectively giving each database role type in the database role type set to each identification data user in the identification data user set for use; adding a call count counter code block in each database interface in the second database interface set; when the usage time reaches the statistical period, counting the number of times each identification data user calls each database interface in the second database interface set according to the call count counter code block, and obtaining a call count data matrix.

8. The data mining-based auxiliary method for technology identification according to claim 7, characterized in that: The S5 comprises the following steps: S51, setting a call count threshold for each database interface in the second database interface set to obtain a call count threshold set; S52, comparing each row of data in the call count data matrix with the call count threshold set; when the call count is greater than or equal to the corresponding call count threshold, granting the call permission of the corresponding database interface to the database role type used by the corresponding authentication data user.

9. A system for implementing the auxiliary method based on data mining for technology identification as described in any one of claims 1 to 8.

Images