Intellectual property infringement detection algorithm based on deep learning
Through the deep learning-based intellectual property infringement detection algorithm, automatically extracting features and performing pattern recognition, solving the problem that traditional detection methods are time-consuming and labor-intensive and difficult to adapt to new forms of infringement, and achieving efficient and accurate automated detection.
Patent Information
- Application Number
- CN202411964437.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-29
AI Technical Summary
Traditional intellectual property infringement detection relies on a large number of manual comparisons and judgments, which is time-consuming and labor-intensive and susceptible to subjective factors, making it difficult to adapt to new forms of infringement.
The intellectual property infringement detection algorithm based on deep learning is adopted, including server, model training module, intellectual property infringement detection model, scope selection module and model evaluation module, and manual intervention is reduced through automatic extraction of features, pattern recognition and comparison analysis.
Automatic intellectual property infringement detection has been realized, the need for manual intervention has been reduced, the infringement method can be adapted to the update of infringement methods, and the detection efficiency and accuracy have been improved.
Smart Images
Figure CN119903490A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intellectual property protection, and more specifically, to an intellectual property infringement detection algorithm based on deep learning. Background Art
[0002] Intellectual property infringement refers to the unauthorized use, copying, dissemination, modification or sale of works, inventions, trademarks, etc. protected by intellectual property without the permission of the intellectual property owner, which infringes upon the legitimate rights and interests of the intellectual property owner.
[0003] Traditional intellectual property infringement detection often relies on a large amount of manual comparison and judgment, which is not only time-consuming and labor-intensive, but also easily affected by subjective factors. The means of intellectual property infringement are constantly being updated and changed, and traditional detection methods are often difficult to adapt to new forms of infringement. Deep learning algorithms can summarize general characteristics and laws by learning a large amount of sample data, and have strong generalization capabilities. Summary of the invention
[0004] In view of the problems existing in the prior art, the purpose of the present invention is to provide an intellectual property infringement detection algorithm based on deep learning.
[0005] In order to solve the above problems, the present invention adopts the following technical solutions:
[0006] Intellectual property infringement detection algorithm based on deep learning, including: server, model training module, intellectual property infringement detection model, scope selection module and model evaluation module;
[0007] The server is used to store existing intellectual property data and update the stored intellectual property data in real time;
[0008] The model training module is used to collect existing intellectual property data from the server for preprocessing, and use the processed intellectual property data as a training set to train the intellectual property infringement detection model;
[0009] The intellectual property infringement detection model is used to input the intellectual property data to be detected, and the model outputs the probability or judgment result of infringement, and determines whether there is infringement based on the set threshold;
[0010] The range selection module is used to select the range of existing intellectual property data in the server according to the time of the intellectual property data to be detected, and input the selected existing intellectual property data into the intellectual property infringement detection model;
[0011] The model evaluation module is used to analyze and evaluate the trained intellectual property infringement detection model, and the unqualified intellectual property infringement detection model continues to be trained through the model training module.
[0012] As a further description of the above technical solution:
[0013] The server includes a database and an update module. The database is used to store existing intellectual property data, and the update module is used to update the intellectual property data in the database.
[0014] As a further description of the above technical solution:
[0015] The training steps of the model training module are as follows:
[0016] Data preprocessing: preprocessing the collected existing intellectual property data, including data cleaning, normalization and feature annotation, and then using it as a training set for training the intellectual property infringement detection model;
[0017] Feature extraction: input the processed training set into the intellectual property infringement detection model to extract features from the data and learn the intrinsic features of the data. The extracted features include the novelty, creativity and practicality of the intellectual property;
[0018] Model training, the extracted features are input into the intellectual property infringement detection model for training, and the parameters of the model are adjusted to minimize the loss function.
[0019] As a further description of the above technical solution:
[0020] The intellectual property infringement detection model includes a recurrent neural network and a long short-term memory network.
[0021] As a further description of the above technical solution:
[0022] The time selection of the scope selection module is based on the application date, publication date and authorization date of the intellectual property to be detected, among which the novelty and creativity comparison selects the existing intellectual property before or after the application date for infringement detection.
[0023] As a further description of the above technical solution:
[0024] The evaluation method of the model evaluation module is to determine the threshold of the relevant similarity between the intellectual property data to be detected and the existing intellectual property data, wherein the relevant similarity includes the similarity of the beneficial effects of the intellectual property, the structural similarity and the content similarity.
[0025] As a further description of the above technical solution:
[0026] The long short-term memory network storage and update algorithm is as follows:
[0027] Forget gate, receives the current input x t and the hidden state h of the previous time step t-1, through a fully connected layer and a sigmoid activation function to calculate the output f of the forget gate t , the value range is between 0 and 1, f t The closer the value is to 1, the more information is retained; the closer it is to 0, the more information is forgotten:
[0028] f t =σ(W f ×[h t-1 ,x t ]+b f );
[0029] Where W f Forget gate weight, b f is the bias term, σ is the sigmoid activation function;
[0030] The input gate also receives the current input xt and the hidden state ht-1 of the previous time step. It calculates the output it of the input gate and the new candidate memory cell state C~t through two fully connected layers, a sigmoid activation function, and a tanh activation function. The calculation formula is as follows:
[0031] i t =σ(W i ×[h t-1 ,x t ]+b i );
[0032] C~t=tanh(W C ×[h t-1 ,x t ]+b C );
[0033] Where W i With W C are the weight matrices of the input gate and candidate memory cell states, b i With b C is the bias term;
[0034] According to the output of the forget gate and the input gate, update the state C of the memory unit t , the calculation formula is as follows:
[0035] C t =f t ⊙C t-1 +i t ⊙C~t
[0036] The output gate, which receives the current input x t and the hidden state h of the previous time step t-1 , through a fully connected layer and a sigmoid activation function to calculate the output o of the output gatet , then o t And the memory cell state C processed by the tanh activation function t Multiply them together to get the hidden state h of the current time step t The calculation formula is as follows:
[0037] o t =σ(W o ×[h t-1 ,x t ]+b o );
[0038] h t =o t ⊙tanh(C t );
[0039] Where W o is the weight matrix of the output gate, b o is the bias term.
[0040] Compared with the prior art, the advantages of the present invention are:
[0041] This solution, based on deep learning algorithms, can automatically extract features, perform pattern recognition and comparative analysis, greatly reducing the need for manual intervention. There is no need for manual review one by one, and it can detect infringements by updating further learning based on the means of infringement. The intellectual property data is divided before detection, which can reduce the computing pressure of the detection model and improve the detection efficiency and accuracy of the detection model. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a schematic diagram of the present invention;
[0043] Figure 2 It is a schematic diagram of the principle of the server of the present invention;
[0044] Figure 3 It is a schematic diagram of the principle of the intellectual property infringement detection model of the present invention;
[0045] Figure 4 Schematic diagram of the training process of the model training module of the present invention.
[0046] Description of the numbers in the figure:
[0047] 1. Server; 11. Database; 12. Update module; 2. Model training module; 3. Intellectual property infringement detection model; 31. Recurrent neural network; 32. Long short-term memory network; 4. Range selection module; 5. Model evaluation module. DETAILED DESCRIPTION
[0048] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention;
[0049] See also Figure 1-4 , the present invention provides the following embodiments:
[0050] The deep learning-based intellectual property infringement detection algorithm includes: a server 1, a model training module 2, an intellectual property infringement detection model 3, a range selection module 4 and a model evaluation module 5; the server 1 is used to store existing intellectual property data and update the stored intellectual property data in real time; the model training module 2 is used to collect the existing intellectual property data of the server 1 for preprocessing, and use the processed intellectual property data as a training set to train the intellectual property infringement detection model 3; the intellectual property infringement detection model 3 is used to input the required intellectual property data to be detected, and the model outputs the probability or judgment result of infringement, and determines whether there is infringement according to the set threshold; the range selection module 4 is used to select the range of the existing intellectual property data in the server 1 according to the time of the intellectual property data to be detected, and input the selected existing intellectual property data into the intellectual property infringement detection model 3; the model evaluation module 5 is used to analyze and evaluate the trained intellectual property infringement detection model 3, and the unqualified intellectual property infringement detection model 3 continues to be trained through the model training module 2.
[0051] Based on the existing intellectual property data stored in the server 1, the infringing samples and non-infringing samples in the existing intellectual property data are used as training sets. After the training set data is processed by the model training module 2, it is input into the intellectual property infringement detection model 3 for detection and analysis. According to the detection and analysis results, the parameters of the intellectual property infringement detection model 3 are continuously adjusted to reduce the loss function. The model evaluation module 5 is used to evaluate whether the trained intellectual property infringement detection model 3 conforms to the target model, and the infringement detection analysis value of the intellectual property is better than or equal to the threshold value, so as to meet the accuracy requirements of the intellectual property infringement detection analysis. Before using the intellectual property infringement detection model 3 to detect the intellectual property data to be detected, the application date of the existing intellectual property data in the server 1 is first distinguished by the range selection module 4 according to the application date of the intellectual property data to be detected. According to the infringement and infringement relationship of the intellectual property data to be detected, the corresponding existing intellectual property data is selected and input into the intellectual property infringement detection model 3 as a comparison party, which greatly reduces the detection workload of the intellectual property infringement detection model 3, improves the detection efficiency of the intellectual property infringement detection model 3, and reduces the probability of false detection, thereby improving the accuracy of intellectual property infringement detection.
[0052] The server 1 includes a database 11 and an update module 12 . The database 11 is used to store existing intellectual property data, and the update module 12 is used to update the intellectual property data in the database 11 .
[0053] The existing intellectual property data is stored in the database 11, which facilitates the retrieval of the training intellectual property infringement detection model 3, and the new intellectual property data is updated by connecting to the Internet through the update module 12, so as to ensure the accuracy of intellectual property infringement detection.
[0054] The training steps of the model training module 2 are as follows: data preprocessing, which includes data cleaning, normalization and feature labeling, and then uses it as a training set for training the intellectual property infringement detection model 3; feature extraction, which includes inputting the processed training set into the intellectual property infringement detection model 3 to extract features from the data and learn the intrinsic features of the data. The extracted features include the novelty, creativity and practicality of the intellectual property; model training, which includes inputting the extracted features into the intellectual property infringement detection model 3 for training and adjusting the parameters of the model to minimize the loss function.
[0055] Based on the infringing samples and non-infringing data in the existing intellectual property data as learning samples, the feature extraction method and direction of the intellectual property infringement detection model 3 are trained, so that the intellectual property infringement detection model 3 can be deeply learned, so as to gradually optimize the detection accuracy and efficiency of the intellectual property infringement detection model 3.
[0056] The intellectual property infringement detection model 3 includes a recurrent neural network 31 and a long short-term memory network 32, wherein the storage and update algorithm of the long short-term memory network 32 is as follows:
[0057] Forget gate, receives the current input x t and the hidden state h of the previous time step t-1 , through a fully connected layer and a sigmoid activation function to calculate the output f of the forget gate t , the value range is between 0 and 1, f t The closer the value is to 1, the more information is retained; the closer it is to 0, the more information is forgotten:
[0058] f t =σ(W f ×[h t-1 ,x t ]+b f );
[0059] Where W f Forget gate weight, b f is the bias term, σ is the sigmoid activation function;
[0060] The input gate also receives the current input xt and the hidden state ht-1 of the previous time step. It calculates the output it of the input gate and the new candidate memory cell state C~t through two fully connected layers, a sigmoid activation function, and a tanh activation function. The calculation formula is as follows:
[0061] i t =σ(W i ×[h t-1 ,x t ]+b i );
[0062] C~t=tanh(W C ×[h t-1 ,x t ]+b C );
[0063] Where W i With W C are the weight matrices of the input gate and candidate memory cell states, b i With b C is the bias term;
[0064] According to the output of the forget gate and the input gate, update the state C of the memory unit t , the calculation formula is as follows:
[0065] C t =f t ⊙C t-1 +i t ⊙C~t
[0066] The output gate, which receives the current input x t and the hidden state h of the previous time step t-1 , through a fully connected layer and a sigmoid activation function to calculate the output o of the output gate t , then o t And the memory cell state C processed by the tanh activation function t Multiply them together to get the hidden state h of the current time step t The calculation formula is as follows:
[0067] o t =σ(W o ×[h t-1 ,x t ]+b o );
[0068] h t =o t ⊙tanh(C t );
[0069] Where Wo is the weight matrix of the output gate, b o is the bias term.
[0070] The long short-term memory network 32 assists the recurrent neural network 31 to avoid the problem of gradient vanishing or gradient exploding faced by the recurrent neural network 31 during the comparison and detection process of a large amount of existing intellectual property data with the data to be detected, thereby improving the logic of the recurrent neural network 31 in the face of long-term learning and detection processes.
[0071] The time selection of the scope selection module 4 is based on the application date, publication date and authorization date of the intellectual property to be detected, among which the novelty and creativity comparison selects the existing intellectual property before or after the application date for infringement detection.
[0072] Through the range selection module 4, the existing intellectual property data can be directly selected according to the detection direction before detection, thereby avoiding the input of all existing intellectual property data into the intellectual property infringement detection model 3, reducing the detection burden of the intellectual property infringement detection model 3, and improving the detection efficiency and accuracy of the intellectual property infringement detection model 3.
[0073] The evaluation method of the model evaluation module 5 is to determine the threshold of the relevant similarity between the intellectual property data to be detected and the existing intellectual property data, wherein the relevant similarity includes the similarity of the beneficial effects of the intellectual property, the structural similarity and the content similarity.
[0074] The detection threshold of the intellectual property infringement detection model 3 is set according to the detection experience and requirements, and the model evaluation module 5 determines whether the deep learning of the intellectual property infringement detection model 3 is qualified according to the threshold.
[0075] The above is only a preferred specific implementation of the present invention; however, the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and its improved conception within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. The deep learning-based intellectual property infringement detection algorithm is characterized by: include: Server (1), model training module (2), intellectual property infringement detection model (3), scope selection module (4) and model evaluation module (5); The server (1) is used to store existing intellectual property data and update the stored intellectual property data in real time; The model training module (2) is used to collect existing intellectual property data of the server (1) for preprocessing, and use the processed intellectual property data as a training set to train the intellectual property infringement detection model (3); The intellectual property infringement detection model (3) is used to input the intellectual property data to be detected, and the model outputs the probability of infringement or the judgment result, and determines whether there is infringement according to the set threshold; The range selection module (4) is used to select the range of existing intellectual property data in the server (1) according to the time of the intellectual property data to be detected, and input the selected existing intellectual property data into the intellectual property infringement detection model (3); The model evaluation module (5) is used to analyze and evaluate the trained intellectual property infringement detection model (3), and the unqualified intellectual property infringement detection model (3) continues to be trained through the model training module (2).
2. The deep learning-based intellectual property infringement detection algorithm according to claim 1 is characterized in that: The server (1) comprises a database (11) and an update module (12); the database (11) is used to store existing intellectual property data; and the update module (12) is used to update the intellectual property data in the database (11).
3. The deep learning-based intellectual property infringement detection algorithm according to claim 1 is characterized in that: The training steps of the model training module (2) are as follows: Data preprocessing: preprocessing the collected existing intellectual property data, including data cleaning, normalization and feature labeling, and then using it as a training set for training the intellectual property infringement detection model (3); Feature extraction: input the processed training set into the intellectual property infringement detection model (3) to extract features from the data and learn the intrinsic features of the data. The extracted features include the novelty, creativity and practicality of the intellectual property; Model training, the extracted features are input into the intellectual property infringement detection model (3) for training, and the parameters of the model are adjusted to minimize the loss function.
4. The deep learning-based intellectual property infringement detection algorithm according to claim 1, characterized in that: The intellectual property infringement detection model (3) includes a recurrent neural network (31) and a long short-term memory network (32).
5. The deep learning-based intellectual property infringement detection algorithm according to claim 1, characterized in that: The time selection of the scope selection module (4) is based on the application date, publication date and authorization date of the intellectual property to be detected, wherein the novelty and creativity comparison selects the existing intellectual property before or after the application date for infringement detection.
6. The deep learning-based intellectual property infringement detection algorithm according to claim 1, characterized in that: The evaluation method of the model evaluation module (5) is to determine the threshold of the relevant similarity between the intellectual property data to be detected and the existing intellectual property data, wherein the relevant similarity includes the similarity of the beneficial effects of the intellectual property, the structural similarity and the content similarity.
7. The deep learning-based intellectual property infringement detection algorithm according to claim 4 is characterized in that: The storage and update algorithm of the long short-term memory network (32) is as follows: Forget gate, receives the current input x t and the hidden state h at the previous time step t-1 , through a fully connected layer and a sigmoid activation function to calculate the output f of the forget gate t , the value range is between 0 and 1, f t The closer the value is to 1, the more information is retained; the closer it is to 0, the more information is forgotten: f t =σ(W f ×[h t-1 ,x t ]+b f ); Where W f Forget gate weight, b f is the bias term, σ is the sigmoid activation function; The input gate also receives the current input xt and the hidden state ht-1 of the previous time step. It calculates the output it of the input gate and the new candidate memory cell state C~t through two fully connected layers, a sigmoid activation function, and a tanh activation function. The calculation formula is as follows: i t =σ(W i ×[h t-1 ,x t ]+b i ); C~t=tanh(W C ×[h t-1 ,x t ]+b C ); Where W i With W C are the weight matrices of the input gate and candidate memory cell states, b i With b C is the bias term; According to the output of the forget gate and the input gate, update the state C of the memory unit t , the calculation formula is as follows: C t =f t ⊙C t-1 +i t ⊙C~t The output gate, which receives the current input x t and the hidden state h of the previous time step t-1 , through a fully connected layer and a sigmoid activation function to calculate the output o of the output gate t , then o t And the memory cell state C processed by the tanh activation function t Multiply them together to get the hidden state h of the current time step t The calculation formula is as follows: the t =σ(W o ×[h t-1 ,x t ]+b o ); h t =o t ⊙tanh(C t ); Where W o is the weight matrix of the output gate, b o is the bias term.