A Deep Learning-Based Digital Project Plagiarism Detection Method Based on Multiple Innovation Theory
By constructing a multi-innovation convolutional neural network based on the multi-innovation theory of deep learning, the problem of duplicate project approval caused by synonym transformation in project application was solved, the efficiency and accuracy of project duplication detection were improved, the recall rate was ensured, and fiscal funds were saved.
Patent Information
- Application Number
- CN202411102648.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-13
AI Technical Summary
Existing technologies struggle to effectively identify synonym transformations in project applications, leading to frequent instances of duplicate project approvals. Traditional deep learning algorithms are computationally intensive and have slow convergence speeds, making it difficult to guarantee the efficiency, accuracy, and recall of project plagiarism detection.
A deep learning method based on multiple innovation theory is adopted to construct a multiple innovation convolutional neural network. By using the synonym judgment matrix and the formula of multiple innovation theory, the convolutional neural network algorithm is fine-tuned to improve the synonym extraction accuracy. The system determines whether the item information dataset and the synonym dataset meet the training conditions, constructs a multiple innovation convolutional neural network, and determines the output value of the fully connected layer to identify duplicate items.
This improved the efficiency and accuracy of project plagiarism detection, ensured a high recall rate, achieved accurate identification of synonyms, avoided duplicate project approvals, and saved public funds.
Smart Images

Figure CN118643820B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of project processing, and in particular to a digital project duplication checking method based on deep learning of multiple innovation theory. Background Art
[0002] With the rapid development and popularization of information technology, the management of digital projects has become an indispensable part of modern urban construction. However, multiple project applications and duplicate project approval have become one of the prominent problems in the field of digital project management. Duplicate project approval has caused a large waste of fiscal funds. Therefore, how to establish an effective and feasible digital project duplicate checking mechanism has become a major issue for government departments at all levels. One of the important tasks.
[0003] The common practice now is to simply compare the keywords in the application with the project database, or use traditional deep learning to check for duplicates, to screen out duplicate projects from a large number of reported projects. However, these methods cannot prevent applicants from deliberately changing synonyms in the title or slightly changing the content of the project application to easily avoid the duplicate checking system. At the same time, traditional deep learning algorithms based on neural networks have many network layers, so the amount of calculation is huge, and the algorithm converges slowly, so the effect of fine-tuning the parameters of each layer is very limited. Summary of the invention
[0004] In view of this, the present invention proposes a digital project duplication checking method based on deep learning of multiple innovation theory, which aims to automatically extract synonyms for the digital project duplication checking method to ensure the efficiency, accuracy and completeness of the project duplication checking.
[0005] The present invention proposes a method for checking duplicate content of digital projects based on deep learning of multi-innovation theory, comprising:
[0006] Acquire a project information data set and a synonym data set, and analyze the project information data set and the synonym data set to determine whether the project information data set and the synonym data set meet the training conditions;
[0007] When it is determined that the project information dataset and the synonym dataset meet the training conditions, constructing a multi-innovation theory formula;
[0008] Determine a pre-trained convolutional neural network algorithm, and fine-tune the convolutional neural network algorithm based on the project information data set, the synonym data set and the multi-innovation theory formula to obtain a multi-innovation convolutional neural network;
[0009] Determine the output value of the fully connected layer of the multi-innovation convolutional neural network, compare the output value with the type value of the project information training set, and determine whether there are duplicate projects based on the comparison result.
[0010] Furthermore, when analyzing the project information data set and the synonym data set to determine whether the project information data set and the synonym data set meet the training conditions, it includes:
[0011] A project information data sequence A is constructed according to the project information data set, wherein A={Ab1, Ab2, Ab3, ..., Ab n1};
[0012] Determine a first semantic vector sequence C corresponding to the project information data sequence A, where C={Cd1, Cd2, Cd3, ..., Cd n2};
[0013] Construct a synonym data sequence E according to the synonym data set, where E={Ef1, Ef2, Ef3, ..., Ef m1};
[0014] Determine a second semantic vector sequence G corresponding to the synonym data sequence E, where G={Gh1, Gh2, Gh3, ..., Gh m2};
[0015] Constructing a synonym judgment matrix K according to the first semantic vector sequence C and the second semantic vector sequence G, and calculating a lexical similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K;
[0016] Among them, the synonym judgment matrix K is:
[0017] ;
[0018] It is determined whether the project information dataset and the synonym dataset meet the training conditions based on the vocabulary similarity coefficient.
[0019] Furthermore, when calculating the vocabulary similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K, it includes:
[0020] The vocabulary similarity coefficient between the project information dataset and the synonym dataset is calculated according to the following formula:
[0021] ;
[0022] Among them, J is the vocabulary similarity coefficient, p1 is the number of values in the synonym judgment matrix K, p2 is the number of rows in the synonym judgment matrix K, and p1=n2, p3 is the number of columns in the synonym judgment matrix K, and p3=m2, Δi=1,2,3,...n2, Δj=1,2,3,...m2.
[0023] Further, when judging whether the project information dataset and the synonym dataset meet the training conditions based on the vocabulary similarity coefficient, it includes:
[0024] Obtaining a preset vocabulary similarity coefficient threshold, and judging whether the project information data set and the synonym data set meet the training conditions according to the relationship between the vocabulary similarity coefficient and the vocabulary similarity coefficient threshold;
[0025] When the vocabulary similarity coefficient is less than the vocabulary similarity coefficient threshold, it is determined that the project information data set and the synonym data set do not meet the training conditions;
[0026] When the vocabulary similarity coefficient is greater than or equal to the vocabulary similarity coefficient threshold, it is determined that the project information data set and the synonym data set meet the training conditions.
[0027] Furthermore, when it is determined that the project information dataset and the synonym dataset meet the training conditions, constructing the multi-innovation theory formula includes:
[0028] ;
[0029] in, is the multi-innovation theory function, t is the current time step or number of iterations, represents the gain matrix, and , R is a real number set, n is the number of rows of the gain matrix, p is the number of columns of the gain matrix, that is, the length of the new information, represents the innovation vector, , and p ≥ 1 represents the innovation length. When p = 1, it degenerates into the standard innovation.
[0030] Furthermore, when determining the pre-trained convolutional neural network algorithm, it includes:
[0031] Determine the nonlinear error function of the convolutional neural network algorithm:
[0032] ;
[0033] in, is the nonlinear error function of the convolutional neural network algorithm, The convolution kernel parameters in the convolution layer, For the network layer, is the weight of the fully connected layer, is the bias value, L is the number of network layers, h n is the output value of the nth network layer, y n is the expected output value of the nth network layer;
[0034] The nonlinear error function is trained based on the back propagation algorithm and the gradient descent method, and the network parameters are adjusted. The parameter update value for each iteration is:
[0035] ;
[0036] in, is the convolution layer parameter, i is the row index of the convolution kernel in the output feature map, and j is the column index of the convolution kernel in the output feature map;
[0037] ;
[0038] in, is the downsampling layer parameter;
[0039] ;
[0040] in, is the parameter of the fully connected layer, and W is the weight matrix.
[0041] Further, when the convolutional neural network algorithm is fine-tuned based on the project information dataset, the synonym dataset and the multi-innovation theory formula to obtain a multi-innovation convolutional neural network, it includes:
[0042] Expanding the output node x of each network layer of the convolutional neural network algorithm into an output vector based on the multi-innovation theory formula;
[0043] ;
[0044] in, is the output vector, t s is the target value t at time step s, and T is the transposition operation;
[0045] Update the convolution layer parameters, the downsampling layer parameters and the fully connected layer according to the output vector to obtain corresponding convolution layer update parameters, downsampling layer update parameters and fully connected layer update parameters;
[0046] A multi-innovation convolutional neural network is obtained based on the convolutional layer update parameters, the downsampling layer update parameters, the fully connected layer update parameters, the synonym data set and the project information data set.
[0047] Further, when the convolution layer parameters, the downsampling layer parameters and the fully connected layer are updated according to the output vector to obtain corresponding convolution layer update parameters, downsampling layer update parameters and fully connected layer update parameters, it includes:
[0048] ;
[0049] in, Update the parameters for the convolution layer, m is the column index;
[0050] ;
[0051] in, Update the parameters of the downsampling layer, and D is the weight of the downsampling layer;
[0052] ;
[0053] in, Update parameters for the fully connected layers.
[0054] Further, when a multi-innovation convolutional neural network is obtained based on the convolution layer update parameters, the downsampling layer update parameters, the fully connected layer update parameters, the synonym dataset and the project information dataset, it includes:
[0055] Inputting the synonym dataset into the output layer;
[0056] The output value of the convolutional layer is obtained according to the convolutional layer update parameters:
[0057] ;
[0058] in, is the output value of the convolutional layer, and M is the number of feature maps output by the convolutional layer;
[0059] The output value of the downsampling layer is obtained according to the downsampling layer update parameters:
[0060] ;
[0061] in, is the output value of the downsampling layer, and down is a weight used to adjust the mapping strength from the input feature map to the output feature map;
[0062] The output value of the fully connected layer is obtained according to the fully connected layer update parameters:
[0063] ;
[0064] in, is the output value of the fully connected layer.
[0065] Further, when determining the output value of the fully connected layer of the multi-innovation convolutional neural network, and comparing the output value with the type value of the project information training set, and judging whether there are duplicate projects based on the comparison result, it includes:
[0066] When the output value is greater than or equal to the type value, it is determined that there is a duplicate item;
[0067] When the output value is less than the type value, it is determined that there is no duplicate item.
[0068] Compared with the prior art, the present invention has the following beneficial effects:
[0069] The present invention discloses a digital project duplicate checking method based on deep learning of multiple-innovation theory, which determines whether a project information data set and a synonym data set meet training conditions; when the training conditions are met, constructs a multiple-innovation theory formula; determines a pre-trained convolutional neural network algorithm, and fine-tunes the convolutional neural network algorithm based on the project information data set, the synonym data set and the multiple-innovation theory formula to obtain a multiple-innovation convolutional neural network; determines an output value of a fully connected layer of the multiple-innovation convolutional neural network, compares the output value with a type value of a project information training set, and determines whether there are duplicate projects. The present invention can reversely fine-tune the parameters of each layer of the pre-trained convolutional neural network algorithm, improve the calculation convergence speed of the convolutional neural network, improve the accuracy of synonym extraction, accurately identify duplicate projects, and ensure the efficiency, accuracy and recall rate of project duplicate checking. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:
[0071] Figure 1 A flowchart of a method for checking duplicate content in a digital project based on deep learning using multi-innovation theory is provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0072] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0073] like Figure 1 As shown, in some embodiments of the present application, this embodiment provides a method for checking duplicate content of a digital project based on deep learning of multi-innovation theory, including:
[0074] S110: Acquire a project information data set and a synonym data set, and analyze the project information data set and the synonym data set to determine whether the project information data set and the synonym data set meet the training conditions;
[0075] In some embodiments of the present application, when analyzing the project information data set and the synonym data set to determine whether the project information data set and the synonym data set meet the training conditions, it includes:
[0076] A project information data sequence A is constructed according to the project information data set, wherein A={Ab1, Ab2, Ab3, ..., Ab n1};
[0077] Determine a first semantic vector sequence C corresponding to the project information data sequence A, where C={Cd1, Cd2, Cd3, ..., Cd n2};
[0078] Construct a synonym data sequence E according to the synonym data set, where E={Ef1, Ef2, Ef3, ..., Ef m1};
[0079] Determine a second semantic vector sequence G corresponding to the synonym data sequence E, where G={Gh1, Gh2, Gh3, ..., Gh m2};
[0080] Constructing a synonym judgment matrix K according to the first semantic vector sequence C and the second semantic vector sequence G, and calculating a lexical similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K;
[0081] Among them, the synonym judgment matrix K is:
[0082] ;
[0083] It is determined whether the project information dataset and the synonym dataset meet the training conditions based on the vocabulary similarity coefficient.
[0084] In this embodiment, n1 is the number of project information data, n2 is the number of first semantic vectors, m1 is the number of synonym data, and m2 is the number of second semantic vectors.
[0085] The beneficial effect of the above technical solution is: the present invention constructs a synonym judgment matrix K according to the first semantic vector sequence C and the second semantic vector sequence G, and calculates the lexical similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K, which can provide a judgment basis for judging whether the project information data set and the synonym data set meet the training conditions, thereby ensuring the judgment accuracy.
[0086] In some embodiments of the present application, when calculating the vocabulary similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K, it includes:
[0087] The vocabulary similarity coefficient between the project information dataset and the synonym dataset is calculated according to the following formula:
[0088] ;
[0089] Among them, J is the vocabulary similarity coefficient, p1 is the number of values in the synonym judgment matrix K, p2 is the number of rows in the synonym judgment matrix K, and p1=n2, p3 is the number of columns in the synonym judgment matrix K, and p3=m2, Δi=1,2,3,...n2, Δj=1,2,3,...m2.
[0090] In some embodiments of the present application, when judging whether the project information dataset and the synonym dataset meet the training conditions based on the vocabulary similarity coefficient, it includes:
[0091] Obtaining a preset vocabulary similarity coefficient threshold, and judging whether the project information data set and the synonym data set meet the training conditions according to the relationship between the vocabulary similarity coefficient and the vocabulary similarity coefficient threshold;
[0092] When the vocabulary similarity coefficient is less than the vocabulary similarity coefficient threshold, it is determined that the project information data set and the synonym data set do not meet the training conditions;
[0093] When the vocabulary similarity coefficient is greater than or equal to the vocabulary similarity coefficient threshold, it is determined that the project information data set and the synonym data set meet the training conditions.
[0094] In this embodiment, the vocabulary similarity coefficient threshold can be set according to actual conditions, and is preferably 0.85.
[0095] The beneficial effect of the above technical solution is: the present invention determines whether the project information data set and the synonym data set meet the training conditions according to the relationship between the vocabulary similarity coefficient and the vocabulary similarity coefficient threshold, which can ensure the accuracy of the digital duplicate checking project and lay the foundation for project duplicate checking.
[0096] S120: When it is determined that the project information dataset and the synonym dataset meet the training conditions, construct a multi-innovation theory formula;
[0097] In some embodiments of the present application, when it is determined that the project information dataset and the synonym dataset meet the training conditions, constructing the multi-innovation theory formula includes:
[0098] ;
[0099] in, is the multi-innovation theory function, t is the current time step or number of iterations, represents the gain matrix, and , R is a real number set, n is the number of rows of the gain matrix, p is the number of columns of the gain matrix, that is, the length of the new information, represents the innovation vector, , and p ≥ 1 represents the innovation length. When p = 1, it degenerates into the standard innovation.
[0100] In this embodiment, the multi-innovation theory is an important theory in the field of system identification, and its core lies in improving the accuracy of state estimation or parameter estimation of the system by using the innovations at multiple moments in the past.
[0101] In this embodiment, the innovation length refers to the number of past time point data used to recursively calculate the predicted value of the future time point in time series analysis.
[0102] The beneficial effect of the above technical solution is that the invention constructs a multi-innovation theory formula, which can accurately fine-tune the pre-trained convolutional neural network algorithm to avoid large errors. The multi-innovation theory formula can make full use of some of the previous useful information, which will have a better recognition effect. This method increases the amount of calculation of the algorithm itself, but improves the estimated convergence speed, exchanging a small time complexity for improved accuracy.
[0103] S130: determining a pre-trained convolutional neural network algorithm, and fine-tuning the convolutional neural network algorithm based on the project information dataset, the synonym dataset and the multi-innovation theory formula to obtain a multi-innovation convolutional neural network;
[0104] In some embodiments of the present application, when determining a pre-trained convolutional neural network algorithm, it includes:
[0105] Determine the nonlinear error function of the convolutional neural network algorithm:
[0106] ;
[0107] in, is the nonlinear error function of the convolutional neural network algorithm, The convolution kernel parameters in the convolution layer, For the network layer, is the weight of the fully connected layer, is the bias value, L is the number of network layers, h n is the output value of the nth network layer, y n is the expected output value of the nth network layer;
[0108] The nonlinear error function is trained based on the back propagation algorithm and the gradient descent method, and the network parameters are adjusted. The parameter update value for each iteration is:
[0109] ;
[0110] in, is the convolution layer parameter, i is the row index of the convolution kernel in the output feature map, and j is the column index of the convolution kernel in the output feature map;
[0111] ;
[0112] in, is the downsampling layer parameter;
[0113] ;
[0114] in, is the parameter of the fully connected layer, and W is the weight matrix.
[0115] The beneficial effect of the above technical solution is that the key to the multi-innovation convolutional neural network lies in the fine-tuning of the parameters of each layer during the back-propagation process. Therefore, the fine-tuning of the parameters of each layer lays the foundation for the construction of the multi-innovation convolutional neural network.
[0116] In some embodiments of the present application, when the convolutional neural network algorithm is fine-tuned based on the project information dataset, the synonym dataset and the multi-innovation theory formula to obtain the multi-innovation convolutional neural network, it includes:
[0117] Expanding the output node x of each network layer of the convolutional neural network algorithm into an output vector based on the multi-innovation theory formula;
[0118] ;
[0119] in, is the output vector, t s is the target value t at time step s, and T is the transposition operation;
[0120] Update the convolution layer parameters, the downsampling layer parameters and the fully connected layer according to the output vector to obtain corresponding convolution layer update parameters, downsampling layer update parameters and fully connected layer update parameters;
[0121] A multi-innovation convolutional neural network is obtained based on the convolutional layer update parameters, the downsampling layer update parameters, the fully connected layer update parameters, the synonym data set and the project information data set.
[0122] The beneficial effect of the above technical solution is: based on the multi-innovation theory formula, the present invention expands the output node x of each network layer of the convolutional neural network algorithm into an output vector, and combines the convolution layer parameters, downsampling layer parameters and fully connected layer to accurately construct a multi-innovation convolutional neural network and improve construction efficiency.
[0123] In some embodiments of the present application, when the convolution layer parameters, the downsampling layer parameters, and the fully connected layer are updated according to the output vector to obtain corresponding convolution layer update parameters, downsampling layer update parameters, and fully connected layer update parameters, it includes:
[0124] ;
[0125] in, Update the parameters for the convolution layer, m is the column index;
[0126] ;
[0127] in, Update the parameters of the downsampling layer, and D is the weight of the downsampling layer;
[0128] ;
[0129] in, Update parameters for the fully connected layers.
[0130] In some embodiments of the present application, when a multi-innovation convolutional neural network is obtained based on the convolution layer update parameters, the downsampling layer update parameters, the fully connected layer update parameters, the synonym dataset and the project information dataset, it includes:
[0131] Inputting the synonym dataset into the output layer;
[0132] The output value of the convolutional layer is obtained according to the convolutional layer update parameters:
[0133] ;
[0134] in, is the output value of the convolutional layer, and M is the number of feature maps output by the convolutional layer;
[0135] The output value of the downsampling layer is obtained according to the downsampling layer update parameters:
[0136] ;
[0137] in, is the output value of the downsampling layer, and down is a weight used to adjust the mapping strength from the input feature map to the output feature map;
[0138] The output value of the fully connected layer is obtained according to the fully connected layer update parameters:
[0139] ;
[0140] in, is the output value of the fully connected layer.
[0141] The beneficial effect of the above technical solution is: the present invention combines the multi-innovation theory to reversely fine-tune the parameters of each layer of the traditional neural network to improve the computational convergence speed of the convolutional neural network and the true accuracy of synonym extraction, and exchanges a small time complexity for improved accuracy. It is in line with the scenario of duplicate checking in digital projects and has practical social value.
[0142] S140: Determine the output value of the fully connected layer of the multi-innovation convolutional neural network, and compare the output value with the type value of the project information training set, and determine whether there are duplicate projects based on the comparison result.
[0143] In some embodiments of the present application, when determining the output value of the fully connected layer of the multi-innovation convolutional neural network, and comparing the output value with the type value of the project information training set, and judging whether there are duplicate projects based on the comparison result, it includes:
[0144] When the output value is greater than or equal to the type value, it is determined that there is a duplicate item;
[0145] When the output value is less than the type value, it is determined that there is no duplicate item.
[0146] In this embodiment, the type value of the project information training set is a specific value, which is set according to the project information training set.
[0147] The beneficial effect of the above technical solution is: the present invention compares the output value with the type value of the project information training set, and determines whether there are duplicate projects based on the comparison results, thereby achieving accurate detection of duplicate projects, improving the detection efficiency of synonym projects, and avoiding duplicate project establishment and causing financial losses.
[0148] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0149] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0150] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0151] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A digital project duplication checking method based on deep learning of multi-innovation theory, characterized in that: include: Acquire a project information data set and a synonym data set, and analyze the project information data set and the synonym data set to determine whether the project information data set and the synonym data set meet the training conditions; When it is determined that the project information dataset and the synonym dataset meet the training conditions, constructing a multi-innovation theory formula; Determine a pre-trained convolutional neural network algorithm, and fine-tune the convolutional neural network algorithm based on the project information data set, the synonym data set and the multi-innovation theory formula to obtain a multi-innovation convolutional neural network; Determine the output value of the fully connected layer of the multi-innovation convolutional neural network, and compare the output value with the type value of the project information training set, and determine whether there are duplicate projects based on the comparison result; When it is determined that the project information dataset and the synonym dataset meet the training conditions, constructing the multi-innovation theory formula includes: ; in, is the multi-innovation theory function, t is the current time step or number of iterations, represents the gain matrix, and , R is a real number set, n is the number of rows of the gain matrix, p is the number of columns of the gain matrix, that is, the length of the new information, represents the innovation vector, , and p ≥ 1 represents the innovation length. When p = 1, it degenerates to the standard innovation. When deciding on a pre-trained convolutional neural network algorithm, include: Determine the nonlinear error function of the convolutional neural network algorithm: ; in, is the nonlinear error function of the convolutional neural network algorithm, The convolution kernel parameters in the convolution layer, For the network layer, is the weight of the fully connected layer, is the bias value, L is the number of network layers, h n is the output value of the nth network layer, y n is the expected output value of the nth network layer; The nonlinear error function is trained based on the back propagation algorithm and the gradient descent method, and the network parameters are adjusted. The parameter update value for each iteration is: ; in, is the convolution layer parameter, i is the row index of the convolution kernel in the output feature map, and j is the column index of the convolution kernel in the output feature map; ; in, is the downsampling layer parameter; ; in, is the parameter of the fully connected layer, W is the weight matrix; When the convolutional neural network algorithm is fine-tuned based on the project information dataset, the synonym dataset and the multi-innovation theory formula to obtain a multi-innovation convolutional neural network, the method includes: Expanding the output node x of each network layer of the convolutional neural network algorithm into an output vector based on the multi-innovation theory formula; ; in, is the output vector, t s is the target value t at time step s, and T is the transposition operation; Update the convolution layer parameters, the downsampling layer parameters and the fully connected layer according to the output vector to obtain corresponding convolution layer update parameters, downsampling layer update parameters and fully connected layer update parameters; Obtaining a multi-innovation convolutional neural network based on the convolution layer update parameters, the downsampling layer update parameters, the fully connected layer update parameters, the synonym dataset, and the project information dataset; When the convolution layer parameters, the downsampling layer parameters and the fully connected layer are updated according to the output vector to obtain corresponding convolution layer update parameters, downsampling layer update parameters and fully connected layer update parameters, the method includes: ; in, Update the parameters for the convolution layer, m is the column index; ; in, Update the parameters of the downsampling layer, and D is the weight of the downsampling layer; ; in, Update parameters for the fully connected layer; When a multi-innovation convolutional neural network is obtained based on the convolution layer update parameters, the downsampling layer update parameters, the fully connected layer update parameters, the synonym dataset and the project information dataset, it includes: Inputting the synonym dataset into the output layer; The output value of the convolutional layer is obtained according to the convolutional layer update parameters: ; in, is the output value of the convolutional layer, and M is the number of feature maps output by the convolutional layer; The output value of the downsampling layer is obtained according to the downsampling layer update parameters: ; in, is the output value of the downsampling layer, and down is a weight used to adjust the mapping strength from the input feature map to the output feature map; The output value of the fully connected layer is obtained according to the fully connected layer update parameters: ; in, is the output value of the fully connected layer.
2. The method for checking duplicate content of digital projects based on deep learning of multiple innovation theory according to claim 1 is characterized in that: When analyzing the project information data set and the synonym data set to determine whether the project information data set and the synonym data set meet the training conditions, it includes: A project information data sequence A is constructed according to the project information data set, wherein A={Ab1, Ab2, Ab3, ..., Ab n1 }; Determine a first semantic vector sequence C corresponding to the project information data sequence A, where C={Cd1, Cd2, Cd3, ..., Cd n2 }; Construct a synonym data sequence E according to the synonym data set, where E={Ef1, Ef2, Ef3, ..., Ef m1 }; Determine a second semantic vector sequence G corresponding to the synonym data sequence E, where G={Gh1, Gh2, Gh3, ..., Gh m2 }; Constructing a synonym judgment matrix K according to the first semantic vector sequence C and the second semantic vector sequence G, and calculating a lexical similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K; Among them, the synonym judgment matrix K is: ; It is determined whether the project information dataset and the synonym dataset meet the training conditions based on the vocabulary similarity coefficient.
3. The method for checking duplicate content of a digital project based on deep learning of multiple innovation theory according to claim 2 is characterized in that: When calculating the vocabulary similarity coefficient between the project information data set and the synonym data set according to the synonym judgment matrix K, it includes: The vocabulary similarity coefficient between the project information dataset and the synonym dataset is calculated according to the following formula: ; Among them, J is the vocabulary similarity coefficient, p1 is the number of values in the synonym judgment matrix K, p2 is the number of rows in the synonym judgment matrix K, and p1=n2, p3 is the number of columns in the synonym judgment matrix K, and p3=m2, Δi=1,2,3,...n2, Δj=1,2,3,...m2.
4. The method for checking duplicate content of a digital project based on deep learning of multiple innovation theory according to claim 2 is characterized in that: When judging whether the project information dataset and the synonym dataset meet the training conditions based on the vocabulary similarity coefficient, it includes: Obtaining a preset vocabulary similarity coefficient threshold, and judging whether the project information data set and the synonym data set meet the training conditions according to the relationship between the vocabulary similarity coefficient and the vocabulary similarity coefficient threshold; When the vocabulary similarity coefficient is less than the vocabulary similarity coefficient threshold, it is determined that the project information data set and the synonym data set do not meet the training conditions; When the vocabulary similarity coefficient is greater than or equal to the vocabulary similarity coefficient threshold, it is determined that the project information data set and the synonym data set meet the training conditions.
5. The method for checking duplicate content of digital items based on deep learning of multi-innovation theory according to claim 1 is characterized in that: When determining the output value of the fully connected layer of the multi-innovation convolutional neural network, and comparing the output value with the type value of the project information training set, and judging whether there is a duplicate project based on the comparison result, it includes: When the output value is greater than or equal to the type value, it is determined that there is a duplicate item; When the output value is less than the type value, it is determined that there is no duplicate item.
