A neural network-based field mapping method, apparatus, and system

Abstract

The application discloses a field mapping method and device based on a neural network, a storage medium and a system. Two neural network convolution models are trained, and inference is simultaneously performed on to-be-mapped data to obtain a first inference result and a second inference result respectively. The obtained vectors are multiplied, and a maximum value in a source object field range is taken as an inference result. The field mapping method, the device, the storage medium and the system improve the accuracy of field mapping.

Description

Technical Field

[0001] This invention relates to the field of neural network-based field mapping technology, and more particularly to a neural network-based field mapping method, apparatus, computer-readable storage medium, and system. Background Technology

[0002] Field mapping is the process of associating a field in one table or database with a field in another table or database. It's a common operation in data processing and analysis, helping to quickly and efficiently integrate, compare, and analyze data. When performing field mapping, the data types and formats of the two fields to be mapped must be determined to ensure correct mapping. Configuration can be challenging when implementers configure data integration schemes or when clients want to modify existing schemes, especially if the target and source objects have significantly different fields (dissimilar encodings and names) and no standard pre-built schemes are available for reference. If similar data integration schemes have been configured in other projects, it's desirable to collect data from these schemes, train a deep learning model, and use this model to recommend field mappings that reuse this experience, thereby reducing implementation and configuration difficulty.

[0003] In existing technologies, matching is typically performed by the encoding or name of fields in the source and target objects.

[0004] However, the existing technology still has the following drawbacks: 1. Using field codes and names for matching has a limited application scenario, and it cannot match when there are differences in field codes and names between the source system and the target system; 2. Relying on implementation personnel to perform field matching manually is inefficient and the skill level varies, and it cannot make full use of existing mapping relationships in other fields.

[0005] Therefore, there is a current need for a field mapping method, apparatus, computer-readable storage medium, and system based on neural networks to overcome the aforementioned deficiencies in the prior art. Summary of the Invention

[0006] This invention provides a neural network-based field mapping method, apparatus, computer-readable storage medium, and system, thereby improving the accuracy of field mapping.

[0007] An embodiment of the present invention provides a field mapping method based on a neural network. The field mapping method includes: acquiring data to be mapped, and acquiring a convolutional model with specified weights and a convolutional model without specified weights from a preset resource center; both the convolutional model with specified weights and the convolutional model without specified weights are neural network convolutional models; inputting the data to be mapped into the convolutional model with specified weights to obtain a first inference result, inputting the data to be mapped into the convolutional model without specified weights to obtain a second inference result, and calculating a field mapping result based on the data to be mapped, the first inference result, and the second inference result.

[0008] As an improvement to the above scheme, before acquiring the data to be mapped and obtaining the specified weight convolutional model and the unspecified weight convolutional model from a preset resource center, the field mapping method further includes: periodically acquiring running data at a preset first time interval and cleaning and sorting the running data to obtain training data; obtaining the specified weight convolutional model and the unspecified weight convolutional model according to a preset text vector model, the training data, and a preset convolutional model framework; and periodically synchronizing the specified weight convolutional model and the unspecified weight convolutional model to a preset resource center at a preset second time interval.

[0009] As an improvement to the above scheme, the running data is cleaned and sorted, specifically including: removing special symbols from the running data, splitting the words in the running data, and removing the first word from the running data according to the frequency of occurrence of the word and a preset frequency range to obtain the second word; and sorting the second word according to a preset importance sorting sequence.

[0010] As an improvement to the above scheme, a field mapping result is calculated based on the data to be mapped, the first inference result, and the second inference result. Specifically, this includes: multiplying the first inference result by the second inference result to obtain a first inference result vector; obtaining all second vectors of the data to be mapped and summing all second vectors to obtain a second inference result vector; multiplying the first inference result vector by the second inference result vector to obtain a final inference result vector, and selecting the maximum value from the final inference result vector as the field mapping result.

[0011] As an improvement to the above scheme, based on a preset text vector model, the training data, and a preset convolutional model framework, a convolutional model with specified weights and a convolutional model without specified weights are obtained. Specifically, this includes: training the training data using the preset text vector model to obtain a word vector matrix; creating a convolutional model with specified weights and a convolutional model without specified weights based on the preset convolutional model framework and the word vector matrix; and training the convolutional model with specified weights and the convolutional model without specified weights respectively using the input parameter set in the training data as input parameters.

[0012] As an improvement to the above scheme, the specified weight convolutional model uses the word vector matrix as the weight of each word, while the unspecified weight convolutional model does not specify the weight of each word and obtains the word vectors through backpropagation.

[0013] As an improvement to the above scheme, after obtaining the specified weight convolutional model and the unspecified weight convolutional model from the preset resource center, the field mapping method further includes: deploying the weight convolutional model and the unspecified weight convolutional model locally.

[0014] As an improvement to the above scheme, the importance ranking sequence is: target field, source object name, target object name, source system name, and target system name.

[0015] As an improvement to the above scheme, the dropout layer is used to randomly shut down neural connections to prevent overfitting.

[0016] Another embodiment of the present invention provides a field mapping device based on a neural network. The field mapping device includes a resource acquisition unit and an inference mapping unit. The resource acquisition unit is used to acquire data to be mapped and to acquire a specified weight convolutional model and an unspecified weight convolutional model from a preset resource center. Both the specified weight convolutional model and the unspecified weight convolutional model are neural network convolutional models. The inference mapping unit is used to input the data to be mapped into the specified weight convolutional model to obtain a first inference result, input the data to be mapped into the unspecified weight convolutional model to obtain a second inference result, and calculate a field mapping result based on the data to be mapped, the first inference result, and the second inference result.

[0017] As an improvement to the above scheme, the field mapping device further includes a model training unit, which is used to: periodically acquire running data at a preset first time interval and perform data cleaning and sorting on the running data to obtain training data; obtain a convolutional model with specified weights and a convolutional model without specified weights according to a preset text vector model, the training data, and a preset convolutional model framework; and periodically synchronize the convolutional model with specified weights and the convolutional model without specified weights to a preset resource center at a preset second time interval.

[0018] As an improvement to the above scheme, the model training unit is also used to: remove special symbols from the running data, split the words in the running data, and remove the first word from the running data according to the frequency of occurrence of the words and a preset frequency range.

[0019] As an improvement to the above scheme, the inference mapping unit is further configured to: multiply the first inference result and the second inference result to obtain a first inference result vector; obtain all second vectors of the data to be mapped, and sum all the second vectors to obtain a second inference result vector; multiply the first inference result vector and the second inference result vector to obtain a final inference result vector, and select the maximum value from the final inference result vector as the field mapping result.

[0020] As an improvement to the above scheme, the model training unit is further configured to: train the training data using a preset text vector model to obtain a word vector matrix; create a weighted convolutional model and an unweighted convolutional model based on a preset convolutional model framework and the word vector matrix; and train the weighted convolutional model and the unweighted convolutional model respectively using the input parameter group in the training data as input parameters.

[0021] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the neural network-based field mapping method as described above.

[0022] One embodiment of the present invention provides a computer program product that, when run on a computer, causes the computer to execute the neural network-based field mapping method described above.

[0023] Another embodiment of the present invention provides a neural network-based field mapping system, the field mapping system including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the neural network-based field mapping method as described above.

[0024] Compared with existing technologies, this technical solution has the following beneficial effects:

[0025] This invention provides a field mapping method, apparatus, computer-readable storage medium, and system based on neural networks. By training two neural network convolutional models, inference is performed simultaneously on the data to be mapped to obtain a first inference result and a second inference result, respectively. Based on the data to be mapped, the first inference result, and the second inference result, a field mapping result is calculated. This field mapping method, apparatus, computer-readable storage medium, and system improve the accuracy of field mapping. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a field mapping method based on a neural network according to an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram illustrating the principle of a field mapping method based on a neural network according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the structure of a field mapping device based on a neural network provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1

[0031] The present invention first describes a field mapping method based on a neural network.

[0032] like Figure 1 As shown, the field mapping method includes:

[0033] S1: Obtain the data to be mapped, and obtain the specified weight convolutional model and the unspecified weight convolutional model from the preset resource center.

[0034] This embodiment aims to achieve efficient field mapping through a model. Field mapping refers to determining the mapping relationship between source data and target data fields during data integration. In this description, the data to be mapped is the source data, which is the metadata used for data integration in the service cloud, used to describe the data model, operations, events, and other information of the source or target system; the field mapping result is the target data.

[0035] In one embodiment, before acquiring the data to be mapped and obtaining the specified weight convolutional model and the unspecified weight convolutional model from a preset resource center, the field mapping method further includes: periodically acquiring running data at a preset first time interval and cleaning and sorting the running data to obtain training data; obtaining the specified weight convolutional model and the unspecified weight convolutional model according to a preset text vector model, the training data, and a preset convolutional model framework; and periodically synchronizing the specified weight convolutional model and the unspecified weight convolutional model to a preset resource center at a preset second time interval.

[0036] The periodically acquired operational data constitutes a data integration scheme, which integrates source and target objects in the service cloud. This scheme is used to configure the mapping relationship and intermediate transformation relationships between source and target objects. The training data consists of cleaned and sorted operational data, including at least: target system name, source system name, target metadata name, source system metadata name, and target system field code.

[0037] In one embodiment, obtaining a weighted convolutional model and an unweighted convolutional model based on a preset text vector model, the training data, and a preset convolutional model framework specifically includes: training the training data using the preset text vector model to obtain a word vector matrix; creating a weighted convolutional model and an unweighted convolutional model based on the preset convolutional model framework and the word vector matrix; and training the weighted convolutional model and the unweighted convolutional model respectively using the input parameter set in the training data as input parameters. The input parameter set includes, but is not limited to, the target system name, the source system name, the target object name, the source object name, and the target field.

[0038] Specifically, a word vector matrix is ​​obtained by training the collected data using the FastText model (described as a "text vector model" in this paper), and two convolutional models, A (described as a "weighted convolutional model" in this paper) and B (described as a "weightless convolutional model" in this paper), are created respectively. After creating the weighted and weightless convolutional models, the target system name, source system name, target object name, source object name, and target field from the collected data are used as input parameters, and the corresponding target field is used as the label. The weighted and weightless convolutional models are then trained respectively. As an improvement to the above scheme, the weighted convolutional model uses the word vector matrix as the weight of each word, while the weightless convolutional model does not specify the weight of each word, and obtains the word vectors through backpropagation.

[0039] To ensure the accuracy of the training data and thus improve the accuracy of the trained model, it is necessary to periodically synchronize the client's data integration plan to the data platform for cleaning and reconstruct the data based on the variable importance obtained from SPSS analysis for training purposes. The target system name, source system name, target object name, source object name, and target field from the collected data are used as input parameters, with the corresponding target field serving as the label for training both weighted and unweighted convolutional models. In the actual analysis, SPSS software analysis revealed that the importance of variables affecting the source field, in descending order, is: target field, source object name, target object name, source system name, and target system name. Therefore, the variables used for prediction are arranged in this order (described in this paper as the "importance ranking sequence").

[0040] In one embodiment, data cleaning and sorting of the running data specifically includes: removing special symbols from the running data; splitting the words in the running data; removing a first word from the running data based on the frequency of occurrence of the words and a preset frequency range to obtain a second word; and sorting the second word according to a preset importance ranking sequence. The first word is a word in the running data whose frequency value is outside the preset frequency range (i.e., removing high-frequency and low-frequency words).

[0041] A word is composed of a number of ordered characters, and a collection of words is called a vocabulary. For example, in the C language, words include identifiers, keywords, operators, delimiters, constants, strings, and comment characters. A field is a common term in computer science, referring to a column in a database table. In a database table, each field defines the type of data in that column, such as text, number, date, etc., and each field has a unique name so that programmers can easily reference them.

[0042] Both the weighted convolutional model and the unweighted convolutional model are recurrent neural network convolutional models. The recurrent neural network convolutional model includes an embedding layer, a fully connected layer, a dropout layer, and a softmax activation function. The embedding layer uses the result of fasttext training as the word vector matrix, and the word vector matrix in the embedding layer is not specified and is determined through backpropagation. The fully connected layer includes an LSTM layer with 200 neurons. The dropout layer is used to randomly close neural connections to prevent overfitting. The softmax activation function is used as the final activation function to ensure that the result is obtained.

[0043] In one embodiment, after obtaining the specified weight convolutional model and the unspecified weight convolutional model from a preset resource center, the field mapping method further includes: obtaining the specified weight convolutional model and the unspecified weight convolutional model from the preset resource center; and deploying the weight convolutional model and the unspecified weight convolutional model locally.

[0044] S2: Input the data to be mapped into the specified weight convolution model to obtain a first inference result, input the data to be mapped into the unspecified weight convolution model to obtain a second inference result, and calculate the field mapping result based on the data to be mapped, the first inference result, and the second inference result.

[0045] In one embodiment, calculating the field mapping result based on the data to be mapped, the first inference result, and the second inference result specifically includes: multiplying the first inference result and the second inference result to obtain a first inference result vector; obtaining all second vectors of the data to be mapped and summing all second vectors to obtain a second inference result vector; multiplying the first inference result vector and the second inference result vector to obtain a final inference result vector, and selecting the maximum value from the final inference result vector as the field mapping result. In one embodiment, obtaining all second vectors of the data to be mapped specifically involves: obtaining the one-hot vector of the field corresponding to the source object.

[0046] In practical applications, the data collected periodically according to the data collection plan is first cleaned. Then, the cleaned data is input into two convolutional models (a weighted convolutional model and an unweighted convolutional model) to obtain a first inference result and a second inference result. The weighted convolutional model requires the FastText pre-trained word vector matrix as the weights for each word during its creation. Next, the first and second inference results are multiplied to obtain an inference result vector. Simultaneously, the one-hot vectors of all fields of the source object are obtained and summed to obtain the second inference result vector. Finally, the first and second inference result vectors are multiplied, the maximum value is taken, and the field mapping result is output through a word segmenter. This process can be completed in a data platform in practice, and the entire field mapping model can be deployed to a service cloud for use by other tenants.

[0047] In practical use, the above methods enable: 1. Process automation: Once the data collection, model training, and deployment processes are established and perfected, no further human intervention is required; 2. Asset accumulation: As more and more customers use the integrated cloud, the on-site configuration mapping relationships will be continuously accumulated and used as assets; 3. Security and reliability: Mapping is performed only on prediction results with high accuracy obtained by the model, and mapping is not based on prediction results when the prediction accuracy is low; 4. Selectivity: Users can modify the mapping relationships obtained by model prediction to adapt to special on-site scenarios.

[0048] In practical applications, there may be significant differences between the target and source objects in the data integration schemes already deployed and the desired modification schemes (the encoding and names may be dissimilar). Furthermore, there may be no standard pre-built scheme to refer to, making it quite difficult for implementers to configure new schemes or modify old ones. This invention addresses this issue by proposing a neural network-based mapping method. By collecting a large amount of data integration scheme data and using it as a training basis, corresponding features are extracted from a large number of field mappings. A field mapping model is trained using deep learning, enabling the implementation to extract features from the data to be mapped and obtain recommended field mapping results from the field mapping model based on these extracted features. This avoids the need for reconfiguration when there are significant differences, thus achieving the reuse of mapping experience.

[0049] This invention describes a field mapping method based on neural networks. By training two neural network convolutional models, inference is performed simultaneously on the data to be mapped to obtain a first inference result and a second inference result, respectively. The obtained vectors are multiplied, and the maximum value within the range of the source object field is taken as the inference result. This field mapping method improves the accuracy of field mapping. Specific Implementation Example 2

[0051] In addition to the methods described above, embodiments of the present invention also disclose a field mapping device based on a neural network. See [link to related documentation]. Figure 2 Another embodiment provides a schematic diagram of the structure of a device.

[0052] like Figure 2 As shown, the field mapping device includes a resource acquisition unit 11 and an inference mapping unit 12.

[0053] The resource acquisition unit 11 is used to acquire the data to be mapped and to acquire the specified weight convolutional model and the unspecified weight convolutional model from the preset resource center. The specified weight convolutional model and the unspecified weight convolutional model are both recurrent neural network convolutional models. The recurrent neural network convolutional model includes an embedding layer, a fully connected layer, a dropout layer, and a softmax activation function. The fully connected layer includes an LSTM layer with 200 neurons.

[0054] The inference mapping unit 12 is used to input the data to be mapped into the specified weight convolution model to obtain a first inference result, input the data to be mapped into the unspecified weight convolution model to obtain a second inference result, and calculate the field mapping result based on the data to be mapped, the first inference result, and the second inference result.

[0055] In one embodiment, the inference mapping unit 12 is further configured to: multiply the first inference result by the second inference result to obtain a first inference result vector; obtain all second vectors of the data to be mapped, and sum all the second vectors to obtain a second inference result vector; multiply the first inference result vector and the second inference result vector to obtain a final inference result vector, and select the maximum value from the final inference result vector as the field mapping result.

[0056] In one embodiment, the field mapping device further includes a model training unit, which is configured to: periodically acquire running data and perform data cleaning and sorting on the running data at a preset first time interval to obtain training data; obtain a convolutional model with specified weights and a convolutional model without specified weights based on a preset text vector model, the training data, and a preset convolutional model framework; and periodically synchronize the convolutional model with specified weights and the convolutional model without specified weights to a preset resource center at a preset second time interval.

[0057] In one embodiment, the model training unit is further configured to: remove special symbols from the running data, split the words in the running data, and remove the first word from the running data according to the occurrence frequency of the word and a preset frequency range to obtain the second word; and sort the second word according to a preset importance sorting sequence.

[0058] In one embodiment, the model training unit is further configured to: train the training data using a preset text vector model to obtain a word vector matrix; and create a weighted convolutional model and an unweighted convolutional model based on a preset convolutional model framework and the word vector matrix.

[0059] If the integrated unit of the field mapping device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the aforementioned neural network-based field mapping method.

[0060] The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0061] An embodiment of the present invention provides a computer program product that, when run on a computer, causes the computer to execute the neural network-based field mapping method as described above, and is able to achieve the same technical effects as described above.

[0062] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between units indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0063] This invention describes a field mapping device and a computer-readable storage medium based on a neural network. By training two neural network convolutional models, inference is performed simultaneously on the data to be mapped to obtain a first inference result and a second inference result, respectively. The obtained vectors are multiplied, and the maximum value within the range of the source object field is taken as the inference result. This field mapping device and computer-readable storage medium improve the accuracy of field mapping. Specific Implementation Example 3

[0065] In addition to the methods and apparatus described above, embodiments of the present invention also describe a field mapping system based on a neural network.

[0066] The field mapping system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the neural network-based field mapping method as described above.

[0067] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the device, connecting various parts of the device via various interfaces and lines.

[0068] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0069] This invention describes a field mapping system based on neural networks. By training two neural network convolutional models, inference is performed simultaneously on the data to be mapped to obtain a first inference result and a second inference result, respectively. The obtained vectors are multiplied, and the maximum value within the range of the source object field is taken as the inference result. This field mapping system improves the accuracy of field mapping.

[0070] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A field mapping method based on a neural network, characterized in that, The field mapping method includes: Obtain the data to be mapped, and obtain a convolutional model with specified weights and a convolutional model without specified weights from a preset resource center; both the convolutional model with specified weights and the convolutional model without specified weights are neural network convolutional models. The data to be mapped is input into the convolutional model with specified weights to obtain a first inference result, and the data to be mapped is input into the convolutional model without specified weights to obtain a second inference result. Based on the data to be mapped, the first inference result, and the second inference result, the field mapping result is calculated. Before acquiring the data to be mapped and obtaining the specified weight convolutional model and the unspecified weight convolutional model from a preset resource center, the field mapping method further includes: Obtain the training data; Based on the preset text vector model, the training data, and the preset convolution model framework, a convolution model with specified weights and a convolution model without specified weights are obtained. Synchronize the specified weight convolutional model and the unspecified weight convolutional model to the preset resource center; The step of calculating the field mapping result based on the data to be mapped, the first inference result, and the second inference result specifically includes: Based on the first reasoning result and the second reasoning result, a first reasoning result vector is obtained; Obtain the second vector of all fields of the data to be mapped, and sum all the second vectors to obtain the second inference result vector; Based on the first inference result vector and the second inference result vector, a final inference result vector is obtained; and field mapping results are selected from the final inference result vector.

2. The field mapping method based on neural networks according to claim 1, characterized in that, The acquisition of training data includes: At a preset first time interval, the running data is periodically acquired, and the running data is cleaned and sorted to obtain the training data.

3. The field mapping method based on neural networks according to claim 1, characterized in that, The step of calculating the field mapping result based on the data to be mapped, the first inference result, and the second inference result specifically includes: Multiply the first reasoning result by the second reasoning result to obtain the first reasoning result vector; Obtain the second vector of all fields of the data to be mapped, and sum all the second vectors to obtain the second inference result vector; wherein, the second vector is the one-hot vector of the corresponding field of the source object; The first inference result vector and the second inference result vector are multiplied to obtain the final inference result vector, and the maximum value in the final inference result vector is selected as the field mapping result.

4. The field mapping method based on neural networks according to claim 1, characterized in that, The specified weight convolutional model assigns weights to each word using a word vector matrix, while the unspecified weight convolutional model does not specify weights for each word and obtains word vectors through backpropagation.

5. The field mapping method based on neural networks according to claim 2, characterized in that, The operational data is cleaned and sorted, specifically including: Remove special symbols from the running data, split the words in the running data, and remove the first word from the running data according to the occurrence frequency of the word and a preset frequency range to obtain the second word; wherein, the first word is a word in the running data whose occurrence frequency value is outside the preset frequency range; The second word is sorted according to a preset importance sorting sequence.

6. The field mapping method based on neural networks according to claim 5, characterized in that, The importance ranking sequence is: target field, source object name, target object name, source system name, and target system name.

7. The field mapping method based on neural networks according to claim 2, characterized in that, Based on a preset text vector model, the training data, and a preset convolutional model framework, a convolutional model with specified weights and a convolutional model without specified weights are obtained, specifically including: The training data is trained using a preset text vector model to obtain a word vector matrix; Based on the preset convolution model framework and the word vector matrix, create a convolution model with specified weights and a convolution model without specified weights. Using the input parameter set in the training data as input parameters, the convolutional model with specified weights and the convolutional model without specified weights are trained respectively.

8. A field mapping device based on a neural network, characterized in that, The field mapping device includes a resource acquisition unit and an inference mapping unit, wherein... The resource acquisition unit is used to acquire the data to be mapped, and to acquire a convolutional model with specified weights and a convolutional model without specified weights from a preset resource center; both the convolutional model with specified weights and the convolutional model without specified weights are neural network convolutional models. The inference mapping unit is used to input the data to be mapped into the specified weight convolution model to obtain a first inference result, input the data to be mapped into the unspecified weight convolution model to obtain a second inference result, and calculate the field mapping result based on the data to be mapped, the first inference result, and the second inference result. The field mapping device further includes a model training unit, which is used to: acquire training data; obtain a convolutional model with specified weights and a convolutional model without specified weights based on a preset text vector model, the training data, and a preset convolutional model framework; and synchronize the convolutional model with specified weights and the convolutional model without specified weights to a preset resource center. The inference mapping unit is further configured to: obtain a first inference result vector based on the first inference result and the second inference result; obtain all second vectors of the data to be mapped, and accumulate all second vectors to obtain a second inference result vector; obtain a final inference result vector based on the first inference result vector and the second inference result vector, and select field mapping results from the final inference result vector.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the neural network-based field mapping method as described in any one of claims 1 to 7.

10. A field mapping system based on a neural network, characterized in that, The field mapping system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the neural network-based field mapping method as described in any one of claims 1 to 7.

11. A computer program product, characterized in that, When the computer program product is run on a computer, it causes the computer to perform the neural network-based field mapping method as described in any one of claims 1 to 7.

Images