Enterprise data management method and system based on deep learning

Abstract

The invention relates to the technical field of enterprise management, in particular to an enterprise data management method and system based on deep learning, and the method comprises the steps: collecting enterprise data, constructing an enterprise database, and carrying out the classified storage to form sub-databases of various types of enterprise data; obtaining a search text of a visitor in response to the data access request; the search text of the visitor is input into an enterprise data matching model, the enterprise data matching model is associated with an enterprise database, the model calculates the matching degree of the search text and each piece of enterprise data within the data range, meeting the visitor permission level, of the enterprise database, and the enterprise data with the matching degree meeting a preset threshold value is output; the model comprises a plurality of hidden layers, the sub-database of each type of enterprise data corresponds to one or more hidden layers of the model, and the number of the hidden layers of the selected enterprise data type is adjusted based on the selection of the visitor for the enterprise data. According to the invention, the real-time access requirement of enterprise employees to enterprise data can be flexibly and efficiently met.

Description

Technical Field

[0001] The present application relates to the field of enterprise management technology, and in particular to an enterprise data management method and system based on deep learning. Background Art

[0002] As enterprises expand in size and their information level improves, they generate and accumulate massive amounts of enterprise data in their daily operations. The types of enterprise data cover finance, human resources, supply chain, customer relationship management, etc. As enterprises expand in size and their business becomes more complex, the types and quantities of enterprise data double year by year. At the same time, higher requirements are placed on data management and utilization, especially in terms of the efficiency of data access and compliance with data security.

[0003] In the prior art, the management of enterprise data mainly relies on the database authority control system. Specifically, when enterprise employees access the enterprise database, they confirm their data access rights through identity authentication and limit the database range they can access. The database authority control system extracts keywords from the data entered by enterprise employees, queries the database and extracts enterprise data from the database range that employees can access in the enterprise database. The existing method controls the access level of data through employee authority, ensures data security, and can meet the needs of data access to a certain extent.

[0004] However, when faced with diverse enterprise data, the existing technology still relies on static query methods for enterprise employees to access enterprise data, that is, the access request and data structure are fixed in design, lacking the ability to dynamically adapt to different data types and access requirements, and failing to optimize according to the data type. At the same time, when providing data feedback, due to the static query method and the sometimes large amount of data, it is often difficult to accurately feedback the enterprise data that the visitor wants to access. Therefore, the existing methods are difficult to flexibly and efficiently meet the real-time access needs of enterprise employees to enterprise data. Summary of the invention

[0005] This application provides a deep learning-based enterprise data management method and system that can flexibly and efficiently meet the real-time access needs of enterprise employees to enterprise data. This application provides the following technical solutions: In a first aspect, the present application provides an enterprise data management method based on deep learning, the method comprising: Collecting enterprise data and its corresponding permission levels and constructing an enterprise database, classifying and storing the enterprise data in the enterprise database according to the type of the enterprise data to form sub-databases for each type of enterprise data; In response to a visitor's request for data access, obtaining the visitor's search text and permission level; Input the visitor's search text and permission level into a preset enterprise data matching model, associate the enterprise data matching model with the enterprise database, calculate the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and output the enterprise data whose matching degree meets the preset threshold; The enterprise data matching model includes several hidden layers, and each sub-database of each type of enterprise data corresponds to one or more hidden layers of the enterprise data matching model. After the enterprise data matching model outputs the enterprise data, the number of hidden layers of the selected enterprise data type is adjusted based on the visitor's choice of the enterprise data.

[0006] In a specific possible implementation scheme, the enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold, including calculating the matching degree between the search text and each enterprise data based on text similarity and structural similarity: The vector representations of the search text and the text of each enterprise data are generated respectively and ; Calculate the absolute difference between the search text and the enterprise data text in the semantic representation space, that is, the Manhattan distance as follows: ; in, Represents each dimension of a vector; Calculate the Jaccard similarity between the search text and the enterprise data text. The formula is as follows: ; in, and Represents the collection of words in the search text and enterprise data text respectively; Use the following formula to calculate the matching degree between the search text and the enterprise data text : ; After calculating the matching degree between the search text and the enterprise data text, the enterprise data matching model outputs the enterprise data whose matching degree is less than or equal to a preset threshold.

[0007] In a specific implementation scheme, the enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold, including calculating the matching degree between the search text and each enterprise data using a method combining a semantic representation model and a text reconstruction metric: Use the autoencoder technology to reconstruct the text of the search text and enterprise data as follows: ; ; in, To search for text from The reconstructed output is, For text from enterprise data The reconstructed output; Use Euclidean distance to calculate the reconstruction error of the text: ; ; in, and Represent the reconstruction errors of the search text and the enterprise data text respectively; the matching degree between the search text and the enterprise data text is calculated by the following formula : ; in is a constant; After calculating the matching degree between the search text and the enterprise data text, the enterprise data matching model outputs enterprise data whose matching degree is greater than or equal to a preset threshold.

[0008] In a specific implementation manner, adjusting the number of hidden layers of the selected enterprise data type based on the visitor's selection of the enterprise data includes: After the enterprise data matching model outputs the enterprise data, the visitor selects the output enterprise data, and the number of hidden layers of the type to which the enterprise data belongs is adjusted according to the enterprise data selected by the visitor; If the visitor selects enterprise data with a high matching degree, the number of hidden layers corresponding to the type of enterprise data will be reduced; If the visitor selects enterprise data with a lower matching degree, the number of hidden layers corresponding to the type of the enterprise data will be increased.

[0009] In a specific implementation manner, adjusting the number of hidden layers of the selected enterprise data type based on the visitor's selection of the enterprise data includes: After the enterprise data matching model outputs the enterprise data, the visitor selects the output enterprise data, and the number of hidden layers of the type to which the enterprise data belongs is adjusted according to the enterprise data selected by the visitor; The adjustment formula for the number of hidden layers is as follows: ; in, is the number of hidden layers after adjustment, is the current number of hidden layers for this data type, The total number of enterprise data output by the enterprise data matching model, The company data selected for the visitor is ranked from high to low in the matching results. is an integer, To adjust the threshold constant, calculate If it is not an integer, the nearest integer is taken as The value of .

[0010] In a specific implementation scheme, after the enterprise data matching model outputs the enterprise data, the following further comprises: Display data based on the visitor's custom sorting rules, such as matching degree sorting, data update time sorting, and data creation time sorting; Allow visitors to combine multiple custom sorting rules for data display.

[0011] In the second aspect, the present application provides an enterprise data management system based on deep learning, which adopts the following technical solutions: An enterprise data management system based on deep learning, comprising: A database construction module, used to collect enterprise data and its corresponding authority levels and construct an enterprise database, in which the enterprise data is classified and stored according to the type of the enterprise data to form a sub-database of each type of enterprise data; A request response module, used to respond to a visitor's request for data access and obtain the visitor's search text and permission level; An enterprise data matching module, used for inputting the search text and permission level of the visitor into a preset enterprise data matching model, the enterprise data matching model is associated with the enterprise database, the enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold; A hidden layer adjustment module is used for the enterprise data matching model, which includes several hidden layers. Each sub-database of enterprise data of each type corresponds to one or more hidden layers of the enterprise data matching model. After the enterprise data matching model outputs the enterprise data, the number of hidden layers of the selected enterprise data type is adjusted based on the visitor's choice of the enterprise data.

[0012] In a third aspect, the present application provides an electronic device, comprising a processor and a memory; the memory stores a program, and the program is loaded and executed by the processor to implement an enterprise data management method based on deep learning as described in the first aspect.

[0013] In a fourth aspect, the present application provides a computer-readable storage medium, wherein a program is stored in the storage medium, and when the program is executed by a processor, it is used to implement an enterprise data management method based on deep learning as described in the first aspect.

[0014] In summary, the beneficial effects of this application include at least: 1) By introducing an innovative matching calculation method that combines Manhattan distance and Jaccard similarity, the matching accuracy between enterprise data and search text is effectively improved. Manhattan distance can measure the difference between texts in semantic space, while Jaccard similarity quantifies the overlap of texts at the structural level. The combination of the two not only considers the semantic similarity of texts, but also the structural matching of texts, thereby improving the accuracy of enterprise data search and extraction, especially adapting to complex and diverse enterprise data types.

[0015] 2) By adopting the text reconstruction error method based on autoencoders, this application effectively enhances the adaptive ability of the data matching model. This method can automatically adapt to different types of text data by learning the implicit features of the data, and can efficiently match both structured and unstructured data. Compared with traditional methods, the text reconstruction error method does not rely on specific weighting rules, but flexibly responds to different text types and data structures through the self-learning ability of the model, thereby improving the robustness and generalization ability of the system.

[0016] 3) Allowing enterprise data matching models to adjust the number of hidden layers for each data type based on user feedback can more flexibly adapt to data matching results. For example, a user's selection of a lower-ranked item indicates that the current number of layers may not be sufficient to capture the characteristics of that type of data, so the number of hidden layers is automatically increased. This dynamic adjustment helps improve the flexibility of the model in matching user needs, allowing the model to be gradually optimized without retraining.

[0017] Through dynamic matching evaluation and gradually adjusted hidden layer strategies, the enterprise data model can be more flexible to adapt to diverse data types and access requirements. With the expansion of enterprise scale and data volume, the existing technology relies on fixed structure and static query methods, which makes it difficult to optimize in real time according to different data types and access rights. This application constructs a permission-based classified enterprise data sub-database, uses semantic and structural matching algorithms to accurately extract the data required by visitors, and introduces a dynamic hidden layer adjustment mechanism to optimize the model structure in real time based on the matching results selected by visitors, thereby improving the accuracy and flexibility of matching. This not only meets the enterprise's requirements for efficient data access and security compliance, but also adapts to diverse data access needs, achieving dynamic adaptability and accuracy of the model.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present application in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the enterprise data management method based on deep learning in an embodiment of the present application.

[0020] Figure 2 It is a schematic diagram of the overall process of the enterprise data management method based on deep learning in an embodiment of the present application.

[0021] Figure 3 It is a structural block diagram of an enterprise data management system based on deep learning in an embodiment of the present application.

[0022] Figure 4 It is a block diagram of an electronic device for enterprise data management based on deep learning in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present application but are not intended to limit the scope of the present application.

[0024] Optionally, the present application takes the deep learning-based enterprise data management method provided in each embodiment as an example used in an electronic device, where the electronic device is a terminal or a server. The terminal may be a computer, a tablet computer, etc. This embodiment does not limit the type of electronic device.

[0025] Reference Figure 1 , is a flowchart of an enterprise data management method based on deep learning provided by an embodiment of the present application, the method includes at least the following steps: Step S101: collect enterprise data and their corresponding authority levels and construct an enterprise database, classify and store the enterprise data according to the type of the enterprise data in the enterprise database, and form sub-databases for each type of enterprise data.

[0026] In step S101, all required enterprise data are collected from various business departments within the enterprise. In the process of enterprise data collection, each enterprise data is associated with the corresponding permission level. The permission level includes multiple levels, such as public access, department-level access, management access, confidential data, etc. The specific settings are defined according to the security needs and data sensitivity of the enterprise. By integrating various types of enterprise data and their permission level information, the enterprise database in this application selects an existing database management system (such as MySQL, MongoDB or Elasticsearch, etc.) to classify and store enterprise data. For example, financial data, human resources data, customer relationship data, supply chain data, etc. are each stored in different sub-databases. Each enterprise data in this application is actually a text, which is convenient for subsequent matching operations.

[0027] Step S102: In response to a visitor's request for data access, obtain the visitor's search text and permission level.

[0028] In step S102, a corporate data access request from a visitor (i.e., an enterprise employee) is received, the request including the search text entered by the visitor through the front-end interactive window and the visitor's identity information, and the visitor's permission level is determined by verifying the visitor's identity information. The search text may be a word, phrase, or sentence, such as "financial report", "customer order details", or "market sales analysis for the third quarter of this year".

[0029] Step S103: input the visitor's search text and permission level into a preset enterprise data matching model, associate the enterprise data matching model with the enterprise database, calculate the matching degree between the search text and each enterprise data within the data range in the enterprise database that meets the visitor's permission level, and output the enterprise data whose matching degree meets the preset threshold.

[0030] It should be noted that the enterprise data matching model in this application is constructed based on the existing deep learning model. In this application, a convolutional neural network model is selected, and then the designed matching algorithm is embedded in each hidden layer of the convolutional neural network model, thereby obtaining the enterprise data matching model in this application. The enterprise data matching model includes an input layer, several hidden layers and an output layer, wherein several hidden layers are associated with the enterprise database to facilitate subsequent matching operations.

[0031] In a feasible embodiment, the matching degree between the search text and each enterprise data is calculated based on text similarity and structural similarity, and an innovative combination of Manhattan distance and semantic overlap is introduced. The core of this method is to use the semantic similarity measure between texts and combine structural similarity (such as phrases, grammatical structures, etc.) to create a new matching measure. Specifically, first use the existing BERT model or other existing models to generate vector representations of the search text and the text of each enterprise data, respectively. and , and then calculate the absolute difference between the search text and the enterprise data text in the semantic representation space, that is, the Manhattan distance. The Manhattan distance is used to measure the absolute difference between two texts in the semantic space. The larger the sum, the greater the difference between the two texts. Manhattan distance The calculation formula is as follows: ; in, Represents each dimension of the vector.

[0032] Then, it is necessary to measure the structural similarity between the search text and the enterprise data text at the phrase or word level. This application chooses to calculate the Jaccard similarity. The Jaccard similarity calculates the ratio of the intersection and union of two sets, reflecting the overlap of the two texts at the word or phrase level. The Jaccard similarity calculation formula for the search text and the enterprise data text is as follows: ; in, and Represents the collection of terms in the search text and enterprise data text respectively.

[0033] Finally, the following formula is used to calculate the matching degree between the search text and the enterprise data text : ; In the above formula, the final match is calculated by combining Manhattan distance and Jaccard similarity. Manhattan distance is used to quantify the semantic difference between two texts, and then Jaccard similarity is used to measure the structural similarity of the texts. In order to ensure that the two can complement each other and strengthen text matching, Jaccard similarity is normalized so that its value range is between 0 and 1, thus avoiding the complexity caused by weighting. Through the product of these two parts, the final match can reflect both semantic differences and structural similarities. Manhattan distance and Jaccard similarity complement each other in the calculation of the final match. Manhattan distance focuses on semantic differences, while Jaccard similarity focuses on the literal overlap of texts. Combining the two by multiplication can effectively balance the two different information, thereby obtaining a more accurate matching result.

[0034] In the implementation, after calculating the matching degree between the search text and the enterprise data text, the enterprise data matching model outputs the enterprise data with a matching degree less than or equal to the preset threshold. Traditional matching methods (such as using only word frequency or cosine similarity) usually only focus on the superficial vocabulary similarity and fail to effectively consider the differences in the semantic and structural levels of the text. For example, a system based on keyword search may miss some semantically related but vocabulary-different data, while matching based solely on word frequency may ignore differences in grammar or phrase structure. The above method not only considers the semantic similarity of the text (quantifying the vector difference of the text through Manhattan distance), but also combines the structural similarity of the text (measuring the overlap of vocabulary and phrases through Jaccard similarity). This dual matching method can fully capture the relationship between the search text and the enterprise data text and improve the matching accuracy. The biggest advantage of combining Manhattan distance and Jaccard similarity is that it can comprehensively evaluate text matching through dual dimensions. Manhattan distance captures differences at the semantic level, while Jaccard similarity focuses on the structural overlap of text. The combination of the two can effectively improve the accuracy and robustness of text matching and avoid the limitations of a single measurement method. This improves the system's adaptability and provides more accurate and flexible query results when processing complex and dynamically changing enterprise data.

[0035] In another feasible embodiment, the degree of match between the search text and each enterprise data can also be calculated by combining a semantic representation model and a text reconstruction metric. Text reconstruction error is an innovative method to measure the degree of match between the search text and the text of each enterprise data. The core idea is to judge their match by the difference in reconstruction between the search text and the text of each enterprise data. Unlike traditional semantic matching methods (such as cosine similarity), the present application quantifies similarity by reconstructing the implicit structure of each text, focusing on the expression differences of the text. Specifically, each text is first reconstructed using existing autoencoder technology. An autoencoder is an unsupervised learning model that learns the core features of a text by encoding the input text into a low-dimensional representation and then reconstructing it back to the original text through a decoder. The present application uses an existing autoencoder structure to reconstruct the text of the search text and enterprise data as follows: ; ; in, To search for text from The reconstructed output is, For text from enterprise data The reconstructed output. Then the text reconstruction error is calculated. The text reconstruction error is a measure of the difference between the original text and the reconstructed text. The smaller the text reconstruction error, the higher the similarity of the text. This application uses the Euclidean distance to calculate the text reconstruction error: ; ; in, and Represent the reconstruction errors of the search text and the enterprise data text respectively. Finally, the matching degree between the search text and the enterprise data text is calculated by the following formula: : ; in, is a constant used to control the result range. In this application, K is 100. In the above formula, Represents the absolute difference between the reconstruction error of the search text and the enterprise data text. The denominator is added with 1 to ensure that the denominator is always greater than zero to avoid the zero division problem. The reconstruction error is a way to measure the difference between the text structure and semantics. For the same text, the smaller the reconstruction error, the more accurately the core information such as the structure and semantics of the text can be reconstructed. and When we compare the difference between the search text and the enterprise data text, we are comparing the difference in "reconstruction accuracy" of the two texts. In other words, if the error difference between the search text and the enterprise data text is small, it means that their core features (structure and semantics) are similar and the match is high. If the difference is large, their core features are very different and the match is low. It can be used to reflect the consistency or difference in expression between two texts, further affecting their matching degree.

[0036] The above method not only considers the semantics of the search text, but also the implicit features of the text, and reconstructs the data features learned by the autoencoder, so as to more accurately capture the intrinsic relationship between the search text and the enterprise data text. In addition, this method does not rely on specific weighting rules, but calculates the matching degree through a new measurement method such as reconstruction error, which enables the method to adapt to different types of text and handle mixed matching of structured and unstructured data.

[0037] In the implementation, after calculating the matching degree between the search text and the enterprise data text, the enterprise data matching model outputs the enterprise data with a matching degree greater than or equal to the preset threshold. In the above method, the core idea of ​​the text reconstruction error is to reconstruct the text through an autoencoder, focusing on capturing the intrinsic characteristics of the text. This method can identify deeper structural and semantic information in the text, not just relying on superficial literal overlap. Compared with traditional methods such as cosine similarity, the latter usually focuses on the angular relationship between word vectors and cannot deeply understand the context and structural characteristics of the text. The text reconstruction error can reflect the core semantics of the text through the reconstruction process, thereby more accurately measuring the similarity between texts.

[0038] In addition, as a preferred embodiment, after the enterprise data matching model outputs the matched enterprise data, the data is displayed according to the visitor's custom sorting rules, such as sorting by matching degree, data update time, data creation time, etc. At the same time, multiple factors can be combined, for example: sorting by matching degree first, then sorting by a certain time of the data, and finally obtaining a comprehensive sorting result. The most appropriate data sorting method is selected according to the actual scenario, thereby improving the practicality and flexibility of the system.

[0039] Step S104: the enterprise data matching model includes several hidden layers. Each type of enterprise data sub-database corresponds to one or more hidden layers of the enterprise data matching model. After the enterprise data matching model outputs the enterprise data, the number of hidden layers of the selected enterprise data type is adjusted based on the visitor's choice of enterprise data.

[0040] In step S104, after the enterprise data matching model outputs the enterprise data, the visitor selects the output enterprise data, and then the number of hidden layers of the type of the enterprise data is adjusted according to the enterprise data selected by the visitor. Specifically, the adjustment formula of the number of hidden layers is as follows: ; in, is the number of hidden layers after adjustment, is the current number of hidden layers for this data type, The total number of enterprise data output by the enterprise data matching model, The company data selected for the visitor is ranked from high to low in the matching results. is an integer, To adjust the threshold constant, used to control the amplitude and direction of the adjustment, the value is 2 in this application. After the formula is calculated If it is not an integer, the nearest integer is taken as In the above formula, is the matching effect factor, is the normalization item, when the selected enterprise data matching result is at the front (i.e. When the matching effect factor is smaller, the matching effect factor value is larger (close to 1), and when the matching result of the selected enterprise data is later (i.e. When the value is larger, the matching effect factor value is smaller (close to 0). In the formula, the result range of the normalization term is shifted from [0,1] to [-0.5,0.5] by subtracting from 0.5.

[0041] For example, if the total number of enterprise data matching results output by the enterprise data matching model is The value is 10, and the matching degree of the enterprise data selected by the visitor is ranked 2, the matching result is good, the current number of hidden layers is 5, calculated according to the above formula 4.4, finally Take 4 as the calculation result. Accordingly, the number of hidden layers corresponding to the enterprise data type selected by the visitor is reduced from 5 to 4. If the matching degree of the enterprise data selected by the visitor is ranked is 8, the matching result is poor, the current number of hidden layers is 5, calculated according to the above formula 5.6, finally The calculation result is 6. Accordingly, the number of hidden layers corresponding to the enterprise data type selected by the visitor increases from 5 to 6. If the matching item is closer (the effect is better), the number of hidden layers increases, and if the matching degree is lower, the number of hidden layers decreases.

[0042] In step S104, in a traditional static model, the number of hidden layers is usually fixed during design and training, resulting in a relatively low adaptability of the model to data. Even if a certain type of matching effect is not good, it is necessary to manually retrain and set parameters, which is inefficient and difficult to quickly respond to the differentiated needs of different data types. This method allows the enterprise data matching model to adjust the number of hidden layers of each data type according to the visitor's feedback, and can adapt to the data matching effect more flexibly. For example, the visitor's choice of a lower ranking item indicates that the current number of layers may not be sufficient to capture the characteristics of this type of data, so the number of hidden layers is automatically increased. This dynamic adjustment helps to improve the flexibility of the model in matching visitor needs, so that the model can be gradually optimized without retraining. At the same time, through the dynamic adjustment of the visitor's selected feedback, the model can be gradually optimized in multiple interactions, which is manifested as dynamically adjusting the number of hidden layers according to the visitor's choice, so that the model effect meets the actual needs. This method enables the model to form a good closed loop with the visitor's needs and improve the visitor experience and matching accuracy.

[0043] In addition, traditional methods usually use a fixed model structure, treat different data types equally, and cannot dynamically allocate computing resources. The use of computing resources for static models is usually inevitably wasteful, while the dynamically adjusted number of hidden layers can effectively control the computational complexity. When the model performs well on a certain data type, automatically reducing the number of hidden layers can reduce the use of computing resources. This mechanism effectively prevents the model structure from being too complex, thereby saving memory and computing resources, and is particularly suitable for enterprise environments with diverse data types. This approach of adjusting the number of hidden layers based on the matching effect factor (calculated by the ratio of the ranking value to the total number) can reduce unnecessary computing overhead while ensuring matching accuracy.

[0044] In summary, combined with Figure 2 , this application constructs a classified enterprise data sub-database for storage, and performs semantic and structural matching within the data range of limited permissions according to the visitor's permissions and search keywords, so as to obtain data with high matching degree. After the visitor selects the result, the system adjusts the number of hidden layers of the corresponding type of data based on the selected matching ranking, and improves the generalization ability of the matching model by adjusting the number of hidden layers. Unlike the traditional static query method, this solution makes the enterprise data model more flexible to adapt to diversified data types and access requirements through dynamic matching evaluation and gradually adjusted hidden layer strategy. With the expansion of enterprise scale and data volume, the existing technology relies on fixed structure and static query method, and it is difficult to optimize in real time according to different data types and access permissions. This application constructs a classified enterprise data sub-database based on permissions, and accurately extracts the data required by the visitor with the help of semantic and structural matching algorithms. At the same time, a dynamic hidden layer adjustment mechanism is introduced to optimize the model structure in real time based on the matching results selected by the visitor, thereby improving the accuracy and flexibility of matching. This not only meets the requirements of enterprises for efficient data access and security compliance, but also can adapt to diversified data access needs, and realize the dynamic adaptability and accuracy of the model.

[0045] Figure 3 This is a block diagram of a deep learning-based enterprise data management system provided by an embodiment of the present application. The system includes at least the following modules: The database construction module is used to collect enterprise data and its corresponding permission levels and build an enterprise database, in which enterprise data is classified and stored according to its type to form sub-databases of each type of enterprise data.

[0046] The request response module is used to respond to the visitor's request for data access and obtain the visitor's search text and permission level.

[0047] The enterprise data matching module is used to input the visitor's search text and permission level into a preset enterprise data matching model. The enterprise data matching model is associated with the enterprise database. The enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold.

[0048] The hidden layer adjustment module is used for the enterprise data matching model, which includes several hidden layers. Each sub-database of enterprise data of each type corresponds to one or more hidden layers of the enterprise data matching model. After the enterprise data matching model outputs the enterprise data, the number of hidden layers of the selected enterprise data type is adjusted based on the visitor's choice of enterprise data.

[0049] For relevant details, refer to the above method embodiment.

[0050] Figure 4 4 is a block diagram of an electronic device provided by an embodiment of the present application. The device at least includes a processor 401 and a memory 402.

[0051] The processor 401 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 401 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 401 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 401 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 401 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0052] The memory 402 may include one or more computer-readable storage media, which may be non-transitory. The memory 402 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 402 is used to store at least one instruction, which is used to be executed by the processor 401 to implement the enterprise data management method based on deep learning provided in the method embodiment of the present application.

[0053] In some embodiments, the electronic device may further optionally include: a peripheral device interface and at least one peripheral device. The processor 401, the memory 402 and the peripheral device interface may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface via a bus, a signal line or a circuit board. Schematically, the peripheral devices include but are not limited to: a radio frequency circuit, a touch display screen, an audio circuit, and a power supply.

[0054] Of course, the electronic device may also include fewer or more components, which is not limited in this embodiment.

[0055] Optionally, the present application also provides a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the deep learning-based enterprise data management method of the above method embodiment.

[0056] Optionally, the present application also provides a computer product, which includes a computer-readable storage medium, in which a program is stored, and the program is loaded and executed by a processor to implement the deep learning-based enterprise data management method of the above method embodiment.

[0057] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0058] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A deep learning-based enterprise data management method, characterized in that: The method comprises: Collecting enterprise data and its corresponding permission levels and constructing an enterprise database, classifying and storing the enterprise data in the enterprise database according to the type of the enterprise data to form sub-databases for each type of enterprise data; In response to a visitor's request for data access, obtaining the visitor's search text and permission level; Input the visitor's search text and permission level into a preset enterprise data matching model, associate the enterprise data matching model with the enterprise database, calculate the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and output the enterprise data whose matching degree meets the preset threshold; The enterprise data matching model includes several hidden layers, and each sub-database of each type of enterprise data corresponds to one or more hidden layers of the enterprise data matching model. After the enterprise data matching model outputs the enterprise data, the number of hidden layers of the selected enterprise data type is adjusted based on the visitor's choice of the enterprise data.

2. The enterprise data management method based on deep learning according to claim 1, characterized in that: The enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold, including calculating the matching degree between the search text and each enterprise data based on text similarity and structural similarity: The vector representations of the search text and the text of each enterprise data are generated respectively and ; Calculate the absolute difference between the search text and the enterprise data text in the semantic representation space, that is, the Manhattan distance as follows: ； in, Represents each dimension of a vector; Calculate the Jaccard similarity between the search text and the enterprise data text. The formula is as follows: ； in, and Represents the collection of words in the search text and enterprise data text respectively; Use the following formula to calculate the matching degree between the search text and the enterprise data text : ； After calculating the matching degree between the search text and the enterprise data text, the enterprise data matching model outputs the enterprise data whose matching degree is less than or equal to a preset threshold.

3. The enterprise data management method based on deep learning according to claim 1, characterized in that: The enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold, including using a method combining the semantic representation model and the text reconstruction metric to calculate the matching degree between the search text and each enterprise data: Use the autoencoder technology to reconstruct the text of the search text and enterprise data as follows: ； ； in, To search for text from The reconstructed output is, For text from enterprise data The reconstructed output; Use Euclidean distance to calculate the reconstruction error of the text: ； ； in, and Represent the reconstruction errors of the search text and the enterprise data text respectively; the matching degree between the search text and the enterprise data text is calculated by the following formula : ； in is a constant; After calculating the matching degree between the search text and the enterprise data text, the enterprise data matching model outputs enterprise data whose matching degree is greater than or equal to a preset threshold.

4. The enterprise data management method based on deep learning according to claim 1, characterized in that: The adjusting the number of hidden layers of the selected enterprise data type based on the visitor's selection of the enterprise data comprises: After the enterprise data matching model outputs the enterprise data, the visitor selects the output enterprise data, and the number of hidden layers of the type to which the enterprise data belongs is adjusted according to the enterprise data selected by the visitor; If the visitor selects enterprise data with a high matching degree, the number of hidden layers corresponding to the type of enterprise data will be reduced; If the visitor selects enterprise data with a lower matching degree, the number of hidden layers corresponding to the type of the enterprise data will be increased.

5. The enterprise data management method based on deep learning according to claim 1, characterized in that: The adjusting the number of hidden layers of the selected enterprise data type based on the visitor's selection of the enterprise data comprises: After the enterprise data matching model outputs the enterprise data, the visitor selects the output enterprise data, and the number of hidden layers of the type to which the enterprise data belongs is adjusted according to the enterprise data selected by the visitor; The adjustment formula for the number of hidden layers is as follows: ； in, is the number of hidden layers after adjustment, is the current number of hidden layers for this data type, The total number of enterprise data output by the enterprise data matching model, The company data selected for the visitor is ranked from high to low in the matching results. is an integer, To adjust the threshold constant, calculate If it is not an integer, the nearest integer is taken as The value of .

6. The enterprise data management method based on deep learning according to claim 1, characterized in that: After the enterprise data matching model outputs the enterprise data, the following further includes: Display data based on the visitor's custom sorting rules, such as matching degree sorting, data update time sorting, and data creation time sorting; Allow visitors to combine multiple custom sorting rules for data display.

7. An enterprise data management system based on deep learning, characterized in that: include: A database construction module, used to collect enterprise data and its corresponding authority levels and construct an enterprise database, in which the enterprise data is classified and stored according to the type of the enterprise data to form a sub-database of each type of enterprise data; A request response module, used to respond to a visitor's request for data access and obtain the visitor's search text and permission level; An enterprise data matching module, used for inputting the search text and permission level of the visitor into a preset enterprise data matching model, the enterprise data matching model is associated with the enterprise database, the enterprise data matching model calculates the matching degree between the search text and each enterprise data within the data range of the enterprise database that meets the visitor's permission level, and outputs the enterprise data whose matching degree meets the preset threshold; A hidden layer adjustment module is used for the enterprise data matching model, which includes several hidden layers. Each sub-database of enterprise data of each type corresponds to one or more hidden layers of the enterprise data matching model. After the enterprise data matching model outputs the enterprise data, the number of hidden layers of the selected enterprise data type is adjusted based on the visitor's choice of the enterprise data.

8. An electronic device, characterized in that: The device includes a processor and a memory; a program is stored in the memory, and the program is loaded and executed by the processor to implement an enterprise data management method based on deep learning as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a program, which, when executed by a processor, is used to implement an enterprise data management method based on deep learning as described in any one of claims 1 to 6.

Images