Digital archive data processing method and system based on AI

Abstract

The invention relates to the technical field of digital archives, and discloses an AI-based digital archive data processing method and system, and the method comprises the steps: reading metadata information of an archive; randomly initializing a visualization strategy; collecting behavior data of the user once in each visual control period, wherein the behavior data of the user comprises a query record of the user and an operation record of the user in the visual environment; constructing a sequential element group based on the collected behavior data of the user; inputting the sequential element group into the visual control neural network, outputting a visual strategy, and executing the visual strategy in the next visual control period; according to the method, a deep learning generation model is utilized, an archive information visualization system is developed, the interactive experience of a user and a digital archive is enhanced through technologies such as virtual reality and augmented reality, and a visualization strategy can be continuously updated according to behavior interaction of the user and the digital archive.

Description

Technical Field

[0001] The present invention relates to the field of digital archive technology, and more specifically, to a method and system for processing digital archive data based on AI. Background Art

[0002] OFD is based on XML technology to describe document data. It is compact in size, open in format, easy to understand, and readable and usable for a long time. OFD expands its features according to the characteristics of various fields in my country, and more deeply meets the application needs. It provides great convenience for the processing of domestic digital archive data. On this basis, many methods for the visualization and utilization of digital archives have been proposed, but they are generally based on fixed rules and arrangement logic for the visualization of digital archives, and cannot achieve adaptive visualization for each user. Summary of the invention

[0003] The present invention provides a method and system for processing digital archive data based on AI to solve technical problems in related technologies.

[0004] The present invention provides a method for processing digital archive data based on AI, comprising:

[0005] Step 100, read the metadata information of the archive;

[0006] Step 200, randomly initialize the visualization strategy, the visualization strategy includes the archive room modeling type, the storage location of the archive, the archive group type, the brightness mark of the archive, and the preview item type of the archive;

[0007] Step 300, collecting user behavior data once in each visualization control cycle, the user behavior data includes the user's query records and the user's operation records in the visualization environment;

[0008] Step 400, construct a sequential element group based on the collected user behavior data, wherein a single-column element grid and a square matrix element grid are stored in an element position of the sequential element group, wherein an element position of the single-column element grid stores association information of an abstract object, and an element position of the square matrix element grid stores a binary number, and the element position of the square matrix element grid can locate two element positions with the same number of rows in the single-column element grid, and the number of rows of the two element positions of the single-column element grid is equal to the number of rows or columns of the element positions of the square matrix element grid. If the binary number stored in the element position of the square matrix element grid is 1, it indicates that there is an abstract connection between the two abstract objects corresponding to the two element positions of the located single-column element grid, otherwise the stored binary number is 0;

[0009] Step 500, input the sequential element group into the visualization control neural network, output the visualization strategy, and the visualization strategy is executed in the next visualization control cycle.

[0010] Furthermore, the archive content feature vectors extracted in advance by the deep learning model are obtained. These feature vectors represent the semantic information of the archive text and image content.

[0011] Furthermore, the archive room modeling type refers to how to represent and display the overall structure and layout of the archive room in the visualization system;

[0012] The storage location of each file is the spatial location of each file in the archive room modeling;

[0013] The archive group type refers to the type of rules followed by archives located in the same area;

[0014] Brightness marking of archives refers to the use of different brightness levels to represent certain attributes or states of archives in visualization;

[0015] The preview item type of a profile refers to the type of profile preview information displayed when a user selects or hovers over a profile in the visualization.

[0016] Furthermore, the method of the present invention is applied in a virtual reality environment.

[0017] Furthermore, abstract objects include users, archives, archive groups, and archive shelves;

[0018] Cases where there are abstract connections between abstract objects include:

[0019] The user views the file, the file belongs to a file group, and the file is located on the file shelf.

[0020] Furthermore, the calculation formula of the first encoder is as follows:

[0021]

[0022] where k i Represents the first identification feature of the i-th element of a single-column element grid, φ i represents the aggregation coefficient of the i-th element of a single-column element grid, M i represents the co-located element bit set of the i-th element bit of a single-column element grid, the element bit of the single-column element grid in the co-located element bit set and the i-th element bit of the single-column element grid are located at the same element bit of the square matrix element grid with a stored binary number of 1, σ represents the activation function, W k represents the state weight coefficient, x j Represents the association information stored in the i-th element position of a single-column element grid;

[0023]

[0024] Z i =W z x i

[0025] Z j =W z x j

[0026] where x i and x j Respectively represent the XXX information of the independent object represented by the i-th and j-th cells of the data matrix, W z represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, and LeakyRelu represents the rectified linear unit function;

[0027] The calculation formula of the second encoder is as follows:

[0028] u (t) =σ(W u X (t) +W u H (t-1) +b u )

[0029] r (t) =σ(W r X (t) +W r H (t-1) +b r )

[0030] C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c )

[0031] H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1)

[0032] Among them, W u , W r , W c represents the weight parameter, b u 、b r 、b c represents the bias parameter, ⊙ represents the dot product, u(t) 、r (t) and C (t) Represent the first, second, and third internal coding features respectively, where M all represents the set of all element bits of a single-column grid of elements, The first identification feature of the i-th element position of the single-column element grid output when the content stored in the t-th element position of the sequential element group is input into the first encoder, H (t) and H (t-1) They represent the t-th and t-1-th second identification features, respectively, n≥t≥1, n represents the total number of element positions in the sequential element group, and when t=1, H (t-1) =X (t) , tanh is the hyperbolic tangent function, and σ represents the sigmoid function.

[0033] Furthermore, the calculation formula of the FC layer is as follows:

[0034] y=σ(W y h n +b y )

[0035] Where y represents a one-dimensional array, h n represents the nth second identification feature, W y is the output layer weight parameter, b y is the output layer bias parameter, and σ represents the sigmoid function.

[0036] Furthermore, the training steps of the visual control neural network are as follows:

[0037] Step 101, copy the visual control neural network to generate the Actor network and the TargetActor network, use the feedforward neural network as the Critic network, copy the Critic network to obtain the TargetCritic network, and randomly initialize the parameters of the Actor network, the TargetActor network, the Critic network, and the TargetCritic network;

[0038] Step 102: Input the sequential element group s of the user collected for the tth time t Go to the Actor network to get the visualization strategy a' t ;

[0039] Visualization strategy a' t Add random noise to obtain visualization strategy a t ;

[0040] The visualization strategy a t Input into the environment and get immediate reward r for feedback tand the next sequential element group s t+1 ;Experience s t ,a t ,r t ,s t+1 Deposit into experience pool;

[0041] Step 103, randomly obtain N experiences from the experience pool;

[0042] Input sequential element group s t+1 Go to the TargetActor network to get the visualization strategy a t+1 ;

[0043] Step 104, calculating the first loss value L1 and the second loss value L2, and updating the Actor network and the Critic network using the gradient descent method;

[0044]

[0045]

[0046] Where Q Critic (s t ,a t ) represents the input sequential element group s t and visualization strategy a t The output of the Critic network, Q Critic (s t ,a' t ) means to convert a' t and t Input the output of the Critic network, Indicates the execution of visualization strategy a t The comprehensive Q value, Q TargetCritic 9s t+1 ,a t+1 ) represents the input sequential element group s t+1 and visualization strategy a t+1 The output of the TargetCritic network;

[0047] γ is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0048] Update the Actor network through L1 and the Critic network through L2;

[0049] Step 105, every fixed number of network updates, update the TargetActor so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network;

[0050] Step 106, the step is terminated when the Actor network and the Critic network converge, or the number of cycles of steps 103-105 reaches a set value, the step is terminated.

[0051] The present invention also provides a system for processing digital archive data based on AI, which can be run on a computer entity and can execute the aforementioned method for processing digital archive data based on AI when it is run.

[0052] The present invention also provides a computer storage medium for storing computer-readable instructions, which can execute the aforementioned AI-based digital archive data processing method when the computer-readable instructions are read.

[0053] The beneficial effects of the present invention are:

[0054] The present invention utilizes deep learning to generate models and develops an archive information visualization system. Through virtual reality, augmented reality and other technologies, the interactive experience between users and digital archives is enhanced, and the visualization strategy can be continuously updated according to the behavioral interaction between users and digital archives to optimize services for each user. There is no need to obtain user preferences and purpose information in advance, so the system has strong applicability and is more flexible. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of a method for processing digital archive data based on AI of the present invention. DETAILED DESCRIPTION

[0056] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0057] At least one embodiment of the present invention discloses a method for processing digital archive data based on AI, such as Figure 1 As shown, the following steps are included:

[0058] Step 100, read the metadata information of the archive, such as title, author, creation time, subject keywords, etc.

[0059] Obtain archive content feature vectors extracted in advance by deep learning models (such as CNN, RNN, etc.). These feature vectors represent the semantic information of archive text, images, etc.

[0060] Step 200, randomly initialize the visualization strategy, the visualization strategy includes the archive room modeling type, the storage location of the archive, the archive group type, the brightness mark of the archive, and the preview item type of the archive;

[0061] Archives Modeling Type:

[0062] Archives room modeling type refers to how to represent and display the overall structure and layout of archives in a visualization system. Common modeling types include modeling types that simulate various types of real archives and archives. One modeling type corresponds to the modeling of a simulated archive or archive.

[0063] Storage location of each archive:

[0064] That is, the spatial position of each archive in the archive room modeling;

[0065] File group type:

[0066] The file group type refers to the type of rules followed by files in the same area.

[0067] Example:

[0068] Partition by subject: Divide the archive into different areas according to the subject or category of the archive, and each area will store archives of the corresponding subject. For example, put the files related to finance in one area and the files related to human resources in another area.

[0069] Timeline layout: Arrange the files along the timeline according to their creation or modification time. The newest files are placed at one end of the timeline and the oldest files are placed at the other end.

[0070] Relevance layout: Place related archives in adjacent locations based on their relevance (such as common authors, citation relationships, etc.). This makes it easier for users to discover and explore related archives.

[0071] Brightness mark of the file:

[0072] Brightness marking of archives refers to using different brightness levels to represent certain attributes or states of archives in visualization. Brightness marking can help users quickly identify and distinguish different types of archives.

[0073] Preview item type for each archive:

[0074] Explanation: The preview item type of a profile refers to the type of profile preview information displayed when a user selects or hovers over a profile in the visualization. The preview item can help users quickly understand the profile's summary content without fully opening the profile.

[0075] Example:

[0076] Metadata preview: Displays key metadata of the archive, such as title, author, creation time, subject keywords, etc.

[0077] Thumbnail preview: For image-type archives, a thumbnail preview of the archive is displayed, allowing users to quickly browse the archive's visual content.

[0078] Summary Preview: Displays a summary or summary of the archive, providing a brief overview of the archive's contents.

[0079] Related archive preview: Displays a list of other archives associated with the current archive, making it easier for users to explore the relationship between archives

[0080] Preferably, the method of the present invention is applied to modeling and visualization of virtual reality environments.

[0081] Step 300, collecting user behavior data once in each visualization control cycle, the user behavior data includes the user's query records and the user's operation records in the visualization environment;

[0082] Step 400, construct a sequential element group based on the collected user behavior data, wherein a single-column element grid and a square matrix element grid are stored in an element position of the sequential element group, wherein an element position of the single-column element grid stores association information of an abstract object, and an element position of the square matrix element grid stores a binary number, and the element position of the square matrix element grid can locate two element positions with the same number of rows in the single-column element grid, and the number of rows of the two element positions of the single-column element grid is equal to the number of rows or columns of the element positions of the square matrix element grid. If the binary number stored in the element position of the square matrix element grid is 1, it indicates that there is an abstract connection between the two abstract objects corresponding to the two element positions of the located single-column element grid, otherwise the stored binary number is 0;

[0083] In one embodiment of the present invention, the abstract objects include users, files, file groups, and file racks;

[0084] Cases where there are abstract connections between abstract objects include:

[0085] The user views the file, the file belongs to a file group, and the file is located on the file shelf.

[0086] Step 500, input the sequential element group into the visualization control neural network, output the visualization strategy, and the visualization strategy is executed in the next visualization control cycle.

[0087] Based on the aforementioned methods, interactive information query and filtering functions can also be provided, allowing users to quickly locate archives of interest through keyword search, attribute screening, etc.

[0088] For the selected archive node, detailed information is displayed, such as complete metadata, archive summary, related archives, etc.

[0089] It supports users to mark, comment, and share archives of interest, promoting collaboration and knowledge sharing.

[0090] The present invention provides a system for processing digital archive data based on AI, which can be run on a computer entity and can execute the aforementioned method for processing digital archive data based on AI when it is run.

[0091] The present invention provides a computer storage medium for storing computer-readable instructions, which can execute the aforementioned AI-based digital archive data processing method when the computer-readable instructions are read.

[0092] In one embodiment of the present invention, the visualization control neural network includes a backbone network, which includes a first encoder, a second encoder and an FC (fully connected layer) layer, wherein the first encoder is used to sequentially input the content stored in each element bit of the sequential element group, and correspondingly output the first identification feature to the second encoder, the second encoder outputs the second identification feature to the FC layer, and the FC layer outputs a one-dimensional array, the c-th number of the one-dimensional array represents the probability of the c-th visualization strategy being selected, and the visualization strategy with the largest probability value is used as the output.

[0093] Specifically, an example of a first encoder is provided, and the calculation formula of the first encoder is as follows:

[0094]

[0095] where k i represents the first identification feature of the i-th element position of a single-column element grid, represents the aggregation coefficient of the i-th element of a single-column element grid, M i represents the co-located element bit set of the i-th element bit of a single-column element grid, the element bit of the single-column element grid in the co-located element bit set and the i-th element bit of the single-column element grid are located at the same element bit of the square matrix element grid with a stored binary number of 1, σ represents the activation function, W k represents the state weight coefficient, x j Represents the association information stored in the i-th element position of a single-column element grid;

[0096]

[0097] Z i =W z x i

[0098] Z j =W z x j

[0099] where x i and x j Respectively represent the XXX information of the independent object represented by the i-th and j-th cells of the data matrix, W z represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, and LeakyRelu represents the rectified linear unit function;

[0100] Specifically, an example of a second encoder is provided, and the calculation formula of the second encoder is as follows:

[0101] u (t) =σ(W u X (t) +W u H (t-1) +b u )

[0102] r (t) =σ(W r X (t) +W r H (t-1) +b r )

[0103] C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c )

[0104] H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1)

[0105] Among them, W u , W r , W c represents the weight parameter, b u 、b r 、b c represents the bias parameter, ⊙ represents the dot product, u (t) 、r (t) and C (t) Represent the first, second, and third internal coding features respectively, where M allrepresents the set of all element bits of a single-column grid of elements, The first identification feature of the i-th element position of the single-column element grid output when the content stored in the t-th element position of the sequential element group is input into the first encoder, H (t) and H (t-1) They represent the t-th and t-1-th second identification features, respectively, n≥t≥1, n represents the total number of element positions in the sequential element group, and when t=1, H (t-1) =X (t) , tanh is the hyperbolic tangent function, and σ represents the sigmoid function.

[0106] Specifically, an example of the FC layer is provided. The calculation formula of the FC layer is as follows:

[0107] y=σ(W y h n +b y )

[0108] Where y represents a one-dimensional array, h n represents the nth second identification feature, W y is the output layer weight parameter, b y is the output layer bias parameter, and σ represents the sigmoid function.

[0109] The training steps of the visual control neural network are as follows:

[0110] Step 101, copy the visual control neural network to generate the Actor network and the TargetActor network, use the feedforward neural network as the Critic network, copy the Critic network to obtain the TargetCritic network, and randomly initialize the parameters of the Actor network, the TargetActor network, the Critic network, and the TargetCritic network;

[0111] Step 102: Input the sequential element group s of the user collected for the tth time t Go to the Actor network to get the visualization strategy a' t ;

[0112] Visualization strategy a' t Add random noise to obtain visualization strategy a t ;

[0113] The visualization strategy a t Input into the environment and get immediate reward r for feedback t and the next sequential element group s t+1 ;Experience s t ,a t ,r t ,s t+1 Deposit into experience pool;

[0114] Step 103, randomly obtain N experiences from the experience pool;

[0115] Input sequential element group s t+1 Go to the TargetActor network to get the visualization strategy a t+1 ;

[0116] Step 104, calculating the first loss value L1 and the second loss value L2, and updating the Actor network and the Critic network using the gradient descent method;

[0117]

[0118] Where Q Critic (s t ,a t ) represents the input sequential element group s t and visualization strategy a t The output of the Critic network, Q Critic (s t ,a' t ) means to convert a' t and t Input the output of the Critic network, Indicates the execution of visualization strategy a t The comprehensive Q value, Q TargetCritic (s t+1 ,a t+1 ) represents the input sequential element group s t+1 and visualization strategy a t+1 The output of the TargetCritic network;

[0119] γ is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0120] Update the Actor network through L1 and the Critic network through L2;

[0121] Step 105, every fixed number of network updates, update the TargetActor so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network;

[0122] Step 106, the step is terminated when the Actor network and the Critic network converge, or the number of cycles of steps 103-105 reaches a set value, the step is terminated.

[0123] In one embodiment of the present invention, r t=γ1r q -γ2b q

[0124] where r q Represents the user's visualization strategy a t Satisfaction after implementation, b q Indicates execution strategy a t is the per-unit value of the data processing amount when , γ1 and γ2 are the expected target weights, and their default values ​​are 0.6 and 0.2 respectively.

[0125] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for processing digital archive data based on AI, characterized in that: include: Step 100, read the metadata information of the archive; Step 200, randomly initialize the visualization strategy, the visualization strategy includes the archive room modeling type, the storage location of the archive, the archive group type, the brightness mark of the archive, and the preview item type of the archive; Step 300, collecting user behavior data once in each visualization control cycle, the user behavior data includes the user's query records and the user's operation records in the visualization environment; Step 400, construct a sequential element group based on the collected user behavior data, wherein a single-column element grid and a square matrix element grid are stored in an element position of the sequential element group, wherein an element position of the single-column element grid stores association information of an abstract object, and an element position of the square matrix element grid stores a binary number, and the element position of the square matrix element grid can locate two element positions with the same number of rows in the single-column element grid, and the number of rows of the two element positions of the single-column element grid is equal to the number of rows or columns of the element positions of the square matrix element grid. If the binary number stored in the element position of the square matrix element grid is 1, it indicates that there is an abstract connection between the two abstract objects corresponding to the two element positions of the located single-column element grid, otherwise the stored binary number is 0; Step 500, input the sequential element group into the visualization control neural network, output the visualization strategy, and the visualization strategy is executed in the next visualization control cycle.

2. The method for processing digital archive data based on AI according to claim 1, characterized in that: Obtain the archive content feature vectors extracted in advance by the deep learning model. These feature vectors represent the semantic information of the archive text and image content.

3. The method for processing digital archive data based on AI according to claim 1, characterized in that: Archives modeling type refers to how to represent and display the overall structure and layout of archives in the visualization system; The storage location of each file is the spatial location of each file in the archive room modeling; The archive group type refers to the type of rules followed by archives located in the same area; Brightness marking of archives refers to the use of different brightness levels to represent certain attributes or states of archives in visualization; The preview item type of a profile refers to the type of profile preview information displayed when a user selects or hovers over a profile in the visualization.

4. The method for processing digital archive data based on AI according to claim 1, characterized in that: The method of the present invention is applied in a virtual reality environment.

5. The method for processing digital archive data based on AI according to claim 1, characterized in that: Abstract objects include users, files, file groups, and file racks; Cases where there are abstract connections between abstract objects include: The user views the file, the file belongs to a file group, and the file is located on the file shelf.

6. The method for processing digital archive data based on AI according to claim 1, characterized in that: The calculation formula of the first encoder is as follows: where k i Represents the first identification feature of the i-th element of a single-column element grid, φ i represents the aggregation coefficient of the i-th element of a single-column element grid, M i represents the co-located element bit set of the i-th element bit of a single-column element grid, the element bit of the single-column element grid in the co-located element bit set and the i-th element bit of the single-column element grid are located at the same element bit of the square matrix element grid with a stored binary number of 1, σ represents the activation function, W k represents the state weight coefficient, x j Represents the association information stored in the i-th element position of a single-column element grid; Z i =W z x i Z j =W z x j where x i and x j Respectively represent the XXX information of the independent object represented by the i-th and j-th cells of the data matrix, W z represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, and LeakyRelu represents the rectified linear unit function; The calculation formula of the second encoder is as follows: u (t) =σ(W u X (t) +W u H (t-1) +b u ) r (t) =σ(W r X (t) +W r H (t-1) +b r ) C (t) =tanh(W c X (t) +W c r (t) ⊙H 9t-1) +b c H (t) =(1u (t) )⊙C (t) +u (t) ⊙H (t-1) Among them, W u , W r , W c represents the weight parameter, b u , b r , b c represents the bias parameter, ⊙ represents the dot product, u (t) 、r (t) and C (t) Represent the first, second, and third internal coding features respectively, where M all represents the set of all element bits of a single-column grid of elements, The first identification feature of the i-th element position of the single-column element grid output when the content stored in the t-th element position of the sequential element group is input into the first encoder, H (t) and H (t-1) They represent the t-th and t-1-th second identification features, respectively, n≥t≥1, n represents the total number of element positions in the sequential element group, and when t=1, H (t-1) =X t) , tanh is the hyperbolic tangent function, and σ represents the sigmoid function.

7. The method for processing digital archive data based on AI according to claim 6, characterized in that: The calculation formula of the FC layer is as follows: y=σ(W y h n +b y ) Where y represents a one-dimensional array, h n represents the nth second identification feature, W y is the output layer weight parameter, b y is the output layer bias parameter, and σ represents the sigmoid function.

8. The method for processing digital archive data based on AI according to claim 7, characterized in that: The training steps of the visual control neural network are as follows: Step 101, copy the visual control neural network to generate the Actor network and the TargetActor network, use the feedforward neural network as the Critic network, copy the Critic network to obtain the TargetCritic network, and randomly initialize the parameters of the Actor network, the TargetActor network, the Critic network, and the TargetCritic network; Step 102: Input the sequential element group s of the user collected for the tth time t Go to the Actor network to get the visualization strategy a' t ; Visualization strategy a' t Add random noise to obtain visualization strategy a t ; The visualization strategy a t Input into the environment and get immediate reward r for feedback t and the next sequential element group s t+1 ;Experience s t ,a t ,r t ,s t+1 Deposit into experience pool; Step 103, randomly obtain N experiences from the experience pool; Input sequential element group s t+1 Go to the TargetActor network to get the visualization strategy a t+1 ; Step 104, calculating the first loss value L1 and the second loss value L2, and updating the Actor network and the Critic network using the gradient descent method; Where Q Critic (s t ,a t ) represents the input sequential element group s t and visualization strategy a t The output of the Critic network, Q Critic (s t ,a' t ) means to convert a' t and t Input the output of the Critic network, Indicates the execution of visualization strategy a t The comprehensive Q value, Q TargetCritic (s t+1 ,a t+1 ) represents the input sequential element group s t+1 and visualization strategy a t+1 The output of the TargetCritic network; γ is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6; Update the Actor network through L1 and the Critic network through L2; Step 105, every fixed number of network updates, update the TargetActor so that its parameters are the same as those of the current Actor network, and update the TargetCritic network so that its parameters are the same as those of the current Critic network; Step 106, the step is terminated when the Actor network and the Critic network converge, or the number of cycles of steps 103-105 reaches a set value, the step is terminated.

9. A system for processing digital archive data based on AI, characterized in that: It can be run on a computer entity, and when it is run, it can execute an AI-based digital archive data processing method as described in any one of claims 1-8.

10. A computer storage medium, characterized in that: It is used to store computer-readable instructions, which, when read, can execute an AI-based digital archive data processing method as described in any one of claims 1-8.

Images