Archive file information processing method, device and equipment

Abstract

The invention provides an archive file information processing method, device and equipment, and relates to the technical field of data processing. The archive file information processing method comprises the steps of obtaining archive file information data to be processed; performing data classification identification processing on the to-be-processed archive file information data through an archive information data processing model to obtain a target classification result corresponding to the to-be-processed archive file information data, and outputting the target classification result; wherein the archive information data processing model is obtained by training a preset large language model according to a target archive file information data training set, and in the training process, weight parameters of multiple perceptron layers of the preset large language model are adjusted according to output of a self-attention mechanism layer of the preset large language model, and target parameters are obtained. According to the scheme, automatic classification processing of the archive file information data is achieved, and meanwhile the method has the advantages of being high in classification processing speed and high in classification processing accuracy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device and equipment for processing archive file information. Background Art

[0002] In the existing archive opening review and processing work, manual review is generally adopted for each file or simple keyword matching to review and process the files that need to be opened. The files are classified through the review and processing work, and then the files that meet the requirements are opened according to the classification results and the opening requirements. Among them, when manual review is adopted for each file, when the volume of files to be reviewed is huge, individuals need to be meticulous at the content level, which is very time-consuming, labor-intensive, manpower-consuming and material-consuming. When simple keyword matching review is adopted, keyword matching based on sensitive word lists is required, and a complete sensitive word list needs to be accumulated and sorted. The word list is updated slowly and iteratively, the cost is high, the degree of assistance is low, and the matching accuracy is low. There are problems such as misjudgment and poor review and processing effect. Summary of the invention

[0003] The present invention provides a method, device and equipment for processing archive file information, which solves the problems of time-consuming, labor-consuming, man-consuming, low efficiency and low precision of classification processing when conducting open review and classification processing of existing archive data.

[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows: An embodiment of the present invention provides a method for processing archive file information, comprising: Obtaining information data of archive files to be processed; Through the archival information data processing model, the archival file information data to be processed is subjected to data classification and identification processing, and a target classification result corresponding to the archival file information data to be processed is obtained and output; Among them, the archive information data processing model is obtained by training a preset large language model according to a target archive file information data training set. During the training process, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model to obtain the target parameters.

[0005] Optionally, a preset large language model is trained according to the target archive file information data to obtain an archive information data processing model, including: Obtain target archive file information data training set; Inputting the target archive file information data training set into the encoding layer of the preset large language model for feature extraction and encoding processing to obtain the output of the encoding layer; The output of the encoding layer is converted according to the embedding layer of the preset large language model to obtain the output of the embedding layer; Performing position encoding processing on the output of the embedding layer according to the position encoding layer of the preset large language model to obtain the output of the position encoding layer; The output of the position encoding layer is weighted according to the self-attention mechanism layer of the preset large language model to obtain the output of the self-attention mechanism layer; The output of the self-attention mechanism layer is propagated and processed according to the multi-layer perceptron layer of the preset large language model to obtain the output of the multi-layer perceptron layer; According to the output layer of the preset large language model, the output of the multi-layer perceptron layer is converted and processed to obtain the prediction classification result corresponding to the input target archive file information data; Determine the archival information data processing model based on the predicted classification results.

[0006] Optionally, obtain a training set of target archive file information data, including: Acquire multiple archive file information data with target vocabulary; Performing text content structured processing on all archive file information data containing target vocabulary to obtain first processed data; According to a preset target vocabulary classification standard, each first processed data is respectively annotated to obtain second processed data; The second processed data is formatted to obtain a target archive file information data training set.

[0007] Optionally, all archive file information data with target vocabulary are processed by text content structured arrangement to obtain first processed data, including: Extracting the contents of all archive file information data containing the target vocabulary to obtain first extracted data; The first extracted data are all stored in a preset text through a first preset format to obtain first processed data.

[0008] Optionally, formatting the second processed data to obtain a target archive file information data training set includes: Formatting the data in the second processed data into a target array; The target array is traversed to obtain the text corpus segment and the corresponding label of each archive file information data in the second processing data, and the text corpus segment and the corresponding label are formatted into the target object format to obtain the target archive file information data training set.

[0009] Optionally, based on the prediction and classification results, determine the archival information data processing model, including: The predicted classification result is back-propagated, and the loss value of the predicted classification result and the actual classification result is calculated through the loss function. When the loss value is less than a preset value, the current model is output to obtain the archive information data processing model.

[0010] Optionally, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model to obtain the target parameters, including: Construct a new weight matrix based on the output of the self-attention mechanism layer and the preset low-rank matrix; The original weight matrix in the multi-layer perceptron layer in the large language model is replaced with a new weight matrix to obtain target parameters.

[0011] An embodiment of the present invention further provides a device for processing archive file information, comprising: An acquisition module is used to acquire information data of archive files to be processed; The processing module is used to perform data classification and identification processing on the archival file information data to be processed through the archival information data processing model, obtain the target classification result corresponding to the archival file information data to be processed, and output it; wherein the archival information data processing model is obtained by training a preset large language model according to a target archival file information data training set, and during the training process, according to the output of the self-attention mechanism layer of the preset large language model, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted to obtain the target parameters.

[0012] An embodiment of the present invention further provides a computing device, comprising: a processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the computer executes the above method.

[0013] An embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the above method.

[0014] The above solution of the present invention includes at least the following beneficial effects: The method for processing archive file information of the present invention obtains archive file information data to be processed; performs data classification and identification processing on the archive file information data to be processed through an archive information data processing model, obtains a target classification result corresponding to the archive file information data to be processed, and outputs it; wherein the archive information data processing model is obtained by training a preset large language model according to a training set of target archive file information data, and during the training process, according to the output of the self-attention mechanism layer of the preset large language model, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted to obtain the target parameters. Automatic classification processing of archive file information data is realized, and at the same time, it has the advantages of fast classification processing speed and high classification processing accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the method for processing archive file information of the present invention; Figure 2 It is a module block diagram of the device for processing archive file information of the present invention. DETAILED DESCRIPTION

[0016] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0017] like Figure 1 As shown, an embodiment of the present invention provides a method for processing archive file information, comprising: Step 11, obtaining the archive file information data to be processed; Step 12, through the archival information data processing model, the archival file information data to be processed is subjected to data classification and identification processing, and the target classification result corresponding to the archival file information data to be processed is obtained and output; wherein, the archival information data processing model is obtained by training a preset large language model according to the target archival file information data training set, and during the training process, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model to obtain the target parameters.

[0018] In this embodiment, the method for processing archival file information is applied to the scenario of archive open review and classification. By optimizing and training the preset large language model, an archival information data processing model is obtained, and then the archival file information data to be processed is input into the archival information data processing model for automatic identification and classification. This solves the problems of time-consuming, labor-intensive, and man-consuming, as well as low classification processing efficiency and low classification processing accuracy when performing open review and classification processing on existing archival data, and realizes automatic classification processing of archival file information data, while having the advantages of fast classification processing speed and high classification processing accuracy.

[0019] In an optional embodiment of the present invention, a preset large language model is trained according to target archive file information data to obtain an archive information data processing model, including: Step 21, obtaining a training set of target archive file information data; Step 22, inputting the target archive file information data training set into the encoding layer of the preset large language model for feature extraction and encoding processing to obtain the output of the encoding layer; Step 23, converting the output of the encoding layer according to the embedding layer of the preset large language model to obtain the output of the embedding layer; Step 24, performing position encoding processing on the output of the embedding layer according to the position encoding layer of the preset large language model to obtain the output of the position encoding layer; Step 25, weighting the output of the position encoding layer according to the self-attention mechanism layer of the preset large language model to obtain the output of the self-attention mechanism layer; Step 26, performing propagation processing on the output of the self-attention mechanism layer according to the multi-layer perceptron layer of the preset large language model to obtain the output of the multi-layer perceptron layer; Step 27, converting the output of the multi-layer perceptron layer according to the output layer of the preset large language model to obtain the prediction classification result corresponding to the input target archive file information data; Step 28, determining the archival information data processing model based on the predicted classification results.

[0020] In this embodiment, the preset large language model is provided with an encoder in the encoding layer, which is used to process the corpus (the target archive file information data in the target archive file information data training set) through the model's encoder, and convert the text data corresponding to the target archive file information data into a format that the model can understand, that is, token. ID sequence; the embedding layer is used to convert the discrete vocabulary (such as words or subwords) of the encoding layer output data into a high-dimensional continuous vector representation; by mapping the vocabulary to the vector space, the input embedding layer provides a numerical representation for subsequent processing, so that the model can understand and process text data; the position encoding layer adds position information to the output data of the embedding layer to solve the problem that the encoder architecture cannot directly process the sequence order; by adding position encoding to the embedding vector, the model can identify the relative or absolute position of the vocabulary in the sequence, thereby capturing the contextual relationship; the self-attention mechanism layer performs weighted processing on the output data of the position encoding layer, so that the model can pay attention to different parts of the input sequence; by calculating the relationship between each vocabulary and other vocabulary, the self-attention mechanism generates a weighted sum representation to capture contextual information, thereby improving the model's ability to understand the context; the multi-layer perceptron layer further processes the output data of the self-attention mechanism layer, which is usually used for nonlinear transformation and feature combination; The perceptron layer is composed of multiple fully connected layers, usually including activation functions (such as ReLU) and random dropout layers to enhance the expressiveness of the model. This layer is responsible for converting the output of the self-attention mechanism into a more complex feature representation; the output layer is used to convert the output data of the multi-layer perceptron layer into the format required for the target task, such as generating text or classification; depending on the task, the output layer may be a softmax layer (for classification tasks) or a linear layer (for generation tasks), which is responsible for mapping the final output of the model to the actual vocabulary or label space; in a preferred embodiment, an enhancement module is provided in the preset large language model, and the enhancement module is a generalized linear model module for enhancing the generation ability and efficiency of the model and improving the performance of the model on specific tasks. The generalized linear model module combines multiple mechanisms (such as self-attention, feedforward network, etc.) to optimize information flow and feature learning, so that the model can more effectively utilize context information in generation tasks.

[0021] In an optional embodiment of the present invention, step 28 may include: The predicted classification result is back-propagated, and the loss value of the predicted classification result and the actual classification result is calculated through the loss function. When the loss value is less than a preset value, the current model is output to obtain the archive information data processing model.

[0022] In this embodiment, step 28 can specifically be that in the processing of each layer, the model calculates the loss value through the loss function, and updates the parameters of the model through the back propagation algorithm. The model sets the number of iterations through the step parameter of each training cycle. In each iteration, the model forward propagates a batch of data, calculates the loss, and then performs back propagation to update the parameters; during the training process, the learning rate is dynamically adjusted by using a cosine annealing strategy or a linear warm-up strategy. At the same time, the save cycle parameters are configured so that the model regularly saves checkpoints so that training can be resumed when needed or used for model evaluation.

[0023] In an optional embodiment of the present invention, step 21 may include: Step 211, obtaining a plurality of archive file information data containing target vocabulary; Step 212, performing text content structured processing on all archive file information data containing the target vocabulary to obtain first processed data; Step 213, according to the preset target vocabulary classification standard, each first processed data is respectively annotated to obtain second processed data; Step 214, formatting the second processed data to obtain a target archive file information data training set.

[0024] In this embodiment, the target vocabulary refers to vocabulary containing sensitive information, such as the leakage of certain information affects the stability of certain information; the preset target vocabulary classification standard can be set according to needs, for example, all archive file information data containing the vocabulary "the leakage of certain information" are classified into the category of "affecting the stability of certain information"; in this embodiment, the second processed data includes the text content of the first processed data and each content has a corresponding classification label.

[0025] In an optional embodiment of the present invention, step 212 may include: Step 2121, extracting the contents of all archive file information data containing the target vocabulary to obtain first extracted data; Step 2122: Store the first extracted data in a preset text using a first preset format to obtain first processed data.

[0026] In this embodiment, step 212 may specifically be to use a preset toolkit to extract the content of the archive file information data with the target vocabulary; the specific extraction code may refer to the following: from asposewords import Documentdoc=Document("document.docx") doc=Document("document.docx") text="" for paragraph in doc.get_child_nodes(doc.is_paragraph,True): text+=paragraph.get_text() document.docx is the file name, and the output text parameter is the content to be extracted.

[0027] In this embodiment, the first preset format is the {""} format, that is, the content text of each archive file information data is wrapped using the {""} format, and then stored in the preset text to obtain the first processed data. The format of the first processed data can be a txt file, for example, a txt file storing the text {"Someone leaked the information under certain circumstances for certain reasons"}.

[0028] In an optional embodiment of the present invention, step 214 may include: Step 2141, formatting the data in the second processed data into a target array; Step 2142, traversing the target array, obtaining the text corpus segment and the corresponding label of each archive file information data in the second processed data, and formatting the text corpus segment and the corresponding label into the target object format to obtain the target archive file information data training set; In this embodiment, step 213 specifically formats the data in the second processed data into a json array; traverses the json array to obtain the text corpus fragments and corresponding labels of each target archive file information data in the second processed data, and formats the text corpus fragments and corresponding labels into a json object format to obtain the target archive file information data training set; specifically, the text and label data in the txt file are formatted according to the json standard, wherein the archive text content is set to its key as input, and the label key is set to output, and in addition, a key instruction is added, and the corresponding value is "You are now a content reviewer, identify and extract sensitive content in the following text, and output it in JSON format". For example, the corpus (text content) is {"someone leaked the secret under certain circumstances for certain reasons"; "affecting the stable archives of someone"} The final output is {"instruction": "You are now a content reviewer, identify and extract sensitive content in the following text, and output it in JSON format", "input": "someone leaked the secret under certain circumstances for certain reasons", "output": "affecting the stable archives of someone"}.

[0029] In an optional embodiment of the present invention, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model to obtain the target parameters, including: Construct a new weight matrix based on the output of the self-attention mechanism layer and the preset low-rank matrix; The original weight matrix in the multi-layer perceptron layer in the large language model is replaced with a new weight matrix to obtain target parameters.

[0030] In this embodiment, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model, and the target parameters are obtained as follows: assuming that the original weight matrix in the multi-layer perceptron layer in the preset large language model is W, and the preset low-rank matrices are A and B, the new weight matrix is ​​constructed as W′=W+A⋅B; wherein the dimensions of A and B are selected as (d, r) and (r, d), wherein r is the rank value of the low rank, and d is the dimension of the original weight matrix. The preset large language model obtained after adjustment only updates the parameters of matrices A and B in the subsequent training process, while keeping W unchanged; this design can effectively reduce the number of parameters that need to be trained by introducing low-rank matrices in the self-attention layer and the multi-layer perceptron layer, thereby improving the training efficiency of the model and maintaining the performance of the model. This method is particularly suitable for adjusting models under resource-constrained conditions to meet the needs of specific tasks. During model training, only the parameters of the additional parallel low-rank matrix are trained. When the "rank value" is much smaller than the original parameter dimension, the number of newly added low-rank matrix parameters is also very small. When tuning downstream tasks, only very small parameters need to be trained, but better performance results can be obtained.

[0031] The specific process of obtaining the archival information data processing model in the archival file information processing method of the present invention is as follows: first, the target archival file information data training set is input into the encoding layer of the preset large language model for corpus encoding, and the corpus (the content of the target archival file information data in the target archival file information data training set) is processed by the encoder of the model to convert the text data into a format that the model can understand, that is, a token ID sequence; the encoded token ID sequence is input into the embedding layer of the model, where the token The ID sequence is converted into a vector representation in a high-dimensional space; then, a position encoding layer is applied to retain the order information of the words in the sequence, which is crucial for understanding the context; the data is then transmitted to the self-attention layer, and the input sequence is mapped to three matrices Q, K, and V through formula calculation; the matrix calculation formula is: Q=XW_Q, K=XW_K, V=XW_V; the X matrix is ​​the matrix of the encoded corpus and the positional encoding of the corpus; W_Q, W_K, and W_V are weight matrices; the attention score is calculated using Q and K, and then normalized by the softmax function to obtain the attention distribution; the normalized value is then multiplied by V to obtain the weighted sum, which represents the importance of each element in the sequence; after obtaining the weighted sum, the data enters the MLP layer, and at the same time, the low-rank matrix is ​​introduced at this layer to adjust The parameters of the entire model; specifically, the update of the weight matrix is ​​expressed as the product of two smaller matrices, that is, ∆W=BA, where B and A are low-rank matrices, and their product is approximately the update of the weight matrix; when backpropagation is performed to update the parameters, only the low-rank matrices B and A will be updated, and the original W_0 weight matrix remains unchanged; at this time, the calculation of the linear layer becomes (W+∆W)x=W_X+(A*B_X), this low-rank matrix update is not only concise, but also can amplify the specific adaptability to new tasks while retaining the original capabilities; finally, the predicted label according to the problem is obtained, and the loss of the predicted label and the actual label of the corpus is calculated, and then backpropagation is performed to update the parameters, and the above steps are repeated until the loss value is less than the preset value, or the number of training cycles is greater than the preset value, at which time the current model is output as the archival information data processing model.

[0032] The method for processing archive file information described in the present invention supports inputting archive file content into an archive information data processing model, and relies on the semantic understanding and text generation capabilities of the archive information data processing model to output structured judgments on sensitive archive information; compared with traditional methods that rely on simple word matching and regular rules, it can greatly improve the understanding of the overall context of the archive and reduce the complexity of the system.

[0033] like Figure 2 As shown, an embodiment of the present invention further provides an archive file information processing device 30, comprising: The acquisition module 31 is used to acquire the information data of the archive file to be processed; The processing module 32 is used to perform data classification and identification processing on the archival file information data to be processed through the archival information data processing model, obtain the target classification result corresponding to the archival file information data to be processed, and output it; wherein the archival information data processing model is obtained by training a preset large language model according to the target archival file information data training set. During the training process, according to the output of the self-attention mechanism layer of the preset large language model, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted to obtain the target parameters.

[0034] Optionally, a preset large language model is trained according to the target archive file information data to obtain an archive information data processing model, including: Obtain target archive file information data training set; Inputting the target archive file information data training set into the encoding layer of the preset large language model for feature extraction and encoding processing to obtain the output of the encoding layer; The output of the encoding layer is converted according to the embedding layer of the preset large language model to obtain the output of the embedding layer; Performing position encoding processing on the output of the embedding layer according to the position encoding layer of the preset large language model to obtain the output of the position encoding layer; The output of the position encoding layer is weighted according to the self-attention mechanism layer of the preset large language model to obtain the output of the self-attention mechanism layer; The output of the self-attention mechanism layer is propagated and processed according to the multi-layer perceptron layer of the preset large language model to obtain the output of the multi-layer perceptron layer; According to the output layer of the preset large language model, the output of the multi-layer perceptron layer is converted and processed to obtain the prediction classification result corresponding to the input target archive file information data; Determine the archival information data processing model based on the predicted classification results.

[0035] Optionally, obtain a training set of target archive file information data, including: Acquire multiple archive file information data with target vocabulary; Performing text content structured processing on all archive file information data containing target vocabulary to obtain first processed data; According to a preset target vocabulary classification standard, each first processed data is respectively annotated to obtain second processed data; The second processed data is formatted to obtain a target archive file information data training set.

[0036] Optionally, all archive file information data with target vocabulary are processed by text content structured arrangement to obtain first processed data, including: Extracting the contents of all archive file information data containing the target vocabulary to obtain first extracted data; The first extracted data are all stored in a preset text through a first preset format to obtain first processed data.

[0037] Optionally, formatting the second processed data to obtain a target archive file information data training set includes: Formatting the data in the second processed data into a target array; The target array is traversed to obtain the text corpus segment and the corresponding label of each archive file information data in the second processing data, and the text corpus segment and the corresponding label are formatted into the target object format to obtain the target archive file information data training set.

[0038] Optionally, based on the prediction and classification results, determine the archival information data processing model, including: The predicted classification result is back-propagated, and the loss value of the predicted classification result and the actual classification result is calculated through the loss function. When the loss value is less than a preset value, the current model is output to obtain the archive information data processing model.

[0039] Optionally, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model to obtain the target parameters, including: Construct a new weight matrix based on the output of the self-attention mechanism layer and the preset low-rank matrix; The original weight matrix in the multi-layer perceptron layer in the large language model is replaced with a new weight matrix to obtain target parameters.

[0040] It should be noted that the device is a device corresponding to the above method, and all implementation methods in the above method are applicable to the embodiments of the device and can achieve the same technical effects.

[0041] The embodiment of the present invention further provides a computing device, comprising: a processor and a memory storing a computer program, wherein the computer program executes the above method when executed by the processor. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0042] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the above method. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0043] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0044] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0045] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0046] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0047] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0048] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

[0049] In addition, it should be pointed out that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understandable that all or any steps or components of the method and apparatus of the present invention can be implemented in hardware, firmware, software or a combination thereof in any computing device (including processors, storage media, etc.) or a network of computing devices, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0050] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. That is to say, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0051] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for processing archive file information, characterized in that: include: Obtaining information data of archive files to be processed; Through the archival information data processing model, the archival file information data to be processed is subjected to data classification and identification processing, and a target classification result corresponding to the archival file information data to be processed is obtained and output; Among them, the archive information data processing model is obtained by training a preset large language model according to a target archive file information data training set. During the training process, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted according to the output of the self-attention mechanism layer of the preset large language model to obtain the target parameters.

2. The method for processing archive file information according to claim 1, characterized in that: The preset large language model is trained according to the target archive file information data to obtain an archive information data processing model, including: Obtain target archive file information data training set; Inputting the target archive file information data training set into the encoding layer of the preset large language model for feature extraction and encoding processing to obtain the output of the encoding layer; The output of the encoding layer is converted according to the embedding layer of the preset large language model to obtain the output of the embedding layer; Performing position encoding processing on the output of the embedding layer according to the position encoding layer of the preset large language model to obtain the output of the position encoding layer; The output of the position encoding layer is weighted according to the self-attention mechanism layer of the preset large language model to obtain the output of the self-attention mechanism layer; The output of the self-attention mechanism layer is propagated and processed according to the multi-layer perceptron layer of the preset large language model to obtain the output of the multi-layer perceptron layer; According to the output layer of the preset large language model, the output of the multi-layer perceptron layer is converted and processed to obtain the prediction classification result corresponding to the input target archive file information data; Determine the archival information data processing model based on the predicted classification results.

3. The method for processing archive file information according to claim 2, characterized in that: Obtain the target archive file information data training set, including: Acquire multiple archive file information data with target vocabulary; Performing text content structured processing on all archive file information data containing target vocabulary to obtain first processed data; According to a preset target vocabulary classification standard, each first processed data is respectively annotated to obtain second processed data; The second processed data is formatted to obtain a target archive file information data training set.

4. The method for processing archive file information according to claim 3, characterized in that: All archive file information data with target vocabulary are processed by text content structured arrangement to obtain first processed data, including: Extracting the contents of all archive file information data containing the target vocabulary to obtain first extracted data; The first extracted data are all stored in a preset text through a first preset format to obtain first processed data.

5. The method for processing archive file information according to claim 3, characterized in that: The second processed data is formatted to obtain a target archive file information data training set, including: Formatting the data in the second processed data into a target array; The target array is traversed to obtain the text corpus segment and the corresponding label of each archive file information data in the second processing data, and the text corpus segment and the corresponding label are formatted into the target object format to obtain the target archive file information data training set.

6. The method for processing archive file information according to claim 2, characterized in that: According to the prediction and classification results, the archival information data processing model is determined, including: The predicted classification result is back-propagated, and the loss value of the predicted classification result and the actual classification result is calculated through the loss function. When the loss value is less than a preset value, the current model is output to obtain the archive information data processing model.

7. The method for processing archive file information according to claim 1, characterized in that: According to the output of the self-attention mechanism layer of the preset large language model, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted to obtain the target parameters, including: Construct a new weight matrix based on the output of the self-attention mechanism layer and the preset low-rank matrix; The original weight matrix in the multi-layer perceptron layer in the large language model is replaced with a new weight matrix to obtain target parameters.

8. A device for processing archive file information, characterized in that: include: An acquisition module is used to acquire information data of archive files to be processed; The processing module is used to perform data classification and identification processing on the archival file information data to be processed through the archival information data processing model, obtain the target classification result corresponding to the archival file information data to be processed, and output it; wherein the archival information data processing model is obtained by training a preset large language model according to a target archival file information data training set, and during the training process, according to the output of the self-attention mechanism layer of the preset large language model, the weight parameters of the multi-layer perceptron layer of the preset large language model are adjusted to obtain the target parameters.

9. A computing device, characterized in that include: A processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is performed.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 7.

Images