Document processing method, document processing apparatus, device, and storage medium

Abstract

Embodiments of the present application provide a document processing method and device based on a target deep learning model, an equipment and a storage medium. The method comprises: obtaining the target deep learning model and a document to be classified, wherein the target deep learning model is composed of a target LSTM model and a target MLP model; performing a feature extraction operation on the document to obtain a corresponding target feature vector; analyzing the target feature vector through the target LSTM model to obtain a subject word of the document, and analyzing the target feature vector through the target MLP model to obtain a classification result of the document. Embodiments of the present application aim to simultaneously obtain the subject word and the classification result of the document through the target deep learning model, thereby improving the processing efficiency of the document. Especially for the electronic medical records or clinical experiment documents in the medical field, the processing efficiency of the electronic medical records or clinical experiment documents can be effectively improved and the labor cost can be reduced.

Description

Technical Field

[0001] This application relates to the field of intelligent healthcare, and more particularly to a document processing method, document processing apparatus, computer device, and computer-readable storage medium for a target deep learning model. Background Technology

[0002] Electronic medical records (EMRs) are digitized patient medical records stored, managed, transmitted, and reproduced using electronic devices. Before processing hospital EMR documents, it's necessary to identify the type of each EMR document. Clinical trial data is ultimately submitted in document form, and the numerous stages involved in the clinical trial process result in a large number and variety of documents. Therefore, it's essential to classify and process electronic medical records and clinical trial documents.

[0003] Since keywords in a document can represent its category, existing methods typically involve first determining the document's keywords and then classifying the document based on those keywords. Because determining the keywords and classifying the document are done separately, different models need to be trained for each task, which reduces document processing efficiency and increases maintenance costs. Summary of the Invention

[0004] This application provides a document processing method, document processing apparatus, computer device, and computer-readable storage medium based on a target deep learning model. The aim is to simultaneously obtain the subject terms and classification results of a document through the target deep learning model, thereby improving document processing efficiency. Especially for electronic medical records or clinical trial documents in the medical field, it can effectively improve the processing efficiency of electronic medical records or clinical trial documents and reduce labor costs.

[0005] To achieve the above objectives, this application provides a document processing method based on a target deep learning model, the method comprising:

[0006] Obtain the target deep learning model and the document to be classified, wherein the target deep learning model consists of a target LSTM model and a target MLP model;

[0007] Perform feature extraction on the document to obtain the corresponding target feature vector;

[0008] The target feature vector is analyzed using the target LSTM model to obtain the subject words of the document, and the target feature vector is analyzed using the target MLP model to obtain the classification result of the document.

[0009] To achieve the above objectives, this application also provides a document processing apparatus, the document processing apparatus comprising:

[0010] The acquisition module is used to acquire the target learning model and the document to be classified, wherein the target deep learning model consists of a target LSTM model and a target MLP model;

[0011] A feature extraction module is used to perform feature extraction operations on the document to obtain the corresponding target feature vector;

[0012] The document processing module is used to analyze the target feature vector using the target LSTM model to obtain the subject words of the document, and to analyze the target feature vector using the target MLP model to obtain the classification result of the document.

[0013] In addition, to achieve the above objectives, this application also provides a computer device, the computer device including a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the steps of the document processing method based on the target deep learning model provided in any of the embodiments of this application.

[0014] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the steps of the document processing method based on a target deep learning model provided in any of the embodiments of this application.

[0015] The document processing method, document processing apparatus, computer equipment, and computer-readable storage medium based on a target deep learning model disclosed in this application can analyze electronic medical records or clinical trial documents awaiting classification using a target deep learning model. The target deep learning model includes a target LSTM model and a target MLP model. After feature extraction from the electronic medical records or clinical trial documents awaiting classification, the target LSTM model can be used to analyze the target feature vectors to obtain the subject terms of the electronic medical records or clinical trial documents, and the target MLP model can be used to analyze the target feature vectors to obtain the classification results of the electronic medical records or clinical trial documents. Thus, a single deep learning model can simultaneously obtain the subject terms and classification results of electronic medical records or clinical trial documents, improving the processing efficiency of electronic medical records or clinical trial documents. Furthermore, because the target deep learning model utilizes the overlapping influence of subject terms and document classification, the classification results obtained based on the target deep learning model are more accurate. Especially for electronic medical records or clinical trial documents in the medical field, this can effectively improve the processing efficiency of electronic medical records or clinical trial documents and reduce labor costs. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating a document processing method based on a target deep learning model provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram illustrating the steps of a document processing method based on a target deep learning model provided in an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of the process for obtaining a target deep learning model according to an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of the process for determining the target loss value provided in an embodiment of this application;

[0021] Figure 5 This is a schematic block diagram of a document processing apparatus provided in one embodiment of this application;

[0022] Figure 6 This is a schematic block diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation. Furthermore, although functional modules are divided in the device diagram, in some cases, a different module division may be used.

[0025] The term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0026] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0027] like Figure 1 As shown, the document processing method based on a target deep learning model provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown includes a terminal device 110 and a server 120, whereby the terminal device 110 can communicate with the server 120 via a network. Specifically, the server 120 obtains the target deep learning model and the document to be classified, wherein the target deep learning model consists of a target LSTM model and a target MLP model; further, the server 120 performs feature extraction on the document to obtain the corresponding target feature vector, analyzes the target feature vector through the target LSTM model to obtain the subject words of the document, and analyzes the target feature vector through the target MLP model to obtain the classification result of the document, which is then sent to the terminal device 110. The server 120 can be a standalone server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. Terminal device 110 may be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be connected directly or indirectly through wired or wireless communication, and this application does not impose any restrictions.

[0028] Please see Figure 2 , Figure 2 This is a schematic diagram illustrating the steps of a document processing method based on a target deep learning model, as provided in an embodiment of this application. This document processing method based on a target deep learning model can be applied to computer devices, thereby simultaneously obtaining the document's subject terms and classification results through the target deep learning model, improving document processing efficiency.

[0029] like Figure 2 As shown, the document processing method based on the target deep learning model includes steps S11 to S13.

[0030] Step S11: Obtain the target deep learning model and the document to be classified.

[0031] The target deep learning model consists of a target LSTM model and a target MLP model.

[0032] The documents to be classified can be medical-related documents such as electronic medical records and clinical trial documents, or documents from other fields. This application does not limit this, and the following explanation will use electronic medical records as an example of documents to be classified.

[0033] It should be noted that the LSTM (Long Short-Term Memory) model is a type of temporal recurrent neural network model, suitable for processing and predicting important events with relatively long intervals and delays in time series. LSTM models have already found various applications in the technology field. For example, systems based on LSTM models can perform tasks such as language learning and translation, robot control, image analysis, document summarization, speech and image recognition, handwriting recognition, chatbot control, disease prediction, click-through rates and stock analysis, and music synthesis. In this embodiment, the topic word extraction of documents to be classified can be achieved based on the LSTM model.

[0034] The MLP (Multi-Layer Perceptron) model is a neural network model that employs forward propagation and backward propagation of errors. During forward propagation, the input signal is processed by the hidden layers and then transmitted to the output layer. If the output layer neurons fail to achieve the desired output, the backward propagation stage begins. The output error is returned to the input layer through the hidden layers in a certain form and "distributed" to the neurons in both the hidden and input layers, thus obtaining the error signals of each layer's neurons as the basis for modifying the weight values ​​of each unit. This process of modifying the weight matrices of each layer through forward and backward propagation is repeated, with the weights continuously modified until the network output error gradually decreases to an acceptable level or the set number of learning iterations is reached, completing the network's learning and training. In the embodiments of this application, electronic medical records can be classified based on the MLP model.

[0035] Step S12: Perform feature extraction on the document to obtain the corresponding target feature vector.

[0036] Feature extraction refers to extracting feature words from electronic medical records and quantifying them to represent textual information.

[0037] It should be noted that this application does not limit the method of document feature extraction. For example, it can be TF-IDF, word frequency method, document frequency method, mutual information, etc. This application uses mutual information as an example for document feature extraction.

[0038] Mutual information, a concept in information theory, represents the relationship between information items and is a numerical representation of the statistical correlation between two random variables. Feature extraction using mutual information theory is based on the assumption that terms that appear frequently in a specific category but less frequently in other categories have a high mutual information with that category. The mutual information between a feature and its category reflects the degree of correlation between the feature and the category, and is a widely used standard for building statistical models of word association. Therefore, for each topic, the greater the mutual information of a feature, the greater its co-occurrence probability with that topic. Thus, when using mutual information as an evaluation criterion for feature extraction, the features with the highest mutual information should be selected.

[0039] In this embodiment of the application, feature extraction operations can be performed on documents such as electronic medical records or clinical trial documents through mutual information to obtain the corresponding feature vectors.

[0040] Step S13: Analyze the target feature vector using the target LSTM model to obtain the document's topic words, and analyze the target feature vector using the target MLP model to obtain the document's classification result.

[0041] Since the target deep learning model includes both a target LSTM model and a target MLP model, the target feature vector can be input into the target deep learning model. The target LSTM model analyzes the target feature vector to obtain the document's topic words, and the target MLP model analyzes the target feature vector to obtain the document's classification results.

[0042] In the embodiments of this application, the documents to be classified can be analyzed using a target deep learning model, wherein the target deep learning model includes a target LSTM model and a target MLP model.

[0043] After feature extraction from the documents to be classified, the target feature vector can be analyzed using a target LSTM model to obtain the document's topic words, and the target MLP model can be used to analyze the target feature vector to obtain the document's classification result. In this way, a single deep learning model can simultaneously obtain the document's topic words and classification results, improving document processing efficiency. Furthermore, because the target deep learning model utilizes the overlapping influence of topic words and document classification, the classification results obtained based on the target deep learning model are more accurate.

[0044] Please continue reading. Figure 3 , Figure 3 This is a schematic diagram illustrating the process of obtaining a target deep learning model according to an embodiment of this application. Figure 3 As shown, the target deep learning model can be obtained through steps S111 to S115.

[0045] Step S111: Obtain training data and an initial deep learning model, and sample the training data to obtain target samples.

[0046] The training data includes multiple texts, and each text corresponds to a category; the initial deep learning models include the initial MLP model and the initial LSTM model.

[0047] It should be noted that the above text may be relevant documents in the medical field, such as electronic medical records or clinical trial documents, and this application does not limit this.

[0048] Furthermore, the training data can be sampled to obtain target samples, namely target electronic medical records or target clinical trial documents. This application uses the target electronic medical record as an example for illustration.

[0049] It should be noted that this application does not limit the number of target electronic medical records, for example, it may include 32 electronic medical records, 64 electronic medical records, etc.

[0050] Step S112: Construct the corresponding set of positive and negative samples based on the target sample.

[0051] Understandably, the set of positive and negative samples includes both positive and negative samples.

[0052] Optionally, positive samples can be constructed as follows: randomly select a document A from the target sample, then randomly select S different words from document A, and replace words at random positions in document A with these S different words, until all S words have been replaced. The replaced document A' is the positive sample of document A.

[0053] Alternatively, negative samples can be constructed by using the BERT model to calculate the similarity between each pair of documents in the target sample, and taking the documents with the highest similarity but belonging to different categories as negative samples.

[0054] In this embodiment, training data and an initial deep learning model can be acquired, and the training data can be sampled to obtain the target electronic medical record. Furthermore, a corresponding set of positive and negative samples can be constructed based on the target electronic medical record. Thus, the initial deep learning model can be trained using the target electronic medical record and the set of positive and negative samples.

[0055] Step S113: Perform feature extraction on the target sample and the set of positive and negative samples to obtain the corresponding first feature vector and second feature vector.

[0056] The first feature vector is the feature vector corresponding to the target electronic medical record; the second feature vector is the feature vector corresponding to the set of positive and negative samples.

[0057] Optionally, before performing feature extraction on the target sample and the set of positive and negative samples to obtain the corresponding first feature vector and second feature vector, the process includes: performing data augmentation on the target sample and the set of positive and negative samples to obtain first augmented data and second augmented data, respectively.

[0058] The first augmented data is the data obtained after augmenting the target sample data; the second augmented data is the data obtained after augmenting the positive and negative sample set data.

[0059] It's important to note that data augmentation, also known as data amplification, essentially involves making limited data generate value equivalent to more data without substantially increasing the amount of data. For example, this can be done by removing stop words and non-Chinese characters from the target electronic medical records and positive / negative sample sets, thus reducing irrelevant features in these data sets.

[0060] Optionally, based on the above embodiments, feature extraction operations are performed on the target sample and the set of positive and negative samples to obtain the corresponding first feature vector and second feature vector, including: performing feature extraction operations on the first enhanced data and the second enhanced data to obtain the corresponding first feature vector and second feature vector.

[0061] Understandably, after obtaining the first and second augmented data, feature extraction operations can be performed based on the first and second augmented data to obtain the first and second feature vectors. In this way, feature extraction operations can be performed on the target electronic medical record and the set of positive and negative samples, extracting relevant features and fundamentally improving the overall performance of the model.

[0062] In this embodiment, feature extraction can be performed on the target electronic medical record and the set of positive and negative samples to obtain corresponding first and second feature vectors. This allows for the training of an initial deep learning model based on the first and second feature vectors. Furthermore, data augmentation can be performed on the target electronic medical record and the set of positive and negative samples to reduce irrelevant features, thereby preventing the extraction of irrelevant features and fundamentally improving the overall performance of the model.

[0063] Step S114: Determine the target loss value of the initial deep learning model based on the first feature vector and the second feature vector.

[0064] Specifically, the loss function of the initial deep learning model can be determined first, and then the first feature vector and the second feature vector can be substituted into the loss function to obtain the target loss value in the initial deep learning model.

[0065] It should be noted that this application does not limit the type of loss function mentioned above. For example, the loss function may include at least one of the following: cross-entropy loss function, NCE (Noise Contrastive Estimation) loss function, and ELOB (Evidence Lower Bound) loss function.

[0066] Please continue reading Figure 4 , Figure 4 This is a schematic diagram of the process for determining the target loss value provided in an embodiment of this application. For example... Figure 4 As shown, the target loss value includes the first loss value, which can be determined through steps S1141 to S1143.

[0067] Step S1141: Perform pooling operations on the first feature vector and the second feature vector respectively to obtain the corresponding third feature vector and fourth feature vector.

[0068] The first loss value is the loss value obtained based on the NCE loss function; the third feature vector is the feature vector obtained by pooling the first feature vector; and the fourth feature vector is the feature vector obtained by pooling the second feature vector.

[0069] It's important to note that pooling mimics the human visual system's dimensionality reduction operation. After building a model, pooling is often used to reduce the feature dimensions of the convolutional layers, effectively reducing network parameters and preventing overfitting. Its main functions include: suppressing noise and reducing information redundancy; improving the model's scale invariance and rotation invariance; reducing computational cost; and preventing overfitting.

[0070] In this embodiment, pooling operations can be performed on the first and second feature vectors respectively to obtain the corresponding third and fourth feature vectors. This allows for improvements in model robustness, reduced computational load, and prevention of overfitting based on the third and fourth feature vectors.

[0071] Step S1142: Based on the third and fourth feature vectors, determine the target similarity between the target sample and the set of positive and negative samples.

[0072] Specifically, the target similarity between the positive and negative sample sets can be determined using the following formula:

[0073]

[0074] Where f(x, y) is the target similarity; g φ(x) The third eigenvector; g φ(y)τ is the fourth eigenvector; τ is the temperature parameter, which is set manually.

[0075] Therefore, the above formula can be used to determine the target similarity between the target sample and the set of positive and negative samples based on the third and fourth feature vectors.

[0076] Step S1143: Determine the first loss value of the initial deep learning model based on the target similarity.

[0077] Furthermore, after determining the target similarity, the first loss value of the initial deep learning model can be determined using the following formula:

[0078]

[0079] Among them, L nce y is the first loss value; X is the target sample; f(x, y) is the target similarity.

[0080] Therefore, the above formula can be used to determine the first loss value of the initial deep learning model based on the similarity of the target.

[0081] Optionally, the target loss value also includes a second loss value. The initial deep learning model includes an initial LSTM model. After performing feature extraction on the target sample to obtain the corresponding first feature vector, it also includes: analyzing the third feature vector through the initial LSTM model to obtain the topic words after training the target sample; and determining the second loss value based on the trained topic words and the manually labeled topic words.

[0082] The second loss value is the loss value obtained based on the ELOB loss function.

[0083] Specifically, the target samples can first be manually labeled to obtain manually labeled topic words. Then, the third feature vector is analyzed using an initial LSTM model to obtain the topic words after training the target samples. Furthermore, the second loss value can be determined based on the trained topic words and the manually labeled topic words using the following formula:

[0084] L ELBO =-E q(z|x) [logp(z|x)]+KL[q(z|x)||p(z)]

[0085] Among them, L ELBO is the second loss value; z is the manually labeled topic words; x is the trained topic words; q(z|x) is the approximate distribution of the manually labeled topic words and the trained topic words; KL is the divergence, manually set; p(z|x) is the conditional probability of the manually labeled topic words and the trained topic words.

[0086] Therefore, the above formula can be used to determine the second loss value based on the trained topic words and the manually labeled topic words.

[0087] Optionally, the target loss value also includes a third loss value. The initial deep learning model includes an initial MLP model. After performing feature extraction on the target sample to obtain the corresponding first feature vector, it also includes: classifying the third feature vector through the initial MLP model to obtain the classification result of the target sample after training; and determining the third loss value based on the classification result after training and the manual classification result.

[0088] The third loss value is the loss value obtained based on the cross-entropy loss function.

[0089] Specifically, the target samples can first be manually classified to obtain the manual classification results. Then, the third feature vector is classified using an initial MLP model to obtain the classification results of the target samples after training. Furthermore, the third loss value can be determined based on the training classification results and the manual classification results using the following formula:

[0090]

[0091] Among them, L crossentropy is the third loss value; x is the classification result after training; y is the manual classification result; p(x|y) is the conditional probability of the classification result after training and the manual classification result; q(x|y) is the approximate distribution of the classification result after training and the manual classification result.

[0092] Therefore, the above formula can be used to determine the third loss value based on the classification results after training and the manual classification results.

[0093] Optionally, based on the above embodiments, determining the target loss value of the initial deep learning model based on the first feature vector and the second feature vector includes: summing the first loss value, the second loss value and the third loss value to obtain the target loss value of the initial deep learning model.

[0094] Specifically, the target loss value of the initial deep learning model can be obtained using the following formula:

[0095] L = L nce +L ELBO +L crossentropy

[0096] Where L is the target loss value; L nce This is the first loss value; L ELBO This is the second loss value; L crossentropy This is the third loss value.

[0097] In this embodiment, a target loss value for the target sample and the set of positive and negative samples can be determined based on a third feature vector and a fourth feature vector for training an initial deep learning model. The target loss value includes at least one of a first loss value, a second loss value, and a third loss value. Furthermore, the first, second, and third loss values ​​can be summed to obtain the target loss value. This allows for the utilization of the overlapping influence of trained keywords and document categories, resulting in a more accurate target loss value, thereby improving the accuracy of the target deep learning model and reducing the cost of subsequent model maintenance.

[0098] Step S115: Train the initial deep learning model using the target loss value to obtain the target deep learning model.

[0099] After obtaining the target loss value, the initial deep learning model can be trained using the target loss value to obtain the target deep learning model.

[0100] Specifically, training can be performed by introducing the target loss value into each intermediate layer of the initial deep learning model, so that each intermediate layer has progressively complex feature representation capabilities after training, and the feature representation capabilities between each layer are consistent. This accelerates training convergence, improves the feature transfer capability of each intermediate layer, achieves an overall improvement in model performance, and thus obtains the target deep learning model.

[0101] In this embodiment, the target deep learning model is based on a target loss function, which utilizes the overlapping effects of trained keywords and document categories. Therefore, the target deep learning model exhibits good training performance and model capabilities. Especially for processing electronic medical records and clinical trial documents in the medical field, it can improve the model's processing efficiency and accuracy.

[0102] Please see Figure 5 , Figure 5 This is a schematic block diagram of a document processing apparatus provided in an embodiment of this application. The document processing apparatus can be configured in a server to execute the aforementioned document processing method based on a target deep learning model.

[0103] like Figure 5 As shown, the document processing device 200 includes: an acquisition module 201, a feature extraction module 202, and a document processing module 203.

[0104] The acquisition module 201 is used to acquire the target learning model and the document to be classified, wherein the target deep learning model consists of a target LSTM model and a target MLP model.

[0105] The feature extraction module 202 is used to perform feature extraction operations on the document to obtain the corresponding target feature vector.

[0106] The document processing module 203 is used to analyze the target feature vector through the target LSTM model to obtain the subject words of the document, and to analyze the target feature vector through the target MLP model to obtain the classification result of the document.

[0107] The acquisition module 201 is further configured to acquire training data and an initial deep learning model, sample the training data to obtain target samples, construct a corresponding set of positive and negative samples based on the target samples, perform feature extraction operations on the target samples and the set of positive and negative samples to obtain corresponding first feature vectors and second feature vectors, determine the target loss value of the initial deep learning model based on the first feature vector and the second feature vector, and train the initial deep learning model using the target loss value to obtain a target deep learning model.

[0108] The acquisition module 201 is further configured to perform pooling operations on the first feature vector and the second feature vector respectively to obtain the corresponding third feature vector and fourth feature vector; determine the target similarity between the target sample and the set of positive and negative samples based on the third feature vector and the fourth feature vector; and determine the first loss value of the initial deep learning model based on the target similarity.

[0109] The acquisition module 201 is further configured to analyze the third feature vector through the initial LSTM model to obtain the topic words after training the target sample; and to determine the second loss value based on the trained topic words and the manually labeled topic words.

[0110] The acquisition module 201 is further configured to classify the third feature vector using the initial MLP model to obtain the classification result of the target sample after training; and to determine the third loss value based on the classification result after training and the manual classification result.

[0111] The acquisition module 201 is further configured to sum the first loss value, the second loss value and the third loss value to obtain the target loss value of the initial deep learning model.

[0112] The acquisition module 201 is further configured to perform data augmentation processing on the target sample and the set of positive and negative samples to obtain first augmented data and second augmented data respectively; and to perform feature extraction operation on the first augmented data and the second augmented data to obtain the corresponding first feature vector and second feature vector.

[0113] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and its modules and units can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0114] The methods and apparatus of this application can be used in a wide variety of general-purpose or special-purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer terminal devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0115] For example, the above-described method and apparatus can be implemented as a computer program, which can be used in, for example... Figure 6 It runs on the computer device shown.

[0116] Please see Figure 6 , Figure 6 This is a schematic diagram of a computer device provided in an embodiment of this application. The computer device may be a server.

[0117] like Figure 6 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include volatile storage media, non-volatile storage media, and internal memory.

[0118] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform steps of any document processing method based on a target deep learning model.

[0119] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0120] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When executed by a processor, the computer program enables the processor to perform steps of any document processing method based on a target deep learning model.

[0121] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that the structure of this computer device is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0122] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0123] In some embodiments, the processor is used to run a computer program stored in a memory to perform the following steps: acquiring the target deep learning model and the document to be classified, wherein the target deep learning model consists of a target LSTM model and a target MLP model; performing feature extraction on the document to obtain the corresponding target feature vector; analyzing the target feature vector through the target LSTM model to obtain the subject words of the document; and analyzing the target feature vector through the target MLP model to obtain the classification result of the document.

[0124] In some implementations, the processor is further configured to construct a corresponding set of positive and negative samples based on the target sample; perform feature extraction operations on the target sample and the set of positive and negative samples to obtain a corresponding first feature vector and a second feature vector; determine the target loss value of the initial deep learning model based on the first feature vector and the second feature vector; and train the initial deep learning model using the target loss value to obtain a target deep learning model.

[0125] In some implementations, the processor is further configured to perform pooling operations on the first feature vector and the second feature vector respectively to obtain corresponding third feature vector and fourth feature vector; determine the target similarity between the target sample and the set of positive and negative samples based on the third feature vector and the fourth feature vector; and determine the first loss value of the initial deep learning model based on the target similarity.

[0126] In some implementations, the processor is further configured to analyze the third feature vector using the initial LSTM model to obtain the topic words after training the target sample; and to determine the second loss value based on the trained topic words and the manually labeled topic words.

[0127] In some implementations, the processor is further configured to classify the third feature vector using the initial MLP model to obtain a classification result of the target sample after training; and to determine the third loss value based on the classification result after training and the manual classification result.

[0128] In some implementations, the processor is further configured to sum the first loss value, the second loss value, and the third loss value to obtain the target loss value of the initial deep learning model.

[0129] In some embodiments, the processor is further configured to perform data augmentation processing on the target sample and the set of positive and negative samples to obtain first augmented data and second augmented data, respectively; and to perform feature extraction operations on the first augmented data and the second augmented data to obtain the corresponding first feature vector and second feature vector.

[0130] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed, implement the steps of any of the document processing methods based on a target deep learning model provided in this application.

[0131] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0132] Furthermore, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc.

[0133] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A document processing method based on a target deep learning model, characterized in that, The method includes: Acquire training data, an initial deep learning model, and documents to be classified; sample the training data to obtain target samples; the initial deep learning model includes an initial LSTM model and an initial MLP model; the documents include electronic medical records and / or clinical trial documents. Construct a corresponding set of positive and negative samples based on the target samples; Perform feature extraction operations on the target sample and the set of positive and negative samples to obtain a first feature vector corresponding to the target sample and a second feature vector corresponding to the set of positive and negative samples. Based on the first feature vector and the second feature vector, the target loss value of the initial deep learning model is determined; the target loss value includes a first loss value, a second loss value, and a third loss value; the first loss value is the loss value corresponding to the initial deep learning model; the second loss value is the loss value corresponding to the initial LSTM model; and the third loss value is the loss value corresponding to the initial MLP model. The initial deep learning model is trained using the target loss value to obtain a target deep learning model; wherein, the target deep learning model consists of a target LSTM model and a target MLP model; Perform feature extraction on the document to obtain the corresponding target feature vector; The target feature vector is analyzed using the target LSTM model to obtain the subject words of the document, and the target feature vector is analyzed using the target MLP model to obtain the classification result of the document.

2. The method according to claim 1, characterized in that, Determining the target loss value of the initial deep learning model based on the first feature vector and the second feature vector includes: Pooling operations are performed on the first feature vector and the second feature vector respectively to obtain the corresponding third feature vector and fourth feature vector; Based on the third feature vector and the fourth feature vector, the target similarity between the target sample and the set of positive and negative samples is determined; The first loss value of the initial deep learning model is determined based on the target similarity.

3. The method according to claim 2, characterized in that, After performing feature extraction on the target sample to obtain the corresponding first feature vector, the method further includes: The third feature vector is analyzed using the initial LSTM model to obtain the topic words of the target sample after training; The second loss value is determined based on the trained topic words and the manually labeled topic words.

4. The method according to claim 3, characterized in that, After performing feature extraction on the target sample to obtain the corresponding first feature vector, the method further includes: The third feature vector is classified using the initial MLP model to obtain the classification result of the target sample after training; The third loss value is determined based on the training classification results and the manual classification results.

5. The method according to claim 4, characterized in that, Determining the target loss value of the initial deep learning model based on the first feature vector and the second feature vector includes: The first loss value, the second loss value, and the third loss value are summed to obtain the target loss value of the initial deep learning model.

6. The method according to claim 1, characterized in that, Before performing feature extraction on the target sample and the set of positive and negative samples to obtain the corresponding first feature vector and second feature vector, the following steps are included: Data augmentation processing is performed on the target sample and the set of positive and negative samples to obtain first augmented data and second augmented data, respectively. The step of performing feature extraction on the target sample and the set of positive and negative samples to obtain the corresponding first feature vector and second feature vector includes: Feature extraction is performed on the first enhanced data and the second enhanced data to obtain the corresponding first feature vector and second feature vector.

7. A document processing apparatus, characterized in that, The document processing device includes: The module acquires training data, an initial deep learning model, and documents to be classified. It samples the training data to obtain target samples, where the documents include electronic medical records and / or clinical trial documents. The initial deep learning model includes an initial LSTM model and an initial MLP model. A corresponding set of positive and negative samples is constructed based on the target samples. Feature extraction is performed on the target samples and the set of positive and negative samples to obtain a first feature vector corresponding to the target samples and a second feature vector corresponding to the set of positive and negative samples. Based on the first and second feature vectors, a target loss value for the initial deep learning model is determined. The target loss value includes a first loss value, a second loss value, and a third loss value. The first loss value corresponds to the loss value of the initial deep learning model. The second loss value corresponds to the loss value of the initial LSTM model. The third loss value corresponds to the loss value of the initial MLP model. The initial deep learning model is trained using the target loss value to obtain a target deep learning model. The target deep learning model consists of a target LSTM model and a target MLP model. A feature extraction module is used to perform feature extraction operations on the document to obtain the corresponding target feature vector; The document processing module is used to analyze the target feature vector using the target LSTM model to obtain the subject words of the document, and to analyze the target feature vector using the target MLP model to obtain the classification result of the document.

8. A computer device, characterized in that, include: A memory and a processor; wherein the memory is connected to the processor and is used to store a program, and the processor is used to implement the steps of the document processing method based on the target deep learning model as described in any one of claims 1-6 by running the program stored in the memory.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the steps of the document processing method based on a target deep learning model as described in any one of claims 1-6.

Images