Information classification method and device, computer device and storage medium

Abstract

The application relates to an information classification method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining to-be-classified information, inputting the to-be-classified information into a preset classification network for classification, and obtaining a classification result. The preset classification network is obtained by training an initial classification network. The initial classification network comprises an initial feature extraction subnetwork, an initial classification subnetwork and an initial sequence generation subnetwork. The initial classification subnetwork is used for title information classification of input features, and the initial sequence generation subnetwork is used for field information classification of input features. In the method, the initial feature extraction subnetwork in the initial classification network continuously learns and extracts features of titles and fields in information in a training process, has the ability of learning multi-level text information, and can greatly improve the classification accuracy.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to an information classification method, apparatus, computer equipment, and storage medium. Background Technology

[0002] With the development of artificial intelligence technology, mobile banking apps have added news functions, generating various news information every day. In order to better manage the news information, it is necessary to classify the news information according to its content.

[0003] Currently, the main method for classifying news information is to categorize the headlines within the news articles to obtain classification results. However, for some complex news information, the existing classification methods suffer from inaccurate classification. Summary of the Invention

[0004] Therefore, it is necessary to provide an information classification method, apparatus, computer equipment, and storage medium that can improve the accuracy of news information classification in response to the above-mentioned technical problems.

[0005] Firstly, this application provides an information classification method, which includes:

[0006] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0007] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0008] In one embodiment, the preset classification network includes a feature extraction subnetwork and a classification subnetwork. Information to be classified is input into the preset classification network for classification to obtain a classification result, including:

[0009] The information to be classified is input into the feature extraction sub-network for feature extraction, and the target features of the information to be classified are obtained. The target features include the title features and field features of the information to be classified.

[0010] The target features of the information to be classified are input into the classification sub-network for classification, and the classification result is the title of the information to be classified.

[0011] In one embodiment, the information classification method further includes: training an initial classification network to obtain a preset classification network.

[0012] In one embodiment, the method for training the initial classification network includes:

[0013] Obtain the first annotation information and the second annotation information; the first annotation information is the information after the sample information has been annotated with the title; the second annotation information is the information after the sample information has been annotated with the fields.

[0014] The initial classification network is trained based on the sample information, the first annotation information, and the second annotation information to obtain the preset classification network.

[0015] In one embodiment, an initial classification network is trained based on sample information, first annotation information, and second annotation information to obtain a preset classification network, including:

[0016] The sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features and obtain the features corresponding to the sample information.

[0017] The features corresponding to the sample information are respectively input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network to obtain the first recognition result and the second recognition result; the first recognition result is the title of the sample information, and the second recognition result is the field of the sample information.

[0018] Based on the first recognition result, the second recognition result, the first annotation information, and the second annotation information, the initial classification network is trained to obtain the preset classification network.

[0019] In one embodiment, an initial classification network is trained based on a first identification result, a second identification result, a first annotation information, and a second annotation information to obtain a preset classification network, including:

[0020] Based on the first identification result and the first annotation information, the first loss is determined;

[0021] Based on the second identification result and the second annotation information, determine the second loss;

[0022] Determine the target loss based on the first loss and the second loss;

[0023] The initial classification network is trained based on the target loss to obtain the preset classification network.

[0024] In one embodiment, an initial classification network is trained to obtain a preset classification network, including:

[0025] The initial classification network is trained to obtain a trained initial classification network;

[0026] Based on the initial feature extraction subnetwork and initial classification subnetwork in the trained initial classification network, a preset classification network is constructed.

[0027] Secondly, this application also provides an information classification device, which includes:

[0028] The acquisition module is used to acquire information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0029] The classification module is used to input the information to be classified into a preset classification network for classification and obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into field information.

[0030] Thirdly, this application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0031] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0032] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0033] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0034] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0035] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0036] Fifthly, this application also provides a computer program product, which includes a computer program that, when executed by a processor, performs the following steps:

[0037] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0038] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0039] The aforementioned information classification method, apparatus, computer equipment, and storage medium acquire information to be classified and input it into a preset classification network for classification, thereby obtaining a classification result. The preset classification network is trained based on an initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify title information based on the input features, and the initial sequence generation subnetwork is used to classify field information based on the input features. In this method, because the initial classification network includes an initial classification subnetwork for classifying titles and an initial sequence generation subnetwork for classifying fields, and the initial classification subnetwork classifies titles based on the features output by the initial feature extraction subnetwork, and the initial sequence generation subnetwork classifies fields based on the features output by the initial feature extraction subnetwork, the initial feature extraction subnetwork continuously learns to extract features from titles and fields during training, thus acquiring the ability to learn multi-level text information. Therefore, using the trained initial feature extraction subnetwork to build a preset classification network for classification based on multi-level text information can greatly improve classification accuracy. Attached Figure Description

[0040] Figure 1 This is a diagram illustrating the application environment of the information classification method in one embodiment;

[0041] Figure 2 This is a flowchart illustrating an information classification method in one embodiment;

[0042] Figure 3 This is a flowchart illustrating the information classification method in another embodiment;

[0043] Figure 4 This is a flowchart illustrating the information classification method in another embodiment;

[0044] Figure 5 This is a flowchart illustrating the information classification method in another embodiment;

[0045] Figure 6 This is a flowchart illustrating the information classification method in another embodiment;

[0046] Figure 7 This is a flowchart illustrating the information classification method in another embodiment;

[0047] Figure 8 This is a flowchart illustrating the information classification method in another embodiment;

[0048] Figure 9 This is a structural block diagram of an information classification device in one embodiment;

[0049] Figure 10 This is a structural block diagram of the information classification device in another embodiment;

[0050] Figure 11 This is a structural block diagram of the information classification device in another embodiment;

[0051] Figure 12 This is a structural block diagram of the information classification device in another embodiment;

[0052] Figure 13 This is a structural block diagram of the information classification device in another embodiment;

[0053] Figure 14 This is a structural block diagram of an information classification device in another embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0055] With the rapid development of the internet, online resources have become a focus of widespread attention, and news and information are an important form of online resource presentation. Mobile banking apps, as a crucial customer service channel, play a vital role in the digital transformation of banks. To further enhance customer experience, mobile banking apps have incorporated news and information functions, generating a vast amount of news and information daily. Due to the constant influx of news, users struggle to find what they are interested in amidst the massive amount of information. To better manage news and information, it is necessary to categorize it based on its content. Currently, the main method for categorizing news and information is direct classification. However, for some complex news and information, existing classification methods are not accurate. This application provides an information classification method aimed at solving the aforementioned technical problems. The following embodiments will specifically illustrate the information classification method described in this application.

[0056] The information classification method provided in this application embodiment can be applied to, for example, Figure 1The terminal shown can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. The terminal can also be a server or server cluster. The terminal includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the terminal is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an information classification method. The terminal's display screen can be an LCD screen or an e-ink screen. The terminal's input device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the terminal's casing, or an external keyboard, touchpad, or mouse, etc.

[0057] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. A specific terminal may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0058] In one embodiment, such as Figure 2 As shown, an information classification method is provided, which can be applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0059] S201, Obtain information to be classified, which includes at least one of text, image, voice, or video information.

[0060] Optionally, the information to be categorized can be news information or online resource information such as newspapers and periodicals. In this embodiment, the terminal uses technical means to crawl the information to be categorized from news websites, or it can receive the information to be categorized sent by other devices, or it can obtain the information to be categorized from the database of a cloud storage server. For example, a mobile application used for browsing news receives a large amount of news information every day, which can be used as the information to be categorized.

[0061] S202, the information to be classified is input into a preset classification network for classification to obtain the classification result; the preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the title information of the input features, and the initial sequence generation subnetwork is used to classify the field information of the input features.

[0062] The classification result can include the title corresponding to the information to be classified after classification. The preset classification network can be a convolutional neural network or a fully connected neural network. This network is used to classify the input information into titles and output a classification result containing the titles. The initial classification network can be an untrained convolutional neural network or a deep convolutional neural network. The initial feature extraction subnetwork can be an untrained convolutional neural network or a deep convolutional neural network. The initial classification subnetwork can be a convolutional neural network or a deep convolutional neural network. The initial sequence generation subnetwork can be a convolutional neural network or a deep convolutional neural network.

[0063] In this embodiment, when the terminal obtains the information to be classified based on the aforementioned steps, it can input the information to be classified into a preset classification network for classification, thereby obtaining a classification result including the title. It should be noted that the terminal can pre-construct an initial feature extraction sub-network, an initial classification sub-network, and an initial sequence generation sub-network, and construct an initial classification network based on these three networks. Specifically, after the initial feature extraction sub-network extracts features from the input information, the extracted features can be input into the initial classification sub-network for title classification and the initial sequence generation sub-network for field classification. Then, the initial classification network is trained based on the output results of the initial classification sub-network and the initial sequence generation sub-network, thereby obtaining the preset classification network in the aforementioned application.

[0064] In addition, there are three training methods for training the initial classification network. The first method involves using pre-trained networks for both the initial classification subnetwork and the corresponding initial sequence generation subnetwork. In this case, training only requires adjusting the parameters of the initial feature extraction subnetwork until the outputs of the initial classification subnetwork and the initial sequence generation subnetwork meet preset conditions, thus obtaining a trained initial feature extraction subnetwork. This trained initial feature extraction subnetwork is then combined with the pre-trained initial classification subnetwork to construct the preset classification network. The second method involves using a pre-trained initial sequence generation subnetwork. In this case, training requires adjusting the parameters of both the initial feature extraction subnetwork and the initial classification subnetwork until the outputs of the initial classification subnetwork and the initial sequence generation subnetwork meet preset conditions, thus obtaining a trained initial feature extraction subnetwork. The first method involves extracting a sub-network and a trained initial classification sub-network. Then, the trained initial feature extraction sub-network and the trained initial classification sub-network are used together to construct a preset classification network. The second method involves using untrained networks for the initial feature extraction sub-network, initial classification sub-network, and initial sequence generation sub-network. During training, the parameters of the initial feature extraction sub-network, initial classification sub-network, and initial sequence generation sub-network need to be adjusted until their outputs meet preset conditions. This results in trained initial feature extraction sub-networks, trained initial classification sub-networks, and trained initial sequence generation sub-networks. Finally, the trained initial feature extraction sub-network and the trained initial classification sub-network are used together to construct a preset classification network.

[0065] The information classification method described in the above embodiments obtains information to be classified and inputs it into a preset classification network for classification, thereby obtaining a classification result. The preset classification network is trained based on an initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify title information based on the input features, and the initial sequence generation subnetwork is used to classify field information based on the input features. In this method, since the initial classification network includes an initial classification subnetwork for classifying titles and an initial sequence generation subnetwork for classifying fields, and the initial classification subnetwork classifies titles based on the features output by the initial feature extraction subnetwork, and the initial sequence generation subnetwork classifies fields based on the features output by the initial feature extraction subnetwork, the initial feature extraction subnetwork continuously learns to extract features from titles and fields during training, thus acquiring the ability to learn multi-level text information. Therefore, using the trained initial feature extraction subnetwork to build a preset classification network for classification based on multi-level text information can greatly improve the accuracy of classification.

[0066] In one example, an implementation method is also provided that uses a pre-defined classification network to classify information to be classified. This pre-defined classification network includes a feature extraction subnetwork and a classification subnetwork, such as... Figure 3 As shown, step S202 above, "inputting the information to be classified into a preset classification network for classification to obtain the classification result," includes:

[0067] S301, the information to be classified is input into the feature extraction subnetwork for feature extraction to obtain the target features of the information to be classified.

[0068] The target features include title features and field features of the information to be classified. The field features are used to describe at least one of the following information: keywords, summary terms, key sentences, summary statements, etc., corresponding to the information to be classified.

[0069] In this embodiment, the terminal can pre-construct an initial feature extraction sub-network based on a neural network or machine learning algorithm and train it to obtain a trained feature extraction sub-network. Then, it acquires the information to be classified and inputs it into the feature extraction sub-network to extract title features and field features, thereby obtaining target features including title features and field features. The field features may include at least one of the following: keyword features, summary / generalization features, key sentence features, and summary / generalization sentence features.

[0070] S302, the target features of the information to be classified are input into the classification sub-network for classification, and the classification result is the title of the information to be classified.

[0071] In this embodiment, the terminal can pre-construct an initial classification sub-network based on a neural network or machine learning algorithm and train it to obtain a trained classification sub-network. Then, based on the aforementioned steps, the target features of the information to be classified are obtained, and the obtained target features are input into the classification sub-network for title classification, thereby obtaining a classification result including the title.

[0072] The feature extraction subnetwork in the information classification method of the above embodiments can extract the title features and field features of the information to be classified, thus possessing multi-level content extraction capabilities and improving the accuracy of classifying the information to be classified.

[0073] In one embodiment, based on the preset classification network in the above embodiments, a method for training an initial classification network to obtain the preset classification network is also provided, that is, the terminal further performs the step of: training the initial classification network to obtain the preset classification network. Figure 4 As shown, the training method includes:

[0074] S401, obtain the first annotation information and the second annotation information.

[0075] The first annotation information is the information after the title of the sample information is annotated; the second annotation information is the information after the field of the sample information is annotated.

[0076] In this embodiment, the terminal uses technical means to crawl information to be categorized from news websites, can receive a large amount of sample information sent by other devices, and can also obtain a large amount of sample information from the database of a cloud storage server. When the terminal obtains a large amount of sample information, it can annotate the titles of some of the sample information to obtain first annotation information. For example, if the sample information is a piece of news information text, the title of the news information text is annotated, and the corresponding title of the news information text is used as the annotation information. Furthermore, the terminal can also annotate the fields of some of the sample information to obtain second annotation information. For example, if the sample information is a piece of news information text, the field of the news information text (such as keywords or key phrases) is annotated, and the corresponding fields of the news information text (such as keywords or key phrases) are used as annotation information.

[0077] S402, the initial classification network is trained based on the sample information, the first annotation information and the second annotation information to obtain the preset classification network.

[0078] In this embodiment, the terminal can pre-construct an initial classification network based on a neural network or machine learning algorithm. Once the initial classification network is constructed, the previously acquired sample information can be input into it for training to obtain classification results. Then, based on the classification results, the first annotation information, and the second annotation information, a training loss (Loss) is determined. The parameters of the preset initial classification network are adjusted according to this training loss until the training loss reaches a preset training condition. For example, the preset training condition includes the training loss value being less than a preset loss threshold, or the training loss converging. Finally, the initial classification network obtained after adjusting the parameters is the preset classification network used in the aforementioned application.

[0079] The training method described in this application, by labeling sample information with both title annotation and field annotation, enables the initial classification network to learn more levels of information from the sample information during the training process, thereby obtaining accurate classification results.

[0080] Furthermore, a specific training method for an initial classification network is provided, wherein the initial classification network includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. When training using this initial classification network, such as... Figure 5 As shown, the corresponding training methods include:

[0081] S501, the sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features and obtain the features corresponding to the sample information.

[0082] In this embodiment, the terminal can pre-construct an initial classification network with the aforementioned structure based on a neural network or machine learning algorithm. Once the initial classification network is constructed, the previously acquired sample information can be input into the initial feature extraction sub-network for title and field feature extraction, thereby obtaining the features corresponding to the sample information. These features include title features and field features. The field features may include at least one of the following: keyword features, summary / generalization features, key statement features, and summary / generalization statement features.

[0083] S502, the features corresponding to the sample information are input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network respectively to obtain the first recognition result and the second recognition result.

[0084] The first recognition result is the title of the sample information, and the second recognition result is the fields of the sample information.

[0085] In this embodiment of the application, when the terminal obtains the features corresponding to the sample information based on the aforementioned steps, it can further input the features corresponding to the sample information into the initial classification sub-network for title recognition to obtain a first recognition result including the title, and simultaneously input them into the initial sequence generation sub-network for field recognition to obtain a second recognition result including the field, wherein the field may include at least one of keywords, summary terms, key statements, and summary statements.

[0086] S503, based on the first recognition result, the second recognition result, the first annotation information and the second annotation information, train the initial classification network to obtain the preset classification network.

[0087] This application embodiment classifies three training scenarios. The first scenario is where the initial feature extraction subnetwork in the initial classification network is an untrained network, while the corresponding initial classification subnetwork and initial sequence generation subnetwork are both trained networks. In this scenario, the terminal can determine the first loss based on the first recognition result and the first annotation information, and determine the second loss based on the second recognition result and the second annotation information. Then, during the training process, the parameters of the initial feature extraction subnetwork are first adjusted using the first loss until the first loss reaches a preset training condition, such as the first loss value being less than a preset loss threshold or the first loss converging, resulting in a pre-trained feature extraction subnetwork. Then, the pre-trained feature extraction subnetwork is adjusted using the second loss until the second loss reaches a preset training condition, such as the second loss value being less than a preset loss threshold or the second loss converging, resulting in a trained feature extraction subnetwork. Finally, the preset classification network used above is constructed based on the feature extraction subnetwork and the aforementioned pre-trained initial classification subnetwork.

[0088] The second scenario involves an initial feature extraction subnetwork and its corresponding initial classification subnetwork that are both untrained, while the initial sequence generation subnetwork is a trained network. In this scenario, the terminal can determine a first loss based on the first recognition result and the first annotation information, and a second loss based on the second recognition result and the second annotation information. During training, the first loss is used to adjust the parameters of the initial feature extraction subnetwork and the initial classification subnetwork until the first loss meets preset training conditions, such as the first loss value being less than a preset loss threshold or the first loss converging. This yields a pre-trained feature extraction subnetwork and a pre-trained classification subnetwork. Then, the second loss is used to adjust these pre-trained feature extraction subnetwork and the pre-trained classification subnetwork until the second loss meets preset training conditions, such as the second loss value being less than a preset loss threshold or the second loss converging. This yields a trained feature extraction subnetwork and a trained classification subnetwork. Finally, the pre-trained sequence generation subnetwork is constructed based on the trained feature extraction subnetwork and the trained classification subnetwork. The preset classification network used is described above; optionally, during the training process, a first loss is first used to adjust the parameters of the initial feature extraction subnetwork and the initial classification subnetwork until the first loss reaches a preset training condition, such as the value of the first loss being less than a preset loss threshold, or the first loss converging, to obtain the initially trained feature extraction subnetwork and the trained classification subnetwork. Then, a second loss is used to adjust the initially trained feature extraction subnetwork until the second loss reaches a preset training condition, such as the value of the second loss being less than a preset loss threshold, or the second loss converging, to obtain the trained feature extraction subnetwork. Finally, the preset classification network used above is constructed based on the trained feature extraction subnetwork and the trained classification subnetwork.

[0089] The third scenario involves an initial untrained feature extraction network, its corresponding initial classification subnetwork, and its corresponding initial sequence generation subnetwork. In this scenario, the terminal determines a first loss based on the first recognition result and the first annotation information, and a second loss based on the second recognition result and the second annotation information. During training, the first loss is used to adjust the parameters of the initial feature extraction network, the initial classification subnetwork, and the initial sequence generation subnetwork until the first loss meets preset training conditions. These preset conditions include the first loss value being less than a preset loss threshold, or the first loss converging. This results in a pre-trained feature extraction network, a pre-trained classification subnetwork, and a pre-trained sequence generation subnetwork. Then, the second loss is used to adjust these networks until they meet preset training conditions. These pre-trained conditions also include the second loss value being less than a preset loss threshold, or the second loss converging. This results in a trained feature extraction network, a trained classification subnetwork, and a trained sequence generation subnetwork. Finally, the trained feature extraction network is used to train the sequence generation subnetwork. The aforementioned preset classification network is constructed by taking the sub-network and the trained classification sub-network. Optionally, during training, the parameters of the initial feature extraction sub-network, the initial classification sub-network, and the initial sequence generation sub-network are first adjusted using a first loss until the first loss reaches a preset training condition, such as the first loss value being less than a preset loss threshold or the first loss converging. This yields the initially trained feature extraction sub-network, the trained classification sub-network, and the trained sequence generation sub-network. Then, the initially trained feature extraction sub-network is adjusted using a second loss until the second loss reaches a preset training condition, such as the second loss value being less than a preset loss threshold or the second loss converging. This yields the trained feature extraction sub-network. Finally, the aforementioned preset classification network is constructed based on the trained feature extraction sub-network and the trained classification sub-network.

[0090] Furthermore, a specific implementation of step S503 above, "training the initial classification network based on sample information, first annotation information, and second annotation information to obtain a preset classification network," is provided, such as... Figure 6 As shown, this implementation method includes:

[0091] S601, Based on the first identification result and the first annotation information, determine the first loss.

[0092] In this embodiment, the terminal can use a corresponding loss function to substitute the first identification result and the first annotation information to calculate the first loss. Optionally, the terminal can also use other loss calculation methods to obtain the first loss based on the first identification result and the first annotation information.

[0093] S602, determine the second loss based on the second identification result and the second annotation information.

[0094] In this embodiment, the terminal can use a corresponding loss function to substitute the second identification result and the second annotation information to calculate the second loss. Optionally, the terminal can also use other loss calculation methods to obtain the second loss based on the second identification result and the second annotation information.

[0095] S603, determine the target loss based on the first loss and the second loss.

[0096] In this embodiment of the application, after the first loss and the second loss are determined, the target loss can be determined based on the first loss and the second loss. Specifically, the first loss and the second loss can be weighted and summed, or substituted into the corresponding loss function for calculation to obtain the target loss.

[0097] S604, Train the initial classification network based on the target loss to obtain the preset classification network.

[0098] In this application embodiment, three training scenarios are classified. The first scenario is where the initial feature extraction subnetwork in the initial classification network is an untrained network, while the corresponding initial classification subnetwork and initial sequence generation subnetwork are both trained networks. In this scenario, the terminal adjusts the parameters of the initial feature extraction subnetwork according to the target loss until the target loss reaches the preset training conditions. For example, the preset training conditions include the target loss value being less than the preset loss threshold, or the target loss converging, to obtain the trained feature extraction subnetwork. Finally, the preset classification network used above is constructed based on the feature extraction subnetwork and the aforementioned trained initial classification subnetwork.

[0099] The second scenario involves an initial feature extraction subnetwork and its corresponding initial classification subnetwork that are both untrained, while the initial sequence generation subnetwork is a trained network. In this scenario, the terminal adjusts the parameters of the initial feature extraction subnetwork and the initial classification subnetwork based on the target loss until the target loss reaches a preset training condition. For example, the preset training condition may include the target loss being less than a preset loss threshold or the target loss converging. This results in a trained feature extraction subnetwork and a trained classification subnetwork. Finally, the preset classification network used above is constructed based on the trained feature extraction subnetwork and the trained classification subnetwork.

[0100] The third scenario involves an initial feature extraction subnetwork, a corresponding initial classification subnetwork, and a corresponding initial sequence generation subnetwork in the initial classification network that are all untrained. In this scenario, the terminal adjusts the parameters of the initial feature extraction subnetwork, the initial classification subnetwork, and the initial sequence generation subnetwork based on the target loss until the target loss reaches the preset training conditions. For example, the preset training conditions include the target loss value being less than a preset loss threshold, or the target loss converging. This results in a trained feature extraction subnetwork, a trained classification subnetwork, and a trained sequence generation subnetwork. Finally, the preset classification network used above is constructed based on the trained feature extraction subnetwork and the trained classification subnetwork.

[0101] In one example, an implementation for obtaining a predefined classification network is also provided, such as... Figure 7 As shown above Figures 4-6 The phrase "training the initial classification network to obtain a preset classification network" as described in any of the embodiments includes:

[0102] S701, train the initial classification network to obtain the trained initial classification network.

[0103] The embodiments of this application relate to the above. Figures 4-6 The training method described in any of the embodiments is detailed in the foregoing content and will not be repeated here.

[0104] S702, construct a preset classification network based on the initial feature extraction subnetwork and initial classification subnetwork in the trained initial classification network.

[0105] In this embodiment, when the terminal obtains the trained initial classification network, it can extract the trained initial feature extraction network as a feature extraction sub-network and the trained initial classification sub-network as a classification sub-network. Finally, the feature extraction sub-network and the initial classification sub-network are used to construct a preset classification network, enabling this preset classification network to classify the information to be classified by title, obtaining a classification result including the title. Furthermore, the feature extraction network in this preset classification network can extract title features and field features from the information to be classified, further enabling the classification sub-network to analyze based on these title and field features to obtain accurate classification results.

[0106] In summary, based on all the above embodiments, an information classification method is also provided, such as... Figure 8 As shown, the method includes:

[0107] S801, obtain the first annotation information and the second annotation information.

[0108] The first annotation information is the information after the title of the sample information is annotated; the second annotation information is the information after the field of the sample information is annotated.

[0109] S802, the sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features, and the features corresponding to the sample information are obtained. The features include the title features and field features of the sample information.

[0110] S803, the features corresponding to the sample information are input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network respectively to obtain the first recognition result and the second recognition result.

[0111] The first recognition result is the title of the sample information, and the second recognition result is the fields of the sample information.

[0112] S804, determine the first loss based on the first recognition result and the first annotation information; determine the second loss based on the second recognition result and the second annotation information; determine the target loss based on the first loss and the second loss; train the initial classification network based on the target loss to obtain the preset classification network.

[0113] S805, Obtain information to be classified.

[0114] S806, input the information to be classified into the feature extraction subnetwork in the preset classification network for feature extraction, and obtain the target features of the information to be classified. The target features include the title features and field features of the information to be classified.

[0115] S807: Input the target features of the information to be classified into the classification subnetwork in the preset classification network for classification, and obtain the classification result, which is the title of the information to be classified.

[0116] Each of the above steps has been described in the foregoing embodiments. For details, please refer to the foregoing descriptions. They will not be repeated here.

[0117] The information classification method described above obtains information to be classified and inputs it into a preset classification network for classification, thereby obtaining a classification result. The preset classification network is trained based on an initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features as title information, and the initial sequence generation subnetwork is used to decode the input features through a series of processes to classify field information. This information classification method, based on a conventional text classification task (i.e., title classification), constructs another task (i.e., field classification) using existing information during model training. During training, the model learns both conventional text classification and the other task, further improving its text classification performance. Furthermore, the feature extraction subnetwork in this information classification method has the ability to extract and integrate multi-level information from news and information text. Therefore, the network constructed using the trained feature extraction subnetwork has a more powerful feature extraction capability, extracts richer feature information, has higher information utilization, and better accuracy.

[0118] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0119] Based on the same inventive concept, this application also provides an information classification device for implementing the information classification method described above. The solution provided by this device is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more information classification device embodiments provided below can be found in the limitations of the information classification method described above, and will not be repeated here.

[0120] In one embodiment, such as Figure 9 As shown, an information classification device is provided, comprising:

[0121] Module 10 is used to acquire information to be classified.

[0122] The information to be classified includes at least one of text, images, audio, or video information.

[0123] The classification module 11 is used to input the information to be classified into a preset classification network for classification and obtain the classification result. The preset classification network is trained based on the initial classification network. The initial classification network includes an initial feature extraction subnetwork, an initial classification subnetwork and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into field information.

[0124] In one embodiment, the above-mentioned classification module 11, such as Figure 10 As shown, it includes:

[0125] The feature extraction unit 110 is used to input the information to be classified into the feature extraction subnetwork for feature extraction to obtain the target features of the information to be classified; the target features include the title features and field features of the information to be classified.

[0126] Input unit 111 is used to input the features of the information to be classified into the classification sub-network for classification, and obtain the classification result, which is the title of the information to be classified.

[0127] In one embodiment, such as Figure 11 As shown, the above-mentioned information classification device also includes:

[0128] Training module 12 is used to train the initial classification network to obtain the preset classification network.

[0129] In one embodiment, such as Figure 12 As shown, the training module 12 mentioned above includes:

[0130] The acquisition unit 120 is used to acquire first annotation information and second annotation information; the first annotation information is the information after the sample information has been annotated with a title; the second annotation information is the information after the sample information has been annotated with fields.

[0131] The first training unit 121 is used to train the initial classification network based on sample information, first annotation information and second annotation information to obtain the preset classification network.

[0132] In one embodiment, such as Figure 13 As shown, the first training unit 121 includes:

[0133] The first input subunit 1210 is used to input sample information into the initial feature extraction subnetwork in the initial classification network for feature extraction, so as to obtain the features corresponding to the sample information.

[0134] The second input subunit 1211 is used to input the features corresponding to the sample information into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network, respectively, to obtain the first recognition result and the second recognition result; the first recognition result is the title of the sample information, and the second recognition result is the field of the sample information.

[0135] The training subunit 1212 is used to train the initial classification network based on the first recognition result, the second recognition result, the first annotation information, and the second annotation information to obtain the preset classification network.

[0136] In one embodiment, the training subunit 1212 is specifically used to determine a first loss based on a first recognition result and a first annotation information; determine a second loss based on a second recognition result and a second annotation information; determine a target loss based on the first loss and the second loss; and train an initial classification network based on the target loss to obtain a preset classification network.

[0137] In one embodiment, such as Figure 14 As shown, the training module 12 includes:

[0138] The second training unit 122 is used to train the initial classification network to obtain a trained initial classification network;

[0139] The construction unit 123 is used to construct a preset classification network based on the initial feature extraction subnetwork and the initial classification subnetwork in the trained initial classification network.

[0140] Each module in the aforementioned information classification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0141] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0142] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0143] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0144] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0145] The information to be classified is input into the feature extraction subnetwork for feature extraction to obtain the target features of the information to be classified; the target features include the title features and field features of the information to be classified.

[0146] The target features of the information to be classified are input into the classification sub-network for classification to obtain the classification result, which is the title of the information to be classified.

[0147] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0148] The initial classification network is trained to obtain the preset classification network.

[0149] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0150] Obtain first annotation information and second annotation information; the first annotation information is the information after title annotation of the sample information; the second annotation information is the information after field annotation of the sample information;

[0151] The initial classification network is trained based on the sample information, the first annotation information, and the second annotation information to obtain the preset classification network.

[0152] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0153] The sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features, thereby obtaining the features corresponding to the sample information;

[0154] The features corresponding to the sample information are respectively input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network to obtain a first recognition result and a second recognition result; the first recognition result is the title of the sample information, and the second recognition result is the field of the sample information.

[0155] Based on the first recognition result, the second recognition result, the first annotation information, and the second annotation information, the initial classification network is trained to obtain the preset classification network.

[0156] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0157] Based on the first identification result and the first annotation information, the first loss is determined;

[0158] Based on the second identification result and the second annotation information, the second loss is determined;

[0159] Based on the first loss and the second loss, determine the target loss;

[0160] The initial classification network is trained based on the target loss to obtain the preset classification network.

[0161] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0162] The initial classification network is trained to obtain a trained initial classification network;

[0163] The preset classification network is constructed based on the initial feature extraction subnetwork and the initial classification subnetwork in the trained initial classification network.

[0164] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0165] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0166] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0167] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0168] The information to be classified is input into the feature extraction subnetwork for feature extraction to obtain the target features of the information to be classified; the target features include the title features and field features of the information to be classified.

[0169] The target features of the information to be classified are input into the classification sub-network for classification to obtain the classification result, which is the title of the information to be classified.

[0170] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0171] The initial classification network is trained to obtain the preset classification network.

[0172] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0173] Obtain first annotation information and second annotation information; the first annotation information is the information after title annotation of the sample information; the second annotation information is the information after field annotation of the sample information;

[0174] The initial classification network is trained based on the sample information, the first annotation information, and the second annotation information to obtain the preset classification network.

[0175] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0176] The sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features, thereby obtaining the features corresponding to the sample information;

[0177] The features corresponding to the sample information are respectively input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network to obtain a first recognition result and a second recognition result; the first recognition result is the title of the sample information, and the second recognition result is the field of the sample information.

[0178] Based on the first recognition result, the second recognition result, the first annotation information, and the second annotation information, the initial classification network is trained to obtain the preset classification network.

[0179] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0180] Based on the first identification result and the first annotation information, the first loss is determined;

[0181] Based on the second identification result and the second annotation information, the second loss is determined;

[0182] Based on the first loss and the second loss, determine the target loss;

[0183] The initial classification network is trained based on the target loss to obtain the preset classification network.

[0184] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0185] The initial classification network is trained to obtain a trained initial classification network;

[0186] The preset classification network is constructed based on the initial feature extraction subnetwork and the initial classification subnetwork in the trained initial classification network.

[0187] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0188] Obtain the information to be classified; the information to be classified includes at least one of text, images, audio, or video information.

[0189] The information to be classified is input into a preset classification network for classification to obtain the classification result. The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features into titles, and the initial sequence generation subnetwork is used to classify the input features into fields.

[0190] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0191] The information to be classified is input into the feature extraction subnetwork for feature extraction to obtain the target features of the information to be classified; the target features include the title features and field features of the information to be classified.

[0192] The target features of the information to be classified are input into the classification sub-network for classification to obtain the classification result, which is the title of the information to be classified.

[0193] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0194] The initial classification network is trained to obtain the preset classification network.

[0195] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0196] Obtain first annotation information and second annotation information; the first annotation information is the information after title annotation of the sample information; the second annotation information is the information after field annotation of the sample information;

[0197] The initial classification network is trained based on the sample information, the first annotation information, and the second annotation information to obtain the preset classification network.

[0198] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0199] The sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features, thereby obtaining the features corresponding to the sample information;

[0200] The features corresponding to the sample information are respectively input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network to obtain a first recognition result and a second recognition result; the first recognition result is the title of the sample information, and the second recognition result is the field of the sample information.

[0201] Based on the first recognition result, the second recognition result, the first annotation information, and the second annotation information, the initial classification network is trained to obtain the preset classification network.

[0202] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0203] Based on the first identification result and the first annotation information, the first loss is determined;

[0204] Based on the second identification result and the second annotation information, the second loss is determined;

[0205] Based on the first loss and the second loss, determine the target loss;

[0206] The initial classification network is trained based on the target loss to obtain the preset classification network.

[0207] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0208] The initial classification network is trained to obtain a trained initial classification network;

[0209] The preset classification network is constructed based on the initial feature extraction subnetwork and the initial classification subnetwork in the trained initial classification network.

[0210] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0211] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0212] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An information classification method, characterized in that, The information classification method includes: Obtain information to be classified; the information to be classified includes at least one of text, images, audio, or video information. The information to be classified is input into a preset classification network for classification to obtain the classification result; the preset classification network includes a feature extraction subnetwork and a classification subnetwork. The step of inputting the information to be classified into a preset classification network for classification to obtain a classification result includes: inputting the information to be classified into the feature extraction subnetwork for feature extraction to obtain the target features of the information to be classified, and inputting the target features of the information to be classified into the classification subnetwork for classification to obtain a classification result; the target features are the title features and field features of the information to be classified, and the classification result is the title of the information to be classified; The preset classification network is trained based on the initial classification network, which includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The feature extraction subnetwork is the trained initial feature extraction subnetwork, and the classification subnetwork is composed of the trained initial classification subnetwork. The initial classification subnetwork is used to classify the input features into titles, the initial sequence generation subnetwork is used to decode the input features and classify fields, and the initial feature extraction subnetwork is used to extract features of multi-level text information in the information to be classified; the features of the multi-level text information include title features, keyword features, summary and generalization features, key sentence features, and summary and generalization sentence features.

2. The information classification method according to claim 1, characterized in that, The information classification method further includes: training the initial classification network to obtain the preset classification network.

3. The information classification method according to claim 2, characterized in that, The method for training the initial classification network includes: Obtain first annotation information and second annotation information; the first annotation information is the information after title annotation of the sample information; the second annotation information is the information after field annotation of the sample information; The initial classification network is trained based on the sample information, the first annotation information, and the second annotation information to obtain the preset classification network.

4. The information classification method according to claim 3, characterized in that, The step of training the initial classification network based on the sample information, the first annotation information, and the second annotation information to obtain the preset classification network includes: The sample information is input into the initial feature extraction subnetwork in the initial classification network to extract features, thereby obtaining the features corresponding to the sample information; The features corresponding to the sample information are respectively input into the initial classification subnetwork and the initial sequence generation subnetwork in the initial classification network to obtain a first recognition result and a second recognition result; the first recognition result is the title of the sample information, and the second recognition result is the field of the sample information. Based on the first recognition result, the second recognition result, the first annotation information, and the second annotation information, the initial classification network is trained to obtain the preset classification network.

5. The information classification method according to claim 4, characterized in that, The step of training the initial classification network based on the first identification result, the second identification result, the first annotation information, and the second annotation information to obtain the preset classification network includes: Based on the first identification result and the first annotation information, the first loss is determined; Based on the second identification result and the second annotation information, the second loss is determined; Based on the first loss and the second loss, determine the target loss; The initial classification network is trained based on the target loss to obtain the preset classification network.

6. An information classification device for implementing the information classification method as described in claim 1, characterized in that, The information classification device includes: An acquisition module is used to acquire information to be classified; the information to be classified includes at least one of text, images, audio or video information. The classification module is used to input the information to be classified into a preset classification network for classification and obtain the classification result. The preset classification network is trained based on an initial classification network. The initial classification network includes an initial feature extraction subnetwork, an initial classification subnetwork, and an initial sequence generation subnetwork. The initial classification subnetwork is used to classify the input features as titles, and the initial sequence generation subnetwork is used to classify the input features as field information.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the information classification method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the information classification method according to any one of claims 1 to 5.

Images