WO2020238053A1 - Neural grid model-based text data category recognition method and apparatus, nonvolatile readable storage medium, and computer device - Google Patents

Neural grid model-based text data category recognition method and apparatus, nonvolatile readable storage medium, and computer device Download PDF

Info

Publication number
WO2020238053A1
WO2020238053A1 PCT/CN2019/117726 CN2019117726W WO2020238053A1 WO 2020238053 A1 WO2020238053 A1 WO 2020238053A1 CN 2019117726 W CN2019117726 W CN 2019117726W WO 2020238053 A1 WO2020238053 A1 WO 2020238053A1
Authority
WO
WIPO (PCT)
Prior art keywords
text data
category
confusion
neural network
predicted
Prior art date
Application number
PCT/CN2019/117726
Other languages
French (fr)
Chinese (zh)
Inventor
金戈
徐亮
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020238053A1 publication Critical patent/WO2020238053A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • This application relates to the field of information processing technology, and in particular to methods and devices for identifying text data categories based on neural grid models, non-volatile readable storage media, and computer equipment.
  • the classification of text data is involved in many application scenarios, and based on increasingly intelligent application scenarios, the demand for classification of text data is also increasing. Therefore, when facing multi-classification problems, the classification of text data is usually based on neural networks.
  • the disadvantage of the prior art is that the recognition accuracy of neural networks under current multi-classification problems is often affected by more types of text data, that is, when there are more types of text data, the recognition accuracy of neural networks will be somewhat different. Reduce; At the same time, when building multiple classifiers or multi-step classifiers, professional technicians are required to proofread the specific construction process to confirm whether it is accurate, so as to complete the subsequent text data category recognition operation.
  • the classifier based on the existing neural network has low recognition accuracy when classifying, and users cannot get more accurate text data classification and recognition results, which affects the user experience, and the existing neural network requires professional Technicians intervene to confirm the specific construction process of the classifier, and the labor and time costs are high.
  • this application provides methods and devices for identifying text data categories based on neural grid models, non-volatile readable storage media, and computer equipment.
  • the main purpose is to solve current classifiers constructed based on existing neural networks.
  • the recognition accuracy is low, and the existing neural network requires professional technical personnel to intervene to confirm the specific construction process of the classifier, and the labor and time costs are high.
  • a method for identifying text data categories based on a neural grid model including:
  • the text data category to be recognized is obtained according to the text data to be recognized.
  • a text data category recognition device based on a neural grid model comprising:
  • the verification module is configured to use a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
  • An adjustment module configured to adjust the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
  • the prediction module is used to use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized.
  • a non-volatile readable storage medium having computer readable instructions stored thereon, and when the program is executed by a processor, the recognition of the text data category based on the neural grid model is realized. method.
  • a computer device including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor
  • the processor executes the program, the above-mentioned method for recognizing the text data category based on the neural grid model is realized.
  • the neural grid model-based text data category recognition method and device, non-volatile readable storage medium, and computer equipment provided in this application are in progress with the existing classifiers constructed by neural networks.
  • the recognition accuracy is low, and the user cannot obtain a more accurate text data classification and recognition result.
  • the multi-class neural network model is used to obtain the predicted classification result of the text data according to the text data in the verification set.
  • the predicted classification result of the text data and the preset category label corresponding to the text data in the verification set the multi-class neural network model is adjusted to the initial classification prediction model, and the initial classification prediction model is used to obtain the text to be recognized according to the text data to be recognized Data category.
  • the multi-class neural network model is improved through automatic verification, so that the improved initial classification prediction model does not require professional technicians to confirm the specific construction process of the classifier, reducing labor and time costs, and the improved initial classification prediction model can be effective Improve the accuracy of text data recognition.
  • FIG. 1 shows a schematic flowchart of a method for identifying text data categories based on a neural grid model provided by an embodiment of the present application
  • FIG. 2 shows a schematic flowchart of another method for identifying text data categories based on a neural grid model provided by an embodiment of the present application
  • Fig. 3 shows a schematic structural diagram of a text data category recognition device based on a neural grid model provided by an embodiment of the present application.
  • This embodiment provides a method for identifying text data categories based on a neural grid model, which can automatically construct an initial classification prediction model, and can effectively improve the recognition accuracy of text data categories. As shown in FIG. 1, the method includes:
  • the verification set is used to verify the accuracy of the multi-class neural network model for the recognition of text data categories. It is constructed based on network behaviors, and social network behaviors are massive and diverse. For example, social network behaviors generated by writing a blog, browsing a group of photos, clicking an advertisement, purchasing a product, subscribing to a specific news topic, etc. can be used for multi-class recognition to realize category recognition of social network behaviors.
  • text data category recognition text data of various behaviors are obtained from social networks, and multi-dimensional features (for example, features such as titles and keywords) in the text data are extracted, and the multi-dimensional features are corresponding
  • the preset category label ie, the preset category of the text data constructs a verification set based on the extracted multi-dimensional features and the preset category label.
  • the multi-class neural network model is adjusted to an initial classification prediction model.
  • the predicted classification results based on the output of the multi-class neural network model are compared with the preset category labels corresponding to the multi-dimensional features in the verification set. According to the comparison results, the network is performed on the basis of the original network structure of the multi-class neural network model Structural adjustments to make the adjusted multi-class neural network model (ie, the initial classification prediction model) more accurate in identifying text data categories, especially for text data similar to the text data in the constructed verification set When recognizing, its recognition accuracy is more ideal.
  • a new verification set can be rebuilt to facilitate Based on the constructed new verification set, the original network structure of the multi-class neural network model is re-adjusted to make it more suitable for the recognition of new text data categories.
  • Obtain the text data to be recognized preprocess the text data to be recognized, extract the multi-dimensional features in the text data to be recognized, and use the initial classification prediction model to obtain the text data category to be recognized based on the multi-dimensional features of the text data to be recognized .
  • the initial classification prediction model is trained, the text data to be recognized is input into the trained initialization classification prediction model (ie, the preset classification prediction model), and the text data category to be recognized is output so that the user can be based on the prediction
  • the text data classified by the classification prediction model processed by the proposed classification prediction model can accurately obtain information with a high degree of matching with the text data category, and improve the user experience.
  • a multi-class neural network model can be used to obtain the predicted classification result of the text data according to the text data in the verification set, and the predicted classification result of the text data and the prediction corresponding to the text data in the verification set Set the category label, adjust the multi-class neural network model to the initial classification prediction model, and use the initialized classification prediction model to obtain the text data category to be recognized according to the text data to be recognized, and perform classification with the current classifier constructed by the existing neural network When dividing, the recognition accuracy is low, and users cannot get more accurate text data classification and recognition results.
  • the multi-class neural network model can be improved through automatic verification, so that the improved initialization classification prediction model does not require professional Technicians intervene to confirm the specific construction process of the classifier, reduce labor and time costs, and the improved initial classification prediction model can effectively improve the accuracy of text data recognition.
  • Another method for identifying text data categories based on a neural grid model includes: The training set trains and initializes the neural network model to obtain the multi-class neural network model.
  • a training set based on network behaviors
  • obtain new text data of various behaviors from social networks that is, different from the text data used to construct the verification set above
  • extract multi-dimensional features in the text data and compare Multi-dimensional features should be preset with category labels
  • the training set should be constructed based on the extracted multi-dimensional features and preset category labels, so that the obtained multi-dimensional features of the text data in the training set are different from the multi-dimensional features of the text data in the validation set
  • the obtained multi-dimensional features and preset category labels of the text data in the training set are different from the multi-dimensional features and preset category labels of the text data in the validation set.
  • the initialization neural network model can be the initialization convolutional neural network model or the recurrent neural network model.
  • the main structure of the initialization neural network model is multiple Connected multi-layer structure, to realize the operation of multi-dimensional features, take text data category recognition as an example, the initialization neural network model mainly includes: input layer, convolutional layer, pooling layer and fully connected layer, and the initialization of each layer The network parameters are all randomly generated.
  • the network structure of the initialized neural network model can also be based on a fully connected layer, and the specific network structure of the initialized neural network model is not limited here.
  • the initialization neural network model is trained according to the obtained training set.
  • the convolutional layer and pooling layer of the initialized neural network model are collectively referred to as the hidden layer.
  • the hidden layer There can be multiple layers.
  • Python's tensorflow library input the multi-dimensional features in the training set to the input layer of the initialization neural network model.
  • the input layer preprocesses the multi-dimensional features and converts the multi-dimensional features into multi-dimensional vectors.
  • the input layer The multi-dimensional vector is input to the hidden layer for processing, the last hidden layer inputs the processing result into the fully connected layer, and the fully connected layer converts the processing result into a classification result in text form.
  • the classification result is compared with the preset category label. If the classification result is inconsistent with the preset category label, the cross entropy is calculated according to the output classification result and the preset category label, and used as the loss function for training the initialization neural network model .
  • FIG. 2 Another method for identifying text data categories based on the neural grid model is provided, as shown in FIG. 2 , The method includes:
  • the input layer converts the multi-dimensional features in the validation set into multi-dimensional vectors and then outputs them to the hidden layer for processing.
  • the last hidden layer inputs the processing results into fully connected Layer, the fully connected layer converts the processing results into text-form predictive classification results, so as to adjust the multi-class neural network model to the predicted classification results output by the multi-class neural network model and the preset category labels of the corresponding text data in the validation set Initialize the classification prediction model.
  • step 203 may specifically include: if the predicted category in the predicted classification result of the text data is inconsistent with the preset category label of the corresponding text data in the verification set, Then, the predicted category in the predicted classification result is determined as a confusion category.
  • the predicted classification result output by the multi-class neural network model is compared with the preset category labels corresponding to the multi-dimensional features in the verification set. If the predicted category in the predicted classification result is compared with the prediction of the corresponding multi-dimensional feature in the verification set. If the category labels are inconsistent, the predicted category is determined as the confusion category based on the multi-class neural network model.
  • the confusion category includes a plurality of preset category label classifications and the preset category labels contained under each preset category label classification Quantity, according to the determined confusion category to realize the improvement of the multi-class neural network model.
  • the prediction category in the prediction classification result is consistent with the preset category label of the corresponding multi-dimensional feature in the validation set, that is, the prediction category includes only one preset category label classification and the preset category label classification included in the preset category label classification
  • the number of category labels, the predicted category will not be reset, that is, it will not be used as a consideration for improving the multi-class neural network model.
  • step 203 may specifically include:
  • Step 2031 Determine multiple text data categories according to the confusion category.
  • step 2031 may specifically include: performing a T-type matrix operation on the confusion matrix used to characterize the confusion category to obtain a similarity matrix; Perform spectral clustering operations to obtain multiple text data categories.
  • a confusion matrix for representing the multiple confusion categories is constructed according to the multiple confusion categories in the determined predicted classification results, and each confusion category contained in the confusion matrix matrix is
  • the number of category labels under a preset category label classification is converted into the proportion of the number of category labels under each preset category label classification to the total number of category labels in the corresponding confusion category, and the converted confusion matrix is T-shaped Matrix operation, that is, use the calculation formula matrix*matrix.T to obtain a similarity matrix containing multiple preset category label classifications, and perform spectral clustering operations on the obtained similarity matrix containing multiple preset category label classifications to obtain the spectrum
  • the output result of the clustering is the classification of multiple preset categories.
  • the T-type matrix is the Toeplitz matrix.
  • the spectral clustering operation is specifically to classify the preset category labels in the similarity matrix as vertices, and use the similarity between the vertices as the feature vector to construct a vector space and perform segmentation to obtain higher intra-class similarity The degree of similarity between the lower class and the lower class, so as to realize the further classification of multiple confusing classes in the predicted classification result.
  • the confusion category is the category with a high degree of confusion in the process of using the multi-class neural network model to predict, that is, the multi-dimensional features corresponding to the multiple preset category label classifications included in each confusion category use multi-classification
  • the prediction categories obtained by the neural network model prediction are easily predicted to become other prediction categories in the confusion category, and the probability of occurrence is high. Therefore, adjusting the network structure of the multi-class neural network model according to the determined confusion category can target and effectively improve the recognition accuracy of the multi-class neural network model for text data categories.
  • it can construct different verification sets for various special application scenarios to achieve targeted improvements to the multi-class neural network model.
  • Step 2032 According to the correspondence between the confusion category and the multiple text data categories, adjust the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model to obtain an initial classification prediction model.
  • a parallel classification neural network is set for the fully connected layer in the multi-class neural network model, that is, on the basis of the multi-class neural network model.
  • the above constructs multiple connected layers parallel to the fully connected layer in the multi-class neural network model. Specifically, the input end of each newly added connection layer is connected to the output end of the connection layer of the corresponding confusion category to realize the expansion of the hidden features in the confusion category, thereby obtaining the multi-category after network structure adjustment Neural network model (ie, initialize the classification prediction model).
  • the previously constructed training set is used to retrain the multi-class neural network model after the network structure is adjusted.
  • the multi-dimensional features in the training set are input to initialize the classification prediction
  • the input layer of the model the input layer preprocesses the multi-dimensional features, converts the multi-dimensional features into multi-dimensional vectors, and inputs the fully connected layer through the hidden layer to obtain the classification result in the form of text.
  • the cross-entropy is calculated according to the classification results output by the fully connected layer and the preset category labels of the corresponding multi-dimensional features in the training set, and is used as the loss function for training the initial classification prediction model.
  • the Adam optimization algorithm is used to minimize the determined loss function In this way, the network parameters in the fully connected layer are optimized, and the network parameters in other hidden layers remain fixed, thereby further improving the classification accuracy of the initial classification prediction model, and the initialized classification prediction model after training is used as the final A preset classification prediction model used to recognize the text data category to be recognized.
  • the multi-dimensional features of the text data to be recognized are extracted, and the extracted multi-dimensional features are input into the preset classification The input layer of the prediction model.
  • the input layer preprocesses the multi-dimensional features and converts the multi-dimensional features into multi-dimensional vectors.
  • the input layer processes the multi-dimensional vectors into the hidden layer and outputs them to the fully connected layer.
  • the fully connected layer obtains the network according to the processing results
  • the classification result of the information that is, the type of text data to be recognized).
  • the network information obtained by the user is network information in the news field
  • extract the multi-dimensional characteristics of the network information in the news field and use the preset classification prediction model for classification processing, and the classification results obtained are respectively international news , Domestic news, sports news, etc.
  • the multi-class neural network model is used to obtain the predicted classification result of the text data according to the text data in the verification set, and the predicted classification result of the text data and the prediction corresponding to the text data in the verification set Set the category label, adjust the multi-class neural network model to the initial classification prediction model, and use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized.
  • the recognition accuracy is low, and the user cannot get a more accurate text data classification and recognition result.
  • the network can realize the multi-class neural network model through automatic verification
  • the structural improvement makes the improved initial classification prediction model eliminate the need for professional technicians to confirm the specific construction process of the classifier, reducing labor and time costs, and the improved initial classification prediction model can effectively improve the accuracy of text data recognition.
  • an embodiment of the present application provides a device for identifying text data categories based on a neural grid model.
  • the device includes: a verification module 31, an adjustment module 32, The prediction module 33.
  • the verification module 31 may be used to obtain a predicted classification result of the text data according to the text data in the verification set by using a multi-class neural network model.
  • the adjustment module 32 may be configured to adjust the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set.
  • the prediction module 33 may be used to use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized.
  • the device further includes: a training module 34.
  • the training module 34 can be used to train and initialize a neural network model using the training set to obtain a multi-class neural network model.
  • the adjustment module 32 may be specifically configured to determine the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set; The confusion category is described, and the network structure of the multi-class neural network model is adjusted to obtain an initial classification prediction model.
  • the confusion category in the predicted classification result is determined, which can also be specifically used if the text If the predicted category in the predicted classification result of the data is inconsistent with the preset category label of the corresponding text data in the verification set, the predicted category in the predicted classification result is determined as a confusion category.
  • the network structure of the multi-class neural network model according to the confusion category to obtain an initialized classification prediction model, which can be specifically used to determine multiple text data categories according to the confusion category; According to the correspondence between the confusion category and the multiple text data categories, the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model is adjusted to obtain an initialized classification prediction model.
  • multiple text data categories are determined according to the confusion category, which can also be specifically used to perform a T matrix operation on a confusion matrix used to characterize the confusion category to obtain a similarity matrix;
  • the degree matrix performs spectral clustering operations to obtain multiple text data categories.
  • the prediction module 33 can also be specifically used to train and initialize the classification prediction model using the training set to obtain the preset classification prediction model; use the preset classification prediction model to recognize the text data to be recognized to obtain the Identify text data categories.
  • an embodiment of the present application also provides a non-volatile readable storage medium on which computer readable instructions are stored, and the program is executed when the processor is executed.
  • the technical solution of the present application can be embodied in the form of a software product, and the software product can be stored in a non-volatile non-volatile readable storage medium (can be CD-ROM, U disk, mobile hard disk) Etc.), including several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in each implementation scenario of this application.
  • a computer device which may be a personal computer, a server, or a network device, etc.
  • the embodiments of the present application also provide a computer device, which can be a personal computer, a server, or a network.
  • the physical device includes a non-volatile readable storage medium and a processor; the non-volatile readable storage medium is used to store computer readable instructions; and the processor is used to execute computer readable instructions to achieve the above Figure 1 and Figure 2 show the method for identifying text data categories based on the neural grid model.
  • the computer device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and so on.
  • the user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like.
  • the network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.
  • a computer device does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or arrange different components.
  • the non-volatile readable storage medium may also include an operating system and a network communication module.
  • the operating system is a program that manages the hardware and software resources of computer equipment, and supports the operation of information processing programs and other software and/or programs.
  • the network communication module is used to implement communication between various components in the non-volatile readable storage medium and communication with other hardware and software in the physical device.
  • the recognition accuracy is low when the classification is divided, and the user cannot obtain a more accurate text data classification and recognition result.
  • the present embodiment can pass Automatic verification improves the multi-class neural network model, so that the improved initialization classification prediction model does not require professional technicians to confirm the specific construction process of the classifier, reducing labor and time costs, and the improved initialization classification prediction model can effectively improve The accuracy of text data recognition.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The present application relates to the technical field of information processing. Disclosed are a neural grid model-based text data category recognition method and apparatus, a nonvolatile readable storage medium, and a computer device, capable of effectively improving the accuracy of text data recognition. The method comprises: according to text data in a validation set, obtaining a predicted classification result of the text data using a multi-classification neural network model; adjusting the multi-classification neural network model into an initialized classification prediction model according to the predicted classification result of the text data and a preset category tag of the text data in the validation set; and obtaining a text data category to be recognized using the initialized classification prediction model according to text data to be recognized. The present application is suitable for classification processing of network information.

Description

基于神经网格模型的文本数据类别的识别方法及装置、非易失性可读存储介质、计算机设备Method and device for identifying text data categories based on neural grid model, non-volatile readable storage medium, and computer equipment
本申请要求与2019年5月31日提交中国专利局、申请号为201910470485.3、申请名称为“基于神经网格模型的文本数据类别的识别方法及装置、存储介质及计算机设备”的中国专利申请的优先权，其全部内容通过引用结合在申请中。This application requires a Chinese patent application filed with the Chinese Patent Office on May 31, 2019, with the application number 201910470485.3, and the application titled "Method and device for identifying text data categories based on neural grid model, storage medium and computer equipment". Priority, the entire content of which is incorporated in the application by reference.
技术领域Technical field
本申请涉及信息处理技术领域，尤其是涉及到基于神经网格模型的文本数据类别的识别方法及装置、非易失性可读存储介质及计算机设备。This application relates to the field of information processing technology, and in particular to methods and devices for identifying text data categories based on neural grid models, non-volatile readable storage media, and computer equipment.
背景技术Background technique
随着科学技术的发展，在很多应用场景下都会涉及到文本数据的分类，且基于越来越智能化的应用场景，对文本数据的类别划分需求也越来越高。因此，在面临多分类问题时，通常基于神经网络来实现对文本数据的分类。With the development of science and technology, the classification of text data is involved in many application scenarios, and based on increasingly intelligent application scenarios, the demand for classification of text data is also increasing. Therefore, when facing multi-classification problems, the classification of text data is usually based on neural networks.
现有技术存在的不足为，目前多分类问题下的神经网络，其识别精度往往会受到文本数据类别较多的影响，即当文本数据的类别较多时，神经网络的识别准确度将会有所降低；同时，在构建多个分类器或者多步分类器时，需要专业的技术人员对具体的构建过程进行校对，以便确认是否准确，从而完成后续的文本数据类别识别操作。The disadvantage of the prior art is that the recognition accuracy of neural networks under current multi-classification problems is often affected by more types of text data, that is, when there are more types of text data, the recognition accuracy of neural networks will be somewhat different. Reduce; At the same time, when building multiple classifiers or multi-step classifiers, professional technicians are required to proofread the specific construction process to confirm whether it is accurate, so as to complete the subsequent text data category recognition operation.
可见，基于现有神经网络所构建的分类器在进行类别划分时，识别准确度较低，用户无法得到较为精确的文本数据分类识别结果，影响用户的使用体验，且现有神经网络需要专业的技术人员介入确认分类器的具体构建过程，人工和时间成本较高。It can be seen that the classifier based on the existing neural network has low recognition accuracy when classifying, and users cannot get more accurate text data classification and recognition results, which affects the user experience, and the existing neural network requires professional Technicians intervene to confirm the specific construction process of the classifier, and the labor and time costs are high.
发明内容Summary of the invention
有鉴于此，本申请提供了基于神经网格模型的文本数据类别的识别方法及装置、非易失性可读存储介质、计算机设备，主要目的在于解决目前基于现有神经网络所构建的分类器识别准确度较低，以及现有神经网络需要专业的技术人员介入确认分类器的具体构建过程，人工和时间成本较高的问题。In view of this, this application provides methods and devices for identifying text data categories based on neural grid models, non-volatile readable storage media, and computer equipment. The main purpose is to solve current classifiers constructed based on existing neural networks. The recognition accuracy is low, and the existing neural network requires professional technical personnel to intervene to confirm the specific construction process of the classifier, and the labor and time costs are high.
根据本申请的一个方面，提供了一种基于神经网格模型的文本数据类别的识别方法，该方法包括：According to one aspect of the present application, there is provided a method for identifying text data categories based on a neural grid model, the method including:
利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果；Using a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型；Adjusting the multi-class neural network model to an initialized classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别。Using the initial classification prediction model, the text data category to be recognized is obtained according to the text data to be recognized.
根据本申请的另一方面，提供了一种基于神经网格模型的文本数据类别的识别装置，该装置包括：According to another aspect of the present application, there is provided a text data category recognition device based on a neural grid model, the device comprising:
验证模块，用于利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果；The verification module is configured to use a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
调整模块，用于根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型；An adjustment module, configured to adjust the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
预测模块，用于利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别。The prediction module is used to use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized.
依据本申请又一个方面，提供了一种非易失性可读存储介质，其上存储有计算机可读指令，所述程序被处理器执行时实现上述基于神经网格模型的文本数据类别的识别方法。According to another aspect of the present application, there is provided a non-volatile readable storage medium having computer readable instructions stored thereon, and when the program is executed by a processor, the recognition of the text data category based on the neural grid model is realized. method.
依据本申请再一个方面，提供了一种计算机设备，包括非易失性可读存储介质、处理器及存储在非易失性可读存储介质上并可在处理器上运行的计算机可读指令，所述处理器执行所述程序时实现上述基于神经网格模型的文本数据类别的识别方法。According to another aspect of the present application, a computer device is provided, including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor When the processor executes the program, the above-mentioned method for recognizing the text data category based on the neural grid model is realized.
借由上述技术方案，本申请提供的基于神经网格模型的文本数据类别的识别方法及装置、非易失性可读存储介质、计算机设备，与目前现有神经网络所构建的分类器在进行类别划分时，识别准确度较低，用户无法得到较为精确的文本数据分类识别结果相比，本申请利用多分类神经网络模型，根据验证集中的文本数据得到该文本数据的预测分类结果，根据该文本数据的预测分类结果和该验证集中与该文本数据对应的预设类别标签，将多分类神经网络模型调整为初始化分类预测模型，并利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别。通过自动验证对多分类神经网络模型进行改进，使得改进后的初始化分类预测模型无需专业的技术人员介入确认分类器的具体构建过程，降低人工和时间成本，且改进后的初始化分类预测模型能够有效提升对文本数据识别的准确度。With the above technical solutions, the neural grid model-based text data category recognition method and device, non-volatile readable storage medium, and computer equipment provided in this application are in progress with the existing classifiers constructed by neural networks. When the classification is divided, the recognition accuracy is low, and the user cannot obtain a more accurate text data classification and recognition result. Compared with this application, the multi-class neural network model is used to obtain the predicted classification result of the text data according to the text data in the verification set. The predicted classification result of the text data and the preset category label corresponding to the text data in the verification set, the multi-class neural network model is adjusted to the initial classification prediction model, and the initial classification prediction model is used to obtain the text to be recognized according to the text data to be recognized Data category. The multi-class neural network model is improved through automatic verification, so that the improved initial classification prediction model does not require professional technicians to confirm the specific construction process of the classifier, reducing labor and time costs, and the improved initial classification prediction model can be effective Improve the accuracy of text data recognition.
上述说明仅是本申请技术方案的概述，为了能够更清楚了解本申请的技术手段，而可依照说明书的内容予以实施，并且为了让本申请的上述和其它目的、特征和优点能够更明显易懂，以下特举本申请的具体实施方式。The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly, it can be implemented in accordance with the content of the specification, and to make the above and other purposes, features and advantages of this application more obvious and understandable. , The following specifically cite the specific implementation of this application.
附图说明Description of the drawings
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：The drawings described here are used to provide a further understanding of the application and constitute a part of the application. The exemplary embodiments and descriptions of the application are used to explain the application and do not constitute an improper limitation of the application. In the attached picture:
图1示出了本申请实施例提供的一种基于神经网格模型的文本数据类别识别方法的流程示意图；FIG. 1 shows a schematic flowchart of a method for identifying text data categories based on a neural grid model provided by an embodiment of the present application;
图2示出了本申请实施例提供的另一种基于神经网格模型的文本数据类别识别方法的流程示意图；FIG. 2 shows a schematic flowchart of another method for identifying text data categories based on a neural grid model provided by an embodiment of the present application;
图3示出了本申请实施例提供的一种基于神经网格模型的文本数据类别识别装置的结构示意图。Fig. 3 shows a schematic structural diagram of a text data category recognition device based on a neural grid model provided by an embodiment of the present application.
具体实施方式Detailed ways
下文中将参考附图并结合实施例来详细说明本申请。需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。Hereinafter, the application will be described in detail with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other if there is no conflict.
针对目前基于现有神经网络所构建的分类器识别准确度较低，以及现有神经网络需要专业的技术人员介入确认分类器的具体构建过程，人工和时间成本较高的问题。本实施例提供了一种基于神经网格模型的文本数据类别的识别方法，能够自动化构建初始化分类预测模型，同时能够有效提高文本数据类别的识别准确度，如图1所示，该方法包括：In view of the low recognition accuracy of the current classifier constructed based on the existing neural network, and the existing neural network requires professional technicians to intervene to confirm the specific construction process of the classifier, the labor and time costs are high. This embodiment provides a method for identifying text data categories based on a neural grid model, which can automatically construct an initial classification prediction model, and can effectively improve the recognition accuracy of text data categories. As shown in FIG. 1, the method includes:
101、利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果。101. Use a multi-class neural network model to obtain a predicted classification result of the text data according to the text data in the verification set.
验证集用于验证多分类神经网络模型对文本数据类别的识别准确度，是基于网络行为构建的，社交网络行为是海量的、多种多样的。例如，撰写一篇博客，浏览一组照片，点击一个广告，购买一件商品，订阅特定新闻话题等所产生的社交网络行为均可用于多分类识别，以实现对社交网络行为的类别识别。The verification set is used to verify the accuracy of the multi-class neural network model for the recognition of text data categories. It is constructed based on network behaviors, and social network behaviors are massive and diverse. For example, social network behaviors generated by writing a blog, browsing a group of photos, clicking an advertisement, purchasing a product, subscribing to a specific news topic, etc. can be used for multi-class recognition to realize category recognition of social network behaviors.
在本实施例中，以文本数据类别识别为例，从社交网络获取各种行为的文本数据，提取文本数据中的多维度特征(例如，标题、关键字等特征)，并对应该多维度特征预设类别标签(即，该文本数据的预设类别)，根据所提取的多维度特征和预设的类别标签构建验证集。In this embodiment, taking text data category recognition as an example, text data of various behaviors are obtained from social networks, and multi-dimensional features (for example, features such as titles and keywords) in the text data are extracted, and the multi-dimensional features are corresponding The preset category label (ie, the preset category of the text data) constructs a verification set based on the extracted multi-dimensional features and the preset category label.
102、根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型。102. According to the predicted classification result of the text data and the preset category label of the text data in the verification set, the multi-class neural network model is adjusted to an initial classification prediction model.
将基于多分类神经网络模型输出的预测分类结果与验证集中多维度特征对应的预设类别标签进行比对，根据比对结果，在多分类神经网络模型原有的网络结构的基础上，进行网络结构调整，以使调整后的多分类神经网络模型(即初始化分类预测模型)对文本数据类别的识别准确度更高，尤其是对与所构建的验证集中的文本数据较为相似的文本数据进行类别识别时，其识别准确度更为理想。The predicted classification results based on the output of the multi-class neural network model are compared with the preset category labels corresponding to the multi-dimensional features in the verification set. According to the comparison results, the network is performed on the basis of the original network structure of the multi-class neural network model Structural adjustments to make the adjusted multi-class neural network model (ie, the initial classification prediction model) more accurate in identifying text data categories, especially for text data similar to the text data in the constructed verification set When recognizing, its recognition accuracy is more ideal.
在具体场景中，若针对特有的文本数据进行类别识别时，为了满足调整后的多分类神经网络模型对特有的文本数据类别的识别准确度较高的需求，可以重新构建新的验证集，以便基于所构建的新的验证集，对多分类神经网络模型原有的网络结构进行重新调整，以使其更加适用于新的文本数据类别的识别。In a specific scenario, if category recognition is performed for unique text data, in order to meet the needs of the adjusted multi-class neural network model for higher recognition accuracy of unique text data categories, a new verification set can be rebuilt to facilitate Based on the constructed new verification set, the original network structure of the multi-class neural network model is re-adjusted to make it more suitable for the recognition of new text data categories.
103、利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别。103. Use the initialized classification prediction model to obtain the text data category to be recognized according to the text data to be recognized.
获取待识别文本数据，对获取到的待识别文本数据进行预处理，提取待识别文本数据中的多维度特征，利用初始化分类预测模型，根据待识别文本数据的多维度特征得到待识别文本数据类别。Obtain the text data to be recognized, preprocess the text data to be recognized, extract the multi-dimensional features in the text data to be recognized, and use the initial classification prediction model to obtain the text data category to be recognized based on the multi-dimensional features of the text data to be recognized .
在本实施例中，对初始化分类预测模型进行训练，将待识别文本数据输入训练后的初始化分类预测模型(即预设的分类预测模型)，输出待识别文本数据类别，以使用户能够基于预设的分类预测模型处理后的文本数据分类，准确地获取到与文本数据类别匹配度较高的信息，提升用户体验。In this embodiment, the initial classification prediction model is trained, the text data to be recognized is input into the trained initialization classification prediction model (ie, the preset classification prediction model), and the text data category to be recognized is output so that the user can be based on the prediction The text data classified by the classification prediction model processed by the proposed classification prediction model can accurately obtain information with a high degree of matching with the text data category, and improve the user experience.
对于本实施例可以按照上述方案，利用多分类神经网络模型，根据验证集中的文本数据得到该文本数据的预测分类结果，根据该文本数据的预测分类结果和该验证集中与该文本数据对应的预设类别标签，将多分类神经网络模型调整为初始化分类预测模型，并利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别，与目前现有神经网络所构建的分类器在进行类别划分时，识别准确度较低，用户无法得到较为精确的文本数据分类识别结果相比，本实施例能够通过自动验证对多分类神经网络模型进行改进，使得改进后的初始化分类预测模型无需专业的技术人员介入确认分类器的具体构建过程，降低人工和时间成本，且改进后的初始化分类预测模型能够有效提升对文本数据识别的准确度。For this embodiment, according to the above solution, a multi-class neural network model can be used to obtain the predicted classification result of the text data according to the text data in the verification set, and the predicted classification result of the text data and the prediction corresponding to the text data in the verification set Set the category label, adjust the multi-class neural network model to the initial classification prediction model, and use the initialized classification prediction model to obtain the text data category to be recognized according to the text data to be recognized, and perform classification with the current classifier constructed by the existing neural network When dividing, the recognition accuracy is low, and users cannot get more accurate text data classification and recognition results. Compared with this embodiment, the multi-class neural network model can be improved through automatic verification, so that the improved initialization classification prediction model does not require professional Technicians intervene to confirm the specific construction process of the classifier, reduce labor and time costs, and the improved initial classification prediction model can effectively improve the accuracy of text data recognition.
进一步的，作为上述实施例具体实施方式的细化和扩展，为了完整说明本实施例的具体实施过程，提供了另一种基于神经网格模型的文本数据类别的识别方法，该方法包括：利用训练集训练初始化神经网络模型，得到多分类神经网络模型。Further, as a refinement and extension of the specific implementation of the above-mentioned embodiment, in order to fully explain the specific implementation process of this embodiment, another method for identifying text data categories based on a neural grid model is provided, which includes: The training set trains and initializes the neural network model to obtain the multi-class neural network model.
具体为，基于网络行为构建训练集，从社交网络获取各种行为的新的文本数据(即，有别于上述用于构建验证集的文本数据)，提取文本数据中的多维度特征，并对应该多维度特征预设类别标签，根据所提取的多维度特征和预设的类别标签构建训练集，以使获取到的训练集中文本数据的多维度特征与验 证集中文本数据的多维度特征不相同，或者，获取到的训练集中文本数据的多维度特征和预设的类别标签与验证集中文本数据的多维度特征和预设的类别标签均不相同。Specifically, construct a training set based on network behaviors, obtain new text data of various behaviors from social networks (that is, different from the text data used to construct the verification set above), extract multi-dimensional features in the text data, and compare Multi-dimensional features should be preset with category labels, and the training set should be constructed based on the extracted multi-dimensional features and preset category labels, so that the obtained multi-dimensional features of the text data in the training set are different from the multi-dimensional features of the text data in the validation set Or, the obtained multi-dimensional features and preset category labels of the text data in the training set are different from the multi-dimensional features and preset category labels of the text data in the validation set.
在对获取到的训练集对初始化神经网络模型进行训练之前构建初始化神经网络模型，初始化神经网络模型可以是初始化卷积神经网络模型，或者初始化循环神经网络模型，初始化神经网络模型的主要结构为多连接的多层结构，以实现针对多维度特征的运算，以文本数据类别识别为例，初始化神经网络模型主要包括：输入层、卷积层、池化层和全连接层，其中各层的初始化网络参数均为随机生成的。Construct the initialization neural network model before training the initialization neural network model on the acquired training set. The initialization neural network model can be the initialization convolutional neural network model or the recurrent neural network model. The main structure of the initialization neural network model is multiple Connected multi-layer structure, to realize the operation of multi-dimensional features, take text data category recognition as an example, the initialization neural network model mainly includes: input layer, convolutional layer, pooling layer and fully connected layer, and the initialization of each layer The network parameters are all randomly generated.
在具体场景中，若分类目标为客户分类，则初始化神经网络模型的网络结构也可以以全连接层为主，此处不对初始化神经网络模型的具体网络结构进行限定。In a specific scenario, if the classification target is customer classification, the network structure of the initialized neural network model can also be based on a fully connected layer, and the specific network structure of the initialized neural network model is not limited here.
以文本数据类别识别为例，根据获取到的训练集对初始化神经网络模型进行训练，具体为，将初始化神经网络模型的卷积层和池化层统称为隐层，根据实际应用的需求，隐层可以为多个，在Python的tensorflow库中，将训练集中的多维度特征输入初始化神经网络模型的输入层，输入层对多维度特征进行预处理，将多维度特征转换成多维向量，输入层将多维向量输入隐层进行处理，最后一个隐层将处理结果输入全连接层，全连接层将处理结果转换成文本形式的分类结果。将分类结果与预设类别标签进行比对，若分类结果与预设类别标签不一致，则根据输出的分类结果与预设类别标签计算得到交叉熵，并作为用于训练初始化神经网络模型的损失函数。Taking text data category recognition as an example, the initialization neural network model is trained according to the obtained training set. Specifically, the convolutional layer and pooling layer of the initialized neural network model are collectively referred to as the hidden layer. According to the needs of actual applications, the hidden layer There can be multiple layers. In Python's tensorflow library, input the multi-dimensional features in the training set to the input layer of the initialization neural network model. The input layer preprocesses the multi-dimensional features and converts the multi-dimensional features into multi-dimensional vectors. The input layer The multi-dimensional vector is input to the hidden layer for processing, the last hidden layer inputs the processing result into the fully connected layer, and the fully connected layer converts the processing result into a classification result in text form. The classification result is compared with the preset category label. If the classification result is inconsistent with the preset category label, the cross entropy is calculated according to the output classification result and the preset category label, and used as the loss function for training the initialization neural network model .
不断迭代上述训练过程，利用自适应矩估计(Adam：Adaptive Moment Estimation)优化算法对所确定的损失函数进行最小化，从而实现对隐层的网络参数的更新，直至训练集中所有文本数据的多维度特征全部训练完成，得到多分类神经网络模型。Continuously iterate the above training process, use the adaptive moment estimation (Adam: Adaptive Moment Estimation) optimization algorithm to minimize the determined loss function, so as to realize the update of the hidden layer network parameters, until the multi-dimensionality of all text data in the training set After all the features are trained, a multi-class neural network model is obtained.
进一步的，作为上述实施例具体实施方式的细化和扩展，为了完整说明本实施例的具体实施过程，提供了另一种基于神经网格模型的文本数据类别的识别方法，如图2所示，该方法包括：Further, as a refinement and expansion of the specific implementation of the above-mentioned embodiment, in order to fully explain the specific implementation process of this embodiment, another method for identifying text data categories based on the neural grid model is provided, as shown in FIG. 2 , The method includes:
201、利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果。201. Using a multi-classification neural network model to obtain a predicted classification result of the text data according to the text data in the verification set.
将验证集中的N个多维度特征输入多分类神经网络模型的输入层，输入层将验证集中的多维度特征转换成多维向量后输出给隐层进行处理，最后一个隐层将处理结果输入全连接层，全连接层将处理结果转换成文本形式的预测分类结果，以便根据多分类神经网络模型输出的预测分类结果和验证集中对应的文本数据的预设类别标签，将多分类神经网络模型调整为初始化分类预测模型。Input the N multi-dimensional features in the validation set into the input layer of the multi-class neural network model. The input layer converts the multi-dimensional features in the validation set into multi-dimensional vectors and then outputs them to the hidden layer for processing. The last hidden layer inputs the processing results into fully connected Layer, the fully connected layer converts the processing results into text-form predictive classification results, so as to adjust the multi-class neural network model to the predicted classification results output by the multi-class neural network model and the preset category labels of the corresponding text data in the validation set Initialize the classification prediction model.
202、根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别。202. Determine a confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set.
为了说明步骤202的具体实施方式，作为一种优选实施例，步骤203具体可以包括：若所述文本数据的预测分类结果中的预测类别与所述验证集中相应文本数据的预设类别标签不一致，则将所述预测分类结果中的预测类别确定为混淆类别。To illustrate the specific implementation of step 202, as a preferred embodiment, step 203 may specifically include: if the predicted category in the predicted classification result of the text data is inconsistent with the preset category label of the corresponding text data in the verification set, Then, the predicted category in the predicted classification result is determined as a confusion category.
在本实施例中，将多分类神经网络模型输出的预测分类结果与验证集中多维度特征对应的预设类别标签进行比对，若预测分类结果中的预测类别与验证集中相应多维度特征的预设类别标签不一致，则将该预测类别确定为基于该多分类神经网络模型的混淆类别，该混淆类别包括多个预设类别标签分类及每个预设类别标签分类下所包含的预设类别标签数量，根据所确定的混淆类别以实现对多分类神经网络模型的改进。In this embodiment, the predicted classification result output by the multi-class neural network model is compared with the preset category labels corresponding to the multi-dimensional features in the verification set. If the predicted category in the predicted classification result is compared with the prediction of the corresponding multi-dimensional feature in the verification set. If the category labels are inconsistent, the predicted category is determined as the confusion category based on the multi-class neural network model. The confusion category includes a plurality of preset category label classifications and the preset category labels contained under each preset category label classification Quantity, according to the determined confusion category to realize the improvement of the multi-class neural network model.
此外，若预测分类结果中的预测类别与验证集中相应多维度特征的预设类别标签一致，即该预测类别中仅包括一个预设类别标签分类及该预设类别标签分类下所包含的预设类别标签数量，则该预测类别不进行重新设定，即不作为对多分类神经网络模型进行改进的考虑因素。In addition, if the prediction category in the prediction classification result is consistent with the preset category label of the corresponding multi-dimensional feature in the validation set, that is, the prediction category includes only one preset category label classification and the preset category label classification included in the preset category label classification The number of category labels, the predicted category will not be reset, that is, it will not be used as a consideration for improving the multi-class neural network model.
203、根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型。203. Perform network structure adjustment on the multi-class neural network model according to the confusion category to obtain an initialized classification prediction model.
为了说明步骤203的具体实施方式，作为一种优选实施例，步骤203具体可以包括：To illustrate the specific implementation of step 203, as a preferred embodiment, step 203 may specifically include:
步骤2031：根据所述混淆类别确定多个文本数据类别。Step 2031: Determine multiple text data categories according to the confusion category.
为了说明步骤2031的具体实施方式，作为一种优选实施例，步骤2031具体可以包括：对用于表征所述混淆类别的混淆矩阵进行T型矩阵运算，得到相似度矩阵；对所述相似度矩阵进行谱聚类运算，得到多个文本数据类别。To illustrate the specific implementation of step 2031, as a preferred embodiment, step 2031 may specifically include: performing a T-type matrix operation on the confusion matrix used to characterize the confusion category to obtain a similarity matrix; Perform spectral clustering operations to obtain multiple text data categories.
在本实施例中，在sklearn库中，根据所确定的预测分类结果中的多个混淆类别构建用于表征该多个混淆类别的混淆矩阵matrix，将混淆矩阵matrix中每个混淆类别包含的每个预设类别标签分类下的类别标签数量转换成每个预设类别标签分类下的类别标签数量占所对应的混淆类别中的类别标签总数量的比例，将转换后的混淆矩阵matrix进行T型矩阵运算，即利用计算公式matrix*matrix.T得到包含多个预设类别标签分类的相似度矩阵，并对得到的包含多个预设类别标签分类的相似度矩阵进行谱聚类运算，得到谱聚类输出结果，即多个预设类别标签分类。其中，T型矩阵即托普利兹矩阵Toeplitz matrix。In this embodiment, in the sklearn library, a confusion matrix for representing the multiple confusion categories is constructed according to the multiple confusion categories in the determined predicted classification results, and each confusion category contained in the confusion matrix matrix is The number of category labels under a preset category label classification is converted into the proportion of the number of category labels under each preset category label classification to the total number of category labels in the corresponding confusion category, and the converted confusion matrix is T-shaped Matrix operation, that is, use the calculation formula matrix*matrix.T to obtain a similarity matrix containing multiple preset category label classifications, and perform spectral clustering operations on the obtained similarity matrix containing multiple preset category label classifications to obtain the spectrum The output result of the clustering is the classification of multiple preset categories. Among them, the T-type matrix is the Toeplitz matrix.
在具体场景中，谱聚类运算具体为，将相似度矩阵中的预设类别标签分类作为顶点，以顶点之间的相似度作为特征向量构造向量空间并进行切分，得到较高类内相似度与较低类间相似度，从而实现对预测分类结果中的多个混淆类别的进一步分类。In a specific scenario, the spectral clustering operation is specifically to classify the preset category labels in the similarity matrix as vertices, and use the similarity between the vertices as the feature vector to construct a vector space and perform segmentation to obtain higher intra-class similarity The degree of similarity between the lower class and the lower class, so as to realize the further classification of multiple confusing classes in the predicted classification result.
其中，混淆类别是在利用多分类神经网络模型进行预测的过程中存在较高混淆程度的类别，即每一个混淆类别中所包括的多个预设类别标签分类所对应的多维度特征利用多分类神经网络模型预测所得到的预测类别，都极易被预测成为混淆类别中其它的预测类别，且发生的概率较高。因此，根据所确定的混淆类别，对多分类神经网络模型进行网络结构调整，能够有针对性地，且有效地提升多分类神经网络模型对文本数据类别的识别准确度。同时，能够针对各种特殊的应用场景，通过构建不同的验证集，以实现对多分类神经网络模型所进行的针对性改进。Among them, the confusion category is the category with a high degree of confusion in the process of using the multi-class neural network model to predict, that is, the multi-dimensional features corresponding to the multiple preset category label classifications included in each confusion category use multi-classification The prediction categories obtained by the neural network model prediction are easily predicted to become other prediction categories in the confusion category, and the probability of occurrence is high. Therefore, adjusting the network structure of the multi-class neural network model according to the determined confusion category can target and effectively improve the recognition accuracy of the multi-class neural network model for text data categories. At the same time, it can construct different verification sets for various special application scenarios to achieve targeted improvements to the multi-class neural network model.
步骤2032：根据所述混淆类别与多个文本数据类别的对应关系，对多分类神经网络模型中所述混淆类别所对应的全连接层进行网络结构调整，得到初始化分类预测模型。Step 2032: According to the correspondence between the confusion category and the multiple text data categories, adjust the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model to obtain an initial classification prediction model.
在本实施例中，根据混淆类别与多个文本数据类别之间一对多的对应关系，为多分类神经网络模型中的全连接层设置平行分类神经网络，即在多分类神经网络模型的基础上构建与多分类神经网络模型中的全连接层并行的多个连接层。具体为，每一新增的连接层的输入端分别与对应的混淆类别的连接层的输出端相连接，以实现对混淆类别中的隐含特征的扩充，从而得到网络结构调整后的多分类神经网络模型(即，初始化分类预测模型)。In this embodiment, according to the one-to-many correspondence between the confusion category and the multiple text data categories, a parallel classification neural network is set for the fully connected layer in the multi-class neural network model, that is, on the basis of the multi-class neural network model The above constructs multiple connected layers parallel to the fully connected layer in the multi-class neural network model. Specifically, the input end of each newly added connection layer is connected to the output end of the connection layer of the corresponding confusion category to realize the expansion of the hidden features in the confusion category, thereby obtaining the multi-category after network structure adjustment Neural network model (ie, initialize the classification prediction model).
204、利用训练集训练初始化分类预测模型，得到预设的分类预测模型。204. Use the training set to train and initialize the classification prediction model to obtain a preset classification prediction model.
在本实施例中，利用在前构建好的训练集对网络结构调整后的多分类神经网络模型进行再次训练，具体为，在Python的tensorflow库中，将训练集中的多维度特征输入初始化分类预测模型的输入层，输入层对多维度特征进行预处理，将多维度特征转换成多维向量，并经由隐层输入全连接层，得到文本形式的分类结果。In this embodiment, the previously constructed training set is used to retrain the multi-class neural network model after the network structure is adjusted. Specifically, in the tensorflow library of Python, the multi-dimensional features in the training set are input to initialize the classification prediction The input layer of the model, the input layer preprocesses the multi-dimensional features, converts the multi-dimensional features into multi-dimensional vectors, and inputs the fully connected layer through the hidden layer to obtain the classification result in the form of text.
根据全连接层输出的分类结果与训练集中相应多维度特征的预设类别标签计算得到交叉熵，并作为用于训练初始化分类预测模型的损失函数，利用Adam优化算法对所确定的损失函数进行最小化，从而实现对全连接层中的网络参数进行优化，且其它隐层中的网络参数保持固定不变，从而进一步提升初始化分类预测模型的分类精度，并将训练后的初始化分类预测模型作为最终用于对待识别文本数据类别进行识别的预设的分类预测模型。The cross-entropy is calculated according to the classification results output by the fully connected layer and the preset category labels of the corresponding multi-dimensional features in the training set, and is used as the loss function for training the initial classification prediction model. The Adam optimization algorithm is used to minimize the determined loss function In this way, the network parameters in the fully connected layer are optimized, and the network parameters in other hidden layers remain fixed, thereby further improving the classification accuracy of the initial classification prediction model, and the initialized classification prediction model after training is used as the final A preset classification prediction model used to recognize the text data category to be recognized.
205、利用预设的分类预测模型对待识别文本数据进行识别，得到待识别文本数据类别。205. Use the preset classification prediction model to recognize the text data to be recognized, and obtain the text data category to be recognized.
在本实施例中，当用户需要对所获取的网络信息(即，待识别文本数据)进行分类处理时，提取待识别文本数据的多维度特征，将提取到的多维度特征输入预设的分类预测模型的输入层，输入层对多维度特征进行预处理，将多维度特征转换成多维向量，输入层将多维向量输入隐层进行处理后输出给全连接层，全连接层根据处理结果得到网络信息的分类结果(即，待识别文本数据类别)。In this embodiment, when the user needs to classify the acquired network information (that is, the text data to be recognized), the multi-dimensional features of the text data to be recognized are extracted, and the extracted multi-dimensional features are input into the preset classification The input layer of the prediction model. The input layer preprocesses the multi-dimensional features and converts the multi-dimensional features into multi-dimensional vectors. The input layer processes the multi-dimensional vectors into the hidden layer and outputs them to the fully connected layer. The fully connected layer obtains the network according to the processing results The classification result of the information (that is, the type of text data to be recognized).
在具体场景中，若用户所获取的网络信息为新闻领域的网络信息，提取新闻领域的网络信息的多维度特征，利用预设的分类预测模型进行分类处理后，得到的分类结果分别是国际新闻、国内新闻、体育新闻等。In a specific scenario, if the network information obtained by the user is network information in the news field, extract the multi-dimensional characteristics of the network information in the news field, and use the preset classification prediction model for classification processing, and the classification results obtained are respectively international news , Domestic news, sports news, etc.
通过应用本实施例的技术方案，利用多分类神经网络模型，根据验证集中的文本数据得到该文本数据的预测分类结果，根据该文本数据的预测分类结果和该验证集中与该文本数据对应的预设类别标签，将多分类神经网络模型调整为初始化分类预测模型，并利用初始化分类预测模型，根据待识别文本数据 得到待识别文本数据类别。与目前现有神经网络所构建的分类器在进行类别划分时，识别准确度较低，用户无法得到较为精确的文本数据分类识别结果相比，能够通过自动验证实现对多分类神经网络模型的网络结构改进，使得改进后的初始化分类预测模型无需专业的技术人员介入确认分类器的具体构建过程，降低人工和时间成本，且改进后的初始化分类预测模型能够有效提升对文本数据识别的准确度。By applying the technical solution of this embodiment, the multi-class neural network model is used to obtain the predicted classification result of the text data according to the text data in the verification set, and the predicted classification result of the text data and the prediction corresponding to the text data in the verification set Set the category label, adjust the multi-class neural network model to the initial classification prediction model, and use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized. Compared with the current classifier constructed by the existing neural network, when the classification is divided, the recognition accuracy is low, and the user cannot get a more accurate text data classification and recognition result. The network can realize the multi-class neural network model through automatic verification The structural improvement makes the improved initial classification prediction model eliminate the need for professional technicians to confirm the specific construction process of the classifier, reducing labor and time costs, and the improved initial classification prediction model can effectively improve the accuracy of text data recognition.
进一步的，作为图1方法的具体实现，本申请实施例提供了一种基于神经网格模型的文本数据类别的识别装置，如图3所示，该装置包括：验证模块31、调整模块32、预测模块33。Further, as a specific implementation of the method in FIG. 1, an embodiment of the present application provides a device for identifying text data categories based on a neural grid model. As shown in FIG. 3, the device includes: a verification module 31, an adjustment module 32, The prediction module 33.
验证模块31，可以用于利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果。The verification module 31 may be used to obtain a predicted classification result of the text data according to the text data in the verification set by using a multi-class neural network model.
调整模块32，可以用于根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型。The adjustment module 32 may be configured to adjust the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set.
预测模块33，可以用于利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别。The prediction module 33 may be used to use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized.
在具体的应用场景中，该装置还包括：训练模块34。In a specific application scenario, the device further includes: a training module 34.
训练模块34，可以用于利用训练集训练初始化神经网络模型，得到多分类神经网络模型。The training module 34 can be used to train and initialize a neural network model using the training set to obtain a multi-class neural network model.
在具体的应用场景中，调整模块32，具体可以用于根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别；根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型。In a specific application scenario, the adjustment module 32 may be specifically configured to determine the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set; The confusion category is described, and the network structure of the multi-class neural network model is adjusted to obtain an initial classification prediction model.
在具体的应用场景中，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别，具体还可以用于，若所述文本数据的预测分类结果中的预测类别与所述验证集中相应文本数据的预设类别标签不一致，则将所述预测分类结果中的预测类别确定为混淆类别。In a specific application scenario, according to the predicted classification result of the text data and the preset category label of the text data in the verification set, the confusion category in the predicted classification result is determined, which can also be specifically used if the text If the predicted category in the predicted classification result of the data is inconsistent with the preset category label of the corresponding text data in the verification set, the predicted category in the predicted classification result is determined as a confusion category.
在具体的应用场景中，根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型，具体还可以用于，根据所述混淆类别确定多个文本数据类别；根据所述混淆类别与多个文本数据类别的对应关系，对多分类神经网络模型中所述混淆类别所对应的全连接层进行网络结构调整，得到初始化分类预测模型。In a specific application scenario, adjust the network structure of the multi-class neural network model according to the confusion category to obtain an initialized classification prediction model, which can be specifically used to determine multiple text data categories according to the confusion category; According to the correspondence between the confusion category and the multiple text data categories, the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model is adjusted to obtain an initialized classification prediction model.
在具体的应用场景中，根据所述混淆类别确定多个文本数据类别，具体还可以用于，对用于表征所述混淆类别的混淆矩阵进行T矩阵运算，得到相似度矩阵；对所述相似度矩阵进行谱聚类运算，得到多个文本数据类别。In a specific application scenario, multiple text data categories are determined according to the confusion category, which can also be specifically used to perform a T matrix operation on a confusion matrix used to characterize the confusion category to obtain a similarity matrix; The degree matrix performs spectral clustering operations to obtain multiple text data categories.
在具体的应用场景中，预测模块33，具体还可以用于，利用训练集训练初始化分类预测模型，得到预设的分类预测模型；利用预设的分类预测模型对待识别文本数据进行识别，得到待识别文本数据类别。In specific application scenarios, the prediction module 33 can also be specifically used to train and initialize the classification prediction model using the training set to obtain the preset classification prediction model; use the preset classification prediction model to recognize the text data to be recognized to obtain the Identify text data categories.
需要说明的是，本申请实施例提供的一种基金的持仓调整装置所涉及各功能单元的其他相应描述，可以参考图1和图2中的对应描述，在此不再赘述。It should be noted that, for other corresponding descriptions of the functional units involved in the fund position adjustment device provided in the embodiment of the present application, reference may be made to the corresponding descriptions in FIG. 1 and FIG. 2, and details are not repeated here.
基于上述如图1和图2所示方法，相应的，本申请实施例还提供了一种非易失性可读存储介质，其上存储有计算机可读指令，该程序被处理器执行时实现上述如图1和图2所示的基于神经网格模型的文本数据类别的识别方法。Based on the above-mentioned method shown in Figure 1 and Figure 2, correspondingly, an embodiment of the present application also provides a non-volatile readable storage medium on which computer readable instructions are stored, and the program is executed when the processor is executed. The above-mentioned method for identifying text data categories based on the neural grid model as shown in FIG. 1 and FIG. 2.
基于这样的理解，本申请的技术方案可以以软件产品的形式体现出来，该软件产品可以存储在一个非易失性非易失性可读存储介质(可以是CD-ROM，U盘，移动硬盘等)中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施场景所述的方法。Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, and the software product can be stored in a non-volatile non-volatile readable storage medium (can be CD-ROM, U disk, mobile hard disk) Etc.), including several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in each implementation scenario of this application.
基于上述如图1、图2所示的方法，以及图3所示的虚拟装置实施例，为了实现上述目的，本申请实施例还提供了一种计算机设备，具体可以为个人计算机、服务器、网络设备等，该实体设备包括非易失性可读存储介质和处理器；非易失性可读存储介质，用于存储计算机可读指令；处理器，用于执行计算机可读指令以实现上述如图1和图2所示的基于神经网格模型的文本数据类别的识别方法。Based on the methods shown in Figures 1 and 2 and the virtual device embodiment shown in Figure 3, in order to achieve the above objectives, the embodiments of the present application also provide a computer device, which can be a personal computer, a server, or a network. The physical device includes a non-volatile readable storage medium and a processor; the non-volatile readable storage medium is used to store computer readable instructions; and the processor is used to execute computer readable instructions to achieve the above Figure 1 and Figure 2 show the method for identifying text data categories based on the neural grid model.
可选的，该计算机设备还可以包括用户接口、网络接口、摄像头、射频(Radio Frequency，RF)电路，传感器、音频电路、WI-FI模块等等。用户接口可以包括显示屏(Display)、输入单元比如键盘(Keyboard)等，可选用户接口还可以包括USB接口、读卡器接口等。网络接口可选的可以包括标准的有线接口、无线接口(如蓝牙接口、WI-FI接口)等。Optionally, the computer device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and so on. The user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), etc., and the optional user interface may also include a USB interface, a card reader interface, and the like. The network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.
本领域技术人员可以理解，本实施例提供的一种计算机设备结构并不构成对该实体设备的限定，可以包括更多或更少的部件，或者组合某些部件，或者不同的部件布置。Those skilled in the art can understand that the structure of a computer device provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or arrange different components.
非易失性可读存储介质中还可以包括操作***、网络通信模块。操作***是管理计算机设备硬件和软件资源的程序，支持信息处理程序以及其它软件和/或程序的运行。网络通信模块用于实现非易失性可读存储介质内部各组件之间的通信，以及与该实体设备中其它硬件和软件之间通信。The non-volatile readable storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of computer equipment, and supports the operation of information processing programs and other software and/or programs. The network communication module is used to implement communication between various components in the non-volatile readable storage medium and communication with other hardware and software in the physical device.
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到本申请可以借助软件加必要的通用硬件平台的方式来实现，也可以通过硬件实现。通过应用本申请的技术方案，与目前现有神经网络所构建的分类器在进行类别划分时，识别准确度较低，用户无法得到较为精确的文本数据分类识别结果相比，本实施例可通过自动验证对多分类神经网络模型进行改进，使得改进后的初始化分类预测模型无需专业的技术人员介入确认分类器的具体构建过程，降低人工和时间成本，且改进后的初始化分类预测模型能够有效提升对文本数据识别的准确度。Through the description of the foregoing implementation manners, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform, or by hardware. By applying the technical solution of the present application, compared with the current classifier constructed by the neural network, the recognition accuracy is low when the classification is divided, and the user cannot obtain a more accurate text data classification and recognition result. The present embodiment can pass Automatic verification improves the multi-class neural network model, so that the improved initialization classification prediction model does not require professional technicians to confirm the specific construction process of the classifier, reducing labor and time costs, and the improved initialization classification prediction model can effectively improve The accuracy of text data recognition.
本领域技术人员可以理解附图只是一个优选实施场景的示意图，附图中的模块或流程并不一定是实施本申请所必须的。本领域技术人员可以理解实施场景中的装置中的模块可以按照实施场景描述进行分布于实施场景的装置中，也可以进行相应变化位于不同于本实施场景的一个或多个装置中。上述实施场景的模块可以合并为一个模块，也可以进一步拆分成多个子模块。Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of preferred implementation scenarios, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing this application. Those skilled in the art can understand that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be changed to be located in one or more devices different from the implementation scenario. The modules of the above implementation scenarios can be combined into one module or further divided into multiple sub-modules.
上述本申请序号仅仅为了描述，不代表实施场景的优劣。以上公开的仅为本申请的几个具体实施场景，但是，本申请并非局限于此，任何本领域的技术人员能思之的变化都应落入本申请的保护范围。The above serial number of this application is only for description, and does not represent the merits of implementation scenarios. The above disclosures are only a few specific implementation scenarios of the application, but the application is not limited to these, and any changes that can be thought of by those skilled in the art should fall into the protection scope of the application.

Claims (20)

  1. 一种基于神经网格模型的文本数据类别的识别方法，其特征在于，包括：A method for identifying text data categories based on a neural grid model, which is characterized in that it includes:
    利用训练集训练初始化神经网络模型，得到多分类神经网络模型；Use the training set to train and initialize the neural network model to obtain a multi-class neural network model;
    利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果；Using a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型；Adjusting the multi-class neural network model to an initialized classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别；Use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized;
    其中，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型，具体包括：Wherein, adjusting the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set includes:
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别；Determine the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型。According to the confusion category, adjust the network structure of the multi-class neural network model to obtain an initialized classification prediction model.
  2. 根据权利要求1所述的方法，其特征在于，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别，具体包括：The method according to claim 1, wherein determining the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set includes:
    若所述文本数据的预测分类结果中的预测类别与所述验证集中相应文本数据的预设类别标签不一致，则将所述预测分类结果中的预测类别确定为混淆类别。If the predicted category in the predicted classification result of the text data is inconsistent with the preset category label of the corresponding text data in the verification set, the predicted category in the predicted classification result is determined as a confusion category.
  3. 根据权利要求1所述的方法，其特征在于，根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型，具体包括：The method according to claim 1, wherein, according to the confusion category, adjusting the network structure of the multi-class neural network model to obtain an initial classification prediction model, which specifically comprises:
    根据所述混淆类别确定多个文本数据类别；Determine multiple text data categories according to the confusion category;
    根据所述混淆类别与多个文本数据类别的对应关系，对多分类神经网络模型中所述混淆类别所对应的全连接层进行网络结构调整，得到初始化分类预测模型。According to the correspondence between the confusion category and the multiple text data categories, the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model is adjusted to obtain an initialized classification prediction model.
  4. 根据权利要求3所述的方法，其特征在于，根据所述混淆类别确定多个文本数据类别，具体包括：The method according to claim 3, wherein determining a plurality of text data categories according to the confusion category specifically includes:
    对用于表征所述混淆类别的混淆矩阵进行T型矩阵运算，得到相似度矩阵；Perform a T-type matrix operation on the confusion matrix used to characterize the confusion category to obtain a similarity matrix;
    对所述相似度矩阵进行谱聚类运算，得到多个文本数据类别。Perform spectral clustering operations on the similarity matrix to obtain multiple text data categories.
  5. 根据权利要求1所述的方法，其特征在于，所述利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别，具体包括：The method according to claim 1, wherein said using the initial classification prediction model to obtain the type of text data to be recognized according to the text data to be recognized specifically comprises:
    利用训练集训练初始化分类预测模型，得到预设的分类预测模型；Use the training set to train and initialize the classification prediction model to obtain the preset classification prediction model;
    利用预设的分类预测模型对待识别文本数据进行识别，得到待识别文本数据类别。Use the preset classification prediction model to recognize the text data to be recognized, and obtain the text data category to be recognized.
  6. 一种基于神经网格模型的文本数据类别的识别装置，其特征在于，包括：A recognition device for text data categories based on neural grid model, characterized in that it comprises:
    训练模块，用于利用训练集训练初始化神经网络模型，得到多分类神经网络模型；The training module is used to use the training set to train and initialize the neural network model to obtain the multi-class neural network model;
    验证模块，用于利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类 结果；The verification module is configured to use a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
    调整模块，用于根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型；An adjustment module, configured to adjust the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    预测模块，用于利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别；The prediction module is used to use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized;
    其中，所述调整模块，具体包括：Wherein, the adjustment module specifically includes:
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别；Determine the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型。According to the confusion category, adjust the network structure of the multi-class neural network model to obtain an initialized classification prediction model.
  7. 根据权利要求6所述的装置，其特征在于，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别，具体包括：7. The device according to claim 6, wherein determining the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set includes:
    若所述文本数据的预测分类结果中的预测类别与所述验证集中相应文本数据的预设类别标签不一致，则将所述预测分类结果中的预测类别确定为混淆类别。If the predicted category in the predicted classification result of the text data is inconsistent with the preset category label of the corresponding text data in the verification set, the predicted category in the predicted classification result is determined as a confusion category.
  8. 根据权利要求6所述的装置，其特征在于，根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型，具体包括：The device according to claim 6, wherein, according to the confusion category, adjusting the network structure of the multi-class neural network model to obtain an initial classification prediction model, which specifically comprises:
    根据所述混淆类别确定多个文本数据类别；Determine multiple text data categories according to the confusion category;
    根据所述混淆类别与多个文本数据类别的对应关系，对多分类神经网络模型中所述混淆类别所对应的全连接层进行网络结构调整，得到初始化分类预测模型。According to the correspondence between the confusion category and the multiple text data categories, the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model is adjusted to obtain an initialized classification prediction model.
  9. 根据权利要求8所述的装置，其特征在于，根据所述混淆类别确定多个文本数据类别，具体包括：The device according to claim 8, wherein determining a plurality of text data categories according to the confusion category specifically includes:
    对用于表征所述混淆类别的混淆矩阵进行T型矩阵运算，得到相似度矩阵；Perform a T-type matrix operation on the confusion matrix used to characterize the confusion category to obtain a similarity matrix;
    对所述相似度矩阵进行谱聚类运算，得到多个文本数据类别。Perform spectral clustering operations on the similarity matrix to obtain multiple text data categories.
  10. 根据权利要求6所述的装置，其特征在于，所述预测模块，具体包括：The apparatus according to claim 6, wherein the prediction module specifically comprises:
    利用训练集训练初始化分类预测模型，得到预设的分类预测模型；Use the training set to train and initialize the classification prediction model to obtain the preset classification prediction model;
    利用预设的分类预测模型对待识别文本数据进行识别，得到待识别文本数据类别。Use the preset classification prediction model to recognize the text data to be recognized, and obtain the text data category to be recognized.
  11. 一种非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述程序被处理器执行时实现基于神经网格模型的文本数据类别的识别方法，包括：A non-volatile readable storage medium having computer readable instructions stored thereon is characterized in that, when the program is executed by a processor, a method for recognizing text data categories based on a neural grid model includes:
    利用训练集训练初始化神经网络模型，得到多分类神经网络模型；Use the training set to train and initialize the neural network model to obtain a multi-class neural network model;
    利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果；Using a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型；Adjusting the multi-class neural network model to an initialized classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别；Use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized;
    其中，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类 神经网络模型调整为初始化分类预测模型，具体包括：Wherein, according to the predicted classification result of the text data and the preset category label of the text data in the verification set, adjusting the multi-class neural network model to an initial classification prediction model specifically includes:
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别；Determine the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型。According to the confusion category, adjust the network structure of the multi-class neural network model to obtain an initialized classification prediction model.
  12. 根据权利要求11所述的非易失性可读存储介质，其特征在于，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别，具体包括：The non-volatile readable storage medium according to claim 11, wherein the predicted classification result of the text data is determined based on the predicted classification result of the text data and the preset category label of the text data in the verification set. Confusion categories, including:
    若所述文本数据的预测分类结果中的预测类别与所述验证集中相应文本数据的预设类别标签不一致，则将所述预测分类结果中的预测类别确定为混淆类别。If the predicted category in the predicted classification result of the text data is inconsistent with the preset category label of the corresponding text data in the verification set, the predicted category in the predicted classification result is determined as a confusion category.
  13. 根据权利要求11所述的非易失性可读存储介质，其特征在于，根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型，具体包括：The non-volatile readable storage medium according to claim 11, wherein, according to the confusion category, adjusting the network structure of the multi-class neural network model to obtain an initialized classification prediction model includes:
    根据所述混淆类别确定多个文本数据类别；Determine multiple text data categories according to the confusion category;
    根据所述混淆类别与多个文本数据类别的对应关系，对多分类神经网络模型中所述混淆类别所对应的全连接层进行网络结构调整，得到初始化分类预测模型。According to the correspondence between the confusion category and the multiple text data categories, the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model is adjusted to obtain an initialized classification prediction model.
  14. 根据权利要求13所述的非易失性可读存储介质，其特征在于，根据所述混淆类别确定多个文本数据类别，具体包括：The non-volatile readable storage medium according to claim 13, wherein determining multiple text data categories according to the confusion category includes:
    对用于表征所述混淆类别的混淆矩阵进行T型矩阵运算，得到相似度矩阵；Perform a T-type matrix operation on the confusion matrix used to characterize the confusion category to obtain a similarity matrix;
    对所述相似度矩阵进行谱聚类运算，得到多个文本数据类别。Perform spectral clustering operations on the similarity matrix to obtain multiple text data categories.
  15. 根据权利要求11所述的非易失性可读存储介质，其特征在于，所述利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别，具体包括：The non-volatile readable storage medium according to claim 11, wherein the use of the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized specifically comprises:
    利用训练集训练初始化分类预测模型，得到预设的分类预测模型；Use the training set to train and initialize the classification prediction model to obtain the preset classification prediction model;
    利用预设的分类预测模型对待识别文本数据进行识别，得到待识别文本数据类别。Use the preset classification prediction model to recognize the text data to be recognized, and obtain the text data category to be recognized.
  16. 一种计算机设备，包括非易失性可读存储介质、处理器及存储在非易失性可读存储介质上并可在处理器上运行的计算机可读指令，其特征在于，所述处理器执行所述程序时实现基于神经网格模型的文本数据类别的识别方法，包括：A computer device, including a non-volatile readable storage medium, a processor, and computer readable instructions stored on the non-volatile readable storage medium and running on the processor, characterized in that the processor The method for realizing the recognition of the text data category based on the neural grid model when executing the program includes:
    利用训练集训练初始化神经网络模型，得到多分类神经网络模型；Use the training set to train and initialize the neural network model to obtain a multi-class neural network model;
    利用多分类神经网络模型，根据验证集中的文本数据得到所述文本数据的预测分类结果；Using a multi-classification neural network model to obtain the predicted classification result of the text data according to the text data in the verification set;
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型；Adjusting the multi-class neural network model to an initialized classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别；Use the initial classification prediction model to obtain the text data category to be recognized according to the text data to be recognized;
    其中，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，将所述多分类神经网络模型调整为初始化分类预测模型，具体包括：Wherein, adjusting the multi-class neural network model to an initial classification prediction model according to the predicted classification result of the text data and the preset category label of the text data in the verification set includes:
    根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结 果中的混淆类别；Determine the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set;
    根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型。According to the confusion category, adjust the network structure of the multi-class neural network model to obtain an initialized classification prediction model.
  17. 根据权利要求16所述的计算机设备，其特征在于，根据所述文本数据的预测分类结果和所述验证集中文本数据的预设类别标签，确定所述预测分类结果中的混淆类别，具体包括：The computer device according to claim 16, wherein determining the confusion category in the predicted classification result according to the predicted classification result of the text data and the preset category label of the text data in the verification set includes:
    若所述文本数据的预测分类结果中的预测类别与所述验证集中相应文本数据的预设类别标签不一致，则将所述预测分类结果中的预测类别确定为混淆类别。If the predicted category in the predicted classification result of the text data is inconsistent with the preset category label of the corresponding text data in the verification set, the predicted category in the predicted classification result is determined as a confusion category.
  18. 根据权利要求16所述的计算机设备，其特征在于，根据所述混淆类别，对所述多分类神经网络模型进行网络结构调整，得到初始化分类预测模型，具体包括：The computer device according to claim 16, characterized in that, according to the confusion category, adjusting the network structure of the multi-class neural network model to obtain an initial classification prediction model, specifically comprising:
    根据所述混淆类别确定多个文本数据类别；Determine multiple text data categories according to the confusion category;
    根据所述混淆类别与多个文本数据类别的对应关系，对多分类神经网络模型中所述混淆类别所对应的全连接层进行网络结构调整，得到初始化分类预测模型。According to the correspondence between the confusion category and the multiple text data categories, the network structure of the fully connected layer corresponding to the confusion category in the multi-class neural network model is adjusted to obtain an initialized classification prediction model.
  19. 根据权利要求18所述的计算机设备，其特征在于，根据所述混淆类别确定多个文本数据类别，具体包括：The computer device according to claim 18, wherein determining a plurality of text data categories according to the confusion category specifically includes:
    对用于表征所述混淆类别的混淆矩阵进行T型矩阵运算，得到相似度矩阵；Perform a T-type matrix operation on the confusion matrix used to characterize the confusion category to obtain a similarity matrix;
    对所述相似度矩阵进行谱聚类运算，得到多个文本数据类别。Perform spectral clustering operations on the similarity matrix to obtain multiple text data categories.
  20. 根据权利要求16所述的计算机设备，其特征在于，所述利用初始化分类预测模型，根据待识别文本数据得到待识别文本数据类别，具体包括：The computer device according to claim 16, wherein said using the initialized classification prediction model to obtain the type of text data to be recognized according to the text data to be recognized specifically comprises:
    利用训练集训练初始化分类预测模型，得到预设的分类预测模型；Use the training set to train and initialize the classification prediction model to obtain the preset classification prediction model;
    利用预设的分类预测模型对待识别文本数据进行识别，得到待识别文本数据类别。Use the preset classification prediction model to recognize the text data to be recognized, and obtain the text data category to be recognized.
PCT/CN2019/117726 2019-05-31 2019-11-13 Neural grid model-based text data category recognition method and apparatus, nonvolatile readable storage medium, and computer device WO2020238053A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910470485.3 2019-05-31
CN201910470485.3A CN110362677B (en) 2019-05-31 2019-05-31 Text data category identification method and device, storage medium and computer equipment

Publications (1)

Publication Number Publication Date
WO2020238053A1 true WO2020238053A1 (en) 2020-12-03

Family

ID=68214983

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/117726 WO2020238053A1 (en) 2019-05-31 2019-11-13 Neural grid model-based text data category recognition method and apparatus, nonvolatile readable storage medium, and computer device

Country Status (2)

Country Link
CN (1) CN110362677B (en)
WO (1) WO2020238053A1 (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112599250A (en) * 2020-12-24 2021-04-02 中国人民解放军总医院第三医学中心 Postoperative data analysis method and device based on deep neural network
CN113254587A (en) * 2021-05-31 2021-08-13 北京奇艺世纪科技有限公司 Search text recognition method and device, computer equipment and storage medium
CN113268596A (en) * 2021-05-24 2021-08-17 康键信息技术（深圳）有限公司 Verification method, device and equipment of department classification model and storage medium
WO2022193171A1 (en) * 2021-03-17 2022-09-22 Huawei Cloud Computing Technologies Co., Ltd. System and method for unsupervised multi-model joint reasoning
CN116738343A (en) * 2023-08-08 2023-09-12 云筑信息科技(成都)有限公司 Material data identification method and device for construction industry and electronic equipment
CN117235270A (en) * 2023-11-16 2023-12-15 中国人民解放军国防科技大学 Text classification method and device based on belief confusion matrix and computer equipment

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110362677B (en) * 2019-05-31 2022-12-27 平安科技（深圳）有限公司 Text data category identification method and device, storage medium and computer equipment
CN111309912B (en) * 2020-02-24 2024-02-13 深圳市华云中盛科技股份有限公司 Text classification method, apparatus, computer device and storage medium
CN113626587B (en) * 2020-05-08 2024-03-29 武汉金山办公软件有限公司 Text category identification method and device, electronic equipment and medium
CN112417111B (en) * 2020-11-04 2022-08-23 厦门快商通科技股份有限公司 Text classification method, question answering system and dialogue robot
CN112818119B (en) * 2021-01-26 2022-12-02 支付宝(杭州)信息技术有限公司 Information processing method, device and equipment
CN113919361A (en) * 2021-09-13 2022-01-11 聚好看科技股份有限公司 Text classification method and device
CN115405363A (en) * 2022-08-30 2022-11-29 华北科技学院 Coal mine rock burst monitoring and early warning system based on LSTM neural network

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013040981A1 (en) * 2011-09-23 2013-03-28 浙江大学 Speaker recognition method for combining emotion model based on near neighbour principles
CN108509596A (en) * 2018-04-02 2018-09-07 广州市申迪计算机***有限公司 File classification method, device, computer equipment and storage medium
CN109376240A (en) * 2018-10-11 2019-02-22 平安科技（深圳）有限公司 A kind of text analyzing method and terminal
CN109376243A (en) * 2018-10-23 2019-02-22 平安科技（深圳）有限公司 File classification method and device
CN109471938A (en) * 2018-10-11 2019-03-15 平安科技（深圳）有限公司 A kind of file classification method and terminal
CN109471944A (en) * 2018-11-12 2019-03-15 中山大学 Training method, device and the readable storage medium storing program for executing of textual classification model
CN110362677A (en) * 2019-05-31 2019-10-22 平安科技（深圳）有限公司 The recognition methods of text data classification and device, storage medium, computer equipment

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6163461B2 (en) * 2014-07-16 2017-07-12 日本電信電話株式会社 Classification device, method, and program
US10482313B2 (en) * 2015-09-30 2019-11-19 Siemens Healthcare Gmbh Method and system for classification of endoscopic images using deep decision networks
US11354565B2 (en) * 2017-03-15 2022-06-07 Salesforce.Com, Inc. Probability-based guider
CN107688823B (en) * 2017-07-20 2018-12-04 北京三快在线科技有限公司 A kind of characteristics of image acquisition methods and device, electronic equipment
CN108776774A (en) * 2018-05-04 2018-11-09 华南理工大学 A kind of human facial expression recognition method based on complexity categorization of perception algorithm
CN109242516A (en) * 2018-09-06 2019-01-18 北京京东尚科信息技术有限公司 The single method and apparatus of processing service
CN109446332B (en) * 2018-12-25 2023-08-25 银江技术股份有限公司 People reconciliation case classification system and method based on feature migration and self-adaptive learning

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013040981A1 (en) * 2011-09-23 2013-03-28 浙江大学 Speaker recognition method for combining emotion model based on near neighbour principles
CN108509596A (en) * 2018-04-02 2018-09-07 广州市申迪计算机***有限公司 File classification method, device, computer equipment and storage medium
CN109376240A (en) * 2018-10-11 2019-02-22 平安科技（深圳）有限公司 A kind of text analyzing method and terminal
CN109471938A (en) * 2018-10-11 2019-03-15 平安科技（深圳）有限公司 A kind of file classification method and terminal
CN109376243A (en) * 2018-10-23 2019-02-22 平安科技（深圳）有限公司 File classification method and device
CN109471944A (en) * 2018-11-12 2019-03-15 中山大学 Training method, device and the readable storage medium storing program for executing of textual classification model
CN110362677A (en) * 2019-05-31 2019-10-22 平安科技（深圳）有限公司 The recognition methods of text data classification and device, storage medium, computer equipment

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112599250A (en) * 2020-12-24 2021-04-02 中国人民解放军总医院第三医学中心 Postoperative data analysis method and device based on deep neural network
WO2022193171A1 (en) * 2021-03-17 2022-09-22 Huawei Cloud Computing Technologies Co., Ltd. System and method for unsupervised multi-model joint reasoning
CN113268596A (en) * 2021-05-24 2021-08-17 康键信息技术（深圳）有限公司 Verification method, device and equipment of department classification model and storage medium
CN113254587A (en) * 2021-05-31 2021-08-13 北京奇艺世纪科技有限公司 Search text recognition method and device, computer equipment and storage medium
CN113254587B (en) * 2021-05-31 2023-10-13 北京奇艺世纪科技有限公司 Search text recognition method and device, computer equipment and storage medium
CN116738343A (en) * 2023-08-08 2023-09-12 云筑信息科技(成都)有限公司 Material data identification method and device for construction industry and electronic equipment
CN116738343B (en) * 2023-08-08 2023-10-20 云筑信息科技(成都)有限公司 Material data identification method and device for construction industry and electronic equipment
CN117235270A (en) * 2023-11-16 2023-12-15 中国人民解放军国防科技大学 Text classification method and device based on belief confusion matrix and computer equipment
CN117235270B (en) * 2023-11-16 2024-02-02 中国人民解放军国防科技大学 Text classification method and device based on belief confusion matrix and computer equipment

Also Published As

Publication number Publication date
CN110362677B (en) 2022-12-27
CN110362677A (en) 2019-10-22

Similar Documents

Publication Publication Date Title
WO2020238053A1 (en) Neural grid model-based text data category recognition method and apparatus, nonvolatile readable storage medium, and computer device
CN110147726B (en) Service quality inspection method and device, storage medium and electronic device
WO2021026805A1 (en) Adversarial example detection method and apparatus, computing device, and computer storage medium
CN113255694B (en) Training image feature extraction model and method and device for extracting image features
CN109145766B (en) Model training method and device, recognition method, electronic device and storage medium
US9053358B2 (en) Learning device for generating a classifier for detection of a target
JP5214760B2 (en) Learning apparatus, method and program
CN105022754B (en) Object classification method and device based on social network
US9613296B1 (en) Selecting a set of exemplar images for use in an automated image object recognition system
WO2020114108A1 (en) Clustering result interpretation method and device
Escalante et al. A naive bayes baseline for early gesture recognition
KR101837262B1 (en) Deep learning type classification method with feature-based weighting
US11935298B2 (en) System and method for predicting formation in sports
De Rosa et al. Online open world recognition
WO2020135756A1 (en) Video segment extraction method, apparatus and device, and computer-readable storage medium
CN111400615A (en) Resource recommendation method, device, equipment and storage medium
CN111062440B (en) Sample selection method, device, equipment and storage medium
CN112668482A (en) Face recognition training method and device, computer equipment and storage medium
WO2020114109A1 (en) Interpretation method and apparatus for embedding result
CN111930885B (en) Text topic extraction method and device and computer equipment
CN107609570B (en) Micro video popularity prediction method based on attribute classification and multi-view feature fusion
US20220254036A1 (en) Interactive Formation Analysis in Sports Utilizing Semi-Supervised Methods
Punyani et al. A comparison study of face, gait and speech features for age estimation
KR102060110B1 (en) Method, apparatus and computer program for classifying object in contents
Heo et al. Curriculum learning for self-supervised speaker verification

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19930931

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19930931

Country of ref document: EP

Kind code of ref document: A1