WO2021119949A1

WO2021119949A1 - Text classification model training method, text classification method and apparatus, and electronic device

Info

Publication number: WO2021119949A1
Application number: PCT/CN2019/125747
Authority: WO
Inventors: 刘园林
Original assignee: 深圳市欢太科技有限公司; Oppo广东移动通信有限公司
Priority date: 2019-12-16
Filing date: 2019-12-16
Publication date: 2021-06-24
Also published as: CN114424186A

Abstract

A text classification model training method, a text classification method and apparatus, and an electronic device. The training method comprises: obtaining a first text sample set (101); inputting the first text sample set to a text classification model to obtain a first prediction result (102); if the first prediction result does not satisfy a preset condition, adjusting the text classification model (105); and inputting a second sample set to the adjusted text classification model until the prediction result of the text classification model satisfies the preset condition (107).

Description

Text classification model training method, text classification method, device and electronic equipment

Technical field

This application relates to the field of data processing technology, and in particular to a text classification model training method, text classification method, device and electronic equipment.

Background technique

With the vigorous development of the Internet and mobile Internet, the number of documents to be analyzed has risen sharply. How to classify texts with different granularities (such as sentences, paragraphs, documents) is of great significance for information discovery, information browsing and analysis. For example, in the content distribution business, the realization of many services needs to rely on fine-grained tags. For example, fine-grained tags can enrich the user portrait tag library, better describe the user's image, and support more fine-grained recommendation of information flow. There are tens of thousands of effective tags that can be discovered in the actual business, and thousands of them are more important. The technology that can mark text with a large-scale multi-level tag collection of the order of thousands is particularly important. .

At present, there is a solution to build its own text classification model for each level of label, and finally to integrate the inference results of multiple models; another solution is to use a neural network to build a multi-level text and multi-label classification model, and get it based on the model. The text category prediction result of the training text; another solution is to input the current text to be classified into multiple trained text classification models, calculate the probability of each layer of text, use the product of the probabilities of each level to infer the last level of label, and use The label relationship between the levels is reversed to obtain the labels of each level.

However, there are several major problems in the existing and model solutions. One is that good prediction results cannot be achieved for a large number of tags; the other is that the model cannot be iteratively learned according to the changes of tags, and the hierarchical relationship of the tags cannot be well performed. Prediction; The third is that the processing of a large number of tags is computationally intensive and time-consuming.

Summary of the invention

The embodiments of the present application provide a text classification model training method, text classification method, device, and electronic equipment, so as to improve the accuracy of classifying multi-level and multi-label text.

In the first aspect, an embodiment of the present application provides a text classification model training method, including:

Obtain the first text sample set;

Inputting the first text sample set into the text classification model to perform text category prediction, so as to obtain a first prediction result corresponding to the first text sample;

Comparing the first prediction result with the real result, and judging whether the first prediction result meets a preset condition;

If the first prediction result does not meet the preset condition, adjusting the text classification model to obtain an adjusted text classification model;

Processing, according to a preset processing manner, the target text in the first text sample set whose first prediction result does not meet the preset condition, to obtain a second text sample set;

The second text sample set is input into the adjusted text classification model to continue training until the prediction result of the text classification model meets the preset condition.

In the second aspect, this application provides a text classification method, including:

Obtain the text set to be classified;

Call a pre-trained text classification model;

Inputting the text set to be classified into the pre-trained text classification model to obtain a classification result of the text to be classified;

The text classification model is a text classification model obtained by using the training method of the text classification model provided in the embodiment of the present application.

In the third aspect, an embodiment of the present application provides a text classification model training device, including:

The first obtaining module is used to obtain the first text sample set;

A prediction module, configured to input the first text sample set into the text classification model for text category prediction, so as to obtain a first prediction result corresponding to the first text sample;

A judging module, configured to compare the first prediction result with the real result, and judge whether the first prediction result meets a preset condition;

An adjustment module, configured to adjust the text classification model to obtain an adjusted text classification model if the first prediction result does not meet the preset condition;

A processing module, configured to process the target text whose first prediction result in the first text sample set does not meet a preset condition according to a preset processing mode, to obtain a second text sample set;

The training module is configured to input the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model meets the preset condition.

In a fourth aspect, an embodiment of the present application provides a text classification device, including:

The second acquisition module is used to acquire a text set to be classified;

The calling module is used to call the pre-trained text classification model;

A classification module, configured to input the text set to be classified into the pre-trained text classification model to obtain a classification result of the text to be classified;

In a fifth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, wherein, when the computer program is executed on a computer, the computer is caused to execute the text classification model training method provided in this embodiment Or text classification method.

In a sixth aspect, an embodiment of the present application provides an electronic device including a memory and a processor, the memory stores a computer program, and the processor invokes the computer program stored in the memory to execute:

Obtain the first text sample set;

Inputting the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model satisfies the preset condition.

In a seventh aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, the memory stores a computer program, and the processor invokes the computer program stored in the memory to execute:

Obtain the text set to be classified;

Call a pre-trained text classification model;

Description of the drawings

The following detailed description of specific implementations of the present application in conjunction with the accompanying drawings will make the technical solutions and other beneficial effects of the present application obvious.

FIG. 1 is a schematic diagram of the first process of a text classification model training method provided by an embodiment of the application.

FIG. 2 is a schematic diagram of the second process of the text classification model training method provided by an embodiment of the application.

FIG. 3 is a schematic diagram of the third process of the text classification model training method provided by an embodiment of the application.

FIG. 4 is a schematic flowchart of a text classification method provided by an embodiment of the application.

Fig. 5 is a schematic structural diagram of a text classification model training method provided by an embodiment of the application.

FIG. 6 is a schematic structural diagram of a text classification device provided by an embodiment of the application.

FIG. 7 is a first structural diagram of an electronic device provided by an embodiment of the present application.

FIG. 8 is a schematic diagram of a second structure of an electronic device provided by an embodiment of the present application.

Detailed ways

The application will be further described in detail below with reference to the drawings and embodiments. It is understandable that the specific embodiments described here are used to explain the application, but not to limit the application. In addition, it should be noted that, for ease of description, the drawings only show a part of the structure related to the present application instead of all of the structure.

The terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments. Unless the context clearly dictates otherwise, the singular forms "a" and "one" used herein are also intended to include the plural. It should also be understood that the terms "including" and/or "comprising" used herein specify the existence of the stated features, integers, steps, operations, units and/or components, and do not exclude the existence or addition of one or more Other features, integers, steps, operations, units, components, and/or combinations thereof.

The embodiments of the present application provide a text classification model training method and text classification method. The text classification model training method and text classification method are applied to electronic devices. Among them, the electronic device can be a smart phone, a tablet computer, a palmtop computer, a notebook computer, or a desktop computer that is equipped with a processor and has processing capabilities.

Please refer to FIG. 1. FIG. 1 is a schematic diagram of a first process of a text classification model training method provided by an embodiment of the present application. The text classification model training method may include the following processes:

101. Obtain a first text sample set.

The first text sample set contains multiple first texts. In the process of obtaining the first text sample set, the text to be processed can be obtained first, and then the text to be processed can be segmented to obtain multiple first texts. Encoding is performed to obtain a plurality of first tags, and the first text and the first tags corresponding to the first text form a first text sample set.

For example, if there are multiple pieces of text in the acquired text to be processed, the text to be processed can be segmented. For example, when browsing articles or news, multiple text content in the browsing page can be captured, and the text content of the text to be processed can be several hundred Words, thousands of words, or tens of thousands of words, these text contents are used as the text to be processed, and multiple first texts can be obtained by performing word segmentation processing on the text to be processed.

In some embodiments, multiple database texts in the database can also be obtained, and then the text to be processed is randomly obtained from the database text, the text to be processed is processed to obtain the target database text, and the target database text and the first text are combined into the first text. Set of text samples.

102. Input the first text sample set into a text classification model to perform text category prediction, so as to obtain a first prediction result corresponding to the first text sample.

After the first text sample set is obtained, the first text sample set is input to the text classification model. The first text classification model can recognize multiple first texts, thereby classifying the first text, and obtaining the first text corresponding to the first text. One prediction result. Among them, each first text has its own corresponding prediction result.

103. Compare the first prediction result with the real result, and determine whether the first prediction result meets a preset condition.

Wherein, after the first prediction result is obtained, it can be judged whether the first prediction result corresponding to the first text meets the preset condition, if it is satisfied, the text classification model has been trained, and step 104 is entered. Able to achieve the expected text classification effect. If the training result does not meet the preset condition, go to step 105.

For example, if the preset condition is to determine whether the first prediction result reaches 80% of the preset result, if the first prediction result reaches 80% of the preset result, it means that the prediction result of the text classification model is accurate; if the first prediction If the result is less than 80% of the preset result, it means that the prediction result of the text classification model is not very accurate, and the text classification model still needs to be trained.

104. If the first prediction result meets the preset condition, the text classification model has been trained.

It is understandable that when the first prediction result meets the preset condition, the training of the text classification model is stopped, and the training of the text classification model has been completed at this time.

In some embodiments, in order to further determine whether the text classification model has been trained, additional text sample sets can be obtained, for example, obtained from a database or randomly obtained through the network, and then input into the text classification model, if the prediction result meets the preset conditions , It means that the text classification model has been trained.

105. If the first prediction result does not meet the preset condition, adjust the text classification model to obtain an adjusted text classification model.

It is understandable that if there is a large difference between the first prediction result and the real result, it may be that the text classification model is not suitable for classifying the first text sample set, and it may also be that the text classification model cannot complete the first set of samples alone. The classification of a text sample set needs to add other models, or delete or add part of the structure in the text classification model, so as to realize the structural adjustment of the text classification model.

In some embodiments, when the text classification model includes a convolutional neural network, the number of convolutional layers or pooling layers of the convolutional neural network can be adjusted to achieve different feature extraction. The final adjustment is The second label classification model.

In some embodiments, if the text classification model includes a neural network, the loss function in the text classification model can be adjusted according to the training result to obtain the second label classification model. Specifically, the loss function can be weighted.

In some embodiments, if a neural network is included in the text classification model, the neural network model parameters can be adjusted according to the training result, the loss function of the neural network, and the preset result. Specifically, the neural network model parameters can be adjusted according to the training result and the loss function of the neural network. And the loss value is obtained from the preset result, and then the adjustment direction of the neural network is determined according to the loss value, and finally the parameters of the neural network are adjusted to obtain the second label classification model.

106. Process the target text whose first prediction result in the first text sample set does not meet the preset condition according to a preset processing mode, to obtain a second text sample set.

In some implementation manners, target texts that do not meet a preset condition in the first text may be obtained, and multiple target texts may be segmented according to a preset length to obtain multiple second texts. For example, one of the first texts is "Yuanmingyuan horse head issued a physical examination report today, with white scale inside". If the preset length is 4 characters, the first text can be divided to obtain "Yuanmingyuan horse head, For the second texts released today, physical examination report, internal attachment, and white gutter, multiple second texts can be combined into a second text sample set.

It should be noted that in actual applications, the preset length can be set to hundreds or thousands of words. For example, when the preset length is 300 words, the first text can be divided according to the preset length of 300. To get multiple second texts.

In some embodiments, it is also possible to obtain target texts that do not meet preset conditions and obtain predictable real text data in the database, and combine the first label and the predictable real text data to form a second text sample set.

107. Input the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model meets the preset condition.

It is understandable that after the second text sample set is obtained, the second text sample set can be input to the adjusted text classification model to continue to predict the second text sample set, and then obtain the second prediction corresponding to the second text As a result, the second prediction result is compared with the actual result corresponding to the second text, and it is judged whether the second prediction result meets the preset condition. If the second prediction result meets the preset condition, the training of the text classification model is stopped. The text classification model can already accurately classify the input files.

If the second prediction result still does not meet the preset condition, then go to step 105 to continue training the text classification model. When the prediction result of the text classification model meets the preset condition, it means that the text classification model has been trained. By changing the data of the input text sample set and adjusting the text classification model, the trained text classification model can accurately classify and recognize multi-label text.

In the process of training the text classification model, by adjusting and optimizing the input text samples, and at the same time adjusting and optimizing the text classification model, the text classification model can accurately achieve the text classification effect in the process of predicting multi-label text. .

It can be seen from the above that by obtaining the first sample set, the first sample set is input into the text classification model for text category prediction; the obtained first prediction result is compared with the real result, and it is judged whether the first prediction result satisfies the prediction. Set conditions; if the preset conditions are not met, the text classification model is adjusted to obtain the adjusted text classification model; according to the preset processing method, the first prediction result in the first text sample set does not meet the preset conditions for the target text Processing is performed to obtain a second text sample set; the second text sample set is input into the adjusted text classification model to continue training until the prediction result of the text classification model meets the preset condition. The trained text classification model can classify multi-label texts and at the same time improve the accuracy of text classification.

Please continue to refer to FIG. 2. FIG. 2 is a schematic diagram of the second process of the training method of the text classification model provided by the embodiment of the present application. Before obtaining the first text sample set, the training method of the text classification model includes the following processes:

201. Obtain a text to be processed, and perform word segmentation processing on the text to be processed to obtain multiple first texts.

In an embodiment, the acquired text to be processed may be words, words, paragraphs, articles, etc., and the number of words in the text to be processed may be hundreds, thousands, or tens of thousands. After the text to be processed is acquired, it can be processed. Text segmentation processing, for example, you can divide the words in the paragraph, or divide the sentences in the article, and so on. You can also perform word segmentation processing on the text to be processed through the word segmenter. After the word segmentation process is performed on the text to be processed, the text after the word segmentation is the first text, and the first text includes multiple ones.

It should be noted that the text to be processed can be processed according to the model specifically adopted by the text classification model. For example, when the text classification model is a model that does not require word segmentation, such as the BERT model, the text does not need to be segmented. For processing, you can directly input multiple texts to be processed as the first text sample into the text classification model; when the text classification model is a model that requires word segmentation, such as a convolutional neural network model, it is required at this time Perform word segmentation processing on the text to be processed, thereby obtaining multiple first text samples.

202. Encode the first text to obtain a corresponding first label.

It is understandable that after obtaining multiple first texts, each first text needs to be coded, so that each first text has its own corresponding label, so that the multiple first texts can be distinguished.

203. Obtain the target first label whose label level is lower than the lowest label level in the preset label levels.

When the number of the first text is large, the tags corresponding to the first text can be integrated, and multiple tags can be spliced and integrated into one tag. However, when there are many first labels of the same type, the target first labels below the lowest label level can be obtained, and these target first labels are processed in subsequent steps.

It is understandable that the actual business data is often unbalanced, and the large number of fine-grained tags requires some upper-level tag collection processing. If the number of business data corresponding to fine-grained tags is 0 or almost 0, the prediction results for these tags will actually affect the overall prediction results. This part of the tags can be called invalid tags or low data volume tags. Collect to its superior label and no longer process it.

204. Classify the target first label into the first label of the lowest label level.

In some embodiments, when there are many first labels of the same type, labels below the lowest label level can be obtained. For example, the preset label level is 5, but there are 7 first labels of the same type. The seven first labels are arranged from high to low as A, B, C, D, E, F, G. When the default label level is 5, label F and label G are lower than the lowest label level. At this time, the label F and the label G can be determined to be the target first label, and the target first label F and the target first label G can be classified into the first label E of the lowest label level. The tail tag is processed in this way.

205. Integrate multiple first tags of the same type according to a preset tag level to obtain a first text sample set.

In some embodiments, tags of the same type can be integrated according to a preset tag hierarchy. For example, tags of the same type can be obtained from multiple first tags, where tags of the same type are A, B, C, D, E The five first labels are integrated according to the preset label hierarchy. The corresponding texts of the five first labels of A, B, C, D, and E are "current affairs", "domestic current affairs", and " "Mainland Current Affairs", "Policies", "San Nong", if the default label level is 5 levels, and the first label needs to be spliced from high level to low level, then the combined label will be "ABCDE". The corresponding text of the label is "Current Affairs-Domestic Current Affairs-Mainland Current Affairs-Policies-Agriculture, Rural Areas and Rural Areas". Similarly, multiple first labels of the same type can be integrated to form multiple integrated labels to reduce the number of labels.

Tag coding may be performed on multiple integrated tags. For example, onehot encoding may be used to code the integrated tags, and finally the integrated tags and the text corresponding to the integrated tags form the first text sample set.

Please continue to refer to FIG. 3, which is a schematic diagram of the third process of the training method of the text classification model provided by the embodiment of the present application. The training method of the text classification model may include the following processes:

301. Obtain a first set of text samples.

Before obtaining the first text sample set, invalid characters in the text to be processed can be filtered and deleted to ensure the authenticity of the text to be processed, and then the filtered text to be processed is segmented to obtain multiple first texts. Multiple first texts are text-encoded to obtain multiple first tags. Considering the large number of first tags, tags of the same type can be integrated to form an integrated tag, and finally multiple integrated tags are tag-encoded, and finally the integrated tags The text corresponding to the integrated label forms a first text sample set.

302. Input the first text sample set into a text classification model to perform text category prediction, so as to obtain a first prediction result corresponding to the first text sample.

After the first text sample set is obtained, the first text sample set is input to the text classification model. The first text classification model can recognize multiple first texts, thereby classifying the first text, and obtaining the first text corresponding to the first text. One prediction result. Among them, each first text has its own corresponding prediction result, and the sigmoid function can be used to calculate the accuracy value of the first prediction result.

303. When the first prediction result does not meet the preset condition, adjust the text classification model.

After the first prediction result is obtained, the first prediction result can be compared with the actual result corresponding to the first text to determine whether the first prediction result meets a preset condition. For example, it can be determined whether the accuracy of the first prediction result reaches The preset accuracy rate, if the accuracy rate of the first prediction result does not reach the preset accuracy rate, the text classification model is adjusted.

In some embodiments, the parameters of the text classification model can be adjusted. For example, the loss function of the text classification model can be obtained, and the first prediction result and the real result can be input into the loss function to obtain the loss value, which is adjusted according to the target loss value. The parameters of the text classification model are adjusted so that the loss value of the text classification model is less than or equal to the target loss value.

In some embodiments, the text classification model can also be concatenated with a preset model to adjust the text classification model. For example, the text classification model before the adjustment uses the BERT model, and the BERT model can also be concatenated. Neural network model to form a new text classification model.

In some embodiments, the structure of the text classification model can be adjusted. For example, the text classification model uses an integrated model composed of a BERT model and a convolutional neural network model. In this case, the convolutional neural network model can be multiplied. The scale convolution kernel may implement multiple different pooling operations, and the number of layers of the convolutional neural network can also be changed.

304. Obtain preset parameters of the text classification model, and set the preset parameters in the adjusted model.

It is understandable that the settings of some parameters in the text classification model before the adjustment are correct, and there is no need to adjust them. These parameters that do not need to be adjusted are preset parameters. During the adjustment of the text classification model, you can Obtain the preset parameters of the text classification model before adjustment, so that the text classification model can perform transfer learning based on these preset parameters to ensure that the training time and number of training are reduced during the training of the text classification model, and based on these preset parameters. Setting parameters can continue to adjust the text classification model, and the finally adjusted text classification model can continue to be used for the next training until the prediction result output by the text classification model meets the preset conditions.

305. Perform text segmentation on the target text according to the preset length to obtain multiple segmented texts.

It is understandable that there must be some texts in the first text sample set that cannot be predicted by the text classification model. At this time, it is necessary to perform the target text according to the preset length for the target text whose first prediction result does not meet the preset conditions. Split to get multiple split texts.

In some embodiments, the target tags are integrated through a preset label level to obtain the tail tags of the target tags, and some of the tail tags of the target tags can be selected to be included in the target tags of the lowest tag level.

In some embodiments, it is also possible to select unlabeled text in the database as input samples, and then input these unlabeled texts into the text classification model before adjustment, and select the text corresponding to the label whose prediction probability is greater than the preset threshold as the high Confidence text, the high-confidence text and the target text are used together as a new text sample set, that is, the second text sample set.

306. Encode multiple segmented texts to obtain a second text sample set.

In some embodiments, the segmented text is encoded to obtain the target label, the target label is integrated through the preset label level, and the tail label of the target label is obtained. The tail labels of some target labels can be selected to be included in the target label of the lowest label level. In the database, high-confidence text and target text can be selected from the database and combined into a second text sample set.

307. Input the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model meets the prediction condition.

If the second prediction result still does not meet the preset condition, repeat the above steps to continue training the text classification model until the prediction result of the text classification model meets the preset condition, indicating that the text classification model has been trained. By changing the data of the input text sample set and adjusting the text classification model, the trained text classification model can accurately classify and recognize multi-label text.

It can be seen from the above that, in the embodiment of the present application, the text category prediction is performed by inputting the first text sample set into the text classification model to obtain the first prediction result corresponding to the first text sample; if the first prediction result does not meet the preset condition In this case, adjust the text classification model; obtain the preset parameters of the text classification model, and set the preset parameters in the adjusted model; perform text segmentation on the target text according to the preset length to obtain multiple segmented texts; A plurality of segmented texts are encoded to obtain a second text sample set; the second text sample set is input into the adjusted text classification model to continue training until the prediction result of the text classification model meets the prediction condition. Thus, a trained text classification model is obtained. The trained text classification model can classify multi-level and multi-label text, and the prediction accuracy of the text classification model can also achieve the expected effect.

Please continue to refer to FIG. 4, which is a text classification method provided by an embodiment of the present application. The method includes the following processes:

401. Obtain a text set to be classified.

The invalid characters in the text to be classified can be filtered and deleted to ensure the authenticity of the text to be classified. The text encoding of multiple texts to be classified can obtain multiple text labels to be classified. Considering the large number of texts to be classified, you can Tag integration is performed on tags of the same type to form a text-to-be-categorized integrated label, and finally a plurality of text-to-be-categorized integration tags are tag-encoded, and finally the text-to-be-categorized integration tags and text to be classified form a text-to-be-categorized set.

402. Call a pre-trained text classification model.

It should be noted that after the text classification model is trained, the text classification model can accurately predict most of the text. However, considering the accuracy of text classification, the text classification sub-model can also be concatenated after the text classification model. The classification sub-model can adjust the accuracy of text classification according to actual needs.

403. Input the text set to be classified into a pre-trained text classification model to obtain a classification result of the text to be classified.

It is understandable that after inputting the text set to be classified into the pre-trained text classification model, the accuracy of the classification result of the text to be classified is not very high, but it has reached a high degree of accuracy and can be used for multi-level and multi-label The text to be classified is classified.

404. Obtain the lowest-level label corresponding to the text to be classified, and continue text classification on the text corresponding to the lowest label-level label.

In some embodiments, the lowest-level tail label of the text label to be classified can be input into the text classification sub-model, so as to realize the text classification corresponding to the text corresponding to the lowest-level tail label of the text label to be classified, thereby achieving more accurate text classification. effect.

It can be seen from the above that in this embodiment of the application, the text set to be classified is obtained; the pre-trained text classification model is invoked; the text set to be classified is input to the pre-trained text classification model to obtain the classification result of the text to be classified; The text corresponding to the lowest-level label, and the text corresponding to the lowest label-level label continues to be text categorized. In this way, the classification of multi-level and multi-label text is realized, and the accuracy of text classification is improved.

Please refer to FIG. 5. FIG. 5 is a schematic structural diagram of a text classification model training device provided by an embodiment of the present application. The text classification model training device 500 includes: a first acquisition module 510, a prediction module 520, a judgment module 530, and an adjustment module 540, a processing module 550, and a training module 560.

The first obtaining module 510 is configured to obtain a first text sample set;

The prediction module 520 is configured to input the first text sample set into the text classification model to perform text category prediction, so as to obtain a first prediction result corresponding to the first text sample;

The judging module 530 is configured to compare the first prediction result with the real result, and determine whether the first prediction result meets a preset condition;

The adjustment module 540 is configured to adjust the text classification model to obtain an adjusted text classification model if the first prediction result does not meet the preset condition;

The processing module 550 is configured to process the target text in the first text sample set whose first prediction result does not meet the preset condition according to a preset processing mode, to obtain a second text sample set;

The training module 560 is configured to input the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model meets the preset condition.

In some embodiments, the first obtaining module 510 is specifically configured to obtain a text to be processed, perform word segmentation processing on the text to be processed to obtain a plurality of the first texts; and encode the first text to obtain all the texts. A first label corresponding to the first text; and a plurality of the first labels are integrated according to a preset label level to obtain the first text sample set.

In some embodiments, the first obtaining module 510 is specifically configured to obtain a target first label whose label level is lower than the lowest label level among the preset label levels; and classify the target first label into the lowest label Among the first tags in the hierarchy; multiple first tags of the same type are integrated according to the preset tag hierarchy to obtain a first text sample set.

In some embodiments, the adjustment module 540 is specifically configured to input the first prediction result and the real result into the loss function of the text classification model to obtain a loss value; The parameters of the text classification model are adjusted.

In some embodiments, the adjustment module 540 is specifically configured to adjust the network structure of the text classification model according to the first prediction result and the real result; connect the adjusted text classification model in series with a preset model .

In some embodiments, the adjustment module 540 is specifically configured to obtain the loss function of the text classification model; and perform weighting processing on the loss function according to the first prediction result and the real result.

In some embodiments, the adjustment module 540 is specifically configured to obtain preset parameters of the text classification model; and set the preset parameters in the adjusted text classification model.

It can be seen from the above that, in the embodiment of the present application, the first sample set is obtained by the first obtaining module 510, and the prediction module 520 inputs the first sample set into the text classification model for text category prediction; the judgment module 530 will obtain the first sample set. 1. The prediction result is compared with the actual result to determine whether the first prediction result meets the preset condition; the adjustment module 540 adjusts the text classification model when the first prediction result does not meet the preset condition to obtain the adjusted text classification model The processing module 550 processes the target text in the first text sample set whose first prediction result does not meet the preset conditions according to a preset processing method to obtain a second text sample set; the training module 560 inputs the second text sample set to the adjustment The subsequent text classification model continues to train until the prediction result of the text classification model meets the preset conditions. The trained text classification model can accurately classify multi-level and multi-label text.

Please continue to refer to FIG. 6, which is a schematic structural diagram of a text classification device provided by an embodiment of the present application. The text classification device 600 specifically includes: a second acquisition module 610, a calling module 620, and a classification module 630.

The second obtaining module 610 is used to obtain a target text set;

The calling module 620 is used to call a pre-trained text classification model;

The classification module 630 is configured to input the target text set into the pre-trained text classification model to obtain a classification result of the target text;

The text classification model is a text classification model obtained by the training method of the text classification model provided in the embodiment of the application.

The classification module 630 is also used to input the lowest-level tail label of the text label to be classified into the text classification sub-model, so as to realize the text classification of the text corresponding to the lowest-level tail label of the text label to be classified, thereby achieving more accurate text Classification effect.

It can be seen from the above that in this embodiment of the present application, the second acquisition module 610 acquires the text set to be classified; the calling module 620 calls the pre-trained text classification model; the classification module 630 inputs the text set to be classified into the pre-trained text classification model to The classification result of the text to be classified is obtained; the classification module 630 may also obtain the lowest-level label corresponding to the text to be classified, and continue to perform text classification on the text corresponding to the label of the lowest label level. In this way, the classification of multi-level and multi-label text is realized, and the accuracy of text classification is improved.

It should be noted that the image attribute recognition device provided in this embodiment of the application belongs to the same concept as the image attribute recognition method in the above embodiment, and the image attribute that can be run on the image attribute recognition device is any one provided in the method embodiment. For the specific implementation process of the method, please refer to the image processing method embodiment, which will not be repeated here.

The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the stored computer program is executed on a computer, the computer executes the network model training method or image provided in the embodiment of the present application.的处理方法。 Treatment methods.

Among them, the storage medium may be a magnetic disk, an optical disc, a read only memory (Read Only Memory, ROM,), or a random access device (Random Access Memory, RAM), etc.

An embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory. The processor is configured to execute the computer program stored in the memory by calling the computer program stored in the memory. Example provides the training method of the network model or the image attribute recognition method.

For example, the above-mentioned electronic device may be a mobile terminal such as a tablet computer or a smart phone. Please refer to FIG. 7. FIG. 7 is a schematic diagram of the first structure of an electronic device provided by an embodiment of this application.

The electronic device 700 may include components such as a memory 701 and a processor 702. Those skilled in the art can understand that the structure of the electronic device shown in FIG. 7 does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure, or a combination of certain components, or different component arrangements.

The memory 701 may be used to store software programs and modules. The processor 702 executes various functional applications and data processing by running the computer programs and modules stored in the memory 701. The memory 701 may mainly include a storage program area and a storage data area. The storage program area may store an operating system, a computer program required by at least one function (such as a sound playback function, an image playback function, etc.), etc.; Data created by the use of electronic equipment, etc.

The processor 702 is the control center of the electronic device. It uses various interfaces and lines to connect the various parts of the entire electronic device. It executes the electronic device by running or executing the application program stored in the memory 701 and calling the data stored in the memory 701. The various functions and processing data of the electronic equipment can be used to monitor the electronic equipment as a whole.

In addition, the memory 701 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 701 may further include a memory controller to provide the processor 702 with access to the memory 701.

In this embodiment, the processor 702 in the electronic device will load the executable code corresponding to the process of one or more application programs into the memory 701 according to the following instructions, and the processor 702 will run and store the executable code in the memory 701. The application program in 701, so as to realize the process:

Obtain the first text sample set;

In some implementation manners, when the processor 702 adjusts the text classification model, it may execute:

Inputting the first prediction result and the real result into the loss function of the text classification model to obtain a loss value;

The parameters of the text classification model are adjusted according to the loss value.

Adjusting the network structure of the text classification model according to the first prediction result and the real result;

A preset model is connected in series with the adjusted text classification model.

Acquiring the loss function of the text classification model;

Perform weighting processing on the loss function according to the first prediction result and the real result.

Specifically, when the processor 702 adjusts the text classification model, it may execute:

Acquiring preset parameters of the text classification model;

The preset parameters are set in the adjusted text classification model.

In some implementation manners, when the processor 702 processes the target text whose first prediction result in the first text sample set does not meet a preset condition according to a preset processing manner, it may execute:

Performing text segmentation on the target text according to a preset length to obtain multiple segmented texts;

Encoding the segmented text to obtain the second text sample set.

In some implementation manners, before acquiring the first text sample set, the processor 702 may execute:

Acquiring a text to be processed, and performing word segmentation processing on the text to be processed to obtain a plurality of the first texts;

Encoding the first text to obtain a first label corresponding to the first text;

Integrating a plurality of the first tags according to a preset tag level to obtain the first text sample set.

In some implementation manners, when the processor 702 executes to integrate a plurality of the first tags according to a preset tag level to obtain the first text sample set, it may execute:

Acquiring the target first label whose label level is lower than the lowest label level in the preset label levels;

Classify the target first label into the first label of the lowest label level;

The multiple first tags of the same type are integrated according to the preset tag level to obtain a first text sample set.

Obtain the text set to be classified;

Call a pre-trained text classification model;

Please refer to FIG. 8. FIG. 8 is a schematic diagram of a second structure of an electronic device provided by an embodiment of the application. The difference from the electronic device shown in FIG. 7 is that the electronic device further includes a camera component 703, a radio frequency circuit 704, an audio circuit 705, and Power supply 706. Among them, the display 703, the radio frequency circuit 704, the audio circuit 705, and the power supply 706 are electrically connected to the processor 702, respectively.

The display 703 may be used to display information input by the user or information provided to the user, and various graphical user interfaces. These graphical user interfaces may be composed of graphics, text, icons, videos, and any combination thereof. The display 703 may include a display panel. In some embodiments, the display panel may be configured in the form of a liquid crystal display (LCD) or an organic light-emitting diode (OLED).

The radio frequency circuit 704 may be used to transmit and receive radio frequency signals to establish wireless communication with network equipment or other electronic equipment through wireless communication, and to transmit and receive signals with the network equipment or other electronic equipment.

The audio circuit 705 may be used to provide an audio interface between the user and the electronic device through a speaker or a microphone.

The power supply 706 can be used to power various components of the electronic device 600. In some embodiments, the power supply 706 may be logically connected to the processor 702 through a power management system, so that functions such as charging, discharging, and power consumption management can be managed through the power management system.

Although not shown in FIG. 8, the electronic device 600 may also include a camera component, a Bluetooth module, etc. The camera component may include an image processing circuit, which may be implemented by hardware and/or software components, and may include defining image signal processing (Image Signal Processing) various processing units of the pipeline. The image processing circuit may at least include: multiple cameras, an image signal processor (Image Signal Processor, ISP processor), a control logic, an image memory, a display, and the like. Each camera can include at least one or more lenses and image sensors. The image sensor may include a color filter array (such as a Bayer filter). The image sensor can obtain the light intensity and wavelength information captured by each imaging pixel of the image sensor, and provide a set of raw image data that can be processed by the image signal processor.

In the above embodiments, the description of each embodiment has its own focus. For parts that are not detailed in an embodiment, please refer to the detailed description of the training method/text classification method of the text classification model above. Go into details again.

The training device/text classification device of the text classification model provided by the embodiment of the application belongs to the same concept as the training method/text classification method of the text classification model in the above embodiments. Any method provided in the training method/text classification method embodiment of the text classification model can be run on the classification device. For the specific implementation process, please refer to the training method/image processing method embodiment of the network model. Go into details again.

It should be noted that for the training method/text classification method of the text classification model described in the embodiment of the present application, a person of ordinary skill in the art can understand all of the training method/image processing method of the network model described in the embodiment of the present application. Or part of the process can be accomplished by controlling the relevant hardware through a computer program. The computer program can be stored in a computer readable storage medium, such as in a memory, and executed by at least one processor. May include the flow of the embodiment of the training method of the network model/the image processing method. Wherein, the storage medium may be a magnetic disk, an optical disc, a read only memory (ROM, Read Only Memory), a random access memory (RAM, Random Access Memory), etc.

For the network model training device/image processing device of the embodiment of the present application, its functional modules can be integrated into one processing chip, or each module can exist alone physically, or two or more modules can be used. Integrated in a module. The above-mentioned integrated modules can be implemented in the form of hardware or software functional modules. If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium, such as a read-only memory, a magnetic disk or an optical disk, etc. .

The text classification model training method, text classification method, device, and electronic equipment provided by the embodiments of the application are described in detail above. Specific examples are used in this article to illustrate the principles and implementation of the application. The description of the above embodiments It is only used to help understand the methods and core ideas of this application; at the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation and scope of application. In summary, this The content of the description should not be construed as a limitation on this application.

Claims

A method for training a text classification model, wherein the method includes:

Obtain the first text sample set;

Inputting the first text sample set into the text classification model to perform text category prediction, so as to obtain a first prediction result corresponding to the first text sample;

Comparing the first prediction result with the real result, and judging whether the first prediction result meets a preset condition;

If the first prediction result does not meet the preset condition, adjusting the text classification model to obtain an adjusted text classification model;

Processing, according to a preset processing manner, the target text in the first text sample set whose first prediction result does not meet the preset condition, to obtain a second text sample set;

Inputting the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model satisfies the preset condition.
The method for training a text classification model according to claim 1, wherein said adjusting said text classification model comprises:

Inputting the first prediction result and the real result into the loss function of the text classification model to obtain a loss value;

The parameters of the text classification model are adjusted according to the loss value.
The method for training a text classification model according to claim 1, wherein said adjusting said text classification model comprises:

Adjusting the network structure of the text classification model according to the first prediction result and the real result;

A preset model is connected in series with the adjusted text classification model.
The method for training a text classification model according to claim 1, wherein said adjusting said text classification model comprises:

Acquiring the loss function of the text classification model;

Perform weighting processing on the loss function according to the first prediction result and the real result.
The method for training a text classification model according to any one of claims 1 to 4, wherein said adjusting the text classification model comprises:

Acquiring preset parameters of the text classification model;

The preset parameters are set in the adjusted text classification model.
The training method of a text classification model according to any one of claims 1 to 4, wherein the first prediction result in the first text sample set does not meet a preset condition for the target text according to a preset processing mode Processing, including:

Performing text segmentation on the target text according to a preset length to obtain multiple segmented texts;

Encoding the segmented text to obtain the second text sample set.
The method for training a text classification model according to any one of claims 1 to 4, wherein, before the obtaining the first text sample set, the method further comprises:

Acquiring a text to be processed, and performing word segmentation processing on the text to be processed to obtain a plurality of the first text samples;

Encoding the first text sample to obtain a first label corresponding to the first text sample;

Integrating a plurality of the first tags according to a preset tag level to obtain the first text sample set.
8. The method for training a text classification model according to claim 7, wherein said integrating a plurality of said first labels according to a preset label level to obtain said first text sample set comprises:

Acquiring the target first label whose label level is lower than the lowest label level in the preset label levels;

Classify the target first label into the first label of the lowest label level;

The multiple first tags of the same type are integrated according to the preset tag level to obtain a first text sample set.
A text classification method, which includes:

Obtain the text set to be classified;

Call a pre-trained text classification model;

Inputting the text set to be classified into the pre-trained text classification model to obtain a classification result of the text to be classified;

The text classification model is a text classification model obtained by using the text classification model training method according to any one of claims 1 to 8.
A training device for a text classification model, which includes:

The first obtaining module is used to obtain the first text sample set;

A prediction module, configured to input the first text sample set into the text classification model for text category prediction, so as to obtain a first prediction result corresponding to the first text sample;

A judging module, configured to compare the first prediction result with the real result, and judge whether the first prediction result meets a preset condition;

An adjustment module, configured to adjust the text classification model to obtain an adjusted text classification model if the first prediction result does not meet the preset condition;

A processing module, configured to process the target text whose first prediction result in the first text sample set does not meet a preset condition according to a preset processing mode, to obtain a second text sample set;

The training module is configured to input the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model meets the preset condition.
A text classification device, which includes:

The second acquisition module is used to acquire a text set to be classified;

The calling module is used to call the pre-trained text classification model;

A classification module, configured to input the text set to be classified into the pre-trained text classification model to obtain a classification result of the text to be classified;

The text classification model is a text classification model obtained by using the text classification model training method according to any one of claims 1 to 8.
A storage medium, wherein a computer program is stored in the storage medium, and when the computer program is run on a computer, the computer is caused to execute the method for training a text classification model according to any one of claims 1 to 8 Or the text classification method of claim 9.
An electronic device, wherein the electronic device includes a processor and a memory, and a computer program is stored in the memory, and the processor is configured to execute:

Obtain the first text sample set;

Inputting the first text sample set into the text classification model to perform text category prediction, so as to obtain a first prediction result corresponding to the first text sample;

Comparing the first prediction result with the real result, and judging whether the first prediction result meets a preset condition;

If the first prediction result does not meet the preset condition, adjusting the text classification model to obtain an adjusted text classification model;

Processing, according to a preset processing manner, the target text in the first text sample set whose first prediction result does not meet the preset condition, to obtain a second text sample set;

Inputting the second text sample set into the adjusted text classification model to continue training until the prediction result of the text classification model satisfies the preset condition.
The electronic device according to claim 13, wherein the processor is configured to execute:

Inputting the first prediction result and the real result into the loss function of the text classification model to obtain a loss value;

The parameters of the text classification model are adjusted according to the loss value.
The electronic device according to claim 13, wherein the processor is configured to execute:

Adjusting the network structure of the text classification model according to the first prediction result and the real result;

A preset model is connected in series with the adjusted text classification model.
The electronic device according to any one of claims 13 to 15, wherein the processor is configured to execute:

Acquiring preset parameters of the text classification model;

The preset parameters are set in the adjusted text classification model.
The electronic device according to any one of claims 13 to 15, wherein the processor is configured to execute:

Performing text segmentation on the target text according to a preset length to obtain multiple segmented texts;

Encoding the segmented text to obtain the second text sample set.
The electronic device according to any one of claims 13 to 15, wherein the processor is configured to execute:

Acquiring a text to be processed, and performing word segmentation processing on the text to be processed to obtain a plurality of the first texts;

Encoding the first text to obtain a first label corresponding to the first text;

Integrating a plurality of the first tags according to a preset tag level to obtain the first text sample set.
The electronic device according to any one of claims 13 to 15, wherein the processor is configured to execute:

Acquiring a text to be processed, and performing word segmentation processing on the text to be processed to obtain a plurality of the first texts;

Encoding the first text to obtain a first label corresponding to the first text;

Integrating a plurality of the first tags according to a preset tag level to obtain the first text sample set.
An electronic device, wherein the electronic device includes a processor and a memory, and a computer program is stored in the memory, and the processor is configured to execute:

Obtain the text set to be classified;

Call a pre-trained text classification model;

Inputting the text set to be classified into the pre-trained text classification model to obtain a classification result of the text to be classified;

The text classification model is a text classification model obtained by using the text classification model training method according to any one of claims 1 to 8.