CN111104514B - Training method and device for document tag model - Google Patents

Abstract

The application discloses a training method and device for a document tag model, and relates to the technical field of document tag prediction. The specific implementation scheme is as follows: obtaining a pre-trained document tag model, wherein the document tag model is obtained by pre-training universal training data of each application scene; acquiring scene training data of an application scene to be applied, wherein the scene training data comprises: a plurality of documents and corresponding tag information under an application scene to be applied; obtaining a sub-model related to an application scene to be applied in a document tag model; training the sub-model by using scene training data to obtain a trained document tag model, so that training data required for training the document tag model in an application scene to be applied can be reduced, and training cost is reduced under the condition of ensuring the accuracy of the document tag model.

Description

Training method and device for document labeling model

technical field

The present application relates to the technical field of data processing, in particular to the technical field of document label prediction, and in particular to a training method and device for a document label model.

Background technique

Currently, label prediction techniques for documents are an important work in document content understanding. For the new document label prediction scenario, there are two main solutions. One is to train a general document label model: the difference between each scene is not considered when training the model, and the general document label model is used in all scenarios. The other is to train the document labeling model separately: prepare the training data for the new scene separately for training.

In the first method, the trained model lacks scene or domain specificity, and the prediction accuracy in a single scene is low. In the second method, a large amount of training data needs to be prepared, and the training cost is high.

Contents of the invention

This application proposes a training method and device for a document labeling model. According to the scene training data of the application scene to be applied, the sub-models related to the application scene to be applied in the pre-trained document labeling model are trained, so as to ensure On the premise of the accuracy of the document labeling model, the training cost of the document labeling model in the applicable application scenario is reduced.

An embodiment of the present application proposes a training method for a document tagging model, including:

Obtaining a pre-trained document label model, which is obtained by pre-training the general training data of each application scenario;

Acquiring scene training data of the application scene to be applied, the scene training data including: multiple documents and corresponding label information under the application scene to be applied;

Acquiring sub-models related to the application scenario to be applied in the document labeling model;

The sub-model is trained by using the scene training data to obtain a trained document label model.

In an embodiment of the present application, the document labeling model includes: a preprocessing layer, a candidate recall layer, a rough sorting layer and a fine sorting layer;

The candidate recall layer includes: a parallel keyword recall sub-model, a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model;

The rough sorting layer includes: a parallel rule sub-model and a semantic matching sub-model;

The sub-models related to the applicable application scenario include: a semantic matching sub-model, and any one or more of the following sub-models: a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model.

In one embodiment of the present application, when the sub-models related to the application scenario to be applied include: semantic matching sub-model, multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model, the Using the scene training data to train the sub-model to obtain a trained document label model, including:

For each document in the scene training data, input the document into the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model respectively, and merge the output results to obtain candidate label results;

Input the document and the candidate label result into the semantic matching sub-model, and obtain the correlation between the document and each candidate label in the candidate label result;

According to the correlation between the document and each candidate label in the candidate label result, and the label information corresponding to the document, the semantic matching sub-model, the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model The coefficients of the model are adjusted to obtain a trained document label model.

In one embodiment of the present application, the scene training data further includes: a label set, and the label set includes: labels that can be predicted by the document label model, so that the document label model combines the label set with the documents in the scene training data. label prediction.

In one embodiment of the present application, the training of the sub-model using the scene training data, before obtaining the trained document label model, further includes:

Initialize the coefficients of the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model in the document label model.

The training method of the document label model in the embodiment of the present application is obtained by obtaining a pre-trained document label model, and the document label model is obtained by pre-training with general training data of each application scenario; the scene training data of the application scene to be applied is obtained, and the scene The training data includes: multiple documents in the application scenario to be applied and the corresponding label information; obtain the sub-model related to the application scenario to be applied in the document label model; use the scene training data to train the sub-model to obtain the trained The document labeling model can reduce the training data required for training the document labeling model in the applicable application scenarios, and reduce the training cost while ensuring the accuracy of the document labeling model.

Another embodiment of the present application proposes a training device for a document labeling model, including:

An acquisition module, configured to acquire a pre-trained document label model, which is obtained by pre-training the general training data of each application scenario;

The obtaining module is also used to obtain scene training data of the application scene to be applied, and the scene training data includes: multiple documents and corresponding label information under the application scene to be applied;

The obtaining module is also used to obtain a sub-model related to the application scenario to be applied in the document label model;

A training module, configured to use the scene training data to train the sub-models to obtain a trained document labeling model.

In an embodiment of the present application, the document labeling model includes: a preprocessing layer, a candidate recall layer, a rough sorting layer and a fine sorting layer;

The candidate recall layer includes: a parallel keyword recall sub-model, a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model;

The rough sorting layer includes: a parallel rule sub-model and a semantic matching sub-model;

The sub-models related to the applicable application scenario include: a semantic matching sub-model, and any one or more of the following sub-models: a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model.

In one embodiment of the present application, when the sub-models related to the application scenario to be applied include: semantic matching sub-model, multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model, the The training module is specifically used for,

For each document in the scene training data, input the document into the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model respectively, and merge the output results to obtain candidate label results;

Input the document and the candidate label result into the semantic matching sub-model, and obtain the correlation between the document and each candidate label in the candidate label result;

According to the correlation between the document and each candidate label in the candidate label result, and the label information corresponding to the document, the semantic matching sub-model, the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model The coefficients of the model are adjusted to obtain a trained document label model.

In one embodiment of the present application, the scene training data further includes: a label set, and the label set includes: labels that can be predicted by the document label model, so that the document label model combines the label set with the documents in the scene training data. label prediction.

In one embodiment of the present application, the device further includes: an initialization module, configured to initialize the coefficients of the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model in the document label model operate.

The training device for the document label model in the embodiment of the present application is obtained by obtaining a pre-trained document label model, and the document label model is obtained by pre-training with general training data of each application scenario; the scene training data of the application scene to be applied is obtained, and the scene The training data includes: multiple documents in the application scenario to be applied and the corresponding label information; obtain the sub-model related to the application scenario to be applied in the document label model; use the scene training data to train the sub-model to obtain the trained The document labeling model can reduce the training data required for training the document labeling model in the applicable application scenarios, and reduce the training cost while ensuring the accuracy of the document labeling model.

Another embodiment of the present application provides an electronic device, including: at least one processor; and a memory connected in communication with the at least one processor; wherein, the memory stores information that can be executed by the at least one processor. instructions, the instructions are executed by the at least one processor, so that the at least one processor can execute the method for training a document label model according to the embodiment of the present application.

Another embodiment of the present application provides a non-transitory computer-readable storage medium storing computer instructions, the computer instructions are used to enable the computer to execute the method for training the document labeling model of the embodiment of the present application.

Other effects of the above optional manner will be described below in conjunction with specific embodiments.

Description of drawings

The accompanying drawings are used to better understand the solution, and do not constitute a limitation to the application. in:

Fig. 1 is a schematic diagram according to the first embodiment of the present application;

Fig. 2 is a schematic diagram of the document labeling model structure.

Fig. 3 is a schematic diagram according to the second embodiment of the present application;

Fig. 4 is a schematic diagram according to a third embodiment of the present application;

Fig. 5 is a block diagram of an electronic device used to implement the training method of the document labeling model of the embodiment of the present application;

Detailed ways

Exemplary embodiments of the present application are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present application to facilitate understanding, and they should be regarded as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The following describes the method and device for training the document labeling model in the embodiments of the present application with reference to the accompanying drawings.

Fig. 1 is a schematic diagram according to the first embodiment of the present application. Wherein, it should be noted that the execution subject of the training method of the document label model provided in this embodiment is the training device of the document label model, which can be realized by software and/or hardware, and which can be configured in a terminal device or In the server, this embodiment does not specifically limit it.

As shown in Figure 1, the training method of this document label model can comprise:

In step 101, a pre-trained document labeling model is acquired, and the document labeling model is pre-trained using general training data of various application scenarios.

In this application, a schematic diagram of the structure of the document labeling model can be shown in FIG. 2 . In FIG. 2 , the document labeling model includes: a preprocessing layer, a candidate recall layer, a rough sorting layer and a fine sorting layer. Among them, the preprocessing layer is used to perform segmentation, sentence segmentation, word segmentation, part-of-speech tagging POS, named entity recognition NER and other processing on the document to obtain preprocessing results; the preprocessing results include: segmentation results, sentence segmentation results, and word segmentation results. results, part-of-speech tagging results, and named entity recognition results.

Among them, the candidate recall layer includes: parallel keyword recall sub-model, multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model. The input of these four recall sub-models is the document and the preprocessing result corresponding to the document; the output result is a plurality of candidate labels. The output results of the 4 recall sub-models are combined to obtain the candidate label results. Among them, the keyword recall sub-model is used to determine candidate tags by analyzing the semantic structure and statistical features of the document. Multi-label classification recall sub-model for NN-based multi-label classification to determine candidate labels. An explicit recall submodel to identify candidate labels based on literal matching and frequency filtering. Implicit recall submodel to determine candidate labels based on principal-subordinate component analysis.

Among them, the rough sorting layer includes: a parallel rule sub-model and a semantic matching sub-model. The rule sub-model is used to determine the candidate tags to be filtered in the candidate tag results according to preset rules. The semantic matching sub-model is used to determine the text correlation between the document and each candidate tag in the candidate tag results, and determine the candidate tags to be filtered according to the text correlation. Filter the candidate tags to be filtered in the candidate tag results to obtain the filtered candidate tag results. Among them, the text correlation refers to the semantic level similarity between the text and the candidate tags.

Among them, the fine sorting layer is used to sort each candidate tag according to the text relevance, tag popularity and tag granularity of each candidate tag in the filtered candidate tag results, and predict the tag information corresponding to the document according to the sorting result. Wherein, the popularity of a tag refers to the popularity of a user's attention to a candidate tag, such as the search popularity of a candidate tag. Tag granularity is calculated based on the word type and length of the candidate tags. The more detailed the content of the candidate tags, the smaller the tag granularity. For example, sort by tag granularity, then Baidu -> Baidu Alliance -> Baidu Alliance Summit; Entertainment -> Entertainment Star.

In this application, application scenarios include re-entity tag prediction for long documents, re-accurate tag prediction for questions and answers, and re-recall tag prediction for user-generated content, etc. Wherein, the prediction objects may include: long documents, questions and answers, user original content, and the like. Forecast needs such as re-recall, re-accuracy, re-entity, re-classification, re-high commercial value, etc.

In this application, the general training data of each application scenario may refer to training data obtained by combining the training data of each application scenario, for example. In this application, before determining the application scenarios to be applied, a large amount of general training data of each application scenario can be used to pre-train the initial document label model, so as to reduce the number of application scenarios to be applied after the application scenarios to be applied are determined. The amount of training data.

Step 102, acquiring scenario training data of the application scenario to be applied, where the scenario training data includes: multiple documents in the application scenario to be applied and corresponding label information.

Step 103, obtaining sub-models in the document tag model related to the application scenarios to be applied.

In this application, the sub-models related to the application scenarios to be applied include: semantic matching sub-model, and any one or more of the following sub-models: multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model Model. In this application, a sub-model may be selected from the above-mentioned sub-models for retraining or fine-tuning according to specific application scenarios to be applied.

Step 104, using scene training data to train the sub-models to obtain a trained document labeling model.

In this application, when the sub-models related to the application scenario to be applied include: semantic matching sub-model, multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model, the training device of the document label model performs steps The process of step 104 can specifically be, for each document in the scene training data, input the document into the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model respectively, and combine the output results of each to obtain the candidate Tag results; input the document and candidate tag results into the semantic matching sub-model to obtain the correlation between the document and each candidate tag in the candidate tag results; according to the correlation between the document and each candidate tag in the candidate tag results, and the corresponding tag information of the document, The coefficients of the semantic matching sub-model, the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model are adjusted to obtain a trained document label model.

In this application, in order to improve the accuracy of the trained document label model, the scene training data may also include: a label set, which includes: labels that can be predicted by the document label model, so that the document label model combines the label set with the scene training data Label prediction for documents in .

In the present application, before step 104, the method may also include the following steps: initialize the coefficients of the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model in the document label model, so as to avoid The coefficients of the above sub-models in the pre-trained document labeling model are subject to interference during training in applicable application scenarios, further improving the accuracy of the document labeling model in applicable application scenarios.

The training method of the document label model in the embodiment of the present application is obtained by obtaining a pre-trained document label model, and the document label model is obtained by pre-training with general training data of each application scenario; the scene training data of the application scene to be applied is obtained, and the scene The training data includes: multiple documents in the application scenario to be applied and the corresponding label information; obtain the sub-model related to the application scenario to be applied in the document label model; use the scene training data to train the sub-model to obtain the trained The document labeling model can reduce the training data required for training the document labeling model in the applicable application scenarios, and reduce the training cost while ensuring the accuracy of the document labeling model.

In order to realize the foregoing embodiments, the embodiments of the present application further provide a training device for a document labeling model.

Fig. 3 is a schematic diagram according to the second embodiment of the present application. As shown in Figure 3, the training device 100 of the document labeling model includes:

The acquisition module 110 is configured to acquire a pre-trained document label model, which is obtained by pre-training the general training data of each application scenario;

The obtaining module 110 is also used to obtain scene training data of the application scene to be applied, and the scene training data includes: multiple documents and corresponding label information in the application scene to be applied;

The obtaining module 110 is also used to obtain a sub-model related to the application scenario to be applied in the document label model;

The training module 120 is configured to use the scene training data to train the sub-models to obtain a trained document label model.

In an embodiment of the present application, the document labeling model includes: a preprocessing layer, a candidate recall layer, a rough sorting layer and a fine sorting layer;

The candidate recall layer includes: a parallel keyword recall sub-model, a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model;

The rough sorting layer includes: a parallel rule sub-model and a semantic matching sub-model;

The sub-models related to the applicable application scenario include: a semantic matching sub-model, and any one or more of the following sub-models: a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model.

In one embodiment of the present application, when the sub-models related to the application scenario to be applied include: semantic matching sub-model, multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model, the The training module 120 is specifically used for,

For each document in the scene training data, input the document into the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model respectively, and merge the output results to obtain candidate label results;

Input the document and the candidate label result into the semantic matching sub-model, and obtain the correlation between the document and each candidate label in the candidate label result;

According to the correlation between the document and each candidate label in the candidate label result, and the label information corresponding to the document, the semantic matching sub-model, the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model The coefficients of the model are adjusted to obtain a trained document label model.

In one embodiment of the present application, the scene training data further includes: a label set, and the label set includes: labels that can be predicted by the document label model, so that the document label model combines the label set with the documents in the scene training data. label prediction.

In one embodiment of the present application, with reference to FIG. 4 , the device further includes: an initialization module 130, for the multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model in the document label model The coefficients are initialized.

Wherein, it should be noted that the foregoing explanations on the training method of the document labeling model are also applicable to the training device of the document labeling model in this embodiment, which will not be repeated here.

The training device for the document label model in the embodiment of the present application is obtained by obtaining a pre-trained document label model, and the document label model is obtained by pre-training with general training data of each application scenario; the scene training data of the application scene to be applied is obtained, and the scene The training data includes: multiple documents in the application scenario to be applied and the corresponding label information; obtain the sub-model related to the application scenario to be applied in the document label model; use the scene training data to train the sub-model to obtain the trained The document labeling model can reduce the training data required for training the document labeling model in the applicable application scenarios, and reduce the training cost while ensuring the accuracy of the document labeling model.

According to the embodiments of the present application, the present application also provides an electronic device and a readable storage medium.

As shown in FIG. 5 , it is a block diagram of an electronic device according to a method for training a document tagging model according to an embodiment of the present application. Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the applications described and/or claimed herein.

As shown in FIG. 5 , the electronic device includes: one or more processors 301 , a memory 302 , and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected using different buses and can be mounted on a common motherboard or otherwise as desired. The processor may process instructions executed within the electronic device, including instructions stored in or on the memory, to display graphical information of a GUI on an external input/output device such as a display device coupled to an interface. In other implementations, multiple processors and/or multiple buses may be used with multiple memories and multiple memories, if desired. Likewise, multiple electronic devices may be connected, with each device providing some of the necessary operations (eg, as a server array, a set of blade servers, or a multi-processor system). One processor 301 is taken as an example in FIG. 5 .

The memory 302 is a non-transitory computer-readable storage medium provided in this application. Wherein, the memory stores instructions executable by at least one processor, so that the at least one processor executes the method for training a document tagging model provided in this application. The non-transitory computer-readable storage medium of the present application stores computer instructions, and the computer instructions are used to cause a computer to execute the document labeling model training method provided in the present application.

The memory 302, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as program instructions/modules corresponding to the training method of the document label model in the embodiment of the present application ( For example, the acquisition module 110 and the training module 120 shown in FIG. 3, and the initialization module 130 shown in FIG. 4). The processor 301 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions and modules stored in the memory 302, that is, realizes the training method of the document labeling model in the above method embodiment.

The memory 302 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function; data etc. In addition, the memory 302 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory 302 may optionally include a memory that is remotely located relative to the processor 301, and these remote memories may be connected to the electronic device for training the document labeling model through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The electronic device of the method for training a document labeling model may further include: an input device 303 and an output device 304 . The processor 301, the memory 302, the input device 303, and the output device 304 may be connected via a bus or in other ways. In FIG. 5, connection via a bus is taken as an example.

The input device 303 can receive the input number or character information, and generate key signal input related to the user setting and function control of the training electronic equipment of the document label model, such as a touch screen, a small keyboard, a mouse, a trackpad, a touchpad, an indication input devices such as sticks, one or more mouse buttons, trackballs, joysticks, etc. The output device 304 may include a display device, an auxiliary lighting device (eg, LED), a tactile feedback device (eg, a vibration motor), and the like. The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.

Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor Can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.

These computing programs (also referred to as programs, software, software applications, or codes) include machine instructions for a programmable processor and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine language calculation program. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or means for providing machine instructions and/or data to a programmable processor ( For example, magnetic disks, optical disks, memories, programmable logic devices (PLDs), including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.

To provide for interaction with the user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user. ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.

The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: Local Area Network (LAN), Wide Area Network (WAN) and the Internet.

A computer system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present application may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present application can be achieved, no limitation is imposed herein.

The above specific implementation methods are not intended to limit the protection scope of the present application. It should be apparent to those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims (10)

1. A training method for a document labeling model, characterized in that, comprising: Obtaining a pre-trained document label model, which is obtained by pre-training the general training data of each application scenario; Acquiring scene training data of the application scene to be applied, the scene training data including: multiple documents and corresponding label information under the application scene to be applied; Acquiring sub-models related to the application scenario to be applied in the document labeling model; Using the scene training data to train the sub-model to obtain a trained document label model; Wherein, when the sub-models related to the application scenario to be applied include: semantic matching sub-model, multi-label classification recall sub-model, explicit recall sub-model and implicit recall sub-model, the use of the scenario training data The sub-model is trained to obtain a trained document label model, including: For each document in the scene training data, input the document into the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model respectively, and merge the output results to obtain candidate label results; Input the document and the candidate label result into the semantic matching sub-model, and obtain the correlation between the document and each candidate label in the candidate label result; According to the correlation between the document and each candidate label in the candidate label result, and the label information corresponding to the document, the semantic matching sub-model, the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model The coefficients of the model are adjusted to obtain a trained document label model. 2. The method according to claim 1, wherein the document labeling model comprises: a preprocessing layer, a candidate recall layer, a rough sorting layer and a fine sorting layer; The candidate recall layer includes: a parallel keyword recall sub-model, a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model; The rough sorting layer includes: a parallel rule sub-model and a semantic matching sub-model; The sub-models related to the applicable application scenario include: a semantic matching sub-model, and any one or more of the following sub-models: a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model. 3. The method according to claim 1, wherein the scene training data further comprises: a tag set, the tag set includes: a label that the document tag model can predict, so that the document tag model combines the tag set pair Documents in the scene training data for label prediction. 4. The method according to claim 1, wherein said adopting said scene training data to train said sub-model, before obtaining the trained document label model, further comprising: Initialize the coefficients of the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model in the document label model. 5. A training device for a document labeling model, comprising: An acquisition module, configured to acquire a pre-trained document label model, which is obtained by pre-training the general training data of each application scenario; The obtaining module is also used to obtain scene training data of the application scene to be applied, and the scene training data includes: multiple documents and corresponding label information under the application scene to be applied; The obtaining module is also used to obtain a sub-model related to the application scenario to be applied in the document label model; A training module, configured to use the scene training data to train the sub-model to obtain a trained document label model; Wherein, when the sub-models related to the application scenario to be applied include: a semantic matching sub-model, a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model, the training module is specifically used for, For each document in the scene training data, input the document into the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model respectively, and merge the output results to obtain candidate label results; Input the document and the candidate label result into the semantic matching sub-model, and obtain the correlation between the document and each candidate label in the candidate label result; According to the correlation between the document and each candidate label in the candidate label result, and the label information corresponding to the document, the semantic matching sub-model, the multi-label classification recall sub-model, the explicit recall sub-model and the implicit recall sub-model The coefficients of the model are adjusted to obtain a trained document label model. 6. The device according to claim 5, wherein the document labeling model comprises: a preprocessing layer, a candidate recall layer, a rough sorting layer and a fine sorting layer; The candidate recall layer includes: a parallel keyword recall sub-model, a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model; The rough sorting layer includes: a parallel rule sub-model and a semantic matching sub-model; The sub-models related to the applicable application scenario include: a semantic matching sub-model, and any one or more of the following sub-models: a multi-label classification recall sub-model, an explicit recall sub-model and an implicit recall sub-model. 7. The device according to claim 5, wherein the scene training data further comprises: a tag set, the tag set includes: tags that can be predicted by the document tag model, so that the document tag model combines the tag set pairs Documents in the scene training data for label prediction. 8. The device according to claim 5, further comprising: an initialization module, configured to recall submodels, explicit recall submodels and implicit recall submodels for multi-label classification in the document label model Coefficients are initialized. 9. An electronic device, characterized in that it comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can perform any one of claims 1-4. Methods. 10. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to make the computer execute the method according to any one of claims 1-4.

Images