WO2021164286A1

WO2021164286A1 - User intention recognition method, apparatus and device, and computer-readable storage medium

Info

Publication number: WO2021164286A1
Application number: PCT/CN2020/122059
Authority: WO
Inventors: 朱海军; 许开河; 王少军
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-02-21
Filing date: 2020-10-20
Publication date: 2021-08-26
Also published as: CN111368045A; CN111368045B

Abstract

The present application relates to the technical field of natural language processing. Disclosed are a user intention recognition method, apparatus and device, and a computer-readable storage medium. The method comprises: receiving a target question input by a user, and inputting the target question into an element recognition model so as to acquire a plurality of historical questions associated with the target question; acquiring historical elements corresponding to the historical questions, and performing duplicate checking processing on the historical elements so as to acquire a plurality of relevant historical elements; sequentially combining each relevant historical element with the target question so as to acquire a plurality of pieces of target data, and sequentially performing prediction on each piece of target data by means of the element recognition model so as to determine whether each piece of target data meets a preset requirement; and taking relevant historical elements in the target data which meets the preset requirement as target question elements corresponding to the target question, and on the basis of the target question elements, determining the intention of the user. The technical problem in the prior art of the accuracy being relatively low when elements are recognized is solved.

Description

User intention recognition method, device, equipment and computer readable storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on February 21, 2020, the application number is CN202010110441.2, and the invention title is "User Intention Recognition Method, Device, Equipment, and Computer-readable Storage Medium". The entire content is incorporated into this application by reference.

Technical field

This application relates to the field of natural language processing technology, and in particular to a method, device, device, and computer-readable storage medium for recognizing user intentions.

Background technique

At present, in the intelligent customer service system, the intent recognition of the user query (question) is a very important part. However, in the current intent recognition scheme for the user query (such as intent classification and intent retrieval, etc.), it is impossible to accurately identify the unknown intent. The query is identified.

technical problem

The inventor realizes that in order to solve the above-mentioned problems, the current method is basically to identify the important elements of the user query, but the current common element identification methods have great limitations. Either they have strict restrictions on the data to be processed, or they cannot be directly processed. Optimize the target of the element recognition, thereby reducing the accuracy of the element recognition, thereby affecting the final result of the user's query intention recognition.

Technical solutions

A method for recognizing user intentions, the method for recognizing user intentions includes:

Receiving a target question input by a user, and inputting the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

Acquire historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

Each of the relevant historical elements and the target question are sequentially combined to obtain each target data, and each of the target data is sequentially predicted through the element recognition model to determine whether each of the target data conforms to a preset Require;

The relevant historical elements in the target data that meet the preset requirements are used as target question elements corresponding to the target question, and the user's intention is determined based on the target question elements.

A user intention recognition device, the user intention recognition device comprising:

The first obtaining module is configured to receive the target question input by the user, and input the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

The second acquisition module is configured to acquire historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

The prediction module is used to sequentially combine each of the relevant historical elements and the target question to obtain multiple target data, and predict each of the target data in turn through the element recognition model to determine each of the target data Whether the target data meets the preset requirements;

The determining module is configured to use the relevant historical elements in the target data that meets the preset requirements as the target question elements corresponding to the target question, and determine the user's intention based on the target question elements.

A user intention recognition device, the user intention recognition device comprising: a memory, a processor, and a computer program stored in the memory and running on the processor, wherein:

When the computer program is executed by the processor, the following steps are implemented:

Each of the relevant historical elements and the target question are sequentially combined to obtain multiple target data, and each target data is sequentially predicted through the element recognition model to determine whether each target data meets the forecast Set requirements

A computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the following steps are implemented:

Description of the drawings

FIG. 1 is a schematic diagram of the terminal\device structure of the hardware operating environment involved in the solution of the embodiment of the present application;

2 is a schematic flowchart of a first embodiment of a method for identifying user intentions according to this application;

FIG. 3 is a schematic diagram of functional modules of the user intention recognition device of this application.

The realization of the objectives, functional characteristics, and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Embodiments of the present invention

It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

As shown in FIG. 1, FIG. 1 is a schematic diagram of a terminal structure of a hardware operating environment involved in a solution of an embodiment of the present application.

The terminal in the embodiment of the present application is a user intention recognition device.

As shown in FIG. 1, the terminal may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to implement connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a magnetic disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

Optionally, the terminal may also include a camera, an RF (Radio Frequency, radio frequency) circuit, a sensor, an audio circuit, a WiFi module, and so on. Among them, sensors such as light sensors, motion sensors and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display screen according to the brightness of the ambient light, and the proximity sensor can turn off the display screen and/or when the terminal device is moved to the ear. Backlight. Of course, the terminal device can also be equipped with other sensors such as gyroscope, barometer, hygrometer, thermometer, infrared sensor, etc., which will not be repeated here.

Those skilled in the art can understand that the terminal structure shown in FIG. 1 does not constitute a limitation on the terminal, and may include more or less components than shown in the figure, or combine some components, or arrange different components.

As shown in FIG. 1, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a user intention recognition program.

In the terminal shown in FIG. 1, the network interface 1004 is mainly used to connect to the back-end server and communicate with the back-end server; the user interface 1003 is mainly used to connect to the client (user side) and communicate with the client; and the processor 1001 can be used to call the user intention recognition program stored in the memory 1005 and perform the following operations:

2, the present application provides a method for recognizing user intent. In an embodiment of the method for recognizing user intent, the method for recognizing user intent includes the following steps:

Step S10, receiving a target question input by a user, and inputting the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

In this embodiment, element identification is converted to element search, thereby simplifying the difficulty of the task. Element search can be achieved through multiple queries (questions) and 0/1 classification of elements. 1 means that the element is contained in the query, and 0 means that the element is not contained in the query. However, because the current manual labeling usually only labels which elements are contained in the query, the 0/1 classification model requires a large amount of negative sample data, that is, the elements required by the 0/1 classification model can only be obtained through the query, but the negative samples cannot be obtained. data. Therefore, a large number of high-quality negative sample data can be constructed through the es (elasticsearch, open source search engine) framework. Among them, the negative sample data may be data that is associated with the query but is not the same.

When the negative sample data and elements are obtained, the self-attention and powerful generalization ability of the bert (language representation model) model can be used to perform 0/1 classification training on the query and elements, which can improve the accuracy of data classification Rate and reduce the demand for training data. And after the bert model has completed the classification training of the query and elements, it can be clearly detected that the bert model has a huge amount of parameters, and the model training takes a long time, which may easily lead to the bert model not being able to meet the real-time needs of intelligent customer service, so it is necessary For the bert model that has been trained, the parameters of the model are compressed as much as possible without losing the effect of the model as much as possible. That is to say, the trained bert model can be compressed through the knowledge distillation technology, and after the compression is completed, the trained and compressed bert model can be released, that is, online personnel can formally apply the released bert model bert model. Among them, the knowledge distillation is to induce the training of the student network through the soft goal related to the teacher network of the quotation, and realize the knowledge transfer. In this embodiment, the trained bert model is knowledge simplified through knowledge distillation, and only the parameters associated with the query are retained.

The target question can be any question entered by the user. The element recognition model can be a model after the bert model has been classified and trained and the knowledge is distilled and compressed and released. When the system terminal receives the target question input by the user or the client, it will input the target question into the element recognition model for model training, and obtain each historical question in the knowledge base corresponding to the element recognition model, and retrieve it in each From the historical questions, various historical questions related to the target question are obtained. Among them, when performing es search, you can determine each historical question related to the target question by checking whether the text in each historical question is the same as the target question; it can also check whether the meaning of the word in each historical question has and the target The problem is the same to determine the various historical problems related to the target problem, etc.

Step S20: Obtain historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

After obtaining each historical issue related to the target issue, it is also necessary to determine the historical elements corresponding to these historical issues, that is, obtain the historical elements corresponding to each historical issue related to the target issue from the knowledge base, and obtain the historical elements Later, these historical elements can be checked for duplicate processing to detect whether there are two or more elements that are completely the same. If they exist, determine these same elements as repeated elements, and after determining each repeated element, It is necessary to select one element from these repetitive elements and non-repetitive elements together as the relevant historical element corresponding to the target question. Among them, historical elements can be elements marked in historical questions.

Step S30: Combine each of the relevant historical elements and the target question in sequence to obtain multiple target data, and use the element recognition model to predict each of the target data in turn to determine each of the target data Whether it meets the preset requirements;

After obtaining each relevant historical element, it is also necessary to sequentially combine each relevant historical element and the target question to obtain the combined target data. For example, when the target question is "Are the ATM transfers free?" and the relevant historical elements obtained are "Teller Machines, Handling Fees, Free", then you can turn "Teller Machines, Handling Fees, Free" and "Teller Machine Transfers" in turn. Is it free?” Combine separately to obtain three target data, so that the three target data can be model-trained through the element recognition model. Then, each target data is predicted through the feature recognition model in turn, and the prediction result is used to determine whether the target data meets the preset requirements, that is, when traversing each target data, the feature recognition model determines whether the currently traversed target data meets the preset requirements Requirements (where the preset requirements can be any requirements set by the user in advance), if the currently traversed target data meets the preset requirements, it can be considered that the target problem in the current traversed target data includes the relevant history in the current traversed target data Elements, and determine the relevant historical elements in the target data currently traversed as the target problem elements corresponding to the target problem. However, if the target data currently traversed does not meet the preset requirements, it can be considered that the target problem in the target data currently traversed does not include the relevant historical elements in the target data currently traversed, and the current traversed target data can be determined accordingly. The relevant historical element is not the target question element corresponding to the target question. It should be noted that in this embodiment, all target data needs to be predicted through the element recognition model.

In step S40, relevant historical elements in the target data that meet the preset requirements are used as target question elements corresponding to the target question, and the user's intention is determined based on the target question elements.

After all the target data are predicted by the feature recognition model, the prediction results of each target data predicted by the feature recognition model are obtained, and based on the prediction results, it is determined which target data meets the preset requirements and which target data is not. Preset requirements, and use the relevant historical elements in the target data that meet the preset requirements as the target problem elements corresponding to the target problem, and when each target problem element is obtained, the user's intention can be calculated based on these target problem elements .

In this embodiment, by receiving the target question input by the user, and inputting the target question into the element recognition model, a plurality of historical questions associated with the target question are obtained; and the historical element corresponding to each of the historical questions is obtained , And perform duplicate checking on each of the historical elements to obtain multiple relevant historical elements; sequentially combine each of the relevant historical elements and the target question to obtain multiple target data, and identify them through the elements The model predicts each of the target data in turn to determine whether each of the target data meets the preset requirements; uses the relevant historical elements in the target data that meets the preset requirements as the target question elements corresponding to the target question, and The user's intention is determined based on the target question element. By determining the relevant historical elements according to the target question, and combining the relevant historical elements and the target question in turn, the target data is obtained, and then each target data is sequentially input into the element recognition model for model training to determine the target problem element, thereby avoiding In the prior art for element identification, the phenomenon that the target of element identification cannot be directly optimized occurs, which improves the accuracy of user intent identification, and because element search is performed in the model according to the problem, the obtained value is improved. The accuracy of the elements solves the technical problem of low accuracy when performing element identification in the prior art.

Further, on the basis of the first embodiment of the present application, a second embodiment of the user intention recognition method of the present application is proposed. This embodiment is step S30 of the first embodiment of the present application. The step of sequentially predicting the target data includes:

Step a: traverse each of the target data in turn, and train the currently traversed target data through the element recognition model to determine whether the target question includes relevant historical elements in the currently traversed target data;

After obtaining each target data, it is necessary to perform model training on each target data through the feature recognition model to determine which target data meets the requirements, that is, to traverse each target data in turn, and use the feature recognition model to analyze the current traversal Model training is performed on the target data, and based on the training results, it is determined whether the target question contains relevant historical elements in the target data currently traversed, and different operations are performed based on different determination results.

Step b, if included, determine that the currently traversed target data meets the preset requirements, and use the relevant historical elements in the currently traversed target data as the target problem element corresponding to the target question until each target Data traversal is complete.

When it is judged that the target question contains the relevant historical elements in the target data currently traversed, the relevant historical elements in the target data currently traversed can be used as the target question elements corresponding to the target question, until each target data traversal is completed, that is, it is necessary Determine whether the relevant historical elements in all target data are the target problem elements corresponding to the target problem. However, if the target problem does not include the relevant historical elements in the target data currently traversed, the traversal of the next target data is continued until all target data traversal is completed.

In this embodiment, by traversing all the target data, and determining the target problem element corresponding to the target problem through the element recognition model, the accuracy of determining the target problem element is improved, and because it is directly determined whether the target problem contains relevant historical elements Therefore, compared with determining the elements of the target problem in the prior art, the process of obtaining the elements of the target problem is simplified.

Further, on the basis of any one of the first to second embodiments of this embodiment, a third embodiment of the user intention identification method of the present application is proposed. This embodiment is step S10 of the first embodiment of the present application. Before the step of inputting the target question and inputting the target question into the element recognition model, it includes:

Step c: Obtain a bert model of a language representation model and a plurality of input training questions, and perform classification training on each of the training questions sequentially through the bert model;

Obtain the preset bert model and each training problem that needs to be trained, and classify and train each training problem in turn through the knowledge base corresponding to the bert model and the bert model to determine which elements the training problem contains, until the training of each training problem is completed .

In step d, the bert model completed after each classification training is used as a training model, and an element recognition model is determined based on the training model.

After the bert model has completed the classification and training of each training problem, the bert model that has been trained for each classification can be used as the training model, and the element recognition model can be determined according to this training model. That is, if the training model is directly released, the released training model can be used as a feature recognition model. However, generally due to the huge amount of parameters of the training model, which does not meet the real-time requirements, the training model can be processed first and then released to obtain the features Identify the model.

In this embodiment, the bert model that has undergone classification training is used as the training model, and the element recognition model is determined according to the training model, thereby ensuring that the obtained element recognition model is effective and usable.

Specifically, the step of sequentially classifying and training each of the training questions through the bert model includes:

Step c1, traverse each of the training questions in turn, determine multiple label elements in the training question currently traversed, and obtain multiple negative sample data associated with the training question currently traversed;

Each training problem is traversed in turn, and each labeling element in the training problem currently traversed is determined, and a large number of high-quality negative sample data associated with the training problem currently traversed can be constructed through es retrieval. Among them, the negative sample data is data that is associated with the training problem but is not annotated elements.

Step c2, based on each of the negative sample data and each of the annotation elements, and perform classification training on the currently traversed training questions through the bert model, until the traversal of each of the training questions is completed.

When a large number of high-quality negative sample data and each labeling element are obtained, the bert model can be used to classify and train the current traversed training problem to determine which elements the current traversed training problem contains until the traversal of each training problem is completed, that is Until all training problems have been classified training.

In this embodiment, the usability of the bert model is determined by obtaining the annotation elements and each negative sample data corresponding to the training problem, and performing classification training through the bert model.

Specifically, the step of determining an element recognition model based on the training model includes:

In step d1, the training model is compressed through knowledge distillation to obtain a compressed model, and the compressed model is published to obtain an element recognition model.

When the training model is obtained, it can be clearly detected that the parameters of the training model are huge, and the training of the model takes a long time, which easily causes the training model to fail to meet the real-time needs of intelligent customer service. Therefore, it is necessary to train the model. Under the premise of not losing the effect of the model as much as possible, compress the parameters of the model as much as possible, that is, the training model can be compressed through knowledge distillation, and the compressed training model can be used as the compression model, and the compressed model will be released. In order to obtain the feature recognition model.

In this embodiment, the training model is compressed and released through knowledge distillation to obtain the element recognition model, so as to avoid too many parameters of the training model, which leads to the phenomenon that the training model determines the target problem element for a long time, and ensures the element recognition model Practicality.

Further, on the basis of any one of the first to third embodiments of this embodiment, a fourth embodiment of the user intention recognition method of this application is proposed. This embodiment is step S20 of the first embodiment of this application, and each The historical elements corresponding to the historical questions, and the process of checking the duplicates of each historical element to obtain multiple relevant historical elements, include:

Step e: Obtain the historical elements corresponding to each of the historical questions, perform duplicate checking processing on each of the historical elements, and determine whether there are duplicate historical elements in each of the historical elements;

Obtain the historical elements corresponding to each historical issue, and perform duplicate checking on each historical element in turn to determine whether there are duplicate historical elements in each historical element, and perform different operations based on different determination results.

In step f, if it does not exist, use each of the historical elements as related historical elements.

When it is found through judgment that there are no repeated historical elements in each historical element, each historical element can be directly used as the relevant historical element corresponding to the target question.

In this embodiment, each historical element is checked for duplicate processing to determine whether there is a repeated historical element. If it does not exist, each historical element is taken as a relevant historical element, thereby ensuring the high quality of the obtained relevant historical element.

Further, after the step of determining whether there are repeated historical elements in each of the historical elements, the method includes:

Step h, if it exists, determine whether the types of the repeated historical elements are the same;

When it is judged that there are duplicate elements in each historical element, it is determined whether the types of each repeated historical element (such as word meaning, pinyin, etc.) are the same, that is, when it is determined that there are duplicate historical elements, determine whether all historical elements are repeated If they are the same, you can directly select one of these repeated historical elements as the relevant historical element. If they are not the same, you need to filter for different types of repeated historical elements.

In step g, if they are not the same, select one of the repeated historical elements of the same type as the demand historical element, and use the non-repetitive historical elements and the demand historical elements in each of the historical elements as related Historical elements, wherein each of the historical elements includes non-repetitive historical elements and repeated historical elements.

When the type of each historical element is found to be different after judgment, one of the repeated historical elements of the same type is randomly selected as the demand historical element, and the non-repetitive historical elements and each demand historical element in each historical element are regarded as Relevant historical elements corresponding to the target question. In this embodiment, the historical elements are either repetitive historical elements or non-repetitive historical elements.

In this embodiment, by determining whether the types of each repeated historical element are the same, if they are not the same, it is necessary to obtain a repeated historical element and a non-repetitive historical element together as the related historical element in each type of repeated historical element to ensure that The degree of conciseness of the relevant historical elements obtained.

In addition, referring to FIG. 3, an embodiment of the present application also proposes a user intention recognition device, and the user intention recognition device includes:

Further, the prediction module is also used for:

Traverse each of the target data in turn, and train the currently traversed target data through the element recognition model to determine whether the target question includes relevant historical elements in the currently traversed target data;

If it does, it is determined that the target data currently traversed meets the preset requirements, and the relevant historical elements in the target data currently traversed are used as target problem elements corresponding to the target question, until each target data traversal is completed .

Further, the first obtaining module is also used for:

Acquiring a language representation model bert model and a plurality of input training questions, and sequentially classifying and training each of the training questions through the bert model;

The bert model completed after each classification training is used as the training model, and the element recognition model is determined based on the training model.

Further, the first obtaining module is also used for:

Traverse each of the training questions in turn, determine multiple label elements in the training question currently traversed, and obtain multiple negative sample data associated with the training question currently traversed;

Based on each of the negative sample data and each of the annotation elements, the training questions currently traversed are classified and trained through a bert model until the traversal of each training problem is completed.

Further, the first obtaining module is also used for:

The training model is compressed through knowledge distillation to obtain a compressed model, and the compressed model is published to obtain an element recognition model.

Further, the second acquisition module is also used for:

Obtain the historical elements corresponding to each of the historical questions, perform duplicate checking processing on each of the historical elements, and determine whether there are duplicate historical elements in each of the historical elements;

If it does not exist, use each of the historical elements as related historical elements.

Further, the second acquisition module is also used for:

If it exists, determine whether the types of the repeated historical elements are the same;

If they are not the same, select one of the repeated historical elements of the same type as the demand historical element, and use the non-repetitive historical elements and the demand historical elements in each of the historical elements as related historical elements, Wherein, each of the historical elements includes non-repetitive historical elements and repeated historical elements.

Among them, the steps implemented by each functional module of the user intention recognition device can refer to the various embodiments of the user intention recognition method of the present application, which will not be repeated here.

The present application also provides a user intention recognition device. The user intention recognition device includes: a memory, a processor, and a user intention recognition program stored on the memory; the processor is used to execute the user intention recognition program to To achieve the following steps:

The present application also provides a computer-readable storage medium, the computer-readable storage medium may be volatile or non-volatile, and the computer-readable storage medium stores one or more programs, The one or more programs may also be executed by one or more processors to implement the following steps:

The specific implementation of the computer-readable storage medium of the present application is basically the same as the embodiments of the element identification method described above, and will not be repeated here.

In another embodiment, the user intention recognition method provided by the present application further ensures the privacy and security of all the above-mentioned data, all the above-mentioned data can also be stored in a node of a blockchain. For example, target data and historical issues, etc., these data can be stored in the blockchain node.

It should be noted that the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.

It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, It also includes other elements that are not explicitly listed, or elements inherent to the process, method, article, or system. Without more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, method, article, or system that includes the element.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority or inferiority of the embodiments.

Through the description of the above implementation manners, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM) as described above. , Magnetic disks, optical disks), including several instructions to make a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of the present application.

The above are only the preferred embodiments of the application, and do not limit the scope of the patent for this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of the application, or directly or indirectly applied to other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims

A method for identifying user intentions, wherein the method for identifying user intentions includes the following steps:

Receiving a target question input by a user, and inputting the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

Acquire historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

Each of the relevant historical elements and the target question are sequentially combined to obtain multiple target data, and each target data is sequentially predicted through the element recognition model to determine whether each target data meets the forecast Set requirements

The relevant historical elements in the target data that meet the preset requirements are used as target question elements corresponding to the target question, and the user's intention is determined based on the target question elements.
5. The user intention recognition method of claim 1, wherein the step of sequentially predicting each of the target data through the element recognition model comprises:

Traverse each of the target data in turn, and train the currently traversed target data through the element recognition model to determine whether the target question includes relevant historical elements in the currently traversed target data;

If it does, it is determined that the target data currently traversed meets the preset requirements, and the relevant historical elements in the target data currently traversed are used as target problem elements corresponding to the target question, until each target data traversal is completed .
5. The user intention recognition method according to claim 1, wherein before the step of receiving the target question input by the user and inputting the target question into the element recognition model, the method comprises:

Acquiring a language representation model bert model and a plurality of input training questions, and sequentially classifying and training each of the training questions through the bert model;

The bert model completed after each classification training is used as the training model, and the element recognition model is determined based on the training model.
8. The user intention recognition method of claim 3, wherein the step of sequentially classifying and training each of the training questions through the bert model comprises:

Traverse each of the training questions in turn, determine multiple label elements in the training question currently traversed, and obtain multiple negative sample data associated with the training question currently traversed;

Based on each of the negative sample data and each of the annotation elements, the training questions currently traversed are classified and trained through a bert model until the traversal of each training problem is completed.
8. The user intention recognition method of claim 3, wherein the step of determining an element recognition model based on the training model comprises:

The training model is compressed through knowledge distillation to obtain a compressed model, and the compressed model is published to obtain an element recognition model.
5. The user intention recognition method according to any one of claims 1 to 5, wherein said obtaining historical elements corresponding to each of said historical questions, and performing duplicate checking processing on each of said historical elements, to obtain a plurality of related histories The elements of the steps include:

Obtain the historical elements corresponding to each of the historical questions, perform duplicate checking processing on each of the historical elements, and determine whether there are duplicate historical elements in each of the historical elements;

If it does not exist, use each of the historical elements as related historical elements.
7. The user intention recognition method according to claim 6, wherein after the step of determining whether there are repeated historical elements in each of the historical elements, the method comprises:

If it exists, determine whether the types of the repeated historical elements are the same;

If they are not the same, select one of the repeated historical elements of the same type as the demand historical element, and use the non-repetitive historical elements and the demand historical elements in each of the historical elements as related historical elements, Wherein, each of the historical elements includes non-repetitive historical elements and repeated historical elements.
A user intention recognition device, wherein the user intention recognition device includes:

The first obtaining module is configured to receive the target question input by the user, and input the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

The second acquisition module is configured to acquire multiple historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

The prediction module is used to sequentially combine each of the relevant historical elements and the target question to obtain multiple target data, and predict each of the target data in turn through the element recognition model to determine each of the target data Whether the target data meets the preset requirements;

The determining module is configured to use the relevant historical elements in the target data that meets the preset requirements as the target question elements corresponding to the target question, and determine the user's intention based on the target question elements.
A user intention recognition device, wherein the user intention recognition device includes a memory, a processor, and a user intention recognition program stored on the memory and running on the processor, and the user intention recognition program is When the processor executes, the following steps are implemented:

Receiving a target question input by a user, and inputting the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

Acquire historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

Each of the relevant historical elements and the target question are sequentially combined to obtain multiple target data, and each target data is sequentially predicted through the element recognition model to determine whether each target data meets the forecast Set requirements

The relevant historical elements in the target data that meet the preset requirements are used as target question elements corresponding to the target question, and the user's intention is determined based on the target question elements.
9. The user intention recognition device according to claim 9, wherein the step of sequentially predicting each of the target data through the element recognition model comprises:

Traverse each of the target data in turn, and train the currently traversed target data through the element recognition model to determine whether the target question includes relevant historical elements in the currently traversed target data;

If it does, it is determined that the target data currently traversed meets the preset requirements, and the relevant historical elements in the target data currently traversed are used as target problem elements corresponding to the target question, until each target data traversal is completed .
9. The user intention recognition device according to claim 9, wherein before the step of receiving the target question input by the user and inputting the target question into the element recognition model, the method comprises:

Acquiring a language representation model bert model and a plurality of input training questions, and sequentially classifying and training each of the training questions through the bert model;

The bert model completed after each classification training is used as the training model, and the element recognition model is determined based on the training model.
The user intention recognition device according to claim 11, wherein the step of sequentially classifying and training each of the training questions through the bert model comprises:

Traverse each of the training questions in turn, determine multiple label elements in the training question currently traversed, and obtain multiple negative sample data associated with the training question currently traversed;

Based on each of the negative sample data and each of the annotation elements, the training questions currently traversed are classified and trained through a bert model until the traversal of each training problem is completed.
The user intention recognition device according to claim 11, wherein the step of determining an element recognition model based on the training model comprises:

The training model is compressed through knowledge distillation to obtain a compressed model, and the compressed model is published to obtain an element recognition model.
The user intention recognition device according to any one of claims 9-13, wherein said obtaining historical elements corresponding to each of said historical questions, and performing duplicate checking processing on each of said historical elements, to obtain a plurality of related histories The elements of the steps include:

Obtain the historical elements corresponding to each of the historical questions, perform duplicate checking processing on each of the historical elements, and determine whether there are duplicate historical elements in each of the historical elements;

If it does not exist, use each of the historical elements as related historical elements.
The user intention recognition device according to claim 14, wherein after the step of determining whether there are repeated historical elements in each of the historical elements, the user intention recognition program is executed by the processor further implementing the following steps:

If it exists, determine whether the types of the repeated historical elements are the same;

If they are not the same, select one of the repeated historical elements of the same type as the demand historical element, and use the non-repetitive historical elements and the demand historical elements in each of the historical elements as related historical elements, Wherein, each of the historical elements includes non-repetitive historical elements and repeated historical elements.
A computer-readable storage medium, wherein a user intention recognition program is stored on the computer-readable storage medium, and the following steps are implemented when the user intention recognition program is executed by a processor:

Receiving a target question input by a user, and inputting the target question into the element recognition model to obtain a plurality of historical questions associated with the target question;

Acquire historical elements corresponding to each of the historical questions, and perform duplicate checking processing on each of the historical elements to obtain multiple related historical elements;

Each of the relevant historical elements and the target question are sequentially combined to obtain multiple target data, and each target data is sequentially predicted through the element recognition model to determine whether each target data meets the forecast Set requirements

The relevant historical elements in the target data that meet the preset requirements are used as target question elements corresponding to the target question, and the user's intention is determined based on the target question elements.
16. The computer-readable storage medium of claim 16, wherein the step of sequentially predicting each of the target data through the element recognition model comprises:

Traverse each of the target data in turn, and train the currently traversed target data through the element recognition model to determine whether the target question includes relevant historical elements in the currently traversed target data;

If it does, it is determined that the target data currently traversed meets the preset requirements, and the relevant historical elements in the target data currently traversed are used as target problem elements corresponding to the target question, until each target data traversal is completed .
16. The computer-readable storage medium according to claim 16, wherein before the step of receiving the target question input by the user and inputting the target question into the element recognition model, comprises:

Acquiring a language representation model bert model and a plurality of input training questions, and sequentially classifying and training each of the training questions through the bert model;

The bert model completed after each classification training is used as the training model, and the element recognition model is determined based on the training model.
18. The computer-readable storage medium of claim 18, wherein the step of sequentially classifying and training each of the training questions through the bert model comprises:

Traverse each of the training questions in turn, determine multiple label elements in the training question currently traversed, and obtain multiple negative sample data associated with the training question currently traversed;

Based on each of the negative sample data and each of the annotation elements, the training questions currently traversed are classified and trained through a bert model until the traversal of each training problem is completed.
18. The computer-readable storage medium of claim 18, wherein the step of determining an element recognition model based on the training model comprises:

The training model is compressed through knowledge distillation to obtain a compressed model, and the compressed model is published to obtain an element recognition model.