CN112559885B

CN112559885B - Training model determining method and device for map interest points and electronic equipment

Info

Publication number: CN112559885B
Application number: CN202011561079.7A
Authority: CN
Inventors: 王昆
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2020-12-25
Filing date: 2020-12-25
Publication date: 2024-01-12
Anticipated expiration: 2040-12-25
Also published as: CN112559885A

Abstract

The application discloses a training model determining method and device for map interest points and electronic equipment, and relates to the technical field of natural language processing and big data in the field of deep learning, wherein the method comprises the following steps: acquiring at least one type of pre-training task related to the POI; determining a training model of the first stage according to the pre-training task; and determining the training model of the second stage according to the parameters of the training model of the first stage and at least one kind of proprietary tasks related to the POI. The method and the system can improve the performance of all downstream natural language processing tasks in the field of map POI names, can be used as a general model parameter for issuing, and greatly improve the processing efficiency of map POIs.

Description

Training model determining method and device for map interest points and electronic equipment

Technical Field

The application discloses a training model determining method, device, equipment, medium and program product of map interest points, in particular to the technical field of deep learning, and particularly relates to natural language processing and big data.

Background

The map application is connected with the real world by using the internet technology, and with the development of 5G and AI, the data of the map industry is more and more huge and rich, wherein the map application also comprises text data such as names of POIs (Point of Interesting, points of interest) on the signboard images and the like, text data in intelligent information and the like, and the text data is directly or indirectly related to the names of the POIs.

The models used in POI domain are diverse, which results in many models not sharing a portion of domain knowledge, e.g., models such as LSTM, all require training related tasks from scratch, and domain knowledge cannot be drawn from the class of tasks related thereto.

The parameters of the existing pre-training model are obtained by repeated training on a general corpus, and the parameters do not have knowledge of the POI field, and many semantic relations and features in the POI field have great differences from the general text. Although the pretraining parameters provided by the system can be used for fine adjustment, the system has poor effect in actual engineering and cannot be shared for migration training.

Disclosure of Invention

The application provides a training model determining method, device, equipment, medium and program product for map interest points.

According to an aspect of the present application, there is provided a training model determining method for map interest points, the method including:

acquiring at least one type of pre-training task related to the POI;

determining a training model of a first stage according to the pre-training task;

and determining a training model of a second stage according to the parameters of the training model of the first stage and at least one kind of special task related to the POI.

According to another aspect of the present application, there is provided a training model determining apparatus for map points of interest, including:

the acquisition module is used for acquiring at least one type of pre-training task related to the POI;

the first processing module is used for determining a training model of a first stage according to the pre-training task;

and the second processing module is used for determining a training model of a second stage according to the parameters of the training model of the first stage and at least one kind of special task related to the POI.

According to another aspect of the present application, there is provided an electronic device including:

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 to enable the at least one processor to perform the method for determining a training model for map points of interest provided herein.

According to another aspect of the present application, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the training model determination method of map points of interest provided herein.

According to another aspect of the present application, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the training model determination method of map points of interest provided herein.

It should be understood that the description of this section is not intended to identify key or critical features of the embodiments of the application or to delineate the scope of the application. Other features of the present application will become apparent from the description that follows.

Drawings

The drawings are for better understanding of the present solution and do not constitute a limitation of the present application. Wherein:

FIG. 1 is a flowchart of a method for determining training models of map interest points provided in the present application;

FIG. 2 is a flowchart of another method for determining training models of map interest points provided in the present application;

fig. 3 is a schematic diagram of a first stage and a second stage implementation flow in a training model determining method for map interest points provided in the present application;

FIG. 4 is a flowchart of a method for determining a training model of a second stage according to an embodiment of the present disclosure;

FIG. 5 is a block diagram of a coordinate embedding layer of a training model determination method for map interest points provided in the present application;

FIG. 6 is a block diagram of a training model determining method device for map interest points provided in the present application;

fig. 7 is a block diagram of an electronic device for implementing a training model determination method for map points of interest in an embodiment of the present application.

Detailed Description

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 should be considered as merely exemplary. Accordingly, one 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 present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The following describes a training model determining method, device, equipment, medium and program product of map interest points according to an embodiment of the application with reference to the accompanying drawings.

Fig. 1 is a flowchart of a training model determining method for map interest points according to an embodiment of the present application.

The training model determining method is configured in a training model determining device for example, and the training model determining device can be applied to any electronic device so that the electronic device can execute a training model determining function.

The electronic device may be a personal computer (Personal Computer, abbreviated as PC), a cloud device, a mobile device, etc., and the mobile device may be a hardware device with various operating systems, such as a mobile phone, a tablet computer, a personal digital assistant, a wearable device, a vehicle-mounted device, etc.

As shown in FIG. 1, the training model determining method for the map interest points can comprise the following steps: step S101, at least one type of pre-training task related to the point of interest POI is acquired.

Step S102, determining a training model of the first stage according to the pre-training task.

Step S103, determining a training model of a second stage according to parameters of the training model of the first stage and at least one kind of proprietary task related to the POI.

This embodiment of the present application performs the pre-training in stages, the training of the first stage obtaining a language model weakly related to the POI domain, with the optimization objective of converting the model with low confusion in the general domain and high confusion in the POI domain into a model with low confusion in the POI domain. Training in the second stage further evolves the deep semantic extraction task of the language model in the POI field, and improves the evaluation index of the corresponding task, so that a universal language model which is strongly related to the POI field is obtained.

Based on the above embodiment, another method for determining a training model of a map interest point is provided.

Fig. 2 is a flowchart of another method for determining a training model of a map interest point according to an embodiment of the present application.

As shown in FIG. 2, the training model determining method for the map interest points may further include the following steps:

step S201, at least one of a mask language model, a next sentence prediction model, and an integer mask model related to the POI is acquired.

Here, the mask language model predicts this masked word by randomly masking the word in the POI name using a mask;

the next sentence prediction model is used for dividing a longer POI name into two text segments by cutting or utilizing the POI name with an alias to form text pairs, inputting the text pairs into a network for encoding, and finally selecting the characteristic corresponding to the first flag bit [ CLS ] as the input characteristic of a classifier for classification;

the whole word masking model is used for uniformly masking two words with word level meanings.

Step S202, encoding the inputted POI name by using a mask language model to obtain a first error value.

In the specific implementation, characters in the POI name are randomly covered by using a covering function, characters at covered positions are predicted, and the predicted characters are compared with actual characters at the covered positions to obtain a first error value.

Step S203, the next statement prediction model is adopted to encode the input POI name, and a second error value is obtained.

In the specific implementation, the POI name is divided into at least two text segments, a second text segment is predicted according to the first text segment, and the predicted second text segment is compared with the real second text segment to obtain a second error value.

Step S204, the whole word masking model is adopted to encode the input POI name, and a third error value is obtained.

In the specific implementation, a plurality of characters of word level in the POI name are randomly covered by using a covering function, words at covered positions are predicted, and the predicted words are compared with actual words at the covered positions to obtain a third error value.

In step S205, a sum of the first error value, the second error value, and the third error value is determined, and the sum is used as an output of the training model in the first stage.

Wherein, the sum=the first error value+the second error value+the third error value. As shown in fig. 3, a flow chart of task implementation performed in a first stage and a second stage in a POI (point of interest) pre-training algorithm, where the POI pre-training algorithm is specifically divided into two training stages, each stage includes a corresponding pre-training task, the first stage directly performs pre-training by using multi-task learning, where MLM is a Mask language model (Mask Language Model), NSP is a next sentence prediction (Next Sentence Prediction), and WWM is a Whole Word Mask (white Word Mask);

here, the mask language model predicts the masked word by randomly masking the word in the POI name using the mask, using the language model, for example, masking the "lanzhou beef stretched noodles" as the "CLS" lanzhou [ mask ] beef stretched noodles [ SEP ], inputting the masked word into the pre-training model for encoding, and predicting the actual word "cow" at the mask position;

the next sentence prediction model is used for dividing a longer POI name into two text segments, or using the POI names with aliases to form text pairs, inputting the text pairs into a network for encoding, finally selecting the characteristic corresponding to the first flag bit [ CLS ] as the input characteristic of a classifier for classifying, for example, encoding 'Lanzhou beef noodles (post factory village') into 'CLS' Lanzhou beef noodles [ SEP ] and then inputting the 'CLS' beef noodles into an encoder for classifying, and naturally, randomly sampling two different POIs for combining;

whole word covering: the method has the similar points with the MLM, and the only difference is that the method uniformly masks two words with word level meanings, adds domain knowledge, and directly masks 'Lanzhou beef stretched noodles' into 'CLS Lanzhou mask SEP'.

It may be noted that, in this embodiment, the first stage pre-training content includes a mask language model, a next sentence prediction model and a whole word mask model, so as to obtain a first error value loss1, a second error value loss2 and a third error value loss3, and the sum of the first error value loss1, the second error value loss2 and the third error value loss3 is used to obtain the output of the first stage model, that is, the formula loss=loss 1+loss2+loss3, where loss is the output of the first stage model.

Step S206, determining the training model of the second stage according to the parameters of the training model of the first stage and at least one kind of proprietary task related to the POI.

It should be noted that, the implementation process of step S206 may refer to the implementation process of step S103 in the above embodiment, which is not described herein.

Based on the above embodiments, when determining the training model of the second stage, the first type of proprietary task related to the POI may be trained under the condition that the parameters of the training model of the first stage are minimum, and the detailed description will be given below with reference to fig. 4, where fig. 4 is a flow chart of a method for determining the training model of the second stage according to the embodiment of the present application.

As shown in fig. 4, the method may further include the steps of:

in step S401, training the first type of proprietary task related to the POI under the condition that the parameters of the training model in the first stage are minimum, so as to obtain a first result.

Step S402, based on the first result, training the input N+1st class of special tasks related to POIs to obtain an N+1st result, wherein N is a positive integer.

In the embodiment of the application, the named entity identification under the POI scene is performed on the input special task related to the N+1st class and the POI, and at least one of POI classification, POI matching, label classification and initial matching processing is performed.

The POI classification includes: resolving whether a POI is a real POI name;

POI matching includes: matching whether the two text sections are the same POI;

the tag classification includes: one label for each POI; the initial matching includes: and splicing the phonetic letter head of the POI name and matching with the POI name.

As shown in fig. 3, the training of the second stage performs multitasking learning through a continuous learning method framework:

performing a pre-training task 1 on the POI names after the first-stage pre-training is completed;

performing a pre-training task 1 and a pre-training task 2;

then, a pre-training task 1, a pre-training task 2 and a pre-training task 3 are carried out;

and finally, performing a pre-training task 1, a pre-training task 2, pre-training tasks 3 and … …, wherein the pre-training task N-1 and the pre-training tasks N and N are positive integers, so as to obtain a satisfactory POI name learning result.

It should be noted that, the pretraining tasks in pretraining task 1, pretraining task 2, pretraining tasks 3, … …, pretraining task N-1, pretraining task N may include: point of interest named entity recognition (POI NER), point of interest classification (POI classification), point of interest matching (POI match), tag classification (Tag classification), first letter matching (caps match).

Pretraining task 1, pretraining task 2, pretraining tasks 3, … …, pretraining task N-1, pretraining task N are POI-related proprietary tasks.

And (3) describing named entity identification, POI classification, POI matching, label classification and initial matching in the POI scene:

named entity recognition (POI NER): the named entity recognition task under the POI scene is input into a conditional random field for probability calculation by extracting the characteristics of the language model, and the category of each word is allocated, wherein a BIO labeling method is generally used, and is the same as the traditional NER task;

POI classification (POI classification): distinguishing whether a POI is a real POI name, wherein positive samples of the POI can be directly selected from a POI database, and then selecting negative samples from a general corpus or confusing phrases;

POI matching (POI match): matching whether the two sections of texts are the same POI or not, wherein the part can use some data enhancement methods to enhance the original POI names, for example, the suffix names are removed, additional noise texts are added and other methods to obtain texts similar to the POI as positive samples, and the negative samples can directly match different POIs or remove core words of the original POI for combination;

tag classification (tag classification): in the POI field, each POI corresponds to a corresponding Tag, for example, a 'spicy soup' corresponds to a 'food' Tag, a 'group light square' corresponds to a 'shopping square' Tag, and Tag information is extremely important for analyzing the properties of the POI, so that a Tag classification task is added into a language model to classify the class of the POI in fine granularity;

initial match (caps match): this part enables adaptation to the scenario where POIs are associated, for example, at WiFi names, by concatenating the pinyin letter header of the POI name and then matching the POI name, for example, matching the @ Wang Jima spicy pot @ with the @ wjmlxg @.

In the embodiment, training a first class of special tasks related to POIs to obtain a first result;

training the input second class of special tasks related to the POI based on the first result to obtain a second result; here, training the input second class of proprietary tasks related to POIs, and classifying POIs;

training the input third class of special tasks related to the POI based on the second result to obtain a third result; here, training the input third class of special tasks related to the POI, and performing POI classification and POI matching;

training the input third class of special tasks related to the POI based on the third result to obtain a third result; here, training the input third class of special tasks related to the POI, and performing POI classification, POI matching and tag classification;

training the input special tasks of the fifth class related to the POI based on the fourth result to obtain a fifth result; here, training the inputted fifth class of proprietary tasks related to the POI, and performing POI classification, POI matching, tag classification and initial matching;

based on the method, the training of the exclusive tasks of the sixth class to the N class related to the POI can be further performed;

here, the training model of the second stage further performs fine tuning training by using the model parameters after the training of the first stage, but here, a continuous pre-training concept is used, the training of the model is progressive, and each time, a task is added, the difficulty is increased. The first stage is to reduce the confusion of the language model to enable the language model to understand and encode the effective information in the POI field, and the second stage is to further train the language model by improving the difficulty and complexity of the task to fine tune the parameters in the language model, so that the language model is more universal and robust.

Based on the foregoing embodiment, in an optional embodiment of the present application, the method for determining a training model of a map interest point may further include:

the position information of the POI is embedded into the training model of the first stage.

Wherein the position information of the POI is determined by: gridding an area within a preset range where the POI is located, and taking the grid where the POI is located as position information of the POI; the size of the grid is determined by the density of POIs within the area.

As shown in fig. 5, according to the coordinate position information in the POI field, an Embedding layer of a coordinate Embedding layer is additionally added in the input Embedding layer of the pre-training model, a limited number of coordinate embeddings are obtained in a limited space by gridding an area, and the area size of the grid can be adaptively changed according to the POI density, so that the POI can be modeled by combining with the real-world position information, and the situations that the POI names are the same in different areas can be effectively distinguished. The method is applicable to occasions requiring position information, such as text matching or searching recommendation fields.

In fig. 5, the coordinate embedding is corresponding coordinate information after gridding for the POI field; the position embedding is the position information of the corresponding POI name; paragraph embedding is paragraph number information corresponding to the corresponding POI name text; word embedding is text information corresponding to the corresponding POI name.

The embodiment of the application can greatly improve the performance of all downstream tasks in the field of map POI names, can be used as a general model parameter for release, greatly improves the efficiency of map POI team cooperation, and saves the research and development cost.

In order to achieve the above embodiment, the present application proposes a training model determining device for map points of interest.

Fig. 6 is a schematic structural diagram of a training model determining device for map interest points according to an embodiment of the present application.

As shown in fig. 6, the training model determining apparatus 600 for map points of interest may include: an acquisition module 610, a first processing module 620, and a second processing module 630.

The acquiring module 610 is configured to acquire at least one type of pre-training task related to the POI.

A first processing module 620 is configured to determine a training model of the first stage according to the pre-training task.

The second processing module 630 is configured to determine a training model of the second stage according to parameters of the training model of the first stage and at least one type of proprietary task related to the POI.

Optionally, the obtaining module 610 is specifically configured to: at least one of a masking language model, a next sentence prediction model, and an integer masking model associated with the POI is obtained.

Optionally, the first processing module 620 may include:

and the first coding unit is used for coding the input POI name by adopting the mask language model to obtain a first error value.

The second coding unit is used for coding the input POI name by adopting the next statement prediction model to obtain a second error value;

the third coding unit is used for coding the input POI name by adopting the whole character masking model to obtain a third error value;

and the determining unit is used for determining the sum of the first error value, the second error value and the third error value, and taking the sum as the output of the training model of the first stage.

Optionally, the first coding unit may be further configured to:

randomly masking the characters in the POI name by using a masking function, predicting the characters at the masked positions, and comparing the predicted characters with the actual characters at the masked positions to obtain a first error value.

Optionally, the second encoding unit may be further configured to:

dividing the POI name into at least two text segments, predicting a second text segment according to the first text segment, and comparing the predicted second text segment with the real second text segment to obtain a second error value.

Optionally, the third encoding unit may be further configured to:

and randomly masking a plurality of characters of the word level in the POI name by using a masking function, predicting the word at the masked position, and comparing the predicted word with the actual word at the masked position to obtain a third error value.

Optionally, the second processing module 620 may further include:

the first training unit is used for training the first type of special tasks related to the POI under the condition that the parameters of the training model in the first stage are minimum, so as to obtain a first result;

the second training unit is used for training the input special task related to the POI in the (n+1) th class based on the first result to obtain an (n+1) th result, wherein N is a positive integer.

Optionally, the second training unit may be further configured to:

carrying out named entity recognition under a POI scene on the input special task related to the N+1th class and the POI, and carrying out at least one of POI classification, POI matching, label classification and initial matching;

the POI classification includes: resolving whether a POI is a real POI name;

POI matching includes: matching whether the two text sections are the same POI;

the tag classification includes: one label for each POI;

the initial matching includes: and splicing the phonetic letter head of the POI name and matching with the POI name.

Optionally, the training model determining device 40 for map interest points may further include:

and the embedding module is used for embedding the position information of the POI into the training model in the first stage. Wherein the position information of the POI is determined by:

gridding an area within a preset range where the POI is located, and taking the grid where the POI is located as position information of the POI; the size of the grid is determined based on the density of POIs within the area.

The device provided in this embodiment can implement each process implemented in the method embodiment shown in the foregoing embodiment, and may achieve the same beneficial effects, so that repetition is avoided, and no further description is given here.

In order to achieve the above embodiments, the present application proposes an electronic device including:

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 to enable the at least one processor to perform the training model determination method for map points of interest described in the above embodiments.

In order to implement the above-described embodiments, the present application proposes a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the training model determining method for map points of interest described in the above-described embodiments.

To achieve the above embodiments, the present application proposes a computer program product comprising a computer program which, when executed by a processor, implements the training model determination method of map points of interest described in the above embodiments.

According to embodiments of the present application, there is also provided an electronic device, a non-transitory computer-readable storage medium storing computer instructions, and a computer program product.

Fig. 7 shows a schematic block diagram of an example electronic device 700 that may be used to implement embodiments of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the application described and/or claimed herein.

As shown in fig. 7, the apparatus 700 includes a computing unit 701 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 702 or a computer program loaded from a storage unit 708 into a Random Access Memory (RAM) 703. In the RAM703, various programs and data required for the operation of the device 700 may also be stored. The computing unit 701, the ROM702, and the RAM703 are connected to each other through a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.

Various components in device 700 are connected to I/O interface 705, including: an input unit 706 such as a keyboard, a mouse, etc.; an output unit 707 such as various types of displays, speakers, and the like; a storage unit 708 such as a magnetic disk, an optical disk, or the like; and a communication unit 709 such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunication networks.

The computing unit 701 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 701 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the respective methods and processes described above, for example, an image sample processing method. For example, in some embodiments, the image sample processing method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 708.

In some embodiments, part or all of the computer program may be loaded and/or installed onto device 700 via ROM702 and/or communication unit 709. When the computer program is loaded into RAM703 and executed by computing unit 701, one or more steps of method 708 described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the image sample processing method by any other suitable means (e.g. by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for carrying out methods of the present application may be written in any combination of one or more programming languages. These program code may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus such that the program code, when executed by the processor or controller, causes the functions/operations specified in the flowchart and/or block diagram to be implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this application, a computer-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The computer readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here 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 a user; and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the internet.

The computer system may include a client and a server. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service ("Virtual Private Server" or simply "VPS") are overcome.

It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present application may be performed in parallel, sequentially, or in a different order, provided that the desired results of the technical solutions disclosed in the present application can be achieved, and are not limited herein.

The above embodiments do not limit the scope of the application. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application are intended to be included within the scope of the present application.

Claims

1. A method for determining a training model of a map point of interest, the method comprising:

acquiring at least one pre-training task in a mask language model, a next sentence prediction model and a whole word masking model related to the POI;

training the first type of special tasks related to the POI under the condition that the parameters of the training model in the first stage are minimum, so as to obtain a first result;

training the input N+1st class of special tasks related to the POI based on the first result to obtain an N+1st result, wherein N is a positive integer;

wherein training the input N+1st class of proprietary tasks related to the POI comprises:

carrying out named entity recognition under a POI scene on the input N+1st class and the related proprietary task of the POI, and carrying out at least one of POI classification, POI matching, label classification and initial matching;

the POI classification includes: resolving whether a POI is a real POI name;

the POI matching includes: matching whether the two text sections are the same POI;

the tag classification includes: one label for each POI;

2. The training model determination method according to claim 1, wherein determining the training model of the first stage according to the pre-training task comprises:

encoding the inputted POI name by adopting the mask language model to obtain a first error value;

adopting the next statement prediction model to encode the input POI name to obtain a second error value;

adopting the whole word masking model to encode the input POI name to obtain a third error value;

and determining the sum of the first error value, the second error value and the third error value, and taking the sum as the output of the training model of the first stage.

3. The training model determination method of claim 2, wherein encoding the inputted POI name using the mask language model to obtain a first error value comprises:

4. The training model determining method according to claim 2, wherein encoding the inputted POI name using the next sentence prediction model to obtain a second error value comprises:

dividing the POI name into at least two text segments, predicting a second text segment according to the first text segment, and comparing the predicted second text segment with a real second text segment to obtain a second error value.

5. The training model determining method according to claim 2, wherein the encoding the inputted POI name using the whole word masking model to obtain a third error value comprises:

6. The training model determination method according to claim 1, further comprising:

and embedding the position information of the POI into the training model of the first stage.

7. The training model determination method according to claim 6, wherein the position information of the POI is determined by:

gridding an area within a preset range where the POI is located, and taking the grid where the POI is located as the position information of the POI; the size of the grid is determined according to the density of POIs in the area.

8. A training model determining apparatus for map points of interest, comprising:

the acquisition module is used for acquiring at least one pre-training task in a mask language model, a next sentence prediction model and a whole word masking model which are related to the POI;

the second training unit is used for training the input special tasks related to the POI in the (n+1) th class based on the first result to obtain an (n+1) th result, wherein N is a positive integer;

wherein, the second training unit is further configured to:

the POI classification includes: resolving whether a POI is a real POI name;

the tag classification includes: one label for each POI;

9. The training model determination apparatus of claim 8 wherein said first processing module comprises:

the first coding unit is used for coding the input POI name by adopting the mask language model to obtain a first error value;

10. The training model determination apparatus of claim 9 wherein the first encoding unit is further configured to:

11. The training model determination apparatus of claim 9 wherein the second encoding unit is further configured to:

12. The training model determination apparatus of claim 9 wherein the third encoding unit is further configured to:

13. The training model determination apparatus of claim 8 wherein said apparatus further comprises:

and the embedding module is used for embedding the position information of the POI into the training model in the first stage.

14. The training model determination apparatus of claim 13 wherein the position information of the POI is determined by:

15. An electronic device, comprising:

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 to enable the at least one processor to perform the training model determination method of map points of interest of any of claims 1-7.

16. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the training model determination method of map points of interest of any one of claims 1-7.