CN117574143A

CN117574143A - Data processing method, device, equipment, medium and product

Info

Publication number: CN117574143A
Application number: CN202311370213.9A
Authority: CN
Inventors: 谢红伟; 马小明; 杨俊�; 宿玲玲; 雷达; 王鸣昊; 韩恒克
Original assignee: Apollo Intelligent Connectivity Beijing Technology Co Ltd
Current assignee: Apollo Intelligent Connectivity Beijing Technology Co Ltd
Priority date: 2023-10-20
Filing date: 2023-10-20
Publication date: 2024-02-20

Abstract

The disclosure provides a data processing method, a device, equipment, a medium and a product, relates to the field of artificial intelligence, and particularly relates to the fields of cloud computing, natural language processing, deep learning and the like. The specific implementation scheme is as follows: acquiring a training data set according to a plurality of preset POI task types, wherein the training data set comprises a plurality of training samples corresponding to the POI task types; determining a POI generation model to be trained, wherein the POI generation model comprises a plurality of task processing modules corresponding to the POI task types respectively; and training the POI generating model by using the training data set to obtain a target POI generating model obtained by training, and performing task processing on POI data to be analyzed in parallel by using a plurality of task processing modules respectively corresponding to the POI task types by using the target POI generating model to obtain a task processing result.

Description

Data processing method, device, equipment, medium and product

Technical Field

The present disclosure relates to the fields of cloud computing, natural language processing, deep learning, and the like in the field of artificial intelligence, and in particular, to a data processing method, apparatus, device, medium, and product.

Background

The POI (Point of Interest, information point or interest point) is point class data in an electronic map, and may include information of various attributes such as a name, an address, coordinates, a category, and the like. And a machine learning model of different tasks such as a POI label classification model, a POI address resolution model and the like can be built based on the POI.

However, the existing machine learning models built based on POIs support single task processing, have weak generalization capability and poor processing capability on complex information.

Disclosure of Invention

The present disclosure provides a data processing method, apparatus, device, medium and product for performing multi-POI tasks.

According to a first aspect of the present disclosure, there is provided a data processing method comprising:

acquiring a training data set according to a plurality of preset POI task types, wherein the training data set comprises a plurality of training samples corresponding to the POI task types;

determining a POI generating model to be trained, wherein the POI generating model comprises task processing modules corresponding to a plurality of POI task types respectively;

training the POI generating model by utilizing the training data set to obtain a target POI generating model obtained by training, and performing task processing on POI data to be analyzed in parallel by using task processing modules respectively corresponding to a plurality of POI task types by the target POI generating model to obtain a task processing result.

According to a second aspect of the present disclosure, there is provided a data processing method comprising:

acquiring a target POI generation model obtained through training, wherein the target POI generation model is obtained through training based on the data processing method of the first aspect, and comprises task processing modules respectively corresponding to a plurality of POI task types;

task processing is carried out on the POI data to be analyzed in parallel through task processing modules respectively corresponding to the task types of a plurality of POIs in the target POI generation model, and task processing results are obtained;

and outputting a task processing result of the POI data to be analyzed.

According to a third aspect of the present disclosure, there is provided a data processing apparatus comprising:

the acquisition unit is used for acquiring a training data set according to a plurality of preset POI task types, wherein the training data set comprises a plurality of training samples corresponding to the POI task types;

the system comprises a determining unit, a training unit and a training unit, wherein the determining unit is used for determining a POI generating model to be trained, and the POI generating model comprises task processing modules respectively corresponding to a plurality of POI task types;

the training unit is used for training the POI generation model by utilizing the training data set to obtain a target POI generation model obtained by training, and the target POI generation model performs task processing on the POI data to be analyzed in parallel by using task processing modules respectively corresponding to a plurality of POI task types to obtain a task processing result.

According to a fourth aspect of the present disclosure, there is provided a data processing apparatus comprising:

the model acquisition unit is used for acquiring a target POI generation model obtained through training, wherein the target POI generation model is obtained through training based on the data processing method of the first aspect, and comprises a task processing module corresponding to a plurality of POI task types respectively;

the task execution unit is used for carrying out task processing on the POI data to be analyzed in parallel through task processing modules respectively corresponding to the task types of the multiple POIs in the target POI generation model to obtain a task processing result;

and the result output unit is used for outputting the task processing result of the POI data to be analyzed.

According to a fifth aspect of the present disclosure, there is provided 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 method of the first aspect or the second aspect.

According to a sixth aspect of the present disclosure there is provided a non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method according to the first or second aspect.

According to a seventh aspect of the present disclosure, there is provided a computer program product comprising: computer program stored in a readable storage medium, from which the computer program can be read by at least one processor of an electronic device, the at least one processor executing the computer program causing the electronic device to perform the method of the first or second aspect.

According to the technology disclosed by the disclosure, a training data set can be obtained according to a plurality of preset POI task types, so that the training data set comprises a plurality of training samples corresponding to the POI task types. And determining a POI generation model to be trained, wherein the POI generation model comprises task processing modules corresponding to a plurality of POI tasks respectively, and the POI generation model can be trained through a training data set to obtain a target POI generation model obtained through training. The training samples corresponding to the POI task types are utilized to train the POI generation model, the target POI generation model obtained through training can support parallel task processing of task processing modules corresponding to the POI task types respectively, the model generalization capability of the target POI generation model is improved, the capability of the target POI generation model on more complex POI data to be analyzed is effectively enhanced, and more accurate task processing results are obtained.

It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the disclosure, nor is it intended to be used to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the following specification.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a system diagram of a data processing system provided in accordance with an embodiment of the present disclosure;

FIG. 2 is a schematic diagram according to a first embodiment of the present disclosure;

FIG. 3 is a schematic diagram according to a second embodiment of the present disclosure;

FIG. 4 is a schematic flow chart of a data processing method provided in accordance with an embodiment of the present disclosure;

FIG. 5 is a schematic diagram according to a third embodiment of the present disclosure;

FIG. 6 is a schematic diagram according to a fourth embodiment of the present disclosure;

FIG. 7 is a schematic diagram according to a fifth embodiment of the present disclosure;

FIG. 8 is a schematic diagram according to a sixth embodiment of the present disclosure;

fig. 9 is a block diagram of an electronic device for implementing a data processing method of an embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure 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 disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The disclosure provides a data processing method and device, which are applied to the fields of cloud computing, natural language processing, deep learning and the like in the artificial intelligence field, so as to train and support a multi-task POI (Point of Interest, information points or interest points) generation model, so that the target POI generation model has stronger generalization capability and stronger processing capability on more complex address data.

In the related art, POIs are point-like data in an electronic map. The models corresponding to tasks such as information extraction, name matching and label classification can be trained by utilizing POI data, and the models corresponding to the tasks are obtained. The model obtained through training in the traditional model training mode supports a single task. However, when the model obtained through training executes the corresponding task, the execution effect is poor, the execution accuracy is about 80%, the execution accuracy of the model task obtained through the traditional training is low, the model task does not have strong generalization capability, and the effect is poor.

Taking a tag classification model as an example, the tag classification model may perform marking for the inputted POI data. The data of the two POIs of the Tengwang pavilion and the Tengwang pavilion are easily distinguished to be the same data, and the tags of the tourist attractions are set.

In order to solve the above-mentioned problems, the present disclosure contemplates constructing a target POI generation model for multitasking, whereas the conventional model has a smaller parameter amount, which is not suitable for simultaneously performing multitasking. The present disclosure thus contemplates the use of large models, by constructing large models with sufficient parameters to achieve the goal of supporting multi-task synchronous learning. And further, the model processing efficiency and accuracy of the target POI generation model are improved. The target POI generation model is obtained through training, so that training samples are required to be built according to the task requirements of the multiple tasks in order to realize the training of the target POI generation model, the training data set can be richer, and the learning requirements of a large model can be repeatedly met.

Therefore, in the technical scheme of the disclosure, the training data set can be obtained according to a plurality of preset POI task types, so that the training data set comprises a plurality of training samples corresponding to the plurality of POI task types. And determining a POI generation model to be trained, wherein the POI generation model comprises task processing modules corresponding to a plurality of POI tasks respectively, and the POI generation model can be trained through a training data set to obtain a target POI generation model obtained through training. The method comprises the steps that a plurality of training samples corresponding to a plurality of POI task types are utilized to train a POI generation model, the target POI generation model obtained through training can support parallel use of task processing modules corresponding to the POI task types respectively, so that the target POI generation model is adapted to different processing tasks, the model generalization capability of the target POI generation model is further improved, the capability of the target POI generation model on more complex POI data to be analyzed is effectively enhanced, and more accurate task processing results are obtained.

FIG. 1 is a system diagram of a data processing system according to an embodiment of the present disclosure, the system may include: an open source node 11, a training node 12 and a user terminal 13. The nodes may be cloud computing server nodes or nodes in a computer cluster, which is not limited in this disclosure.

The open source node 11 may serve as a base node of the data processing system, provide a trained LLM model to the data processing system, and provide the trained LLM model to the training node 12.

The training node 12 may configure the model training method of the present disclosure, and perform the model training method of the present disclosure using the LLM model on the basis of the open source node 11. For example, the LLM model is used to acquire an initial tag in the process of acquiring a training data set, the LLM model is used to determine an initial network model of the POI generating model in the process of training the POI generating model, and the like.

After training node 12 to obtain the target POI generative model. The user terminal 13 may execute tasks in parallel by using task processing modules corresponding to the task types of multiple POIs in the target POI generation model.

Fig. 2 is a schematic diagram of a first embodiment of the disclosure, and referring to the model training method shown in fig. 2, the model training apparatus may be configured as a model training apparatus, and the model training apparatus may be located in an electronic device. The electronic device may be, for example, a node of a training layer carried in a data processing system, and may be other types of devices, which the present disclosure does not limit.

The data processing method as shown in fig. 2 may include the steps of:

201. according to a plurality of preset POI task types, a training data set is obtained, wherein the training data set comprises a plurality of training samples corresponding to the POI task types.

Optionally, the POI task type may refer to a task type corresponding to the POI task processing module, which is used to represent a type of PIO task. To facilitate distinguishing between different POI task types, a task identifier may be set for each POI task type.

In the present disclosure, the training samples may include POI base data and target tags corresponding to the POI base data.

Further, at least one training sample may be obtained for each POI task type and at least one training sample for each of the plurality of POI task types may be determined as a plurality of training samples.

Further, a minimum sample number of POI task types may be set, and when at least one training sample is obtained for each POI task type, at least one training sample may be obtained for each POI task type according to the minimum sample number.

202. And determining a POI generation model to be trained, wherein the POI generation model comprises a task processing module corresponding to a plurality of POI task types respectively.

The task processing modules corresponding to the multiple POI task types respectively can be located at an output layer of the POI generation model.

In the present disclosure, the POI generation model may include an input layer, an intermediate layer, and an output layer.

Wherein the input layer is used for receiving POI data. The intermediate layer includes a plurality of neurons for feature computation on the training samples to obtain sample features. Sample features of the intermediate layer are sent to the output layer.

Optionally, the output layer may further include a task discriminating module, where the plurality of task processing modules in the output layer respectively perform task computation on the sample features to obtain prediction results corresponding to the plurality of task processing results. The prediction results corresponding to the task processing modules are input to the task judging module.

The task judging module is used for selecting a target predicted result from the predicted results respectively output by the task processing modules, wherein the target predicted result is the task processing result of the POI generation model.

In this embodiment, the task processing module of each POI task type may refer to a decoder configured based on the processing task of the POI task type, so that the decoder can output a prediction result corresponding to the POI task type.

In general, the POI generating model to be trained can be built in advance or in real time, and the construction time of the POI generating model to be trained in the application is not limited too.

Illustratively, step 202 may include: and constructing a POI generation model based on task processing modules respectively corresponding to the multiple POI task types. And realizing real-time construction of the POI generation model.

Illustratively, step 202 may further comprise: and acquiring a POI generation model which is constructed in advance based on task processing modules respectively corresponding to the multiple POI task types.

203. Training the POI generating model by utilizing the training data set to obtain a target POI generating model obtained by training, and performing task processing on POI data to be analyzed in parallel by using task processing modules respectively corresponding to a plurality of POI task types by the target POI generating model to obtain a task processing result.

The task processing result may refer to a target prediction result generated by a target task processing module in the task processing modules corresponding to the multiple POI task types respectively.

Optionally, training the POI generating model by using the training data set to obtain a target POI generating model obtained by training may include: and training the POI generating model by using the training data set until the POI generating model is detected to meet the target ending condition, and determining the POI generating model meeting the target ending condition as a target POI generating model.

The detecting that the POI generating model meets the target ending condition may include:

and acquiring the detection accuracy of the POI generation model in the training data set, and when the detection accuracy reaches an accuracy threshold, enabling the POI generation model to meet the target ending condition.

Or acquiring the iteration times of the POI generation model, and if the iteration times are equal to the time threshold, the POI generation model meets the target ending condition.

Or, based on the loss function, calculating a loss error corresponding to the POI generation model in the training data set, and if the loss error is determined to be smaller than or equal to a preset error threshold value, enabling the POI generation model to meet a target ending condition.

Or if the training time length of the POI generation model reaches a preset time length threshold value, determining that the POI generation model meets the target ending condition.

Further, a predicted result of the POI generating model on POI basic data corresponding to the training samples can be obtained.

According to the prediction results and the target labels respectively corresponding to the POI basic data, determining the accurate quantity of the prediction accuracy in the POI basic data, calculating the quotient of the accurate quantity and the total data quantity of the POI basic data, and obtaining the detection accuracy of the POI generation model in the training data set.

The predicted result of each POI basic data and the target label are input into the loss function, the loss value of the loss function for outputting each POI basic data is obtained, and the loss error of the POI generation model corresponding to the training data set can be calculated by using the loss value of each POI basic data. The loss error of the POI generation model may be calculated, for example, by means of mean calculation, variance, etc.

In the technical scheme of the disclosure, a training data set can be obtained according to a plurality of preset POI task types, so that the training data set comprises a plurality of training samples corresponding to the plurality of POI task types. And determining a POI generation model to be trained, wherein the POI generation model is constructed and obtained based on task processing modules corresponding to a plurality of POI tasks respectively, and the POI generation model can be trained through a training data set to obtain a target POI generation model obtained through training. The training samples corresponding to the POI task types are utilized to train the POI generation model, the target POI generation model obtained through training can support task processing modules corresponding to the POI task types respectively, the model generalization capability of the target POI generation model is improved, the capability of the target POI generation model on more complex POI data to be analyzed is effectively enhanced, and more accurate task processing results are obtained.

In order for the reader to more fully understand the principles of implementation of the present disclosure, the embodiment shown in fig. 2 will now be further refined in conjunction with fig. 3-5 below.

Further, on the basis of the foregoing embodiment, acquiring the training data set according to the preset multiple POI task types may include:

and carrying out corpus analysis on the plurality of basic material information through the training-obtained pre-training model to obtain a plurality of POI basic data.

Based on the plurality of POI basis data, a training data set is obtained.

Alternatively, the Pre-training model (Pre-training) may be an unsupervised model obtained by training. A large number of unlabeled data training acquisitions may be used. The pre-training model may capture underlying structures and patterns of data, learn basic features of grammar, semantics, and contextual information of the language, and the like.

Further, in order to correlate the output of the pre-training model with the POI domain, the training samples of the pre-training model may be historical base material information.

That is, the pre-training model can be trained by taking the historical base material information as a training sample, so as to obtain the pre-training model adapting to the POI field.

In this embodiment, the corpus analysis of the basic material information by the pre-training model may include: the pre-training model reserves texts related to the POIs in the basic material information, and filters the texts unrelated to the POIs to obtain POI basic data. Related to a POI may refer to related to a POI entity, e.g. a POI name, belonging to address intelligence, belonging to text of a key attribute. The key attributes may include, for example, at least one of telephone, business hours, address meaning, status, domain, etc.

In this embodiment, obtaining the training data set based on the POI base data may include: and acquiring target labels of the POI basic data respectively, taking each POI basic data and the corresponding target label as a training sample, and determining the training sample corresponding to the POI basic data respectively as a training data set.

Further, target labels respectively corresponding to the POI basic data can be obtained through the LLM model obtained through training. And the POI basic data can be input to an interface of the LLM model, and a result output by the LLM model is obtained as a target label.

In addition, the target labels corresponding to the POI basic data can be obtained through manual labeling.

According to the technical scheme, corpus analysis is carried out on a plurality of basic material information by utilizing the pre-training model obtained through training, the pre-training model can capture the bottom layer result and mode of data, basic material information can be adjusted, the quality of the obtained POI basic data is higher, and a training data set more suitable for POI training is obtained. Through training data sets closely related to the POI field, training of the POI generation model can learn more effective knowledge, and the training data sets have higher task processing capacity.

Optionally, the corpus analysis is performed on the plurality of basic material information through the pre-training model obtained through training, and before the plurality of POI basic data are obtained, the method further comprises the steps of: and acquiring a plurality of basic material information.

The basic material information may include at least one of address information, user comment information, POI basic information, super content information, web page content information, POI data specification, POI quality standard, region division information, brand name information, voice dialogue information, associated data with father-son relationship, and the like.

In this embodiment, various types of base material information may include a plurality of base material information, and each base material information is of a higher order of magnitude, so that the POI generation model has enough training samples.

Wherein, obtaining the plurality of base material information may include: a plurality of base material information is read from a plurality of third party data systems. The third party data system may be, for example, a data system corresponding to various applications such as instant messaging software, search software, various social software, travel applications, merchant clients, shopping software, and the like. Of course, the plurality of basic material information related to the present disclosure may not relate to the personal information of the user, or may be obtained on the basis of obtaining the authorization of the user when the personal information is related, and all the data are obtained in compliance with the rules of law.

Further, the base material information in the present embodiment may refer to information containing address contents. After the information is primarily read from the third party data system, the information that does not include address content may be filtered to obtain a plurality of base material information.

For ease of understanding, table 1 shows a plurality of pieces of base material information in the present embodiment.

Further, obtaining a training data set based on the plurality of POI basis data, comprising:

and acquiring initial labels respectively corresponding to the POI basic data.

And adjusting the initial labels of the POI basic data by using a supervised adjustment algorithm to obtain target labels corresponding to the POI basic data.

And determining the POI basic data and the target label of the POI basic data as a training sample, obtaining a plurality of training samples corresponding to the POI basic data, and determining the training samples as a training data set.

Optionally, acquiring the initial tags corresponding to the plurality of POI basic data respectively may include acquiring the initial tags corresponding to the plurality of basic data respectively through the LLM model.

The initial tag may be a task processing result preliminarily set for POI basic data. The target label can be a task processing result with higher precision obtained by optimizing the label on the basis of the initial label of the POI basic data. That is, the target tag may be the true task processing result of the POI base data, and the target tag may be used as a true value to determine the model error.

The supervised Fine-Tuning (SFT) algorithm in this embodiment may refer to Tuning to a specific NLP task by performing adjustments on the marked supervision data after PT (Pre-training) large models. In the adjustment process, parameters of the model can be adjusted according to the supervision signals of the task so as to improve the performance of the task. SFT typically requires a relatively small set of labeled data because the large model PT has been pre-trained on large-scale unlabeled data with some generalization capability. SFT allows models to accomplish various NLP tasks through supervised learning without having to train a completely new model from the beginning.

In the technical scheme of the embodiment, initial labels corresponding to the POI basic data are acquired, and the initial acquisition of the labels of the POI basic data is realized. And then, the initial labels of the POI basic data are adjusted through a supervised adjustment algorithm to obtain target labels corresponding to the POI basic data respectively, so that optimization of the target labels of the POI basic data is realized, the predicted targets of the POI basic data can be more accurately represented, the target labels of the POI basic data and the POI basic data are further determined to be a training sample, after the training samples corresponding to the POI basic data are obtained, when the POI generating model is trained through the training samples, the POI generating model can be enabled to learn the target labels of the POI basic data more effectively, network learning with higher precision is realized, and network training precision is further improved.

Further, on the basis of any one of the embodiments, adjusting the initial tag of each of the plurality of POI basic data by using a supervised adjustment algorithm to obtain a target tag corresponding to each of the plurality of POI basic data, including:

and determining the exclusive POI basic data belonging to the POI task type from the POI basic data according to the POI basic data and the initial labels respectively corresponding to the POI basic data.

And performing supervised adjustment on the trained LLM by using the exclusive POI basic data of the POI task type to obtain a target LLM of the POI task type.

And extracting the labels of the exclusive POI basic data of the POI task type by utilizing the target LLM of the POI task type to obtain target labels of the exclusive POI basic data, so that each exclusive POI basic data is determined to be the target label corresponding to the POI basic data, and the target labels respectively corresponding to the POI basic data are obtained.

Optionally, determining the exclusive POI base data belonging to the POI task type from the plurality of POI base data may include: according to the task processing types of the initial tags corresponding to the POI basic data respectively, determining target initial tags with the task processing types matched with the POI task types, and determining the POI basic data of training data of the target initial tags corresponding to the POI task types as exclusive POI basic data of the POI task types.

Further, matching a task processing type with a POI task type may mean that the task processing type is the same as the POI task type.

In this embodiment, each POI task type may train a corresponding target LLM. The target LLM corresponding to each POI task type can be used for re-obtaining the tag of the exclusive POI basic data corresponding to the POI task type, so as to obtain the target tag of the exclusive POI basic data. That is, the dedicated POI basic data of each POI task type may be input to the target LLM of the POI task type, and the target tag of the dedicated POI basic data may be obtained.

Alternatively, the different POI task types may each correspond to proprietary POI base data. The tag expression of the target tags of the proprietary POI base data belonging to different POI task types may be different.

For example, the POI task type is an address resolution type, and the target label of the training sample belonging to the POI task type is an address resolution result. The POI task type is an intention recognition type, and the target label of the training sample belonging to the POI task type is a text understanding result. The POI task type is a label classification type, and the target labels of training samples belonging to the POI task type are the categories of the samples. The POI task type is a name matching type, and the target labels of the training samples belonging to the POI task type are name matching results with the same or different names. The POI task type is an address generation type, and a target label of a training sample belonging to the POI task type is the generated address.

The POI task type is an information extraction type, and a target label of a training sample belonging to the POI task type is extracted key information. The POI task type is a production coaching type, and a target label of a training sample belonging to the POI task type is a coaching instruction of a production process. The POI task type is a content understanding type, and the target label of the training sample belonging to the POI task type is a content understanding result. The POI task type is an information auditing type, and a target label of a training sample belonging to the POI task type is an auditing result of passing or failing of information auditing. The POI task type is a geocoding type, and the target label of the training sample belonging to the POI task type is a geocoding result. The POI task type is a chain finger hooking type, and a target label of a training sample belonging to the POI task type is a link result of an associated task. The POI task type is an address complement type, and the target label of the training sample belonging to the POI task type is a complemented address.

Of course, the above-described manner of forming the plurality of POI task types and the target tags of the respective POI task types is merely exemplary, and does not constitute a specific limitation on the POI task types and the target tags of the respective POI task types.

In the technical scheme of the embodiment, the exclusive POI basic data belonging to each POI task type is determined according to the plurality of POI basic data and the initial tag of each POI basic data, and further the exclusive POI basic data of the POI task type is utilized to perform supervised adjustment on the LLM of the POI task type, so that the LLM of each task type can learn the structure and the characteristics of the POI basic data, and the method has a stronger task processing effect.

Further, on the basis of any one of the above embodiments, obtaining initial tags corresponding to the POI basic data respectively includes:

acquiring initial labels respectively corresponding to the POI basic data by utilizing the trained LLM;

or, in response to a labeling operation performed by the user on part of the POI basic data in the POI basic data, obtaining an initial tag of the part of the POI basic data.

Optionally, downloading or reading the trained LLM from the open source node is also included.

Optionally, after the initial labels corresponding to the plurality of POI basic data are obtained by using the obtained LLM, the initial labels corresponding to the plurality of POI basic data may be displayed. And detecting labeling operation which is executed by the user for part of POI basic data in the POI basic data.

Wherein the training derived LLM may include an input interface. And the input interface can be used for inputting the POI basic data into the LLM to obtain an understanding result output by the LLM, and determining the understanding result as an initial tag of the POI basic data.

According to the technical scheme, the initial labels respectively corresponding to the POI basic data are acquired through the trained LLM, and the acquiring efficiency and accuracy of the POI basic data can be improved by utilizing the trained LLM. Or the initial tag of the POI basic data can be obtained by using manual labeling, the personalized characteristic of the tag can be reflected by the POI basic data through manual labeling, and the labeling accuracy is improved.

Fig. 3 shows a schematic diagram of a second embodiment of the present disclosure, differing from the previous embodiments in that step 201 is further refined, the method may include:

301. and acquiring a plurality of basic material information.

302. And carrying out corpus analysis on the plurality of basic material information through the training-obtained pre-training model to obtain a plurality of POI basic data.

The Pre-training model may be a Pre-training (PT) large model.

303. And acquiring initial labels respectively corresponding to the POI basic data.

304. And determining the exclusive POI basic data belonging to the POI task type from the POI basic data according to the POI basic data and the initial labels respectively corresponding to the POI basic data.

305. And performing supervised adjustment on the trained LLM by using the exclusive POI basic data of the POI task type to obtain a target LLM of the POI task type.

306. And extracting the labels of the exclusive POI basic data of the POI task type by utilizing the target LLM of the POI task type to obtain target labels of the exclusive POI basic data, so that each exclusive POI basic data is determined to be the target label corresponding to the POI basic data, and the target labels respectively corresponding to the POI basic data are obtained.

307. And determining the POI basic data and the target label of the POI basic data as a training sample, obtaining a plurality of training samples corresponding to the POI basic data, and determining the training samples as a training data set.

In this embodiment, data adjustment is performed through PT to obtain POI basic data with higher text expression strength, and the FST is used for various POI task processing modules, so that the obtained POI basic data and target labels of the POI basic data are more suitable for the various POI task processing modules, and when training the POI generating model by using the training data set formed by the POI basic data and the target labels thereof, the POI generating model can learn characteristics related to various task processing, so that various task processing can be more effectively and accurately processed, the target POI generating model obtained by training has stronger generalization capability, and more efficient and accurate task processing functions can be provided.

In one possible design, training the POI generating model using the training data set to obtain a target POI generating model obtained by training may include:

determining an initial network model of the POI generation model based on the trained LLM;

training the POI generation model by using the training data set and the initial network model until the POI generation model meets the preset target ending condition, and determining the obtained POI generation model as a target POI generation model.

Optionally, determining the initial network model of the POI generation model based on the training obtained LLM may include: based on the trained LLM, reading the parameters of the input layer and the parameters of the middle layer of the trained LLM, determining the parameters of the input layer of the LLM as the parameters of the input layer of the POI generating network, determining the parameters of the middle layer of the LLM as the parameters of the middle layer of the POI generating model, initializing the parameters of the output layer of the POI generating model, and obtaining an initial network model composed of the input layer, the middle layer and the output layer with known parameters.

Optionally, training the POI generating model by using the training data set and the initial network model until the POI generating model meets a preset target ending condition, and determining the obtained POI generating model as the target POI generating model may include: and inputting the POI basic data in the training data set into the initial network model to obtain a prediction result of the initial network model on the POI basic data. And judging that the POI generation model meets a preset target ending condition based on a predicted result of the POI basic data and a target label, and if the POI generation model meets the preset target ending condition, determining the initial network model as a target POI generation model. If not, updating the initial network model, and returning to inputting the POI basic data in the training data set into the initial network model for continuous execution.

Further, based on the predicted result and the target label of the POI basic data, the determining that the POI generating model meets the preset target ending condition may include: calculating a loss value of the initial network model based on a predicted result of the POI basic data and a target label and combining a loss function, and if the loss value is smaller than or equal to an error threshold value, determining that the POI generation model meets a preset target ending condition; if the loss value is larger than the error threshold value, determining that the POI generation model does not meet the preset target ending condition.

The loss function and the error threshold in the present disclosure may be set according to the use requirement, which is not limited in this embodiment. For example, the loss function may be any one of a mean square error function and a cross entropy error function.

Further, based on the predicted result and the target label of the POI basic data, the determining that the POI generating model meets the preset target ending condition may include: calculating the accuracy of the initial network model based on the predicted result of the POI basic data and the target label, and if the accuracy is greater than or equal to an accuracy threshold, determining that the POI generation model meets a preset target ending condition; if the accuracy is smaller than the accuracy threshold, determining that the POI generation model does not meet the preset target ending condition.

According to the technical scheme, the obtained LLM after training can be determined to be the initial network model of the POI generation model, the obtained LLM after training is reused in the training process of the POI generation model, the training difficulty of the POI network model is reduced, the training expense is reduced, the existing LLM is trained through the training data set, the existing LLM can further learn knowledge of a plurality of training samples corresponding to a plurality of POI task types, the obtained target POI generation model after training can support processing tasks corresponding to the plurality of POI task types respectively, and the training efficiency of the target POI generation model is effectively improved.

Further, on the basis of any one of the embodiments, the task processing module corresponding to each of the plurality of POI task types includes any one or more of a tag classification module, a name matching module, an address generation module, an information extraction module, an address resolution module, an intention recognition module, a production coaching module, a content understanding module, an information auditing module, a geocoding module, a chain finger hooking module, and an address completion module.

The multiple POI task types may each correspond to a respective training sample. The POI basic data of each training sample may refer to a text for describing the POI field, and the text may include at least one attribute data such as POI name, address, phone call, business hours, status, etc.

The tag classification model may be used to perform tag setting on the inputted POI data. The name matching model may be used to determine whether the two POI data entered are identical. The address generation model may be used to generate detailed address information traffic for the incoming POI base data. The information extraction model may be used to extract address key information from the inputted POI base data in a format. The address resolution model may be used to address the incoming POI base data. The intent recognition model may be used to logically analyze the inputted POI base data, understand the obtained understanding text. The production coaching model can be used for assisting the instruction of equipment, products or safety and other information contained in the POI basic data. The content understanding model can be used for carrying out content analysis on POI basic data to obtain content analysis results. The information auditing model can be used for auditing the compliance of POI basic data. For example, auditing of laws and regulations may be performed. The geocoding model may be used to convert entity location and attribute information in the POI base data into computer-storable coding.

The chain finger hooking model can be used for establishing task links for a plurality of subtasks in the project or judging whether more than two subtasks have a link relation. The detailed information, state and progress of each task can be quickly oriented and tracked through chain finger hanging. The chain refers to a hooking relation for establishing basic data of two POIs. The link finger effect can influence the services such as POI basic data quality, POI data coverage, commercialized income, navigation promotion, behavior, super-content supply and the like to a great extent.

Optionally, the target tag of the dedicated POI basic data corresponding to the tag classification model may be the category to which the POI basic data belongs.

Illustratively, the specific POI base data corresponding to the tag classification model is, for example: judging POI classification according to the following POI names and address information: POI name: the appliance brand name P1; address intelligence: p2 is available in the city of P3. The target tags are, for example: a household appliance.

Optionally, the target tag of the proprietary POI basic data corresponding to the information extraction model is key information extracted from the POI basic data.

Illustratively, the proprietary POI base data corresponding to the information extraction model is, for example: POI base data: please read POI names, addresses, cities, phones and business hours from the following unstructured text: new store in city a, camping wind coffee shop [ DRIXX BASE ], address: line a1 and line 4 in city a have marking positions and are open in business hours: 12:00-21:00. Target tag of the POI base data: POI name: [ DRIX BASE ]; address: line a1, line 4, has a marking bit; business hours: 12:00-21:00".

Alternatively, the proprietary POI base data corresponding to the name matching model may be two POI base data. The target label is a judging result of whether the input two POI basic data are the same.

Illustratively, the two POI base data are POI1 and POI2, respectively. Wherein, POI 1: tengwang, POI2: tengwang pavilion. Target tag: different from each other.

Optionally, the specific POI basic data corresponding to the address resolution model may be address information. The target tag may be the result of the identification of the address entity. The address entity may be identified by, for example, province, city, county, village/street, village/community, road, house number, point of interest, building number, unit number, floor number, house number, etc.

Illustratively, suppose that the proprietary POI base data corresponding to the address resolution model is: address resolution is performed according to the following addresses: f5 and F6 are intersected to 300 m northly F7 company in F3 region F4 of F2 City of F1. The target labels are as follows: f1 province [ province tag ], F2 city [ city tag ], F3 region [ region tag ], F4 office [ point of interest tag ], F5 way [ way tag ], F6 way [ way tag ], intersection [ cross tag ], east [ direction tag ], 300 meters [ distance tag ], north road F7 company [ point of interest tag ].

Alternatively, the intent recognition model may be used, for example, to detect invalidity of the user's evaluation information. Illustratively, the proprietary POI base data of the intent recognition model may be: "judging whether the failure event occurs according to the following information: the coffee shop that needs to queue the reservation, said that the coffee is drunk well, when the next reservation goes to try to get down ", the target label is: 0, i.e. no failure event. If the specific POI basic data of the intention recognition model is: "judging whether the event is failure according to the following information: once again, and once again, both doors are closed ", the target tag may be 1, i.e., a failure event.

Of course, the foregoing examples are merely illustrative of the technical solutions of the present disclosure, and are not to be construed as being particularly limiting.

According to the technical scheme, through the arrangement of any plurality of task processing modules, the target POI generation model can simultaneously support processing tasks corresponding to various POI task types, has stronger generalization capability, can analyze POI data to be analyzed of any task type, can realize more complex POI data analysis, expands the application scene of the target POI generation model, and further improves the utilization rate of the target POI generation model.

For the purpose of illustrating the technical solution of the present disclosure in detail, fig. 4 shows a schematic flowchart of the data processing method provided by the present disclosure. Referring to the data processing method shown in fig. 4, the training data set includes: the technical scheme of the disclosure is described in detail by taking an information extraction sample, a label classification sample and a name matching sample as examples.

The information extraction sample may be, for example, "POI basic data: please read POI names, addresses, cities, phones and business hours from the following unstructured text: new store in city a, camping wind coffee shop [ DRIXX BASE ], address: line a1 and line 4 in city a have marking positions and are open in business hours: 12:00-21:00. Target tag: POI name: [ DRIX BASE ]; address: line a1, line 4, has a marking bit; business hours: 12:00-21:00".

The label classification sample may be, for example: "POI base data: judging the classification of the POI according to the learned knowledge of you and the name, address and comment information of the POI: name: women's brand name M I address information: female clothes in B1 area of B city. Target tag: women's dress shop).

The name matching samples may be, for example: "POI base data: judging whether the following two POI names are the same according to the learned knowledge: POI 1: tengwang, POI2: tengwang pavilion. Target tag: different from each other.

And carrying out model updating on the trained LLM by a plurality of training samples in the training data set to obtain a target POI generation model. After the target POI generation model is obtained, engineering deployment can be carried out on the target POI generation model, and an information extraction module in the target POI generation model is utilized to execute an information extraction task, a label classification model is utilized to execute a label classification task and a name matching model is utilized to execute a name matching task.

Of course, the sample types and task types are only illustrative, and in practical applications, the sample types are more abundant and the sample numbers are more huge. Further, the target POI generating model supporting multitasking can be trained by using a richer sample data set.

Fig. 5 shows a schematic diagram of a third embodiment of the present disclosure, which is different from the foregoing embodiment in that after obtaining the training-obtained target POI generation model, the method further includes:

501. according to a plurality of preset POI task types, a test data set is obtained, wherein the test data set comprises a plurality of test samples corresponding to the POI task types, and the test samples comprise test POI basic data for testing and target labels corresponding to the test POI data.

In one possible design, the test data set may include partitioning a portion of the training samples from the training data set and transferring the partitioned portion of the training samples to the test samples, the partitioned portion of the training samples no longer belonging to the training samples in the training data set.

Further, step 501 may include: and carrying out corpus analysis on the training material information through the training-obtained pre-training model to obtain training POI basic data. Thereafter, a target LLM may be utilized to obtain a target tag for testing the POI base data. The target LLM is the LLM updated using the training samples in the previous embodiment.

502. And carrying out test processing on the target POI generation model according to the test data set to obtain a test result of the target POI generation model.

Optionally, according to the test data set, performing test processing on the target POI generation model, including: and inputting the test POI basic data of the test sample in the test data set into the target POI generation model to obtain a task processing result of the target POI generation model on the test POI basic data.

Further, obtaining the test result of the target POI generation model may include: and carrying out result analysis processing on the task processing results of each training sample in the test data set to obtain the test result of the target POI generation model.

503. And if the test result of the target POI generation model meets the target test condition, releasing the target POI generation model.

Optionally, step 502 may include: and carrying out test processing on the target POI generation model according to the test data set to obtain the generation accuracy of the target POI generation model. And if the generation accuracy is greater than or equal to a preset generation accuracy threshold, determining that the target test condition is met as a test result of the target POI generation model.

Alternatively, the target test condition may refer to a condition for constraining the test result. Specifically, the test index meets a preset index threshold. For example, when the test index is an accuracy rate, the target test condition may refer to the accuracy rate being greater than or equal to a preset accuracy rate threshold. In addition, the test processing of the target POI generation model can also test indexes such as recall rate, precision and the like. The target test conditions may be set according to the test index, which is not limited in this embodiment.

Further, publishing the target POI generation model may include deploying the target POI generation model through engineering.

Wherein, the publishing the target POI generation model through engineering deployment can comprise: and establishing application interfaces according to task processing modules respectively corresponding to the multiple POI task types in the target POI generation model, and accessing each application interface into the user terminal. Furthermore, the user terminal can input POI data to be analyzed into the target POI generation model through the application interface, so that the target POI generation model can perform task processing on the POI data received through the application interface, and a task processing result is obtained.

According to the technical scheme, the target POI generation model is tested through the test data set, verification of the using effect of the target POI generation model is achieved, and the target POI generation model is issued when verification passes. Compared with the strategy of obtaining the target POI generation model, namely the release strategy, the strategy of re-releasing the target POI generation model through use verification can be realized through testing, invalid model release can be avoided, the failure rate of model release is reduced, and the user experience is influenced.

Further, on the basis of any one of the above embodiments, according to the test data set, performing test processing on the target POI generating model to obtain a test result of the target POI generating model, including:

And performing task processing on the test POI basic data in the test sample by using the target POI generation model to obtain a task processing result corresponding to the test POI basic data.

And determining the test score of the test POI basic data according to the task processing result of the test POI basic data and the target label of the test POI basic data.

And calculating test total scores corresponding to the test data sets according to the test scores respectively corresponding to the test POI basic data, and determining the test total scores as test results of the target POI generation model.

Optionally, the task processing result of the target POI generating model on the test POI basic data may be a processing result corresponding to the target task processing module in the task processing modules corresponding to the multiple POI task types respectively.

Determining the test score of the test POI basic data according to the task processing result of the test POI basic data and the target label of the test POI basic data may include performing similarity calculation on the task processing result of the test POI basic data and the target label of the test POI basic data to obtain the similarity of the task processing result and the target label, and determining the test score of the test POI basic data based on the similarity.

Further, the similarity may be multiplied by a preset full score value, and the obtained product is determined as a test score of the test POI basic data. For example, assuming that the similarity is 80% and the full score is 100, 80% by 100 is given as 80 score, which is the test score.

Determining a test score of the test POI basic data according to the task processing result of the test POI basic data and the target label of the test POI basic data, wherein the determination may include determining the first numerical value as the test score of the test POI basic data if the task processing result of the test POI basic data is the same as the target label of the test POI basic data; and if the task processing result of the test POI basic data is different from the target label of the test POI basic data, determining the second numerical value as the test score of the test POI basic data. Wherein the first value may be 1 and the second value may be 0. Of course, in practical applications, the values of the first value and the second value may also be other values, which are not limited in this disclosure.

According to the technical scheme, task processing is carried out on test POI basic data in a test sample through a target POI generation model, a task processing result corresponding to the test POI basic data is obtained, and testing of the test POI basic data is completed. And then determining the test scores of the test POI basic data by using the task processing results of the test POI data and the target labels. The test scores corresponding to the test samples can be used for calculating a test total score corresponding to the test data set, and the test total score can reflect the test result of the whole test data set. And determining the test result of the target POI generation model through the test score of the whole test data set, and realizing the numerical acquisition of the test result of the target POI generation model. The using effect of the target POI generating model can be comprehensively reflected through the test total score, and then when the target POI generating model is judged by the test total score, the judgment of the target test condition can be rapidly and accurately finished, and the testing efficiency and the accuracy of the target POI generating model are improved.

Further, according to the test scores respectively corresponding to the plurality of test samples, calculating a test total score corresponding to the test data set, including:

determining sample weights of the test samples corresponding to the test POI basic data by using the test complexity of the test samples;

and respectively corresponding sample weights to the POI basic data, and carrying out weighted summation on the test scores respectively corresponding to the POI basic data to obtain a test total score corresponding to the test data set.

Alternatively, the test complexity of the test sample may be obtained by a complexity acquisition model. Further, the complexity acquisition model may be a supervised model. The sample complexity training may be obtained using existing POI samples and existing POI samples. The specific training process may refer to the related art, and will not be described in detail herein.

Alternatively, the test complexity of the test sample may also be obtained by manual setting, for example, a complexity setting page of the test sample may be displayed, and the sample complexity of the test sample may be obtained in response to a complexity setting operation performed by the user on the complexity setting of the test sample.

Alternatively, the test total score for a test dataset may be calculated by the following formula:

Wherein, task _i The method is a POI task processing module, and N is the number of tasks. M is the number of test samples of the ith task processing module. Question_weight _(i,j) And (5) the sample weight of the jth POI basic data in the ith POI task. Prect_Scare _(i,j) And (5) scoring the test of the jth POI basic data in the ith POI task. Poi_score is the total Score of the test corresponding to the test dataset.

According to the technical scheme, according to the test complexity of each test sample, the sample weight of the test POI basic data corresponding to each test sample is determined, and the sample weights corresponding to the POI basic data are respectively involved in the calculation of the test total score of the test data set, so that the test total score of the test data set is obtained by calculation according to the test complexity of the test samples. The test samples with different test complexity have different weight influences on the test total score of the test data set, so that the calculation process of the test total score of the test data set is finer, and the more accurate test total score is obtained.

Fig. 6 shows a schematic diagram of a fourth embodiment of the present disclosure, referring to a data processing method shown in fig. 6, the method may comprise the steps of:

601. the method comprises the steps of obtaining a target POI generation model obtained through training, wherein the target POI generation model is obtained through training in any one of the above embodiments, and comprises task processing modules corresponding to a plurality of POI task types respectively.

602. And performing task processing on the POI data to be analyzed in parallel through task processing modules respectively corresponding to the task types of the multiple POIs in the target POI generation model to obtain a task processing result.

Alternatively, the POI data to be analyzed may be data associated with any POI task type. The target task processing module may include at least one.

Further, step 602 may include: and acquiring processing results respectively corresponding to the POI data to be analyzed in at least one target task processing module, and determining the processing results respectively corresponding to the at least one target task processing module as task processing results of the POI data to be analyzed.

603. And outputting a task processing result of the POI data to be analyzed.

In the technical scheme, the target POI generation model obtained through training can be used for performing task processing on the POI data to be analyzed, and a task processing result is obtained. The target POI generating model supporting the task processing modules corresponding to the POI task types can be used for more accurately analyzing POI data to be analyzed, and more complex task processing can be supported compared with the traditional POI task model, so that task processing efficiency and accuracy are improved.

Optionally, step 602 may further include: and acquiring POI data to be analyzed.

Further, acquiring the POI data to be analyzed may include: and acquiring POI data to be analyzed received by the application interface based on the application interfaces associated with the task processing modules respectively corresponding to the multiple POI task types.

Optionally, the task processing modules corresponding to the multiple POI task types may use the same application interface, and the application interface may be opened to the user terminal. The user can provide POI data to be analyzed through the user terminal, the POI data to be analyzed is received by the application interface and is sent to the target POI generation model through the application interface, and the target POI generation model performs task processing on the POI data to be analyzed through target processing in the task processing modules respectively corresponding to the task types of the POIs, so that a task processing result is obtained.

Taking the information extraction task as an example, a user can provide any one or more of question and answer information, voice dialogue information and the like of a user and a bank staff, a board image of business hours of a bank business point, business bulletin files issued by a bank and the like as POI data to be analyzed to a target POI generation model through a user terminal. The target POI generation model can extract information such as bank address, business hours, open time and the like from the POI data to be analyzed.

Taking a tag classification task as an example, a user can provide address description text to be classified to a target POI generation model through terminal equipment. The target POI generation model can divide the input address description text into category labels, and the POI data to be analyzed can be, for example, "POI information is classified according to the learned knowledge: paint brand 1 (AA road store) address information: AAAA address. The target POI generation model may output the store category of the POI data to be analyzed: and (5) a coating shop.

Further, the target POI generation model can analyze POI data to be analyzed through the plurality of task processing modules to obtain processing results respectively corresponding to the POI data to be analyzed in at least one task processing module, so that task processing results obtained by analyzing the POI data to be analyzed more comprehensively are higher in accuracy.

In the process of executing the name matching task on the POI data by the target POI generation model, a chain finger hooking task can be used for confirming whether task association exists between two POI data so as to improve the matching precision of the name matching task.

For example, in the process of executing the address resolution task on the POI data by the target POI generation model, a chain finger hooking task, an address coding task, an address mining task and the like can be used to obtain an address resolution result with higher precision.

For example, the POI data to be analyzed may be a plurality of pieces of address information, and the address information associated with each piece of address information may be acquired by the address mining task. And address fusion analysis is carried out through the address information and the address information related to the address information, so that a more accurate analysis result is obtained.

Further, on the basis of any one of the above embodiments, determining a POI generation model to be trained includes:

respectively connecting task processing modules corresponding to the task types of the POIs with a task judging module to obtain a plurality of task processing modules and an output layer determined by the task judging module;

determining a POI generation model based on an input layer and an intermediate layer of the LLM and combining an output layer;

the task judging module is used for selecting a predicted result from the results respectively output by the task processing modules and determining a target predicted result as a task processing result of the POI generation model.

Wherein, based on the input layer and the middle layer of the LLM model, combining with the output layer, determining the POI generation model comprises: connecting an input layer of the LLM model with an intermediate layer, connecting the intermediate layer with an output layer, and obtaining a POI generating network formed by the input layer, the intermediate layer and the output layer.

Alternatively, the input layers and middle layers of the LLM model may be those of an existing large language model, which the present embodiment is not limited to. The input layer of the LLM model may receive the inputted POI data. The middle layer can perform feature calculation on the input POI data to obtain POI features. The POI features are input into the output layer, task processing is carried out on the POI features through a plurality of task processing modules of the output layer, and results generated by the task processing modules are obtained.

In one possible design, each task processing module may also output a probability value corresponding to the predicted outcome. The prediction results and probability values generated by the task processing modules are input into the task judging module, and the task judging module can determine the first N maximum probability values and determine that the task processing modules corresponding to the first N maximum probability values are target task processing modules. The target prediction results generated by the target task processing modules can be determined as the task processing results of the POI generation model.

In yet another possible design, the task discrimination module may determine the task of the input prompt (prompt)

) And performing task type detection to obtain N target task types of the input POI data, determining task processing modules corresponding to the N target task types as target task processing modules, and determining target prediction results generated by the N target task processing modules as task processing results of the POI generation model. The hint word may be extracted from the inputted POI data by a hint word acquisition algorithm. The Prompt word acquisition algorithm may be any Prompt word acquisition algorithm, such as a preset promt function.

Wherein N is a positive integer greater than or equal to 1. N is less than or equal to the number of modules of the plurality of task processing modules.

In the technical scheme of the disclosure, task processing modules corresponding to a plurality of POI task types are connected with a task judging module to obtain an output layer formed by the task processing modules and the task judging module. And connecting the input layer, the middle layer and the output layer of the LLM model to obtain the POI generation model. The task judging module is used for selecting a predicted result from output results respectively corresponding to the task processing modules, wherein the predicted result is the output of the POI generation model. The task processing modules corresponding to the POI task types can be used for realizing multitasking, and the task judging module can be used for deciding the final output, so that the prediction result is output while the multitasking is realized, the task processing scene of the POI generation model is expanded, and the accurate prediction result is obtained.

Fig. 7 is a schematic diagram of a fifth embodiment of the present disclosure, and referring to a data processing apparatus shown in fig. 7, the data processing apparatus may include the following units:

the acquiring unit 701 is configured to acquire a training data set according to a plurality of preset POI task types, where the training data set includes a plurality of training samples corresponding to the plurality of POI task types.

The determining unit 702 is configured to determine a POI generation model to be trained, where the POI generation model includes task processing modules corresponding to a plurality of POI task types respectively.

The training unit 703 is configured to train the POI generating model by using a training data set to obtain a target POI generating model obtained by training, and the target POI generating model uses task processing modules corresponding to a plurality of POI task types to perform task processing on the POI data to be analyzed in parallel, so as to obtain a task processing result.

As one embodiment, the acquisition unit includes:

the data acquisition module is used for carrying out corpus analysis on the basic material information through a pre-training model obtained through training to obtain POI basic data;

and the training acquisition module is used for acquiring a training data set based on the POI basic data.

As yet another embodiment, a training acquisition module includes:

the initial acquisition sub-module is used for acquiring initial labels respectively corresponding to the POI basic data;

the label adjustment sub-module is used for adjusting the initial labels of the POI basic data through a supervised adjustment algorithm to obtain target labels corresponding to the POI basic data respectively;

the training acquisition sub-module is used for determining the POI basic data and the target label of the POI basic data as a training sample, obtaining a plurality of training samples corresponding to the POI basic data, and determining the training samples as a training data set.

As yet another embodiment, the tag adjustment sub-module is specifically configured to:

determining exclusive POI basic data belonging to the POI task type from the POI basic data according to the POI basic data and initial labels respectively corresponding to the POI basic data;

utilizing the exclusive POI basic data of the POI task type to carry out supervision adjustment on the trained LLM to obtain a target LLM of the POI task type;

As yet another embodiment, the initial acquisition sub-module includes:

As a further embodiment, the training unit comprises:

the initial acquisition module is used for determining an initial network model of the POI generation model based on the trained LLM;

The model training module is used for training the POI generating model by utilizing the training data set and the initial network model until the POI generating model meets the preset target ending condition, and determining the obtained POI generating model as a target POI generating model.

As yet another embodiment, the plurality of POI task types includes any of a tag classification task, a name matching task, an address generation task, an information extraction task, an address resolution task, an intent recognition task, a production coaching task, a content understanding task, an information auditing task, a geocoding task, a chain finger hooking task, an address completion task.

As yet another embodiment, further comprising:

the test acquisition unit is used for acquiring a test data set according to a plurality of preset POI task types, wherein the test data set comprises a plurality of test samples corresponding to the POI task types, and the test samples comprise test POI basic data for testing and target labels corresponding to the test POI data;

the model test unit is used for carrying out test processing on the target POI generation model according to the test data set to obtain a test result of the target POI generation model;

and the model issuing unit is used for issuing the target POI generation model if the test result of the target POI generation model meets the target test condition.

As yet another embodiment, a model test unit includes:

the result acquisition module is used for performing task processing on the test POI basic data in the test sample by utilizing the target POI generation model to obtain a task processing result corresponding to the test POI basic data;

the score determining module is used for determining the test score of the test POI basic data according to the task processing result of the test POI basic data and the target label of the test POI basic data;

the result determining module is used for calculating the total test score corresponding to the test data set according to the test scores respectively corresponding to the plurality of test POI basic data, and determining the total test score as the test result of the target POI generation model.

As yet another embodiment, the result determination module includes:

the weight determining submodule is used for determining sample weights of the test samples corresponding to the test POI basic data by using the test complexity of the test samples;

and the weighted summation sub-module is used for carrying out weighted summation on the test scores corresponding to the POI basic data respectively according to the sample weights corresponding to the POI basic data respectively to obtain the test total score corresponding to the test data set.

As still another embodiment, the determining unit includes:

The output determining unit is used for respectively connecting the task processing modules corresponding to the task types of the POIs with the task judging module to obtain the output layers determined by the task processing modules and the task judging module;

the model determining unit is used for determining a POI generation model based on the input layer and the middle layer of the LLM and combining with the output layer;

the task judging module is used for selecting a target predicted result from the predicted results respectively output by the task processing modules and determining the target predicted result as a task processing result of the POI generation model.

Fig. 8 is a schematic diagram of a sixth embodiment of the disclosure, and referring to a data processing apparatus shown in fig. 8, the data processing apparatus may include the following units:

the model obtaining unit 801 is configured to obtain a target POI generating model obtained by training, where the target POI generating model is obtained by training based on the data processing method provided by any of the embodiments, and the target POI generating model includes task processing modules corresponding to a plurality of POI task types respectively;

the task execution unit 802 is configured to perform task processing on the POI data to be analyzed in parallel through task processing modules corresponding to the multiple POI task types in the target POI generation model, where the task processing result refers to a target prediction result generated by a target task processing module in the task processing modules corresponding to the multiple POI task types;

And a result output unit 803 for outputting a task processing result of the POI data to be analyzed.

Note that, the target POI generation model in this embodiment is not a generation model for a specific user, and cannot reflect personal information of a specific user. It should be noted that, the base material information in this embodiment comes from the public data set.

In the technical scheme of the disclosure, the related processes of collecting, storing, using, processing, transmitting, providing, disclosing and the like of the personal information of the user accord with the regulations of related laws and regulations, and the public order colloquial is not violated.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

According to an embodiment of the present disclosure, the present disclosure also provides a computer program product comprising: a computer program stored in a readable storage medium, from which at least one processor of an electronic device can read, the at least one processor executing the computer program causing the electronic device to perform the solution provided by any one of the embodiments described above.

Fig. 9 shows a schematic block diagram of an example electronic device 900 that may be used to implement embodiments of the present disclosure. 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 disclosure described and/or claimed herein.

As shown in fig. 9, the apparatus 900 includes a computing unit 901 that can perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 902 or a computer program loaded from a storage unit 908 into a Random Access Memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other by a bus 904. An input/output (I/O) interface 905 is also connected to the bus 904.

Various components in device 900 are connected to I/O interface 905, including: an input unit 906 such as a keyboard, a mouse, or the like; an output unit 907 such as various types of displays, speakers, and the like; a storage unit 908 such as a magnetic disk, an optical disk, or the like; and a communication unit 909 such as a network card, modem, wireless communication transceiver, or the like. The communication unit 909 allows the device 900 to exchange information/data with other devices through a computer network such as the internet and/or various telecommunications networks.

The computing unit 901 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of computing unit 901 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 901 performs the respective methods and processes described above, such as a data processing method. For example, in some embodiments, the data processing method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as the storage unit 908. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device 900 via the ROM 902 and/or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the data processing method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform the data 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), complex 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 disclosure 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 disclosure, a machine-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 machine-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. The server may also be a server of a distributed system or a server that incorporates a blockchain.

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 recited in the present disclosure may be performed in parallel or sequentially or in a different order, provided that the desired results of the technical solutions of the present disclosure are achieved, and are not limited herein.

The above detailed description should not be taken as limiting the scope of the present disclosure. 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 disclosure are intended to be included within the scope of the present disclosure.

Claims

1. A data processing method, comprising:

determining a POI generation model to be trained, wherein the POI generation model comprises a plurality of task processing modules corresponding to the POI task types respectively;

and training the POI generating model by using the training data set to obtain a target POI generating model obtained by training, and performing task processing on POI data to be analyzed in parallel by using a plurality of task processing modules respectively corresponding to the POI task types by using the target POI generating model to obtain a task processing result.

2. The method of claim 1, wherein the acquiring the training data set according to the preset plurality of POI task types includes:

carrying out corpus analysis on a plurality of basic material information through a pre-training model obtained through training to obtain a plurality of POI basic data;

and obtaining the training data set based on a plurality of POI basic data.

3. The method of claim 2, wherein the obtaining the training data set based on the plurality of POI base data comprises:

acquiring initial labels respectively corresponding to the POI basic data;

adjusting the initial labels of the POI basic data by a supervised adjustment algorithm to obtain target labels corresponding to the POI basic data;

and determining the POI basic data and the target label of the POI basic data as a training sample, obtaining a plurality of training samples corresponding to the POI basic data, and determining the training samples as the training data set.

4. The method according to claim 3, wherein the adjusting, by a supervised adjustment algorithm, the initial tag of each of the POI base data to obtain the target tag corresponding to each of the POI base data includes:

performing supervised adjustment on the trained LLM by using the exclusive POI basic data of the POI task type to obtain a target LLM of the POI task type;

and extracting labels from the exclusive POI basic data of the POI task type by utilizing the target LLM of the POI task type to obtain target labels of the exclusive POI basic data, so that each exclusive POI basic data is determined to be the target label of the corresponding POI basic data, and a plurality of target labels respectively corresponding to the POI basic data are obtained.

5. The method according to claim 3 or 4, wherein the obtaining initial tags corresponding to the POI base data respectively includes:

obtaining initial labels corresponding to the POI basic data respectively by utilizing the trained LLM;

or, in response to labeling operation performed by the user on part of POI basic data in the POI basic data, obtaining an initial tag of the part of POI basic data.

6. The method according to any one of claims 1-5, wherein training the POI generation model using the training dataset to obtain a training obtained target POI generation model comprises:

and training the POI generating model by utilizing the training data set and the initial network model until the POI generating model meets the preset target ending condition, and determining the obtained POI generating model as the target POI generating model.

7. The method of any of claims 1-6, wherein the task processing module to which the plurality of POI task types respectively correspond comprises any of a tag classification module, a name matching module, an address generation module, an information extraction module, an address resolution module, an intent recognition module, a production coaching module, a content understanding module, an information auditing module, a geocoding module, a chain finger hooking module, an address completion module.

8. The method according to any one of claims 1-7, wherein after the obtaining the training obtained target POI generation model, further comprising:

according to a plurality of preset POI task types, a test data set is obtained, the test data set comprises a plurality of test samples corresponding to the POI task types, and the test samples comprise test POI basic data for testing and target labels corresponding to the test POI data;

According to the test data set, testing the target POI generation model to obtain a test result of the target POI generation model;

and if the test result of the target POI generation model meets the target test condition, releasing the target POI generation model.

9. The method of claim 8, wherein the performing test processing on the target POI generating model according to the test data set to obtain a test result of the target POI generating model includes:

performing task processing on the test POI basic data in the test sample by using the target POI generation model to obtain a task processing result corresponding to the test POI basic data;

determining a test score of the test POI basic data according to the task processing result of the test POI basic data and the target label of the test POI basic data;

10. The method of claim 9, wherein the calculating a test total score corresponding to the test data set according to the test scores corresponding to the test POI base data, respectively, comprises:

and respectively corresponding to the sample weights according to the POI basic data, and carrying out weighted summation on the test scores respectively corresponding to the POI basic data to obtain the test total score corresponding to the test data set.

11. The method of any of claims 1-9, wherein the determining a POI generation model to be trained comprises:

respectively connecting a plurality of task processing modules corresponding to the POI task types with a task judging module to obtain a plurality of task processing modules and output layers determined by the task judging module;

determining the POI generation model based on an input layer and an intermediate layer of LLM and combining the output layer;

12. A data processing method, comprising:

obtaining a target POI generation model obtained through training, wherein the target POI generation model is obtained through training based on the data processing method of the claims 1-11, and comprises a task processing module corresponding to a plurality of POI task types respectively;

Performing task processing on the POI data to be analyzed in parallel through task processing modules respectively corresponding to a plurality of POI task types in the target POI generation model to obtain task processing results;

and outputting a task processing result of the POI data to be analyzed.

13. A data processing apparatus comprising:

the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a training data set according to a plurality of preset POI task types, and the training data set comprises a plurality of training samples corresponding to the POI task types;

the determining unit is used for determining a POI generating model to be trained, wherein the POI generating model comprises a plurality of task processing modules corresponding to the POI task types respectively;

the training unit is used for training the POI generating model by utilizing the training data set to obtain a target POI generating model obtained by training, and the target POI generating model carries out task processing on POI data to be analyzed in parallel by using a plurality of task processing modules respectively corresponding to the POI task types to obtain a task processing result.

14. The apparatus of claim 13, wherein the acquisition unit comprises:

And the training acquisition module is used for acquiring the training data set based on the POI basic data.

15. The apparatus of claim 14, wherein the training acquisition module comprises:

and the training acquisition sub-module is used for determining the POI basic data and the target label of the POI basic data as a training sample, obtaining a plurality of training samples corresponding to the POI basic data, and determining the training samples as the training data set.

16. The apparatus of claim 15, wherein the tag adjustment sub-module is specifically configured to:

17. The apparatus of claim 15 or 16, wherein the initial acquisition sub-module comprises:

18. The apparatus of any of claims 13-17, wherein the training unit comprises:

and the model training module is used for training the POI generating model by utilizing the training data set and the initial network model until the POI generating model meets the preset target ending condition, and determining the obtained POI generating model as the target POI generating model.

19. The apparatus of any of claims 13-18, wherein the task processing module to which the plurality of POI task types respectively correspond comprises any of a tag classification module, a name matching module, an address generation module, an information extraction module, an address resolution module, an intent recognition module, a production coaching module, a content understanding module, an information auditing module, a geocoding module, a chain finger hooking module, an address completion module.

20. The apparatus of any of claims 13-19, further comprising:

21. The apparatus of claim 20, wherein the model test unit comprises:

and the result determining module is used for calculating the total test score corresponding to the test data set according to the test scores respectively corresponding to the test POI basic data, and determining the total test score as the test result of the target POI generation model.

22. The apparatus of claim 21, wherein the result determination module comprises:

23. The apparatus according to any of claims 13-22, wherein the determining unit comprises:

the output determining unit is used for respectively connecting the task processing modules corresponding to the POI task types with the task judging module to obtain a plurality of task processing modules and output layers determined by the task judging module;

the model determining unit is used for determining the POI generation model based on the input layer and the middle layer of the LLM and combining the output layer;

24. A data processing apparatus comprising:

the system comprises a model acquisition unit, a data processing unit and a data processing unit, wherein the model acquisition unit is used for acquiring a target POI generation model obtained through training, the target POI generation model is obtained through training based on the data processing method of the claims 1-11, and the target POI generation model comprises a plurality of task processing modules corresponding to the POI task types respectively;

the task execution unit is used for carrying out task processing on the POI data to be analyzed in parallel through task processing modules respectively corresponding to the multiple POI task types in the target POI generation model to obtain a task processing result;

25. 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 method of any one of claims 1-11 or 12.

26. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-11 or 12.

27. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the method of any of claims 1-11 or 12.