Multi-task prediction model establishment method and device and prediction method and device

Abstract

The invention relates to the technical field of artificial intelligence, and provides a multi-task prediction model establishment method and device and a prediction method and device. The method comprises the following steps: acquiring first target business data; determining a primary task and a secondary task corresponding to the first target business data according to the importance degree of the task; training a first integrated tree model by utilizing the first target business data to obtain a main task model; screening out second target business data matched with the secondary task from the first target business data; and training the main task model by using the second target business data to obtain a main task model after increment fine adjustment, and taking the main task model after increment fine adjustment as a multi-task prediction model. According to the embodiment of the invention, the performance of the multi-task prediction model can be improved, and the multi-task prediction effect is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for establishing a multi-task prediction model, and a prediction method and device. Background Art

[0002] Using a multi-task model to predict targets can improve the stability of the overall model and reduce the cost of computing, storage, and maintenance. It can be applied to multi-target prediction in fields such as risk control and computer vision. Common multi-task model algorithms include multi-task logistic regression, multi-task neural network, and multi-task tree model. However, multi-task logistic regression has limitations in its ability to learn shared feature representations between tasks, and it is difficult to fully explore the feature associations and information sharing between different tasks. Multi-task neural networks have high data requirements and model complexity, and poor interpretability. Multi-task tree models can fully explore the feature association information of different tasks and have good interpretability, but the loss function of the existing tree model is mainly cross-entropy loss, which is not suitable for modeling class imbalanced samples and difficult samples. In addition, the existing multi-task tree model realizes multi-task learning by sharing features, and there is a "seesaw" phenomenon: when there are differences or conflicts between multiple tasks, optimizing one task may cause the performance of other tasks to drop significantly, and it is difficult to balance the performance of multiple tasks. Therefore, a method for establishing a multi-task prediction model is urgently needed to solve the problem that the existing multi-task tree model is difficult to effectively balance the performance of the model on multiple tasks, so as to improve the performance of the multi-task prediction model and enhance the multi-task prediction effect. Summary of the invention

[0003] In view of the problem that the current multi-task tree model is difficult to effectively balance the performance of the model on multiple tasks, this solution is proposed to overcome the above problem or at least partially solve the above problem.

[0004] On the one hand, an object of some embodiments of this specification is to provide a multi-task prediction model establishment method, the method comprising:

[0005] Acquire first target business data;

[0006] Determine the primary task and the secondary task corresponding to the first target business data according to the importance of the task;

[0007] Using the first target business data to train a first integrated tree model to obtain a main task model;

[0008] Filtering the first target business data to obtain second target business data matching the secondary task;

[0009] The main task model is trained using the second target business data to obtain an incrementally fine-tuned main task model, where the incrementally fine-tuned main task model is a multi-task prediction model.

[0010] Further, determining the primary task and the secondary task corresponding to the first target business data according to the importance of the task includes:

[0011] Determine, according to business scenarios and business requirements, multiple tasks corresponding to the first target business data;

[0012] Based on the importance of the tasks, the main tasks and the secondary tasks are screened from the multiple tasks.

[0013] Furthermore, based on the importance of the tasks, the main tasks and the secondary tasks are screened from the multiple tasks, including:

[0014] Randomly selecting a preset number of business data from the first target business data;

[0015] Using the preset amount of business data to establish training samples;

[0016] For the multiple tasks, respectively use the training samples corresponding to each task to iteratively train the second ensemble tree model to obtain the loss corresponding to each task;

[0017] The importance of each task is determined according to the loss corresponding to each task, and the main task and the sub-task are screened from the multiple tasks according to the importance of each task, wherein the importance of the main task is higher than that of the sub-task.

[0018] Further, the first target business data is used to train a first integrated tree model to obtain a main task model, including:

[0019] According to the business type of the main task, feature derivation is performed on the first target business data to obtain a first feature derivation result;

[0020] Dividing the first feature derived results and their corresponding label information to obtain a first training sample set and a first verification sample set;

[0021] The first training sample set and the first validation sample set are used to train a first ensemble tree model to obtain the main task model.

[0022] Furthermore, the feature derivation results and their corresponding label information are divided to obtain a first training sample set and a first verification sample set, including:

[0023] Constructing a classification sample set according to the feature-derived results and their corresponding label information;

[0024] Using the classification sample set to train a binary classifier;

[0025] Using the trained binary classifier to perform classification prediction on the classification sample set to obtain the classification prediction probability of each sample;

[0026] The samples whose classification prediction probability is less than a preset value are screened from the classification sample set to obtain a screening result, wherein the screening result is a first verification sample set, and the set consisting of other samples in the classification sample set except the screening result is a first training sample set.

[0027] Furthermore, the main task model is trained using the second target business data to obtain an incrementally fine-tuned main task model, wherein the incrementally fine-tuned main task model is a multi-task prediction model, including:

[0028] Obtaining a second training sample set and a second verification sample set according to the second target business data and its corresponding label information;

[0029] Selecting a preset number of training samples from the second training sample set to obtain a local sample training set, and selecting a preset number of verification samples from the second verification sample set to obtain a local sample verification set;

[0030] Using the local sample training set to train the main task model;

[0031] Inputting the local sample verification set into the trained main task model to obtain the corresponding output result;

[0032] Calculate the accuracy rate according to the output result corresponding to the trained main task model and the label information in the local sample verification set;

[0033] If the accuracy rate is not less than a preset threshold, the training of the main task model is completed to obtain the main task model after incremental fine-tuning, and the main task model after incremental fine-tuning is the multi-task prediction model.

[0034] Furthermore, in some embodiments, it also includes:

[0035] If the accuracy rate is less than a preset threshold, removing the local sample training set from the second training sample set to update the second training sample set, and selecting a preset number of training samples from the updated second training sample set as a new local sample training set;

[0036] Using a new local sample training set to train the trained main task model, so as to update the accuracy rate according to the training result;

[0037] Repeat the above steps of training the trained main task model to update the accuracy rate until the accuracy rate is not less than a preset threshold, and obtain the main task model after incremental fine-tuning, wherein the main task model after incremental fine-tuning is a multi-task prediction model.

[0038] Furthermore, obtaining a second training sample set and a second verification sample set according to the second target business data and its corresponding label information further includes:

[0039] According to the business type of the secondary task, feature derivation is performed on the second target business data and its corresponding label information to obtain a second feature derivation result;

[0040] The second feature derived results and their corresponding label information are divided to obtain a second training sample set and a second verification sample set.

[0041] On the other hand, some embodiments of the present specification further provide a multi-task prediction model establishment device, the device comprising:

[0042] An acquisition module, used to acquire first target business data;

[0043] A task determination module, used to determine the primary task and secondary task corresponding to the first target business data according to the importance of the task;

[0044] A training module, used to train a first integrated tree model using the first target business data to obtain a main task model;

[0045] A screening module, used for screening the first target business data to obtain the second target business data matching the secondary task;

[0046] The incremental fine-tuning module is used to train the main task model using the second target business data to obtain the main task model after incremental fine-tuning, and the main task model after incremental fine-tuning is a multi-task prediction model.

[0047] Based on the same inventive concept, some embodiments of this specification further provide a multi-task prediction method, the method comprising:

[0048] Receiving business data to be predicted;

[0049] The business data to be predicted is input into the multi-task prediction model trained by the method described in any of the aforementioned embodiments to predict a multi-task prediction result.

[0050] Based on the same inventive concept, on the other hand, some embodiments of this specification further provide a multi-task prediction device, the device comprising:

[0051] A receiving module, used for receiving task data to be predicted;

[0052] The prediction module is used to input the task data to be predicted into the multi-task prediction model trained by the method described in any of the above embodiments to predict the multi-task prediction results.

[0053] On the other hand, some embodiments of the present specification further provide a computer device, including a memory, a processor, and a computer program stored in the memory, wherein when the computer program is executed by the processor, the instructions of the above method are executed.

[0054] On the other hand, some embodiments of the present specification further provide a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor of a computer device, the instructions of the above method are executed.

[0055] On the other hand, some embodiments of the present specification further provide a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor of a computer device, the instructions of the above method are executed.

[0056] One or more technical solutions provided by some embodiments of this specification have at least the following technical effects:

[0057] The embodiments of the present specification first automatically obtain the first target business data. Since the first target business data can be used to perform predictive modeling on multiple tasks, it is not necessarily possible to establish an integrated multi-task prediction model for multiple tasks. Therefore, it is necessary to determine the main task and the secondary task corresponding to the first target business data according to the importance of the task, so as to establish a multi-task prediction model corresponding to the main task and the secondary task. The establishment process first uses the first target business data to train the first integrated tree model to obtain the main task model. Since the main task model can only realize the prediction of the main task, it is also necessary to filter out the second target business data that matches the secondary task from the first target business data, and use the second target business data to train the main task model, and fine-tune the parameters of the main task model so that the incrementally fine-tuned main task model can realize the prediction of the secondary task, and use the incrementally fine-tuned main task model as the multi-task prediction model, so as to balance the performance of the model on multiple tasks and ensure the prediction effect of the multi-task prediction model on multiple tasks.

[0058] The above description is only an overview of the technical solutions of some embodiments of this specification. In order to more clearly understand the technical means of some embodiments of this specification, they can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of some embodiments of this specification more obvious and easy to understand, the specific implementation methods of some embodiments of this specification are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate some embodiments of this specification or technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or prior art descriptions. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0060] Figure 1 A schematic diagram of an implementation system of a multi-task prediction model establishment method in some embodiments of this specification is shown;

[0061] Figure 2 A flowchart of a method for establishing a multi-task prediction model in some embodiments of this specification is shown;

[0062] Figure 3 This is a schematic diagram of the steps of data preprocessing in some embodiments of this specification;

[0063] Figure 4 A schematic diagram of steps for determining a main task and a secondary task in some embodiments of this specification;

[0064] Figure 5 A schematic diagram of the steps of selecting the main task and the secondary task in some embodiments of this specification;

[0065] Figure 6 A flowchart of a multi-task prediction method in some embodiments of this specification is shown;

[0066] Figure 7 A schematic diagram of the structure of a multi-task prediction model building device in some embodiments of this specification;

[0067] Figure 8 This is a schematic diagram of the structure of a multi-task prediction device in some embodiments of this specification;

[0068] Fig. 9 This is a schematic diagram of the computer device structure provided in some embodiments of this specification.

[0069] [Description of Reference Numerals]

[0070] 101. Terminal;

[0071] 102. Server;

[0072] 701. Get module;

[0073] 702, task determination module;

[0074] 703, training module;

[0075] 704. Screening module;

[0076] 705. Incremental fine-tuning module;

[0077] 801, receiving module;

[0078] 802, prediction module;

[0079] 902. Computer equipment;

[0080] 904, processor;

[0081] 906. Memory;

[0082] 908, driving mechanism;

[0083] 910, input / output interface;

[0084] 912. Input device;

[0085] 914. Output device;

[0086] 916. Presentation equipment;

[0087] 918. Graphical user interface;

[0088] 920, network interface;

[0089] 922, communication link;

[0090] 924. Communication bus. DETAILED DESCRIPTION

[0091] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in some embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on some embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0092] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0093] It should be noted that the acquisition, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of relevant laws and regulations.

[0094] It should be noted that in the embodiments of this specification, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0095] like Figure 1 The diagram is a schematic diagram of an implementation system of a method for establishing a multi-task prediction model according to an embodiment of the present invention, which may include: a terminal 101 and a server 102, wherein the terminal 101 and the server 102 communicate with each other through a network, and the network may include a local area network (LAN), a wide area network (WAN), the Internet, or a combination thereof, and is connected to a website, a user device (such as a computing device), and a back-end system. A staff member may send a request for establishing a multi-task prediction model to the server 102 through the terminal 101, and after receiving the request for establishing a multi-task prediction model, the server 102 calls the target business data in the database for calculation and processing, obtains a modeling result, and sends the modeling result to the terminal 101, so that the staff member performs multi-task prediction according to the modeling result, thereby processing the business.

[0096] In the embodiments of this specification, the server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms.

[0097] In an optional embodiment, the terminal 101 may include but is not limited to electronic devices such as self-service terminal devices, desktop computers, tablet computers, laptop computers, smart wearable devices, etc. Optionally, the operating system running on the electronic device may include but is not limited to Android system, IOS system, Linux, Windows, etc. Of course, the terminal 101 is not limited to the above-mentioned electronic devices with a certain entity, and it can also be software running in the above-mentioned electronic devices.

[0098] In addition, it should be noted that Figure 1 What is shown is only an application environment provided by the present disclosure. In actual application, multiple terminals 101 may be included, and this specification does not limit this.

[0099] Figure 2 It is a flowchart of a method for establishing a multi-task prediction model provided by an embodiment of the present invention. This specification provides method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or device product is executed, it can be executed in the order of the method shown in the embodiment or the accompanying drawings or in parallel. Specifically, Figure 2 As shown, applied to the above-mentioned server side, the method may include:

[0100] S201: Acquire first target business data;

[0101] Specifically, in actual production and life, there are various tasks that need to be predicted. The first target business data can be understood as a data set for multi-task learning, which refers to the training data used to predict the task to be predicted in the target business scenario. The task to be predicted is often multiple related tasks. For example, the target business scenario can be a computer vision scenario, a natural language processing scenario, and a biomedicine scenario. In the computer vision scenario, when the task to be predicted is object recognition, the first target business data can be multiple image data, which can be used to identify multiple related objects. In the natural language processing scenario, when the task to be predicted is text sentiment analysis, the first target business data can be text data, which can be used for sentiment analysis of multiple text entities. In the biomedicine scenario, when the task to be predicted is drug discovery, the first target business data can be pharmacogenomics data, which can be used to predict the performance of multiple related drug combinations.

[0102] S202: Determine a primary task and a secondary task corresponding to the first target business data according to the importance of the task;

[0103] In multi-task learning, although the first target business data can be used to predict multiple related tasks, the prediction targets corresponding to different tasks are different. When using the first target business data for multi-task learning to predict multiple related tasks, the training and optimization directions of different tasks are different. If only one prediction model is used for multi-task prediction, it is necessary to balance the joint learning effect between multiple tasks to avoid over-training and optimization for a single task, which will affect the multi-task prediction effect. Therefore, it is necessary to distinguish between primary tasks and secondary tasks according to the importance of the tasks, and to divide the training order of tasks according to the importance of the tasks, so as to accurately and effectively adjust the network parameters of the multi-task prediction model and ensure the prediction accuracy of the subsequent multi-task prediction model.

[0104] Furthermore, in some embodiments, multiple tasks with correlation may be manifested as: the prediction targets corresponding to the multiple tasks have multiple identical features, or the training samples required to achieve the prediction of the prediction targets have multiple identical features. The correlation between different tasks can be obtained by performing correlation analysis on the tasks, or can be calculated by correlation evaluation methods such as the Gini coefficient and information entropy, which is not limited in this article.

[0105] The higher the importance of a task, the earlier the training sequence corresponding to the task should be. In some embodiments, the importance of a task can be measured by the loss incurred by the task during the prediction process. Furthermore, after distinguishing between the main task and the secondary task, multi-task learning is performed in the order of the main task first and the secondary task later. This is conducive to training and optimizing the multi-task prediction model task by task, capturing more general and generalized feature patterns, learning the generalized representation between learning tasks to avoid overfitting, and learning the unique features of each task to avoid underfitting, thereby enhancing the generalization ability of the multi-task prediction model and facilitating the priority of the prediction accuracy of the main task and the secondary task. In some embodiments, a detailed description of the scheme for determining the main task and the secondary task corresponding to the first target business data according to the importance of the task is described below. Figure 4 Description.

[0106] Furthermore, it should be noted that the multi-task prediction model can be used for at least the prediction of two tasks, namely the primary task and the secondary task. In some embodiments, other tasks such as the third task can also be added according to business needs. However, when the multi-task prediction model is used for the prediction of three or more tasks, namely the primary task, the secondary task and the third task, negative transfer may occur when training the multi-task prediction model, resulting in the prediction accuracy of the multi-task prediction model being affected. Therefore, the multi-task prediction model is often used for the prediction of two tasks, namely the primary task and the secondary task. In some embodiments, it can also be used for the prediction of three or more tasks under the premise of satisfying the prediction accuracy. This is not limited in this article.

[0107] In some embodiments, taking the first target business data as multiple pictures, and the two tasks to be predicted as identifying cats (unlimited breed) and identifying Garfield cats (limited breed) in pictures, respectively, as an example, since the loss generated by training the multi-task prediction model to identify cats (unlimited breed) in pictures is greater than the loss generated by training the multi-task prediction model to identify Garfield cats (limited breed), the importance of identifying cats (unlimited breed) in pictures is higher than the importance of identifying Garfield cats (limited breed) in pictures. Therefore, the main task is to identify cats (unlimited breed) in pictures, and the secondary task is to identify Garfield cats (limited breed) in pictures.

[0108] S203: Using the first target business data to train a first integrated tree model to obtain a main task model;

[0109] Continuing with the example that the first target business data is multiple pictures, the main task is to identify cats in pictures (not limited to breed), and the secondary task is to identify Garfield cats in pictures (limited breed), since the main task is more important, the first target business data is first used to train the first integrated tree model to obtain the main task model that can be used to predict the main task.

[0110] It should be noted that when the amount of data of the first target business data is large and the samples are rich enough, when actually training the first integrated tree model, a part of the data in the first target business data may be sufficient to ensure the prediction accuracy of the main task model for the main task. Furthermore, the method of obtaining a part of the data in the first target business data may be random sampling, etc., which is not limited in this article.

[0111] S204: Filtering the first target business data to obtain second target business data matching the secondary task;

[0112] In order to ensure the prediction accuracy and adjust the parameters of the main task model, the training samples required when the main task is used as the training target and the secondary task are used as the training target are different. Before training the main task model, it is necessary to screen the second target business data that matches the secondary task from the first target business data. The data volume of the second target business data is smaller than the first target business data, and the sample composition of the second target business data is different from the sample composition of the first target business data.

[0113] Continuing with the example that the first target business data is multiple pictures, the main task is to identify cats in pictures (not limited to breeds), and the secondary task is to identify Garfield cats in pictures (limited breeds), the first target business data needs to satisfy that the proportion of pictures containing various types of cats reaches a preset threshold, and the second target business data is a data set mainly composed of pictures containing Garfield cats, that is, the proportion of pictures containing Garfield cats in the second target business data needs to reach another preset threshold, so as to ensure the training effect when the second target business data is subsequently used to train the main task model.

[0114] S205: Using the second target business data to train the main task model to obtain an incrementally fine-tuned main task model, where the incrementally fine-tuned main task model is a multi-task prediction model.

[0115] When the main task model is trained using the second target business data, in order to avoid overfitting of the main task model, it is necessary to implement incremental fine-tuning of the main task model. Incremental fine-tuning uses incremental training to enable the main task model to learn the sample features of the secondary task based on the sample features of the main task that have been learned, and fine-tune the model parameters of the main task model to ensure that the incrementally fine-tuned main task model can take into account the prediction accuracy of the main task and the secondary task, and training in an incremental manner can reduce the training cost. Therefore, using the incrementally fine-tuned main task model as a multi-task prediction model can balance the performance of the multi-task prediction model on multiple tasks and ensure the prediction effect of the multi-task prediction model on multiple tasks.

[0116] Refer to the attached Figure 3 In some embodiments, after obtaining the first target service data, the following may also be included:

[0117] S301: Cleaning the first target business data to remove incomplete data and abnormal data;

[0118] S302: Standardize the cleaned first target business data.

[0119] In some embodiments, incomplete data refers to incomplete data. The incomplete data may be the missing of the entire data or the missing of local information in the data. Abnormal data refers to some data points in the data set that violate conventional standards or do not meet expectations. Abnormal data may be caused by errors in data collection, storage and processing, noise in data capture, changes in data generation, and the like. In some embodiments, after obtaining the first target business data, the first target business data can also be cleaned to improve the data quality of the first target business data by removing incomplete data and abnormal data. The data standardization process is to scale the data in proportion so that it falls into a small specific interval. The data standardization process can remove the unit restrictions of the data and convert it into a dimensionless pure value, so that indicators of different units or magnitudes can be compared and weighted. Specifically, some examples of data standardization processing methods are data normalization processing, and the data normalization process realizes data standardization and normalization by uniformly mapping the data to the [0,1] interval.

[0120] Before training the multi-task prediction model, it is necessary to distinguish the main task and the secondary task in the multi-task. Figure 4 In some embodiments, determining the primary task and the secondary task corresponding to the first target business data according to the importance of the task may include:

[0121] S401: Determine multiple tasks corresponding to the first target business data according to business scenarios and business requirements;

[0122] S402: Based on the importance of the tasks, filter out the main tasks and the secondary tasks from the multiple tasks.

[0123] It can be understood that in some embodiments, business scenarios and business requirements are used to reflect the usage scenarios and associated requirements of products or services, respectively. For the first target business data, under different business scenarios and business requirements, the tasks that need to be predicted may be different. For example, assuming that the first target business data is multiple image data, if the business scenario is a computer vision scenario, when the business requirement is to identify cats, the corresponding prediction task is to identify cats in pictures (regardless of type), and when the business requirement is to identify Garfield, the corresponding prediction task is to identify Garfield in pictures; if the business scenario is a natural language processing scenario, when the business requirement is negative sentiment analysis of text, the corresponding prediction task is to perform negative sentiment analysis on the text, and when the business requirement is positive sentiment analysis, the corresponding prediction task is to perform positive sentiment analysis on the text. Therefore, multiple tasks corresponding to the first target business data can be obtained according to the business scenario and business requirements. There is a correlation between these tasks, and they can all be predicted using methods such as machine learning, that is, these tasks are predictable.

[0124] After obtaining multiple tasks that are relevant and predictable, since the multi-task prediction model cannot guarantee a balanced joint learning effect among multiple tasks, it is necessary to screen and sort the tasks according to their importance to determine the main tasks with higher importance and the secondary tasks with relatively lower importance. The training samples required to predict the secondary tasks can be used to predict the main tasks, but the training samples required to predict the main tasks may not necessarily be used to predict the secondary tasks.

[0125] Further, in some embodiments, under the premise of satisfying the prediction accuracy, the multi-task prediction model can also be used for task prediction of more than two tasks. For example, the multi-task prediction model is used to predict the main task, the secondary task and the third task, and the importance of the main task>the importance of the secondary task>the importance of the third task. Assume that the first target business data is a plurality of pictures, the main task is to identify the cat in the picture (not limited to the type), the secondary task is to identify the Garfield cat in the picture, and the third task can be to identify the heterochromatic Garfield cat in the picture. After obtaining the incremental fine-tuning main task model using the training samples corresponding to the secondary task, the incremental fine-tuning main task model is incrementally trained using the training samples corresponding to the third task, and the incremental training method is the same as the incremental training method described in any of the above embodiments to achieve secondary incremental fine-tuning, and obtain the secondary incremental fine-tuning main task model, which can be used to achieve the prediction of the main task, the secondary task and the third task. It should be noted that when the multi-task prediction model is used for task prediction of more than two tasks, it is necessary to determine whether the multi-task prediction model can be used for task prediction of more than two tasks according to the actual prediction accuracy of the multi-task prediction model.

[0126] After determining multiple relevant and predictable tasks corresponding to the first target business data according to the business scenario and business needs, refer to the attached Figure 5 In some embodiments, based on the importance of the tasks, selecting the main tasks and the secondary tasks from the plurality of tasks may include:

[0127] S501: Randomly selecting a preset number of business data from the first target business data;

[0128] S502: Using the preset amount of business data to create training samples;

[0129] S503: for the multiple tasks, respectively use the training samples corresponding to each task to iteratively train the second ensemble tree model to obtain the loss corresponding to each task;

[0130] S504: Determine the importance of each task according to the loss corresponding to each task, and select the main task and the sub-task from the multiple tasks according to the importance of each task, wherein the importance of the main task is higher than the importance of the sub-task.

[0131] It can be understood that in some embodiments, each task will have a loss due to recognition errors, which may be time, network resources, etc. The losses caused by recognition errors between different tasks can be compared. The importance of a task is determined by the loss of task recognition errors. The greater the cost of recognition errors, the more important the task is. The higher the task ranking in the multi-task prediction model, the more iterations are required. Therefore, the importance of the task can be measured by calculating the output values ​​of the loss functions of different tasks. In an embodiment of the present invention, since the first target business data is a large data set, when calculating the output value of the loss function of the task, a small number of business data samples can be randomly selected from the first target business data to calculate the corresponding loss function output value to reduce the calculation cost.

[0132] After randomly selecting a preset number of business data, obtain the labels of the main task and the labels of the secondary task, and use the label labeling method to label the selected preset number of business data to obtain labeled data, so as to classify the labeled data according to the labels of the main task and the labels of the secondary task, and obtain the corresponding training samples of the main task and the secondary task. After that, and the training samples required for the main task and the secondary task are different, the second integrated tree model is iteratively trained for a preset number of times using the training samples corresponding to each task, so as to calculate the loss function output value after iterative training according to the loss function corresponding to each task, so as to achieve the ranking of the importance of each task, so as to quickly and accurately determine the main task and the secondary task. In some embodiments, on the basis of having obtained the first target business data and determining the main task and the secondary task corresponding to the first target business data according to the importance of the task, using the first target business data to train the first integrated tree model to obtain the main task model can include:

[0133] According to the business type of the main task, feature derivation is performed on the first target business data to obtain a first feature derivation result;

[0134] Dividing the first feature derived results and their corresponding label information to obtain a first training sample set and a first verification sample set;

[0135] The first training sample set and the first validation sample set are used to train a first ensemble tree model to obtain the main task model.

[0136] It can be understood that, in some embodiments, when the first integrated tree model is trained using the first target business data matching the main task, the original first target business data may be insufficiently accurate when directly used for model training. Therefore, the first target business data can be subjected to feature derivation according to the business type of the main task. Feature derivation can create new features from the original data. Good features can provide more information, so that the algorithm can better capture the features in the data and the connections between the features, thereby improving the accuracy of prediction or classification. Common feature derivation methods include feature selection, feature transformation, feature creation, and feature extraction.

[0137] For example, the main task is to predict the customer's credit risk, and the secondary task is to predict the customer's long-term credit risk. The original first target business data is the customer's application flow data, that is, the customer's application behavior at a certain point in time. When deriving features, cross-derivation can be performed based on time dimensions (such as 1 / 3 / 6 / 12 months), business dimensions (the position of the applicant institution in the industry, the type of license held by the applicant institution, the type of product applied for), and statistical dimensions (total number of applications, maximum value, minimum value, mean, standard deviation, coefficient of variation), etc. The first feature derivation result can include the total number of applications in licensed consumer finance institutions within 12 months, the monthly standard deviation of the number of applications for personal business loan products within 3 months, etc., so that the first training sample set and the first verification sample set are obtained according to the first feature derivation result and its corresponding label information, so as to ensure the division effect of the first training sample set and the first verification sample set, improve the sample distribution difference between the first training sample set and the first verification sample set, and avoid the problem of unbalanced sample division.

[0138] In some embodiments, using the first training sample set and the first validation sample set to train the first ensemble tree model to obtain the main task model may include: using the first training sample set to train the first ensemble tree model;

[0139] Inputting the feature derivation results in the first verification sample set into the trained first ensemble tree model to obtain corresponding output results;

[0140] Calculate a first accuracy rate according to the output result corresponding to the trained first integrated tree model and the label information in the first verification sample set;

[0141] If the first accuracy rate is less than a first threshold, retraining the first ensemble tree model;

[0142] If the first accuracy rate is not less than the first threshold, the training of the first integrated tree model is completed to obtain the main task model.

[0143] It can be understood that the first integrated tree model is trained using the first training sample set, and the first verification sample set is used to enhance the training effect of the first integrated tree model. Before the first accuracy rate reaches the first threshold, the first integrated tree model is trained and iterated multiple times until the first accuracy rate reaches the first threshold, and the first integrated tree model is updated. The model is then fitted using the first training sample set to ensure the fitting accuracy, and the results of the model fitting are evaluated using the first verification sample set. The main task model is repeatedly iterated based on the evaluation results to output the main task model with the best prediction accuracy, so as to ensure the prediction accuracy of the main task model.

[0144] Furthermore, in some embodiments, dividing the feature derivation results and their corresponding label information to obtain a first training sample set and a first verification sample set may include:

[0145] Constructing a classification sample set according to the feature-derived results and their corresponding label information;

[0146] Using the classification sample set to train a binary classifier;

[0147] Using the trained binary classifier to perform classification prediction on the classification sample set to obtain the classification prediction probability of each sample;

[0148] The samples whose classification prediction probability is less than a preset value are screened from the classification sample set to obtain a screening result, wherein the screening result is a first verification sample set, and the set consisting of other samples in the classification sample set except the screening result is a first training sample set.

[0149] It can be understood that, in some embodiments, the division of feature-derived results and their corresponding label information is based on adversarial verification. The purpose of adversarial verification is to ensure that samples are evenly distributed in the training set and the test set. The core idea of ​​adversarial verification is: if the importance of a feature in the current model is higher, the distribution of the feature in the training set and the test set is more different, and the distribution difference corresponding to the feature needs to be adjusted to make the distribution of the feature even in the training set and the test set. In the embodiments of this specification, the classification sample set is divided to obtain a suitable first training sample set and a first verification sample set. The division process is actually a binary classification process based on adversarial verification, that is, a sample should belong to a training sample or a verification sample. Therefore, the corresponding binary classifier can be established and trained with the goal of distinguishing between training samples and verification samples.

[0150] Specifically, after the classification sample set is constructed, the binary classifier is first trained using a preset portion of samples in the classification sample set. The preset portion of samples can be obtained by randomly selecting samples from the classification sample set, and the trained binary classifier is used to perform classification prediction on the samples in the classification sample set except the preset portion of samples to obtain the classification prediction probability of each sample. The classification prediction probability is used to measure the possibility that the sample belongs to the first training sample set. If the classification prediction probability of a sample is greater than a preset value, the possibility that the sample belongs to the first training sample set is high, and the sample can be considered to belong to the first training sample set. Thus, by training a binary classifier with the goal of distinguishing between training samples and verification samples, the first training sample set and the first verification sample set can be quickly and accurately divided.

[0151] In some embodiments, in addition to using the above-mentioned screening results as the first verification sample set and using the set consisting of other samples in the classification sample set except the above-mentioned screening results as the first training sample set, the samples in the classification sample set can also be sorted in order from low to high according to the classification prediction probability, and a preset proportion of samples in the front sorting are selected to construct the first verification sample set, and then the set consisting of other samples in the classification sample set except the first verification sample set is used as the first training sample set. This article does not limit this.

[0152] Furthermore, compared with training the first integrated tree model using the first target business data, in order to avoid the problem of over-training and optimization for a single task, the training method of training the main task model using the second target business data is different. In some embodiments, the main task model is trained using the second target business data to obtain the main task model after incremental fine-tuning. The main task model after incremental fine-tuning is a multi-task prediction model, which may include:

[0153] Obtaining a second training sample set and a second verification sample set according to the second target business data and its corresponding label information;

[0154] Selecting a preset number of training samples from the second training sample set to obtain a local sample training set, and selecting a preset number of verification samples from the second verification sample set to obtain a local sample verification set;

[0155] Using the local sample training set to train the main task model;

[0156] Inputting the local sample verification set into the trained main task model to obtain the corresponding output result;

[0157] Calculate the accuracy rate according to the output result corresponding to the trained main task model and the label information in the local sample verification set;

[0158] If the accuracy rate is not less than a preset threshold, the training of the main task model is completed to obtain the main task model after incremental fine-tuning, and the main task model after incremental fine-tuning is the multi-task prediction model.

[0159] It can be understood that, in some embodiments, the method of training the main task model using the second target business data is different from the method of training the first integrated tree model using the first target business data. When the second target business data is used to perform incremental fine-tuning on the main task model, the incremental training method used is to perform model training one by one by selecting different training samples multiple times to update the local sample training set. Such an incremental training method optimizes the model parameters of the main task model on the premise of minimizing the adjustment range of the model parameters of the main task model, and achieves fine-tuning of the model parameters of the main task model, so that the incrementally fine-tuned main task model can not only ensure the recognition accuracy of the main task, but also take into account the recognition accuracy of the secondary task. Compared with the traditional method of predicting multiple tasks through multiple prediction models, it can also reduce the training cost and accelerate the model convergence speed.

[0160] Furthermore, in some embodiments, if the accuracy rate is less than a preset threshold, it means that the prediction accuracy of the main task model after incremental fine-tuning has not reached expectations and does not meet the convergence conditions. It is necessary to adjust the model parameters of the main task model. Specifically, since the current local sample training set has been used for incremental training of the main task model, it is necessary to first remove the current local sample training set from the second training sample set to update the second training sample set, and select a preset number of training samples from the updated second training sample set as a new local sample training set to update the local sample training set. Afterwards, the trained main task model is trained using the new local sample training set, thereby updating the model parameters of the main task model to update the accuracy rate according to the training results.

[0161] Repeat the above steps to train the trained main task model to update the accuracy until the accuracy is not less than the preset threshold. At this time, the prediction accuracy of the trained main task model reaches the expected level and meets the convergence condition, thereby obtaining the main task model after incremental fine-tuning. The main task model after incremental fine-tuning is a multi-task prediction model.

[0162] In addition, in the process of training the first integrated tree model using the first target business data, it is necessary to determine the corresponding first training sample set and the first verification sample set according to the first target business data. Similarly, in the process of training the main task model using the second target business data, it is necessary to determine the corresponding second training sample set and the second verification sample set according to the second target business data. Specifically, the second training sample set and the second verification sample set are obtained according to the second target business data and its corresponding label information, further including:

[0163] According to the business type of the secondary task, feature derivation is performed on the second target business data and its corresponding label information to obtain a second feature derivation result;

[0164] The second feature derived results and their corresponding label information are divided to obtain a second training sample set and a second verification sample set.

[0165] It can be understood that, in some embodiments, the specific implementation method of obtaining the second training sample set and the second verification sample set based on the second target business data and its corresponding label information is similar to the specific implementation method of obtaining the first training sample set and the first verification sample set based on the first target business data and its corresponding label information, and will not be elaborated in this article.

[0166] Further, based on the same inventive concept, some embodiments of this specification also provide a multi-task prediction method, referring to Figure 6 As shown, the method includes:

[0167] S601: receiving business data to be predicted;

[0168] S602: Input the business data to be predicted into the multi-task prediction model trained by the method described in any of the aforementioned embodiments to obtain a multi-task prediction result.

[0169] In some embodiments, the business data to be predicted may be multiple picture data, and the tasks to be predicted are respectively identifying the cat (not limited to breed) in the picture and identifying the Garfield cat (limited breed) in the picture. By inputting the multiple pictures into the corresponding multi-task prediction model, the cat (not limited to breed) and Garfield cat in the multiple picture data can be identified respectively, thereby realizing target recognition quickly and accurately and obtaining multi-task prediction results.

[0170] In some embodiments, the business data to be predicted may be user comment text data, and the tasks to be predicted may be positive sentiment analysis and negative sentiment analysis of the target product, respectively. By inputting the user comment text data into the corresponding multi-task prediction model, the positive sentiment analysis results and negative sentiment analysis results of the target product may be obtained, respectively, thereby quickly and accurately realizing sentiment analysis and obtaining multi-task prediction results.

[0171] In some embodiments, the business data to be predicted may be pharmacogenomics data, and the tasks to be predicted are to predict the performance of multiple drug combinations (such as side effects, half-life, etc.). By inputting the pharmacogenomics data into the corresponding multi-task prediction model, the performance of multiple drug combinations can be predicted, thereby quickly and accurately realizing the performance analysis of the drug combinations and obtaining multi-task prediction results.

[0172] It should be noted that, although the operations of the method of the present invention are described in a specific order in the above embodiments and the accompanying drawings, this does not require or imply that the operations must be performed in the specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0173] Corresponding to the above-mentioned multi-task prediction model establishment method, some embodiments of this specification also provide a multi-task prediction model establishment device, referring to Figure 7 As shown, in some embodiments, the apparatus may include:

[0174] An acquisition module 701 is used to acquire first target business data;

[0175] A task determination module 702, configured to determine a primary task and a secondary task corresponding to the first target business data according to the importance of the task;

[0176] A training module 703 is used to train a first integrated tree model using the first target business data to obtain a main task model;

[0177] A screening module 704 is used to screen the first target business data to obtain second target business data matching the secondary task;

[0178] The incremental fine-tuning module 705 is used to train the main task model using the second target business data to obtain the main task model after incremental fine-tuning, so as to use the main task model after incremental fine-tuning as a multi-task prediction model.

[0179] Based on the same inventive concept, corresponding to the above-mentioned multi-task prediction method, some embodiments of this specification further provide a multi-task prediction device, referring to Figure 8 As shown, in some embodiments, the apparatus may include:

[0180] A receiving module 801 is used to receive task data to be predicted;

[0181] The prediction module 802 is used to input the task data to be predicted into the multi-task prediction model trained by the method described in any of the above embodiments to predict the multi-task prediction results.

[0182] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0183] It should be noted that in the embodiments of this specification, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user and fully authorized by all parties.

[0184] It should be noted that the computer program product described in this specification is a software product that mainly implements the method described in this specification through a computer program.

[0185] The embodiments of this specification also provide a computer device. Fig. 9 As shown, in some embodiments of the present specification, the computer device 902 may include one or more processors 904, such as one or more central processing units (CPUs) or graphics processing units (GPUs), and each processing unit may implement one or more hardware threads. The computer device 902 may also include any memory 906, which is used to store any kind of information such as code, settings, data, etc. In a specific embodiment, the computer program on the memory 906 and can be run on the processor 904, when the computer program is run by the processor 904, the instructions of the method described in any of the above embodiments may be executed. Non-limiting, for example, the memory 906 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 902. In one case, when the processor 904 executes an associated instruction stored in any memory or a combination of memories, the computer device 902 may perform any operation of the associated instruction. The computer device 902 also includes one or more drive mechanisms 908 for interacting with any storage, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0186] The computer device 902 may also include an input / output interface 910 (I / O) for receiving various inputs (via input devices 912) and for providing various outputs (via output devices 914). A specific output mechanism may include a presentation device 916 and an associated graphical user interface 918 (GUI). In other embodiments, the input / output interface 910 (I / O), input device 912, and output device 914 may not be included, and the computer device 902 may be used as a computer device in a network. The computer device 902 may also include one or more network interfaces 920 for exchanging data with other devices via one or more communication links 922. One or more communication buses 924 couple the components described above together.

[0187] The communication link 922 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 922 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0188] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), computer-readable storage media, and computer program products of some embodiments of the present specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processor to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processor generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0189] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processor to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0190] These computer program instructions can also be loaded onto a computer or other programmable data processor so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0191] In a typical configuration, a computer device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0192] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0193] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computer device. As defined in this specification, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0194] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, the embodiments of this specification may take the form of complete hardware embodiments, complete software embodiments or embodiments combining software and hardware. Moreover, the embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0195] The present specification embodiments may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present specification embodiments may also be practiced in distributed computing environments where tasks are performed by remote processors connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0196] It should also be understood that in the embodiments of this specification, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0197] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0198] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiments of this specification. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0199] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for establishing a multi-task prediction model, characterized in that: The method comprises: Acquire first target business data; Determine the primary task and the secondary task corresponding to the first target business data according to the importance of the task; Using the first target business data to train a first integrated tree model to obtain a main task model; Filtering the first target business data to obtain second target business data matching the secondary task; The main task model is trained using the second target business data to obtain an incrementally fine-tuned main task model, where the incrementally fine-tuned main task model is a multi-task prediction model.

2. The method according to claim 1, characterized in that Determining the primary task and the secondary task corresponding to the first target business data according to the importance of the task includes: Determine, according to business scenarios and business requirements, multiple tasks corresponding to the first target business data; Based on the importance of the tasks, the main tasks and the secondary tasks are screened from the multiple tasks.

3. The method according to claim 2, characterized in that Based on the importance of the tasks, the main tasks and the secondary tasks are selected from the multiple tasks, including: Randomly selecting a preset number of business data from the first target business data; Using the preset amount of business data to establish training samples; For the multiple tasks, respectively use the training samples corresponding to each task to iteratively train the second ensemble tree model to obtain the loss corresponding to each task; The importance of each task is determined according to the loss corresponding to each task, and the main task and the sub-task are screened from the multiple tasks according to the importance of each task, wherein the importance of the main task is higher than that of the sub-task.

4. The method according to claim 1, characterized in that: Using the first target business data to train a first integrated tree model to obtain a main task model includes: According to the business type of the main task, feature derivation is performed on the first target business data to obtain a first feature derivation result; Dividing the first feature derived results and their corresponding label information to obtain a first training sample set and a first verification sample set; The first training sample set and the first validation sample set are used to train a first ensemble tree model to obtain the main task model.

5. The method according to claim 4, characterized in that The feature derivative results and their corresponding label information are divided to obtain a first training sample set and a first verification sample set, including: Constructing a classification sample set according to the feature-derived results and their corresponding label information; Using the classification sample set to train a binary classifier; Using the trained binary classifier to perform classification prediction on the classification sample set to obtain the classification prediction probability of each sample; The samples whose classification prediction probability is less than a preset value are screened from the classification sample set to obtain a screening result, wherein the screening result is a first verification sample set, and the set consisting of other samples in the classification sample set except the screening result is a first training sample set.

6. The method according to claim 1, characterized in that The main task model is trained using the second target business data to obtain an incrementally fine-tuned main task model, wherein the incrementally fine-tuned main task model is a multi-task prediction model, including: Obtaining a second training sample set and a second verification sample set according to the second target business data and its corresponding label information; Selecting a preset number of training samples from the second training sample set to obtain a local sample training set, and selecting a preset number of verification samples from the second verification sample set to obtain a local sample verification set; Using the local sample training set to train the main task model; Inputting the local sample verification set into the trained main task model to obtain the corresponding output result; Calculate the accuracy rate according to the output result corresponding to the trained main task model and the label information in the local sample verification set; If the accuracy rate is not less than a preset threshold, the training of the main task model is completed to obtain the main task model after incremental fine-tuning, and the main task model after incremental fine-tuning is the multi-task prediction model.

7. The method according to claim 6, characterized in that Also includes: If the accuracy rate is less than a preset threshold, removing the local sample training set from the second training sample set to update the second training sample set, and selecting a preset number of training samples from the updated second training sample set as a new local sample training set; Using a new local sample training set to train the trained main task model, so as to update the accuracy rate according to the training result; Repeat the above step of training the trained main task model to update the accuracy rate until the accuracy rate is not less than a preset threshold, and obtain the main task model after incremental fine-tuning, wherein the main task model after incremental fine-tuning is a multi-task prediction model.

8. The method according to claim 6, characterized in that Obtaining a second training sample set and a second verification sample set according to the second target business data and its corresponding label information further includes: According to the business type of the secondary task, feature derivation is performed on the second target business data and its corresponding label information to obtain a second feature derivation result; The second feature derived results and their corresponding label information are divided to obtain a second training sample set and a second verification sample set.

9. A multi-task prediction method, characterized in that: The method comprises: Receiving business data to be predicted; The business data to be predicted is input into the multi-task prediction model trained by the method according to any one of claims 1 to 8 to obtain a multi-task prediction result.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the computer program is executed by the processor, the computer program executes the instructions of the method according to any one of claims 1 to 9.

11. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor of a computer device, the computer program executes the instructions of the method according to any one of claims 1 to 9.

12. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program executes instructions of the method according to any one of claims 1 to 9.

Images