WO2021213156A1

WO2021213156A1 - Method and related apparatus for generating task label on basis of relationship graph convolutional network

Info

Publication number: WO2021213156A1
Application number: PCT/CN2021/084223
Authority: WO
Inventors: 张楠; 王健宗; 瞿晓阳
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-11-25
Filing date: 2021-03-31
Publication date: 2021-10-28
Also published as: CN112464042B; CN112464042A

Abstract

Provided are a method and related apparatus for generating a task label on the basis of a relationship graph convolutional network, relating to the field of artificial intelligence. The method comprises: according to user data and task data, generating user task graph data; inputting the user task graph data into a relational graph convolutional network to obtain at least one user feature outputted by the relational graph convolutional network for the user task graph data, and at least one task feature; according to the user feature and task feature, constructing a user task matrix; according to the probability distribution, on at least one task label, of user nodes contained in the user task matrix, generating a target task label of the user corresponding to the user node. By means of the described method, it is unnecessary for an administrator to manually assign tasks to users, improving the efficiency of assigning tasks; furthermore, the generated task labels are more relevant to users, improving the degree of matching of task assignment.

Description

Task label generation method and related device based on relation graph convolutional network

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on November 25, 2020, the application number is 202011342170.X, and the application title is "Task label generation method and related device based on relational graph convolutional network", which The entire content is incorporated into this application by reference.

Technical field

This application belongs to the field of artificial intelligence technology, and in particular relates to a task label generation method and related devices based on a relational graph convolutional network.

Background technique

Deep learning is to learn the inherent laws and representation levels of sample data. The information obtained in the learning process is of great help to the interpretation of data such as text, images and sounds. An important process of generating sample data is to label the data, that is, to label the data. The labeled data can be used as a pilot experience for deep learning.

At present, data labeling is usually carried out by labeling systems or platforms. The administrator distributes the labeling task to a large number of users for labeling through the labeling system, and the same labeling task is usually distributed to multiple users, and the answers are summarized as the labeling result.

However, the inventor found that in the above-mentioned solution, the task dispatch process usually relies on the administrator’s subjective judgment and experience to select annotators or users for the target task. The allocation efficiency is low and may result in poor matching between tasks and annotators, reducing the efficiency of annotation. .

technical problem

According to the above technical problems, the present application provides a method and related device for generating task tags based on a relational graph convolutional network, which avoids the administrator from manually assigning tasks to users, improves the efficiency of task assignment, and the generated task tags are consistent with The user's relevance is stronger, which improves the matching degree of task assignment.

Technical solutions

According to an aspect of the embodiments of the present application, there is provided a task label generation method based on a relationship graph convolutional network, including:

According to user data and task data, generate user task graph data. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data, and the task node corresponds to the task contained in the task data. information;

Input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolution network for the user task graph data and at least one task feature, the user feature corresponding to the user node contained in the user task graph data , The task feature corresponds to the task node contained in the user task graph data;

According to user characteristics and task characteristics, construct a user task matrix, the user task matrix contains the probability distribution of user nodes on at least one task label, and the task labels correspond to the task nodes one-to-one;

According to the probability distribution of the user node contained in the user task matrix on at least one task label, the target task label of the user corresponding to the user node is generated.

According to another aspect of the embodiments of the present application, there is provided an apparatus for generating task tags, including:

The user task graph data module is used to generate user task graph data according to user data and task data. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. The task node Correspond to the task information contained in the task data;

The feature representation output module is used to input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolutional network for the user task graph data and at least one task feature, the user feature corresponding to the user task For the user nodes contained in the graph data, the task features correspond to the task nodes contained in the user task graph data;

The user task matrix building module is used to construct a user task matrix according to user characteristics and task characteristics. The user task matrix contains the probability distribution of user nodes on at least one task label, and the task labels correspond to the task nodes one by one;

The task label generating module is used to generate the target task label of the user corresponding to the user node according to the probability distribution of the user node contained in the user task matrix on at least one task label.

According to another aspect of the embodiments of the present application, there is provided a task tag generation device based on a relational graph convolutional network. The device includes: a processor; and a memory for storing executable instructions of the processor; the processor executes The following steps are performed when the computer-readable instructions are:

According to user data and task data, user task graph data is generated. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. The task node Corresponding to the task information contained in the task data;

Input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolution network for the user task graph data, and at least one task feature, the user feature corresponding to all The user node contained in the user task graph data, and the task feature corresponds to the task node contained in the user task graph data;

Constructing a user task matrix according to the user characteristics and the task characteristics, the user task matrix containing the probability distribution of the user node on at least one task label, and the task label corresponds to the task node one-to-one;

According to the probability distribution of the user node on at least one task label contained in the user task matrix, a target task label of the user corresponding to the user node is generated.

According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, the computer-readable storage medium stores at least one instruction, and the following steps are executed when the at least one instruction is executed by a processor:

According to user data and task data, user task graph data is generated. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. Corresponding to the task information contained in the task data;

Beneficial effect

In the embodiment of the present application, the relationship between the user and the user, the relationship between the task and the task, and the relationship between the user and the task, and the relationship between the user and the task are learned through the relationship graph convolutional network. The administrator manually assigns tasks to users, which improves the efficiency of assigning tasks, and the generated task tags are more relevant to users, which improves the matching of task assignments.

Description of the drawings

Figure 1 is a schematic diagram of an exemplary system architecture of the technical solution of the present application in an application scenario.

Fig. 2 is a flowchart of a method for generating task labels based on a relational graph convolutional network according to an embodiment of the present application.

Fig. 3 is a schematic diagram of an example of user graph data in an embodiment of the present application.

FIG. 4 is a schematic diagram of an example of task graph data in an embodiment of the present application.

Fig. 5 is a schematic diagram of an example of user task graph data in an embodiment of the present application.

Fig. 6 is a block diagram of the composition of a task tag generating device in an embodiment of the present application.

Fig. 7 is a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present application.

Embodiments of the present invention

Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms, and should not be construed as being limited to the examples set forth herein; on the contrary, the provision of these embodiments makes this application more comprehensive and complete, and fully conveys the concept of the example embodiments To those skilled in the art.

In addition, the described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to give a sufficient understanding of the embodiments of the present application. However, those skilled in the art will realize that the technical solutions of the present application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be used. In other cases, well-known methods, devices, implementations or operations are not shown or described in detail in order to avoid obscuring various aspects of the present application.

The block diagrams shown in the drawings are merely functional entities, and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in the form of software, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and/or processor devices and/or microcontroller devices. entity.

The flowchart shown in the drawings is only an exemplary description, and does not necessarily include all contents and operations/steps, nor does it have to be performed in the described order. For example, some operations/steps can be decomposed, and some operations/steps can be combined or partially combined, so the actual execution order may be changed according to actual conditions.

Figure 1 is a schematic diagram of an exemplary system architecture of the technical solution of the present application in an application scenario. As shown in FIG. 1, in an exemplary embodiment, the implementation environment may include: a terminal 101 and a server 102. A labeling system is deployed on the terminal 101 and the server 102. The labeling system is a system for labeling data. The effective labeling data generated by the labeling system provides source data for deep learning training models with strong generalization capabilities. Common labeling tasks in the labeling system include text labeling, voice labeling, translation labeling, image labeling and so on. In the labeling system, the server distributes the corresponding labeling task to each client, and the user completes the labeling task through the client and sends it back to the server. In some embodiments, the tagging system will score the user according to the task completion status of the user, such as the speed and number of tasks completed, and the review status of the tagging results.

Wherein, a wired or wireless communication connection is established between the terminal 101 and the server 102, and data transmission with the server 102 is realized through this communication connection.

A client terminal is running in the terminal 101, and the client terminal can provide a user interaction interface. User interaction can be triggered on the user interaction interface for data annotation. Exemplarily, the client may be an image labeling client, and in the process of image labeling, the user receives the distribution through the client.

It should be noted that, in this implementation environment, the terminal 101 may be a mobile phone, a tablet computer, a notebook computer, a desktop computer, or any other electronic device that can be operated by the client, and there is no restriction here. The client can be an application client or a web client, and there are no restrictions here.

The server 102 is used to provide data support for user interaction operations triggered in the client, so that the client can run normally in the terminal 101. Still taking the above-mentioned image annotation as an example, the server 102 may send the image to be annotated and the corresponding candidate task tag to the client according to the image annotation operation triggered in the client, or may receive the result of the image annotation from the client to provide The customer assigns the content of the follow-up marking task.

The server 102 may be a single server device or a server group composed of multiple server devices, which is not limited here.

The technical solutions provided by this application will be described in detail below in conjunction with specific implementations.

Fig. 2 is a flowchart of a method for generating task labels based on a relational graph convolutional network according to an embodiment of the present application. This method can be executed by the server 103 in the application scenario shown in FIG. 1. As shown in Figure 2, in one embodiment, the method may include the following steps:

Step S210. Generate user task graph data according to the user data and task data. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data, and the task node corresponds to the task data. Contains task information.

The main manifestation of user task graph data is an undirected weighted graph containing user nodes and task nodes, which express the association relationship and degree of association between users, tasks, and between users and tasks involved in the labeling system.

User data includes user information and completion of tasks. User information includes user attributes such as age, resume, profession, professional skills, and foreign language level. The task completion status includes the status of the tasks completed by the user, including information such as the number, type, and time of the task. In the user task graph data, user nodes contain user information of users, and the edges between user nodes represent the association relationship between users with respect to the completed tasks. The greater the weight of the edge, the closer the association relationship.

The task data includes task information and task records. The task information mainly includes task type and task content. Task types include, for example, text annotation tasks, voice annotation tasks, translation annotation tasks, image annotation tasks, and so on. The task content is the task goal of the corresponding task. For example, for the translation and labeling task, the task content is the original text and the translated text, as well as the introduction information about the translation and labeling task itself. In the user task graph data, task nodes include task information about tasks, and the edges between task nodes represent the relationship between tasks relative to the user who completes the task. The greater the weight of the edge, the closer the relationship.

The edge between the user node and the task node represents the association relationship between the user and the task. For example, the task completed by the user belongs to the association relationship between the user and the task. The greater the weight of the edge, the closer the association relationship.

According to user data and task data, the association relationships between users, tasks, and users and tasks can be determined. Based on these association relationships, user task graph data can be generated.

Step S220. Input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolution network for the user task graph data, and at least one task feature, the user feature corresponding to the user task graph data contained The task feature corresponds to the task node contained in the user task graph data.

Relational graph convolutional network Graph Convolutional Network (RGCN) is an extension of large-scale relational data based on graph convolutional networks. It is a graph convolutional network that aggregates local neighbor information. In RGCN, a propagation model is defined to calculate the forward update of nodes or entities in a relational multi-graph:

Among them, the index of the neighbor set of node i in relation r is a regularization constant, which can be learned or extracted, and sparse matrix multiplication can be used to avoid explicit summation of neighbors. In RGCN, the conversion of relationship and feature is introduced, which depends on the type and direction of the edge. The conversion can be a linear message conversion, and other more flexible functions, such as a multilayer neural network, can be used, but at the same time it will increase the amount of calculation required.

In this application, the relational graph convolutional network calculates the corresponding feature representation for each user and each task according to the user information and task information in the user task graph data and the relationship between the user and the task. The feature representation is usually a one-dimensional vector, and may be a sparse feature vector, for example, a one-hot vector. User feature representation and task feature representation are representations related to user data and task data. Then, learn user embedding (embedding) and task embedding (embedding) through RGCN.

The calculation process of RGCN for each node in the user task graph can be, for example, taking any node as a central node, performing a convolution on the central node, and updating the representation of the central node by aggregating information of neighbor nodes. Among them, the aggregation of neighbor nodes is classified according to the edge type, and the corresponding conversion is performed according to the different edge types. The collected information undergoes a regularized summation, and finally passes through the activation function. The information of each vertex updates the shared parameters, parallel calculation, and also includes self-connection, that is to say, includes the node's own representation.

Step S230. Construct a user task matrix according to the user characteristics and the task characteristics. The user task matrix contains the probability distribution of the user nodes on at least one task label, and the task labels correspond to the task nodes one-to-one.

Specifically, the user task matrix can be constructed by matrix multiplication. If there are 3 users and 3 tasks, a 3x3 user task matrix will be constructed, and each element in the matrix represents the probability that the corresponding user will be better at the corresponding task. The user task matrix is normalized by softmax. For example, the value of the probability distribution of each user on all tasks is between 0 and 1, and the sum is 1.

Step S240. According to the probability distribution of the user nodes contained in the user task matrix on at least one task label, generate the target task label of the user corresponding to the user node.

According to the user task matrix, the probability distribution of each user for all task labels can be determined, and the task label corresponding to each user can be generated according to the predetermined task allocation principle. For example, if a predetermined number of tasks are assigned to each user, the task labels can be sorted in descending order according to the probability quantity and the task labels can be taken as the result according to the predetermined number, or all task labels with probabilities greater than a predetermined threshold can be taken as the result, or for Each user can select the task corresponding to the highest probability for the user as the task label.

After determining the task tag of the user, the server can dispatch the task corresponding to the task tag to the user.

In some embodiments of the present application, on the basis of the above embodiments, in the above step S210. Before generating user task graph data according to the user data and task data, the method may further include the following steps:

Step S201. Generate historical user task graph data based on historical task data and historical user data. The historical user data corresponds to multiple known task tags, and the known task tags correspond to the historical task information in a one-to-one manner;

Step S202. Training the relational graph convolutional network based on historical user task graph data and known task labels.

Among them, the correspondence between historical user data and task tags can be determined according to the task records in the historical task data. For example, for a certain user included in historical user data, the task label that appears most frequently among the tasks completed by the user is counted as its corresponding task label. The content contained in the historical user task data is the same as the above-mentioned user task diagram, and will not be repeated here. Generally, historical user data and historical task data can directly use the historical task record information stored in the annotation system.

Based on historical user task graph data and known task labels, the relation graph convolutional network can be trained. Specifically, in the training process, historical user task graph data and known task labels are input into the relationship graph convolutional network to obtain a user feature representation and a task feature representation. According to the obtained results and the correspondence between historical user data and known task tags, the solution of the loss function can be determined. The loss function can use various classification loss functions such as cross entropy. The solution of the loss function is minimized by adjusting the parameters in order to determine the value of each parameter in the graph convolutional network.

In the embodiment of the present application, building historical user task graph data and using the historical user task graph data to train the relational graph convolutional network can obtain a specific model that conforms to the task situation of the labeling system and the user's situation, which is beneficial to improve the relational graph convolution. The accuracy of network output results, thereby improving the accuracy of task assignment.

In some embodiments of the present application, on the basis of the above embodiments, the above step S210. Generate user task graph data according to user data and task data, including:

Step S211. Generate user graph data according to the user data, the user graph data includes at least one user node, and generate task graph data according to the task data, the task graph data includes at least one task node;

Step S212. The user map data and the task map data are merged to obtain the user task map data.

Specifically, the user data graph is an undirected weighted graph with user information as a node and a collaborative relationship between users relative to a task as an edge. The user graph data includes at least one user node. The task graph is an undirected weighted graph with task information as nodes and the relationship between tasks relative to users as edges. The task graph data includes at least one task node.

After the user map data and the task map data are generated, merge the two to obtain the user task map data. Specifically, while keeping the relationship between the nodes and edges in the user graph data and the task graph data unchanged, an edge is added between the task node and the user node according to the user's completion of the task to indicate the relationship between the user and the task. The relationship between them. Traverse each user node in the user graph data, determine its association relationship with each task node in the task graph data, and establish corresponding edges according to the association dangling relationship to obtain the user task graph data.

In the embodiment of the present application, the user task graph is obtained by merging the user graph data and the task graph data. While retaining the original data of the user graph data and the task graph data, the relationship between the user and the task is added to maintain The integrity of the data is conducive to improving the accuracy of the graph convolutional network.

In some embodiments of the present application, on the basis of the above embodiments, generating user graph data according to user data in step S211 includes:

Step S2111. Obtain the user information contained in the user data and the task record information corresponding to each user information;

Step S2112. According to the task record information corresponding to each user information, determine the same task completed between users corresponding to each user information;

Step S2113. Generate user nodes in the user graph data according to the user information, and generate edges between the user nodes according to the same tasks completed between users corresponding to the user information to obtain the user graph data.

Among them, in an embodiment, the task record information mainly includes the task type of each task completed by the user. The same task indicates tasks of the same task type. According to the task type, the number of tasks of the same type completed by two users can be determined. In this embodiment, the edges between users in the user graph data represent the number of tasks of the same type completed by two users.

Specifically, please refer to FIG. 3, which is a schematic diagram of an example of user graph data in an embodiment of the present application. As shown in Figure 3, in this example, there is an edge between user A and user B, which means that user A and user B have completed the same type of task, and the weight value of the edge represents the number of tasks of the same type. For example, if user A has completed 30 voice labeling tasks, and user B has completed 15 voice labeling tasks, the weight value of the edge is 15. For another example, if user A has completed 7 voice labeling tasks and 8 image labeling tasks, and user B has also completed 7 voice labeling tasks and 8 image labeling tasks, the weight value of the edge is also 15, that is, edge The weight value includes the number of all tasks of the same type.

In another embodiment, the task record information includes task identification of each task completed by the user. The same task indicates tasks with the same task ID, that is, identical tasks. Correspondingly, the edges between users in the user graph data represent the number of tasks completed by two users with the same task identifier. For example, if user B has completed 4 image labeling tasks, user C has completed 5 image labeling tasks and the pictures in 3 tasks are the same as the labeling task completed by user B, then there is a margin between user B and user C , And the weight value of the edge is 3.

In this embodiment, the user map data is established based on the same task completed by the user information and the users corresponding to each user information, which fully reflects the collaborative relationship between multiple users with respect to the task, and is beneficial to dispatching multiple users on the same task. The rationality of distribution at the time.

In some embodiments of the present application, on the basis of the above embodiments, generating task graph data according to the task data in step S211 includes:

Step S2114. Obtain the task information contained in the task data and the task history data corresponding to each user information;

Step S2115. Determine each task information performed by the same user according to the task history data corresponding to each user information;

Step S2116. The task nodes in the task graph data are generated according to the task information, and the edges between the task nodes are generated according to the information of each task performed by the same user to obtain the task graph data.

Among them, in one embodiment, the task history data mainly includes the information of the user who completed the task and the task completion time. When determining the information of each task performed by the same user, a time threshold can be set. When the time difference between two tasks completed by the same user is within the time threshold, the two tasks are considered to be related to the user. In this embodiment, the edges between tasks in the task graph data represent the number of users who have completed two tasks.

Specifically, please refer to FIG. 4, which is a schematic diagram of an example of task graph data in an embodiment of the present application. As shown in Figure 4, in this example, there is an edge between task 1 and task 2, which means that task 1 and task 2 are completed by the same user, and the weight value of the edge indicates that the user who has completed task 1 and task 2 quantity. For example, if 40 users have completed task 1, and 30 of them have completed task 2, the weight value of the edge is 30.

In this embodiment, the user graph data is established based on the task information and the information of each task performed by the same user, which fully reflects the association relationship between multiple tasks with respect to the user, and distributes tasks based on the user’s characteristics and skills. So that the user completes the task efficiency.

In some embodiments of the present application, on the basis of the above embodiments, the above step S212. The user map data and the task map data are merged to obtain the user task map data, including:

Step S2121. Obtain the task record information corresponding to the user node in the user graph data, and obtain the task history data corresponding to the task node in the task graph data;

Step S2122. Determine the association relationship between the user node and the task node according to the task record information and the task history data, and the association relationship indicates that the user corresponding to the user node completes the task corresponding to the task node;

Step S2123. According to the association relationship between the user node and the task node, an edge between the user node and the task node is constructed between the user graph data and the task graph data to obtain the user task graph data.

Specifically, according to the task record information and the task history data, the specific tasks completed by each user and the number of times each specific task have been completed can be specifically determined. According to the determined situation of each user completing the task, the association relationship between the user node and the task node can be determined, and for the two nodes that have the association relationship, based on the user graph data and the task graph data, the user can be constructed The edge between the node and the task node. The edge represents the situation where the user completes the corresponding task, and the weight of the edge represents the number of times the user completes the task.

For ease of introduction, please refer to FIG. 5 in conjunction with FIG. 3 and FIG. 4. FIG. 5 is a schematic diagram of an example of user task graph data in an embodiment of the present application. Assuming that task 1 represents a translation and annotation task, if user A has completed 10 different translation and annotation tasks, correspondingly, there is an edge between user A and task 1 on the way of the user task, and its weight is 10.

In another embodiment, task 1 represents a specific translation and annotation task, for example, annotating the translation of a certain article, and task 2 represents another specific translation and annotation task. User A completes task 1 10 times, and user B completes task 2 30 times. In the user relationship graph, there is an edge between user A and task 1, with a weight of 10, and user B and task 2 There is an edge between, and its weight value is 30.

In this embodiment, the user task graph is generated based on the user graph data and task graph data based on the user completing the task, which is beneficial to fully consider the relationship between the user and the task in the calculation process of the relation graph convolutional network, and improve The degree of association between the task and the user in turn makes the calculation of the characteristic representation of the user and the task more accurate.

In some embodiments of the present application, on the basis of the above embodiments, the above step S240. According to the probability distribution of the user node contained in the user task matrix on at least one task label, generate the target task label of the user corresponding to the user node ,include:

Step S241. Determine the maximum probability corresponding to the user node according to the user task matrix;

Step S242. If the maximum probability is greater than the preset probability distribution threshold, the task label corresponding to the maximum probability is used as the target task label of the user corresponding to the user node.

Specifically, the user task matrix contains the assignment probability of each task for each user. If the user task matrix includes 3 users and 3 types of tasks, for one user, there is a 3-dimensional vector, and each feature value in the vector represents the probability that the corresponding task is assigned to the user. For example, the three tasks are text annotation, voice annotation, and image annotation. For user X, the three-dimensional vector can be {0.7, 0.2, 0.1}.

In the vector for a specific user, the maximum probability can be determined, and if the maximum probability is greater than the probability distribution threshold, it means that the user and the corresponding task are more closely related, and the task label can be used as the The user's task label. For example, for the aforementioned user X, the maximum probability is 0.7, and if the probability distribution threshold is 0.5, it can be determined that the text is labeled as the task tag of the user.

In this embodiment, the user's task label is determined by comparing the maximum probability of the user node with the probability distribution threshold, which helps to avoid forcibly restricting the user's task type when the user's own tendency is not obvious, and helps to improve the accuracy of task assignment. sex.

It should be noted that although the various steps of the method in this application are described in a specific order in the drawings, this does not require or imply that these steps must be performed in the specific order, or that all the steps shown must be performed to achieve the expectation. the result of. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and/or one step may be decomposed into multiple steps for execution, etc.

The following describes the implementation of the device of the present application, which can be used to execute the task label generation method based on the relational graph convolutional network in the foregoing embodiment of the present application. Fig. 6 is a block diagram of the composition of a task tag generating device in an embodiment of the present application. As shown in FIG. 6, the task tag generating apparatus 300 may mainly include:

The user task graph data module 310 is used to generate user task graph data according to user data and task data. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. The node corresponds to the task information contained in the task data;

The feature representation output module 320 is configured to input user task graph data into the relationship graph convolutional network to obtain at least one user feature output by the relationship graph convolution network for the user task graph data, and at least one task feature, the user feature corresponding to the user The user nodes contained in the task graph data, and the task characteristics correspond to the task nodes contained in the user task graph data;

The user task matrix construction module 330 is configured to construct a user task matrix according to user characteristics and task characteristics. The user task matrix contains the probability distribution of user nodes on at least one task label, and the task labels correspond to the task nodes one-to-one;

The task label generating module 340 is configured to generate the target task label of the user corresponding to the user node according to the probability distribution of the user node contained in the user task matrix on at least one task label.

In some embodiments of the present application, according to the above technical solutions, the task tag generating device 300 further includes:

The historical user task graph data generation module is used to generate historical user task graph data based on historical task data and historical user data. The historical user data corresponds to multiple known task tags, and the known task tags and historical task information are one by one correspond;

The relational graph convolutional network training module is used to train the relational graph convolutional network based on historical user task graph data and known task labels.

In some embodiments of the present application, according to the above technical solutions, the user task graph data module 310 may include:

The user graph data generating unit is configured to generate user graph data according to user data, the user graph data includes at least one user node, and the task graph data is generated according to task data, and the task graph data includes at least one task node;

The merging processing unit is used for merging the user map data and the task map data to obtain the user task map data.

In some embodiments of the present application, according to the above technical solutions, the user graph data generating unit may include:

The user information acquisition subunit is used to acquire the user information contained in the user data and the task record information corresponding to each user information;

The same task determination subunit is used to determine the same task completed between users corresponding to each user information according to the task record information corresponding to each user information respectively;

The user node generating subunit is used to generate user nodes in the user graph data according to the user information, and generate edges between the user nodes according to the same tasks completed between users corresponding to the user information to obtain the user graph data.

The task information acquisition subunit is used to acquire the task information contained in the task data and the task history data corresponding to each user information;

The same user determination subunit is used to determine each task information performed by the same user according to the task history data corresponding to each user information;

The task node generating subunit is used to generate task nodes in the task graph data according to task information, and generate edges between task nodes according to the information of each task performed by the same user to obtain task graph data.

In some embodiments of the present application, according to the above technical solutions, the merging processing unit may include:

The task record information obtaining subunit is used to obtain the task record information corresponding to the user node in the user graph data, and obtain the task history data corresponding to the task node in the task graph data;

The association relationship determination subunit is used to determine the association relationship between the user node and the task node according to the task record information and the task history data, and the association relationship instructs the user corresponding to the user node to complete the task corresponding to the task node;

The user task graph data construction subunit is used to construct an edge between the user node and the task node between the user graph data and the task graph data according to the association relationship between the user node and the task node to obtain the user task graph data.

In some embodiments of the present application, according to the above technical solutions, the task tag generation module 340 may include:

The maximum probability determining unit is used to determine the maximum probability corresponding to the user node according to the user task matrix;

The task label determining unit is configured to, if the maximum probability is greater than the preset probability distribution threshold, use the task label corresponding to the maximum probability as the target task label of the user corresponding to the user node.

It should be noted that the device provided in the foregoing embodiment and the method provided in the foregoing embodiment belong to the same concept, and the specific manners for performing operations of each module have been described in detail in the method embodiment, and will not be repeated here.

It should be noted that the computer system 400 of the electronic device shown in FIG. 7 is only an example, and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

As shown in FIG. 7, the computer system 400 includes a central processing unit (Central Processing Unit (CPU) 401, which can execute according to a program stored in a read-only memory (Read-Only Memory, ROM) 402 or a program loaded from the storage part 408 to a random access memory (Random Access Memory, RAM) 403 Various appropriate actions and processing. In RAM 403, various programs and data required for system operation are also stored. The CPU 401, the ROM 402, and the RAM 403 are connected to each other through a bus 404. An input/output (Input/Output, I/O) interface 405 is also connected to the bus 404.

The following components are connected to the I/O interface 405: an input part 406 including a keyboard, a mouse, etc.; an output part 407 including a cathode ray tube (Cathode Ray Tube, CRT), a liquid crystal display (LCD), etc., and speakers 407 ; A storage part 408 including a hard disk, etc.; and a communication part 409 including a network interface card such as a LAN (Local Area Network) card, a modem, and the like. The communication section 409 performs communication processing via a network such as the Internet. The driver 410 is also connected to the I/O interface 405 as needed. A removable medium 411, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 410 as required, so that the computer program read therefrom is installed into the storage part 408 as required.

In particular, according to the embodiments of the present application, the processes described in the flowcharts of the various methods may be implemented as computer software programs. For example, the embodiments of the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from the network through the communication part 409, and/or installed from the removable medium 411. When the computer program is executed by the central processing unit (CPU) 401, various functions defined in the system of the present application are executed.

It should be noted that the computer-readable medium shown in the embodiments of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be non-volatile or Can be volatile. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or a combination of any of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Erasable Programmable Read Only Memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this application, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in a baseband or as a part of a carrier wave, and a computer-readable program code is carried therein. This propagated data signal can take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium, and the computer-readable medium may send, propagate, or transmit the program for use by or in combination with the instruction execution system, apparatus, or device . The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation of the system architecture, functions, and operations of the system, method, and computer program product according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more for realizing the specified logic function. Executable instructions. It should also be noted that, in some alternative implementations, the functions marked in the block may also occur in a different order from the order marked in the drawings. For example, two blocks shown one after another can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and a combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by It is realized by a combination of dedicated hardware and computer instructions.

It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above may be embodied in one module or unit. Conversely, the features and functions of a module or unit described above can be further divided into multiple modules or units to be embodied.

Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described here can be implemented by software, or can be implemented by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk, etc.) or on the network , Including several instructions to make a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) execute the method according to the embodiment of the application.

Those skilled in the art will easily think of other embodiments of the present application after considering the specification and practicing the application disclosed herein. This application is intended to cover any variations, uses, or adaptive changes of this application. These variations, uses, or adaptive changes follow the general principles of this application and include common knowledge or customary technical means in the technical field that are not disclosed in this application. .

It should be understood that the present application is not limited to the precise structure that has been described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the application is only limited by the appended claims.

Claims

A task label generation method based on a relational graph convolutional network, which includes:

According to user data and task data, user task graph data is generated. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. Corresponding to the task information contained in the task data;

Input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolution network for the user task graph data, and at least one task feature, the user feature corresponding to all The user node contained in the user task graph data, and the task feature corresponds to the task node contained in the user task graph data;

Constructing a user task matrix according to the user characteristics and the task characteristics, the user task matrix containing the probability distribution of the user node on at least one task label, and the task label corresponds to the task node one-to-one;

According to the probability distribution of the user node on at least one task label contained in the user task matrix, a target task label of the user corresponding to the user node is generated.
The method according to claim 1, wherein before said generating user task graph data based on user data and task data, the method further comprises:

Generating historical user task graph data according to historical task data and historical user data, where the historical user data corresponds to a plurality of known task tags, and the known task tags correspond to the historical task information in a one-to-one correspondence;

Training the relational graph convolutional network according to the historical user task graph data and the known task label.
The method according to claim 1, wherein said generating user task graph data according to user data and task data comprises:

Generating user graph data according to the user data, the user graph data including the at least one user node, and generating task graph data according to the task data, the task graph data including the at least one task node;

The user map data and the task map data are combined to obtain the user task map data.
The method according to claim 3, wherein said generating user graph data according to said user data comprises:

Acquiring user information contained in the user data and task record information corresponding to each user information;

Determine, according to the task record information corresponding to the respective user information, the same tasks completed by the users corresponding to the respective user information;

The user nodes in the user graph data are generated according to the user information, and the edges between the user nodes are generated according to the same tasks completed between users corresponding to the user information to obtain the user graph data.
The method according to claim 3, wherein the generating task graph data according to the task data comprises:

Acquiring task information contained in the task data and task historical data corresponding to each user information;

Determine each task information performed by the same user according to the task history data corresponding to each user information;

The task nodes in the task graph data are generated according to the task information, and the edges between the task nodes are generated according to the information of each task performed by the same user to obtain the task graph data.
The method according to claim 3, wherein said merging process of said user map data and said task map data to obtain said user task map data comprises:

Obtaining task record information corresponding to the user node in the user graph data, and obtaining task history data corresponding to the task node in the task graph data;

Determining an association relationship between the user node and the task node according to the task record information and the task history data, the association relationship instructing the user corresponding to the user node to complete the task corresponding to the task node;

According to the association relationship between the user node and the task node, an edge between the user node and the task node is constructed between the user graph data and the task graph data to obtain the user task graph data.
The method according to claim 1, wherein the generating the target task label of the user corresponding to the user node according to the probability distribution of the user node contained in the user task matrix on at least one task label comprises :

Determine the maximum probability corresponding to the user node according to the user task matrix;

If the maximum probability is greater than the preset probability distribution threshold, the task label corresponding to the maximum probability is used as the target task label of the user corresponding to the user node.
A task tag generating device, which includes:

The user task graph data module is used to generate user task graph data according to user data and task data. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user data contained in The user information of the task node corresponds to the task information contained in the task data;

The feature representation output module is configured to input the user task graph data into the relation graph convolutional network to obtain at least one user feature and at least one task feature output by the relation graph convolutional network for the user task graph data, The user feature corresponds to the user node contained in the user task graph data, and the task feature corresponds to the task node contained in the user task graph data;

The user task matrix construction module is configured to construct a user task matrix according to the user characteristics and the task characteristics, the user task matrix contains the probability distribution of the user node on at least one task label, and the task label is The task nodes have a one-to-one correspondence;

The task label generating module is configured to generate the target task label of the user corresponding to the user node according to the probability distribution of the user node on at least one task label contained in the user task matrix.
A task tag generating device, wherein the task tag generating device includes a memory, a processor, and computer-readable instructions stored in the memory and running on the processor, and the processor executes the computer Perform the following steps when the instructions are readable:

According to user data and task data, user task graph data is generated. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. Corresponding to the task information contained in the task data;

Input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolution network for the user task graph data, and at least one task feature, the user feature corresponding to all The user node contained in the user task graph data, and the task feature corresponds to the task node contained in the user task graph data;

Constructing a user task matrix according to the user characteristics and the task characteristics, the user task matrix containing the probability distribution of the user node on at least one task label, and the task label corresponds to the task node one-to-one;

According to the probability distribution of the user node on at least one task label contained in the user task matrix, a target task label of the user corresponding to the user node is generated.
9. The task label generating device according to claim 9, wherein, before the user task graph data is generated based on the user data and task data, the processor further executes the following steps when executing the computer readable instruction:

Generating historical user task graph data according to historical task data and historical user data, where the historical user data corresponds to a plurality of known task tags, and the known task tags correspond to the historical task information in a one-to-one correspondence;

Training the relational graph convolutional network according to the historical user task graph data and the known task label.
The task label generating device according to claim 9, wherein said generating user task graph data according to user data and task data comprises:

Generating user graph data according to the user data, the user graph data including the at least one user node, and generating task graph data according to the task data, the task graph data including the at least one task node;

The user map data and the task map data are combined to obtain the user task map data.
The task label generating device according to claim 11, wherein said generating user map data according to said user data comprises:

Acquiring user information contained in the user data and task record information corresponding to each user information;

Determine, according to the task record information corresponding to the respective user information, the same tasks completed by the users corresponding to the respective user information;

The user nodes in the user graph data are generated according to the user information, and the edges between the user nodes are generated according to the same tasks completed between users corresponding to the user information to obtain the user graph data.
The task label generating device according to claim 11, wherein said generating task graph data according to said task data comprises:

Acquiring task information contained in the task data and task historical data corresponding to each user information;

Determine each task information performed by the same user according to the task history data corresponding to each user information;

The task nodes in the task graph data are generated according to the task information, and the edges between the task nodes are generated according to the information of each task performed by the same user to obtain the task graph data.
The task label generating device according to claim 11, wherein the merging process of the user map data and the task map data to obtain the user task map data comprises:

Obtaining task record information corresponding to the user node in the user graph data, and obtaining task history data corresponding to the task node in the task graph data;

Determining an association relationship between the user node and the task node according to the task record information and the task history data, the association relationship instructing the user corresponding to the user node to complete the task corresponding to the task node;

According to the association relationship between the user node and the task node, an edge between the user node and the task node is constructed between the user graph data and the task graph data to obtain the user task graph data.
The task label generating device according to claim 14, wherein the target task of the user corresponding to the user node is generated based on the probability distribution of the user node on at least one task label contained in the user task matrix Labels, including:

Determine the maximum probability corresponding to the user node according to the user task matrix;

If the maximum probability is greater than the preset probability distribution threshold, the task label corresponding to the maximum probability is used as the target task label of the user corresponding to the user node.
A computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and the following steps are executed when the at least one instruction is executed by a processor:

According to user data and task data, user task graph data is generated. The user task graph data includes at least one user node and at least one task node. The user node corresponds to the user information contained in the user data. Corresponding to the task information contained in the task data;

Input the user task graph data into the relation graph convolutional network to obtain at least one user feature output by the relation graph convolution network for the user task graph data, and at least one task feature, the user feature corresponding to all The user node contained in the user task graph data, and the task feature corresponds to the task node contained in the user task graph data;

Constructing a user task matrix according to the user characteristics and the task characteristics, the user task matrix containing the probability distribution of the user node on at least one task label, and the task label corresponds to the task node one-to-one;

According to the probability distribution of the user node on at least one task label contained in the user task matrix, a target task label of the user corresponding to the user node is generated.
16. The computer-readable storage medium according to claim 16, wherein, before the user task graph data is generated according to the user data and task data, the following steps are performed when the at least one instruction is executed by the processor:

Generating historical user task graph data according to historical task data and historical user data, where the historical user data corresponds to a plurality of known task tags, and the known task tags correspond to the historical task information in a one-to-one correspondence;

Training the relational graph convolutional network according to the historical user task graph data and the known task label.
The computer-readable storage medium according to claim 16, wherein said generating user task graph data according to user data and task data comprises:

Generating user graph data according to the user data, the user graph data including the at least one user node, and generating task graph data according to the task data, the task graph data including the at least one task node;

The user map data and the task map data are combined to obtain the user task map data.
18. The computer-readable storage medium according to claim 18, wherein said generating user graph data according to said user data comprises:

Acquiring user information contained in the user data and task record information corresponding to each user information;

Determine, according to the task record information corresponding to the respective user information, the same tasks completed by the users corresponding to the respective user information;

The user nodes in the user graph data are generated according to the user information, and the edges between the user nodes are generated according to the same tasks completed between users corresponding to the user information to obtain the user graph data.
18. The computer-readable storage medium of claim 18, wherein the generating task graph data according to the task data comprises:

Acquiring task information contained in the task data and task historical data corresponding to each user information;

Determine each task information performed by the same user according to the task history data corresponding to each user information;

The task nodes in the task graph data are generated according to the task information, and the edges between the task nodes are generated according to the information of each task performed by the same user to obtain the task graph data.