Task prediction method and device, electronic equipment, storage medium and computer program product

Abstract

The invention provides a task prediction method and device, electronic equipment, a storage medium and a computer program product. The method comprises the following steps: determining a correlation coefficient between a first data feature of a first data source and a first data feature of a second data source; based on the correlation coefficient, updating the first data feature of the first data source and the first data feature of the second data source to obtain a second data feature of the first data source and a second data feature of the second data source; performing feature fusion on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature; and based on the fused data features, predicting the target task to obtain a prediction result of the target task. Through the task prediction method and device, the task prediction accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a task prediction method, device, electronic device, storage medium and computer program product. Background Art

[0002] The target task prediction method is an important research direction in the field of artificial intelligence and machine learning. It aims to predict the performance, trend or behavior of a specific task or goal in the future by analyzing and learning historical data. In many technical fields such as economy, meteorology, medicine, and transportation, the target task prediction method plays a vital role. With the advent of the big data era, enterprises and organizations have generated massive amounts of data, which contain valuable information. By analyzing this information, decision-making can be assisted, business processes can be optimized, and work efficiency can be improved. By analyzing historical data, the target task prediction method can discover the inherent connections and patterns between data, which can provide support for future decision-making.

[0003] In the related art, the data features used for target task prediction have a single data source. Summary of the invention

[0004] The embodiments of the present application provide a task prediction method, device, electronic device, storage medium and computer program product, which can improve the accuracy of task prediction.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present application provides a task prediction method, the method comprising:

[0007] Determine a correlation coefficient between a first data feature of a first data source and a first data feature of a second data source;

[0008] Based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source;

[0009] Performing feature fusion on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature;

[0010] Based on the fused data features, the target task is predicted to obtain a prediction result of the target task.

[0011] The present application provides a task prediction device, including:

[0012] A coefficient determination module, used to determine a correlation coefficient between a first data feature of a first data source and a first data feature of a second data source;

[0013] a feature updating module, configured to update the first data feature of the first data source and the first data feature of the second data source based on the correlation coefficient to obtain the second data feature of the first data source and the second data feature of the second data source;

[0014] A feature fusion module, used for fusing the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature;

[0015] The task prediction module is used to predict the target task based on the fused data features to obtain the prediction result of the target task.

[0016] An embodiment of the present application provides an electronic device, the electronic device comprising:

[0017] A memory for storing computer executable instructions or computer programs;

[0018] The processor is used to implement the task prediction method provided in the embodiment of the present application when executing the computer executable instructions or computer programs stored in the memory.

[0019] An embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions or a computer program for implementing the task prediction method provided in the embodiment of the present application when executed by a processor.

[0020] An embodiment of the present application provides a computer program product, including computer executable instructions or a computer program. When the computer executable instructions or the computer program are executed by a processor, the task prediction method provided by the embodiment of the present application is implemented.

[0021] The embodiments of the present application have the following beneficial effects:

[0022] By using the embodiment of the present application, the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source is determined, and then based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source, and the second data feature of the first data source and the second data feature of the second data source are further fused to obtain fused data features, and based on the fused data features, the target task is predicted to obtain the prediction result of the target task. In this way, the first data features of multiple data sources can be constructed, and the synergistic fusion effect between the first data features of multiple data sources can be combined to improve the accuracy of task prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of an application mode of the task prediction method provided in an embodiment of the present application;

[0024] Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0025] Figure 3A It is a first flow chart of the task prediction method provided in an embodiment of the present application;

[0026] Figure 3B It is a second flow chart of the task prediction method provided in an embodiment of the present application;

[0027] Figure 3C is a third flow chart of the task prediction method provided in an embodiment of the present application;

[0028] Figure 3D 4 is a schematic diagram of a fourth process flow of the task prediction method provided in an embodiment of the present application;

[0029] Figure 4 It is a schematic diagram of a multi-source heterogeneous time series model provided in an embodiment of the present application.

[0030] It should be pointed out that the above-mentioned "first" and "second" are only used to distinguish different solutions, and do not represent the degree of superiority or inferiority of the solutions or the priority in the implementation process. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0032] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0033] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0034] The relevant data collection and processing in the embodiments of this application should be strictly in accordance with the requirements of relevant laws and regulations when applied in examples, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of authorization of laws and regulations and the personal information subject.

[0035] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0036] Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application have the same meanings as those commonly understood by those skilled in the art. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0037] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0038] 1) Multi-source heterogeneity: refers to the diversified data sources and different data structures.

[0039] 2) Gated Recurrent Unit (GRU): It is a recurrent unit in deep learning, which combines a gated unit (such as a gated unit in a recurrent neural network) and a reversible unit (such as some units in a recurrent neural network and a time-recurrent neural network). It is usually used to generate hidden layers in a neural network model to improve the performance of the model when processing sequence data.

[0040] 3) Time series features: It is a combination of features arranged in chronological order.

[0041] 4) Embedding: It is a vector used to embed input data into a high-dimensional space, so that the input data can adapt to the needs of various machine learning algorithms. It is usually expressed as a weighted sum of vectors, where each input data is regarded as an independent feature and encoded by an embedding matrix.

[0042] 5) Multilayer Perceptron (MLP): It is a feedforward artificial neural network model consisting of at least three layers of nodes: input layer, hidden layer and output layer. Each node can be a neuron. In addition to the input node, each node in the hidden layer and the output layer has a nonlinear activation function. The multilayer perceptron breaks the limitations of linear models by using nonlinear activation functions in the hidden layer and the output layer, so that it can capture and simulate complex relationships in the data.

[0043] 6) Activation function (Sigmod): It is a nonlinear activation function widely used in artificial neural networks, and the output value is between 0 and 1.

[0044] The embodiments of the present application provide a task prediction method, device, electronic device, storage medium and computer program product, which can improve the accuracy of task prediction.

[0045] At present, when performing user demand prediction tasks in a certain business scenario, most of the technical solutions that combine embedded vectors and multi-layer perceptrons are used. Before performing task prediction, data features for task prediction are constructed. Data features are usually sequence features, and the data source of sequence features is relatively single. The embedded vectors of data features pass through the multi-layer perceptron network to output the target task prediction results. Therefore, the data source of the data features used for target task prediction is single, and there is a lack of synergistic fusion between multiple data features, and the accuracy of task prediction is not high.

[0046] The following describes an exemplary application of the electronic device provided by the embodiment of the present application. The electronic device provided by the embodiment of the present application can be implemented as various types of terminals such as a laptop computer, a tablet computer, a desktop computer, a set-top box, a smart phone, a smart speaker, a smart watch, a smart TV, and a vehicle-mounted terminal, and can also be implemented as a server. The following describes an exemplary application when the device is implemented as a server.

[0047] See also Figure 1 , Figure 1 Schematic diagram of the application mode of the task prediction method provided in the embodiment of the present application, for example, Figure 1 The server 200, network 300, terminal 400 and database 500 are involved. The terminal 400 is connected to the server 200 via the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two. The database 500 stores multiple different data sources required for task prediction.

[0048] In the task prediction process, the terminal 400 responds to the user's task prediction instruction and sends a task prediction request to the server 200. The server 200 responds to the task prediction request, obtains multiple different data sources from the database 500, and determines the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source. Based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source. The second data feature of the first data source and the second data feature of the second data source are feature fused to obtain the fused data feature. Based on the fused data feature, the target task is predicted to obtain the prediction result of the target task, and the prediction result is returned to the terminal 400. The terminal 400 receives the prediction result and displays it. In this way, the user can learn about the prediction result of the target task through the terminal 400.

[0049] The embodiments of the present application can be implemented through database technology. In short, a database can be regarded as an electronic file cabinet where electronic files are stored. Users can add, query, update, delete, etc. data in the files. The so-called "database" is a collection of data that is stored together in a certain way, can be shared with multiple users, has as little redundancy as possible, and is independent of the application program.

[0050] A database management system (DBMS) is a computer software system designed for managing databases. It generally has basic functions such as storage, retrieval, security, and backup. Database management systems can be classified according to the database model it supports, such as relational, Extensible Markup Language (XML); or according to the type of computer it supports, such as server clusters, mobile phones; or according to the query language used, such as Structured Query Language (SQL), XQuery; or according to the performance focus, such as maximum scale, maximum operating speed; or other classification methods. Regardless of the classification method used, some database management systems can cross categories, for example, supporting multiple query languages ​​at the same time.

[0051] The embodiments of the present application can also be implemented through cloud technology. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model application. It can form a resource pool, which is used on demand and is flexible and convenient. Cloud computing technology will become an important support. The background services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, as well as the promotion of search services, social networks, mobile commerce and open collaboration, each item may have its own hash code identification mark in the future, and all need to be transmitted to the background system for logical processing. Data of different levels will be processed separately, and all kinds of industry data require strong system backing support, which can only be achieved through cloud computing.

[0052] The embodiments of the present application can also be implemented through machine learning. Machine Learning (ML) is a multi-disciplinary cross-disciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning by teaching.

[0053] In some embodiments, the server (e.g., server 200) may 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), and big data and artificial intelligence platforms. The terminal 400 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a car terminal, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0054] See also Figure 2 , Figure 2 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, the electronic device may be a terminal or a server, Figure 2The electronic device 600 shown includes: at least one processor 410, a memory 450, and at least one network interface 420. The various components in the electronic device 600 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 440 is not described in detail. Figure 2 Various buses are labeled as bus system 440 .

[0055] The processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0056] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0057] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0058] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0059] The operating system 451 includes system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic businesses and process hardware-based tasks.

[0060] The network communication module 452 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420. Exemplary network interfaces 420 include: Bluetooth, Wireless Authentication (WiFi), and Universal Serial Bus (USB).

[0061] In some embodiments, the device provided in the embodiments of the present application can be implemented in software. Figure 2 A task prediction device 455 stored in a memory 450 is shown, which may be software in the form of a program or plug-in, including the following software modules: a coefficient determination module 4551, a feature update module 4552, a feature fusion module 4553, and a task prediction module 4554. These modules are logical, and therefore may be arbitrarily combined or further split according to the functions implemented. The functions of each module will be described below.

[0062] In some embodiments, the terminal or server can implement the task prediction method provided in the embodiment of the present application by running various computer executable instructions or computer programs. For example, computer executable instructions can be microprogram-level commands, machine instructions or software instructions. The computer program can be a native program or software module in the operating system; it can be a local (Native) application (APPlication, APP), that is, a program that needs to be installed in the operating system to run, such as a cloud computing APP or an instant messaging APP; it can also be a small program that can be embedded in any APP, that is, a program that can be run only by downloading it to a browser environment. In short, the above-mentioned computer executable instructions can be instructions in any form, and the above-mentioned computer program can be an application, module or plug-in in any form.

[0063] The task prediction method provided in the embodiment of the present application will be explained in combination with the exemplary application and implementation of the server device provided in the embodiment of the present application.

[0064] The following describes the task prediction method provided by the embodiment of the present application. As mentioned above, the electronic device that implements the task prediction method of the embodiment of the present application can be a terminal, a server, or a combination of the two. Next, the task prediction method provided by the embodiment of the present application is described by taking the electronic device as a server as an example. Figure 3A , Figure 3A is a first flow chart of the task prediction method provided in the embodiment of the present application, which will be combined with Figure 3A The steps shown are explained.

[0065] In step 301, a correlation coefficient between a first data feature of a first data source and a first data feature of a second data source is determined.

[0066] Here, the first data feature is the data feature required for predicting the target task, and the first data source and the second data source may be different data sources among n (n is an integer greater than 1) data sources, and the first data source and the second data source may also be the same data source among n data sources. For any first data source and second data source among the n data sources, the task prediction method provided by the embodiment of the present application can be executed. In some embodiments, the first data source (and / or the second data source) may be any one of the n data sources, and of course, the first data source (and / or the second data source) may also be m (m is an integer greater than 1 and less than n) data sources among the n data sources. In some embodiments, the first data source and the second data source may constitute the n data sources. The data source may be a diversified data source, and each data source may have a different data structure. The data source is divided according to a preset dimension, and one preset dimension corresponds to one data source. For example, the data source may be a behavior, such as login behavior, click behavior, etc., and different behaviors may correspond to different data sources. For example, the target task prediction may be a user demand prediction, and the login platform data features in the user login behavior data source and the click commodity data features in the user click behavior data source are obtained. The target task prediction may be information recommendation prediction, and the user type data features in the user category data source and the user identification data features in the user identification data source are obtained.

[0067] In some embodiments, see Figure 3B , Figure 3B is a second flow chart of the task prediction method provided in the embodiment of the present application. Figure 3A Before step 301 shown, you can also perform Figure 3B Steps 3001 to 3002 are described in detail below.

[0068] In step 3001, time series characteristics of a first data source and time series characteristics of a second data source are obtained.

[0069] Here, time series features refer to features extracted from time series data that can be used for analysis and modeling. Time series features represent specific attributes or patterns of time series. The first data source and the second data source include continuous time series data and discrete time series data. Continuous time series data needs to be discretized to obtain time series features, and non-enumerable time series features are converted into enumerable time series feature values. For example, the payment amount within a period of time is non-enumerable. After discretization, the payment amount is divided into enumerable values: [100, 10000, 20000], and the corresponding index value is used as the payment amount in the current time period.

[0070] In step 3002, time series prediction is performed on the time series features of the first data source and the time series features of the second data source to obtain the first data features of the first data source and the first data features of the second data source.

[0071] Here, time series prediction is a statistical analysis method that predicts future values ​​by analyzing time series features. Time series features are a set of data recorded in chronological order, and each data represents the value of the target object at a specific time point. Through the gated recurrent unit (GRU) in the neural network model, time series prediction is performed on the time series features of the first data source and the time series features of the second data source to obtain the first data feature for target task prediction. The time series features of each data source correspond to a gated recurrent unit, and the time series features of n data sources correspond to n gated recurrent units.

[0072] Example, reference Figure 4 , Figure 4 It is a schematic diagram of a multi-source heterogeneous time series model provided in an embodiment of the present application. In the collaborative layer 402, time series feature 1, time series feature i and time series feature n from different data sources are input, and time series feature 1 corresponds to the gated recurrent unit GRU-1 model. The gated recurrent unit GRU-1 model is used to perform time series prediction on time series feature 1, and outputs the first data feature e(1). Time series feature i corresponds to the gated recurrent unit GRU-i model. The gated recurrent unit GRU-i model is used to perform time series prediction on time series feature i, and outputs the first data feature e(i). Time series feature n corresponds to the gated recurrent unit GRU-n model. The gated recurrent unit GRU-n model is used to perform time series prediction on time series feature n, and outputs the first data feature e(n).

[0073] Based on this step 3001 to step 3002, by obtaining the time series characteristics of the first data source and the time series characteristics of the second data source, and performing time series prediction on the time series characteristics of the first data source and the time series characteristics of the second data source, the first data characteristics for target task prediction are obtained, so that the data sources for task prediction are diversified, and the sequence characteristics of multiple data sources are constructed to accurately express the characteristic changes of the target prediction object.

[0074] In some embodiments, the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source may be determined by performing the following steps: performing correlation processing on the first data feature of the first data source and the first data feature of the second data source to obtain a correlation coefficient.

[0075] Here, a data source currently being processed is used as the first data source, and the first data feature of the first data source is subjected to correlation processing with the first data feature of the second data source to obtain the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source. The same data source currently being processed can also be used as the first data source and the second data source, and the first data feature of the first data source is subjected to correlation processing with the first data feature of the second data source to obtain the autocorrelation coefficient between the first data feature of the first data source and the first data feature of the second data source. The correlation coefficient can be obtained by using the Pearson correlation coefficient calculation formula to analyze the linear relationship between the first data features. The correlation coefficient can be obtained by using the Spearman correlation coefficient calculation formula to analyze the hierarchical relationship between the first data features. The correlation coefficient can be obtained by using the Kendall correlation coefficient calculation formula to analyze the correlation between the first data features.

[0076] In some embodiments, correlation processing is performed on the first data feature of the first data source and the first data feature of the second data source to obtain a correlation coefficient, which can be performed through the following steps: multiplying the first data feature of the first data source by the first data feature of the second data source to obtain a first product result of the data source; adding the first data feature of the first data source and the first data feature of the second data source to obtain a first summation result; multiplying the first data feature of the first data source by the first summation result to obtain a second product result; determining the ratio of the first product result to the second product result, and using the ratio as the correlation coefficient.

[0077] In some embodiments, the first data source and the second data source constitute n data sources, and the correlation coefficient calculation formula between the first data features of the n data sources is as follows:

[0078]

[0079] Among them, e i is the first data feature of the ith data source (also referred to as the first data source) among the n data sources, e j is the first data feature of the jth data source (also known as the second data source) among the n data sources, α ij is the correlation coefficient between the first data feature of the i-th data source and the first data feature of the j-th data source. Figure 3A In step 302, based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source.

[0080] Here, the first data feature of the first data source and the first data feature of the second data source are updated based on the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source, and the second data feature is the data feature after the first data feature is updated.

[0081] In some embodiments, the first data feature of the first data source and the first data feature of the second data source can be updated based on the correlation coefficient to obtain the second data feature of the first data source and the second data feature of the second data source by performing the following steps: when the correlation coefficient is lower than the correlation threshold, the first data feature of the first data source and the first data feature of the second data source are updated based on the correlation coefficient to obtain the second data feature of the first data source and the second data feature of the second data source.

[0082] Here, the correlation threshold is a preset correlation coefficient threshold. When the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source is lower than the correlation threshold, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source.

[0083] In some embodiments, the following steps may be performed: when the correlation coefficient is not lower than the correlation threshold, any one of the first data source and the second data source is selected as the third data source, and a target correlation coefficient between the first data feature of the third data source and the first data feature of the fourth data source is determined; when the target correlation coefficient is lower than the correlation threshold, based on the target correlation coefficient, the first data feature of the third data source and the first data feature of the fourth data source are updated to obtain the second data feature of the third data source and the second data feature of the fourth data source; the second data feature of the third data source and the second data feature of the fourth data source are feature fused to obtain the target fused data feature; based on the target fused data feature, a target task is predicted to obtain a target prediction result of the target task

[0084] Here, when the correlation coefficient is not lower than the correlation threshold, that is, the correlation between the first data feature of the first data source and the first data feature of the second data source is too high, and the first data source and the second data source contain repeated first data features, remove any one of the first data source and the second data source, use the remaining data source as the third data source, and then select any one of the n data sources as the fourth data source. The target correlation coefficient is the correlation coefficient between the first data feature of the third data source and the first data feature of the fourth data source. When the target correlation coefficient is lower than the correlation threshold, the first data feature of the third data source and the first data feature of the fourth data source are updated to obtain the second data feature of the third data source and the second data feature of the fourth data source, and the second data feature of the third data source and the second data feature of the fourth data source are used to predict the subsequent target task.

[0085] In an embodiment of the present application, when the correlation coefficient is not lower than the correlation threshold, one data source is selected from the first data source and the second data source as the third data source, so that the third data source does not contain repeated first data features, and data processing is subsequently performed based on the first data features of the third data source and the first data features of the fourth data source, thereby reducing the amount of data processing and improving the prediction efficiency of the target task. In some embodiments, the first data feature of the first data source and the first data feature of the second data source can be updated based on the correlation coefficient by performing the following steps: using the correlation coefficient as the weight of the first data feature of the first data source, performing data processing on the first data feature of the first data source, and obtaining the second data feature of the first data source; using the correlation coefficient as the weight of the first data feature of the second data source, performing data processing on the first data feature of the second data source, and obtaining the second data feature of the second data source.

[0086] Here, the first data feature of the first data source and the first data feature of the second data source are updated, and the correlation coefficient is used as the weight of the first data feature of the first data source and the first data feature of the second data source, and the first data feature of the first data source and the first data feature of the second data source are processed to obtain the second data feature of the first data source and the second data feature of the second data source. The data processing can be to use the correlation coefficient as the weight to perform weighted summation processing on the first data feature of the first data source to obtain the second data feature of the first data source. The data processing can also be to use the correlation coefficient as the weight to perform weighted summation on the first data feature of the first data source to obtain the first data processing result, and then average the first data processing result to obtain the second data feature of the first data source. The data processing process of the first data feature of the second data source is the same as above, and will not be repeated here.

[0087] In some embodiments, the first data source and the second data source constitute n data sources, and the second data feature calculation formula of the i-th data source (also referred to as the first data source or the second data source) among the n data sources is as follows:

[0088]

[0089] Among them, e′ i is the second data feature of the i-th data source, α i is the correlation coefficient corresponding to the ith data source, e i is the first data feature of the i-th data source, and α i As the weight of the first data feature of the i-th data source, the first data feature of the i-th data source is weighted and summed to obtain the second data feature of the i-th data source.

[0090] Continue to refer Figure 3A In step 303, feature fusion is performed on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature.

[0091] Here, feature fusion is to splice the second data feature of the first data source and the second data feature of the second data source, and the fused data feature is the data feature obtained by splicing the second data feature of each data source.

[0092] In some embodiments, see Figure 3C , Figure 3C It is a third flow chart of the task prediction method provided in an embodiment of the present application. Figure 3A Step 303 shown may be performed by Figure 3C Steps 3031 to 3033 are implemented as described in detail below.

[0093] In step 3031, a first influencing parameter value of the first data source on the prediction result is determined, and a second influencing parameter value of the second data source on the prediction result is determined.

[0094] Here, the influence parameter value indicates the influence degree of the data source on the prediction result, or the importance degree of the data source on the prediction result. According to the actual needs of task prediction, the influence parameter value of each data source on the prediction result can be set to determine the influence parameter value of the first data source and the second data source on the prediction result respectively, the first influence parameter value is the influence parameter value of the first data source on the prediction result, and the second influence parameter value is the influence parameter value of the second data source on the prediction result.

[0095] In step 3032, the first influencing parameter value is used as the weight of the first data source, and the second influencing parameter value is used as the weight of the second data source.

[0096] Here, the influence parameter value of the first data source on the prediction result is used as the weight of the first data source, and the influence parameter value of the second data source on the prediction result is used as the weight of the second data source.

[0097] In step 3033, based on the weight of the first data source and the weight of the second data source, feature fusion is performed on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature.

[0098] Here, based on the weight of the first data source and the weight of the second data source, the second data feature of the first data source and the second data feature of the second data source are spliced ​​to obtain a fused data feature.

[0099] Based on this step 3031 to step 3033, by taking the influencing parameter value as the weight of the data source and based on the weights of the n data sources, the second data features of the n data sources are fused, which takes into account the degree of influence of different data sources on the prediction results of the target task, thereby improving the accuracy of the prediction results of the target task.

[0100] Continue to refer Figure 3A , in step 304, the target task is predicted based on the fused data features to obtain the prediction result of the target task.

[0101] In some embodiments, see Figure 3D , Figure 3D 4 is a schematic diagram of a fourth process flow of the task prediction method provided in the embodiment of the present application. Figure 3A Before step 304 shown, you can also perform Figure 3D Step 305 is described in detail below.

[0102] In step 305, basic object features of the target object are obtained.

[0103] Here, the target task is for the target object, and the target object can be a certain attribute object or a certain individual object. The basic characteristics of the object include at least one of the following characteristics: an identification data feature of the identification data source, and a category data feature of the category data source. The identification data source can be an identification type data source such as a user identification data source and a commodity identification data source. The identification data feature is the corresponding user identification data feature and commodity identification data feature. The identification data feature refers to a string of numbers that uniquely marks a certain attribute or individual. The category data source can be a gender category data source and a place of origin category data source. The category data feature is the corresponding gender category data feature and place of origin category data feature. The category data feature refers to the text that marks the category or place of origin of the individual. The gender category data feature is represented by "male and female", and the place of origin category data feature is represented by the text of the province to which it belongs.

[0104] Continue to refer Figure 3D , after step 305, Figure 3A Step 304 shown may be performed by Figure 3D Steps 3041 to 3042 are implemented as described in detail below.

[0105] In step 3041, the object basic features and the fused data features are spliced ​​to obtain spliced ​​data features.

[0106] Here, the basic features of the object are spliced ​​with the fused data features. This can be done by splicing the embedded vector of the identification data features with the embedded vector of the fused data features to obtain the spliced ​​data features; or by splicing the embedded vector of the category data features with the embedded vector of the fused data features to obtain the spliced ​​data features; or by splicing the embedded vector of the identification data features, the embedded vector of the category data features and the embedded vector of the fused data features to obtain the spliced ​​data features.

[0107] The embedded vector of the identification data feature is to convert the identification data feature into a vector representation using appropriate embedding methods, such as the word vector model Word2Vec, the global logarithmic bilinear regression model GloVe, etc., according to the type and requirements of the target task, and learn the similarity of words or phrases through statistical methods to generate the embedded vector of the identification data feature. The embedded vector of the category data feature is to convert the discrete category data feature into a vector representation using appropriate embedding methods, such as the training word vector model Skip-Gram or the continuous bag of words model CBOW, according to the type and requirements of the target task, and infer the embedded vector of the category data feature through context information. The fused data feature is a time series feature. By using the time embedding method, the fused data feature can be divided into multiple windows, each of which can represent the state at a time point. Time embedding can be performed using a fixed window size or a sliding window size method. For each window, an embedding method can be used to convert it into a vector representation. The embedding method can be linear embedding, cosine embedding, etc. By combining these window vectors together, the embedded vector of the fused data feature can be obtained.

[0108] Example, reference Figure 4 , Figure 4 4 is a schematic diagram of a multi-source heterogeneous time series model provided by an embodiment of the present application. In the overlay layer 403, the embedded vector of the identification data feature 405, the embedded vector of the category data feature 406 and the embedded vector of the fusion sequence feature 409 (fusion data feature) are spliced ​​to obtain a spliced ​​embedded vector 411, i.e., a spliced ​​data feature.

[0109] In step 3042, the target task is predicted based on the spliced ​​data features to obtain a prediction result of the target task.

[0110] Here, the concatenated data features are input into a multilayer perceptron for target task prediction, and the prediction results of the target task are obtained. The multilayer perceptron is a feedforward artificial neural network model, which consists of at least three layers of nodes: input layer, hidden layer, and output layer. Each node can be a neuron, and each node in the hidden layer and output layer, except for the input node, has a nonlinear activation function Sigmod. The multilayer perceptron breaks the limitations of the linear model by using nonlinear activation functions in the hidden layer and the output layer, so that it can capture and simulate complex relationships in the data.

[0111] In the input layer of the multi-layer perceptron, the concatenated data features are directly input into the neural network. Since the concatenated data features are embedded vectors, the high-dimensional identification or category information has been mapped to a low-dimensional vector space, so the concatenated data features can be directly input as nodes of the input layer. For example, if an embedded vector dimension is 100, then there will be 100 neurons in the input layer, and each neuron corresponds to one dimension in the embedded vector.

[0112] In the hidden layer of the multi-layer perceptron, the concatenated data features will undergo a series of operations and transformations. These transformations include: linear transformation or nonlinear transformation. For linear transformation, each hidden layer neuron will affect all neurons in the input layer. The output of the hidden layer can be calculated by the weight matrix, bias vector and activation function Sigmoid of the hidden layer. Hidden layer neurons usually use nonlinear activation functions to further transform the results of linear transformation to introduce nonlinear characteristics. Commonly used activation functions include Sigmoid, ReLU and Tanh.

[0113] In the output layer of the multi-layer perceptron, the processing of the spliced ​​data features depends on the type of target task. For regression tasks, each neuron in the output layer usually corresponds to a predicted value. For classification tasks, the number of neurons in the output layer is usually equal to the number of categories, and each neuron corresponds to one category. The output result of the output layer can be calculated through the weight matrix, bias vector and normalization function Softmax of the output layer.

[0114] In an embodiment of the present application, by splicing the basic features of the object with the fused data features, a spliced ​​data feature is obtained, so that the spliced ​​data feature can accurately represent complex features, thereby performing target task prediction based on the spliced ​​data feature to obtain prediction results of various target tasks.

[0115] In some embodiments, the target task includes m subtasks, where m is an integer greater than 1. Based on the fused data features, the target task is predicted to obtain a prediction result of the target task, which can be obtained by performing the following steps: Based on the fused data features, the m subtasks are predicted to obtain respective subtask prediction results of the m subtasks.

[0116] Here, the target task includes multiple subtasks, and each subtask is predicted based on the fused data features to obtain the subtask prediction results corresponding to each subtask. For example, the target task may include a task to predict a user's click on a product, and a task to predict the probability of purchasing the product after clicking, that is, the probability of a user clicking on a product and the probability of purchasing the product can be predicted at the same time. For a target task including one subtask and a target task including multiple subtasks, the first data features of the n data sources used for target task prediction can be shared, but the target task prediction model structure including multiple subtasks is different from the target task prediction model structure including one subtask. The target task prediction model including multiple subtasks can be a combination of target task prediction models of multiple subtasks, and the weight of the first data feature used for each subtask prediction can also be adjusted.

[0117] In some embodiments, the task prediction method provided by the embodiments of the present application can be applied in the financial field. When predicting the user's financial management needs, obtain the first data features of the user in n different data sources, such as: obtaining the login platform data features in the user login behavior data source in the financial management APP, and the click financial product data features in the user click behavior data source. Determine the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source, based on the correlation coefficient, update the first data feature of the first data source and the first data feature of the second data source to obtain the second data feature of the first data source and the second data feature of the second data source, perform feature fusion on the second data feature of the first data source and the second data feature of the second data source, obtain the fused data feature, and predict the user's financial management needs based on the fused data feature to obtain the user's financial management needs prediction result. In this way, when predicting the user's financial management needs, construct the first data features of multiple data sources, and combine the synergistic fusion effect between the first data features of multiple data sources to improve the accuracy of the user's financial management needs prediction.

[0118] In some embodiments, the task prediction method provided by the embodiments of the present application can be applied in the field of information recommendation. When making information recommendation predictions, obtain the first data features of the user in n different data sources, such as: obtaining the data features of the click recommendation information in the user's click behavior data source, and the purchase data features of the recommended objects (such as virtual resources (such as coupons), commodities, etc.) in the user's purchase behavior data source. Determine the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source, based on the correlation coefficient, update the first data feature of the first data source and the first data feature of the second data source to obtain the second data feature of the first data source and the second data feature of the second data source, perform feature fusion on the second data feature of the first data source and the second data feature of the second data source, obtain the fused data feature, and perform information recommendation prediction based on the fused data feature to obtain the information recommendation prediction result. In this way, when making information recommendation predictions, construct the first data features of multiple data sources, combine the synergistic fusion effect between the first data features of multiple data sources, and improve the accuracy of information recommendation predictions.

[0119] By using the above-mentioned embodiment of the present application, the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source is determined, and based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source, and the second data feature of the first data source and the second data source are subjected to feature fusion to obtain fused data features, and based on the fused data features, the target task is predicted to obtain the prediction result of the target task. In this way, the first data features of multiple data sources can be constructed, and the synergistic fusion effect between the first data features of multiple data sources can be combined to improve the accuracy of task prediction.

[0120] The following will take n data sources for task prediction as an example to illustrate an exemplary application of the task prediction method provided in an embodiment of the present application in an actual user demand prediction scenario.

[0121] At present, when performing user demand prediction tasks in certain business scenarios, most of them use a technical solution that combines embedded vectors with multi-layer perceptrons. Before obtaining embedded vectors, prediction features are constructed. Prediction features are usually divided into statistical features and sequence features. The data source of sequence features is mainly a single sequence feature. For example, the sequence feature is constructed by counting the number of logins or clicks according to a certain user behavior. The prediction features pass through the input layer of the multi-layer perceptron, then the hidden layer training parameters, and finally the output layer outputs the prediction results.

[0122] In the process of predicting user demand forecasting tasks, there are the following problems:

[0123] (1) The data source for constructing sequence features is relatively single, and a single sequence feature cannot accurately describe customer characteristics.

[0124] (2) There is a lack of mutual synergy and fusion between multiple sequence features, and it is impossible to obtain the user's historical behavior sequence characteristics from different angles.

[0125] In view of the problems existing in the related art, the embodiment of the present application proposes a task prediction method, which includes the following improvements compared with the related art:

[0126] (1) Construct heterogeneous sequence features from multiple data sources (the first data features of n data sources). The heterogeneous sequence features from multiple data sources can more accurately characterize the changing characteristics of user needs in the same period. For example, in the payment scenario, according to the sequence feature statistics method of a single data source, the number of user logins over time is counted as a sequence feature. According to the heterogeneous sequence feature statistics method from multiple data sources, not only can the number of user logins, login frequency and other related sequence features be counted, but also the sequence features of user click behaviors over a period of time can be counted according to different events. Constructing heterogeneous sequence features from multiple data sources can not only express the changes in user interests based on the sequence features related to login and click behaviors, but also characterize the user's preference for specific events based on the number of clicks.

[0127] (2) Fuse the sequence features of multiple data sources (fuse the second data features of n data sources), combine the synergy of multiple sequences to learn the user's historical behavior sequence features from different angles, and enhance the ability of the target prediction model to learn deeper features. For example, in the payment scenario, the heterogeneous sequence features of multiple data sources are constructed according to the frequency of user login to the application platform and the application click interaction. Different weights can be assigned based on the correlation between the two sequences. In actual scenarios, it is more expected that users log in and have interactive behaviors, so the interactive sequence features are assigned a larger weight, thereby achieving a synergistic effect between different sequences.

[0128] Example, reference Figure 4 , Figure 4 Schematic diagram of a multi-source heterogeneous time series model provided in an embodiment of the present application. Figure 4 The model structure in the example is used to explain the task prediction method provided in the embodiment of the present application.

[0129] As an example, the multi-source heterogeneous time series model structure includes an input layer 401 , a collaborative layer 402 , an overlay layer 403 and a multi-layer perceptron layer 404 .

[0130] In the input layer 401 , identification data features 405 , category data features 406 , and multiple sequence features in region 407 are input, and the multiple sequence features include: sequence 1, sequence i, and sequence n.

[0131] The identification data feature 405 can be a user identification data feature, a commodity identification data feature, a contract identification data feature, etc., which is a string of numbers that uniquely marks a certain attribute or individual. The category data feature 406 refers to the text that marks the category or range to which the individual belongs, for example: gender data feature: male or female, or data features such as the province of origin. Sequence 1, sequence i and sequence n are sequence features after the discretization of continuous features. Discretization of continuous features refers to converting the original non-enumerable feature values ​​into enumerable feature values. For example: the payment amount is non-enumerable. After discretization, the payment amount is divided into different interval values: [100, 10000, 20000], and the interval value corresponding to the index value is used as the discrete value of the payment amount. The identification data feature 405 can be used as a data source alone, and the category data feature 406 can be used as part of multiple sequence features according to the actual business scenario requirements.

[0132] In order to better predict user needs, multiple sequence features (the first data features of the first data source and the first data features of the second data source mentioned above) are added in the embodiment of the present application. For example, within a period of time, by month, the changes in the user's login features on the application platform are counted over time, as well as the user's click sequence features based on different events in the application, and the user's payment behavior-related sequence features over a period of time are counted.

[0133] In the collaborative layer 402, the input is sequence 1, sequence i and sequence n from different data sources, and each sequence feature corresponds to a gated recurrent unit GRU model. Multiple sequence features correspond to multiple gated recurrent unit GRU models, sequence 1 corresponds to the gated recurrent unit GRU-1 model, sequence i corresponds to the gated recurrent unit GRU-i model, and sequence n corresponds to the gated recurrent unit GRU-n model. After the gated recurrent unit GRU model is trained, sequence 1 outputs the predicted sequence feature e(1), sequence i outputs the predicted sequence feature e(i), and sequence n outputs the predicted sequence feature e(n), and the correlation coefficient between different predicted sequence features is calculated. The correlation coefficient calculation formula is shown in area 408:

[0134]

[0135] Among them, e i is the current prediction sequence feature, e j is a prediction sequence feature different from the current prediction sequence feature, α ij It is the correlation coefficient between the current prediction sequence features and other prediction sequence features.

[0136] Based on the correlation coefficients between different prediction sequence features, different prediction sequence features are updated. The prediction sequence feature update calculation formula is shown in area 408:

[0137]

[0138] Among them, α i is the correlation coefficient corresponding to the current prediction sequence feature, and α i As the weight of the current prediction sequence feature, the weighted sum of the current prediction sequence features is performed, e′ i is the updated current prediction sequence feature.

[0139] As shown in region 407, after weighted summation of the predicted sequence feature e(1) corresponding to sequence 1, an updated sequence feature e′1 is obtained, and after weighted summation of the predicted sequence feature e(i) corresponding to sequence i, an updated sequence feature e′ is obtained. i , the predicted sequence feature e(n) corresponding to sequence n is weighted and summed to obtain the updated sequence feature e′ n The multiple updated prediction sequence features are spliced ​​to obtain the fused sequence feature 409. This not only utilizes the serialization representation of multi-source heterogeneous data, but also plays a role in the synergistic fusion between different sequences, making the obtained fused sequence feature information more diverse.

[0140] In the overlay layer 403, the embedded vector of the identification data feature 405, the embedded vector of the category data feature 406 and the embedded vector of the fused sequence feature 409 are concatenated to obtain a concatenated embedded vector 411 (the concatenated data feature mentioned above).

[0141] In the multi-layer perceptron layer 404, the concatenated embedded vector 411 is input into the multi-layer perceptron 412. The concatenated embedded vector 411 is trained by the multi-layer deep neural network in the multi-layer perceptron and processed by the activation function sigmod to obtain the prediction result for output.

[0142] For example, the multi-source heterogeneous time series model provided in the embodiment of the present application can also adopt a multi-objective prediction model to increase the complexity of the prediction model. The multi-objective prediction task can be a multi-objective prediction of the probability of users clicking on the goods and purchasing the goods after clicking. In the payment scenario, for example, the probability of users clicking on payment-related coupons and the probability of using them can be predicted. The input sequence features of the multi-objective prediction task and the single-objective prediction task in the prediction model can be shared, and there is no different sequence features for different tasks. However, the model structure of the multi-objective prediction task is different from the model structure of the single-objective prediction task. The model structure of the multi-objective prediction task can be a combination of multiple single-objective prediction tasks, and only the weights of the sequence features used for each prediction task are adjusted.

[0143] In the above-mentioned application scenario of user demand prediction, by obtaining the sequence features of multiple data sources and fusing them, in the process of constructing the sequence features of multiple data sources, starting from different feature sources and feature refinement dimensions, the breadth and depth of the feature sources are taken into account. The sequence features of multiple data sources can more accurately represent the evolution of individual behavior over a period of time, improve the ability of the prediction model to capture feature changes, and memorize them into the model parameters. For example, the prediction model uses the time series features of multiple data sources to learn the user's login status on the platform within a period of time, as well as the periodic click status and payment initiation status. After the prediction model learns the periodic behavior characteristics, it can more effectively predict the payment probability in the next period.

[0144] The following is a description of an exemplary structure of the task prediction device 455 provided in the embodiment of the present application implemented as a software module. In some embodiments, Figure 2 As shown, the software modules stored in the task prediction device 455 of the memory 450 may include: a coefficient determination module 4551, which is used to determine the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source; a feature update module 4552, which is used to update the first data feature of the first data source and the first data feature of the second data source based on the correlation coefficient to obtain the second data feature of the first data source and the second data feature of the second data source; a feature fusion module 4553, which is used to perform feature fusion on the second data feature of the first data source and the second data feature of the second data source to obtain fused data features; a task prediction module 4554, which is used to predict the target task based on the fused data features to obtain a prediction result of the target task.

[0145] In some embodiments, the coefficient determination module 4551 is also used to obtain the time series characteristics of the first data source and the time series characteristics of the second data source before determining the correlation coefficient between the first data characteristics of the first data source and the first data characteristics of the second data source; perform time series prediction on the time series characteristics of the first data source and the time series characteristics of the second data source to obtain the first data characteristics of the first data source and the first data characteristics of the second data source.

[0146] In some embodiments, the coefficient determination module 4551 is further used to perform correlation processing on the first data feature of the first data source and the first data feature of the second data source to obtain a correlation coefficient.

[0147] In some embodiments, the coefficient determination module 4551 is also used to multiply the first data feature of the first data source by the first data feature of the second data source to obtain a first product result of the data source; add the first data feature of the first data source to the first data feature of the second data source to obtain a first summation result; multiply the first data feature of the first data source by the first summation result to obtain a second product result; determine the ratio of the first product result to the second product result, and use the ratio as the correlation coefficient.

[0148] In some embodiments, the feature update module 4552 is also used to update the first data feature of the first data source and the first data feature of the second data source based on the correlation coefficient when the correlation coefficient is lower than the correlation threshold, so as to obtain the second data feature of the first data source and the second data feature of the second data source.

[0149] In some embodiments, the feature update module 4552 is also used to select any data source from the first data source and the second data source as the third data source when the correlation coefficient is not lower than the correlation threshold, and determine the target correlation coefficient between the first data feature of the third data source and the first data feature of the fourth data source; when the target correlation coefficient is lower than the correlation threshold, based on the target correlation coefficient, update the first data feature of the third data source and the first data feature of the fourth data source to obtain the second data feature of the third data source and the second data feature of the fourth data source.

[0150] In some embodiments, the feature fusion module 4553 is further used to perform feature fusion on the second data feature of the third data source and the second data feature of the fourth data source to obtain a target fused data feature.

[0151] In some embodiments, the task prediction module 4554 is also used to predict the target task based on the target fusion data features to obtain a target prediction result of the target task.

[0152] In some embodiments, the feature update module 4552 is also used to use the correlation coefficient as the weight of the first data feature of the first data source, perform data processing on the first data feature of the first data source, and obtain the second data feature of the first data source; use the correlation coefficient as the weight of the first data feature of the second data source, perform data processing on the first data feature of the second data source, and obtain the second data feature of the second data source.

[0153] In some embodiments, the feature fusion module 4553 is also used to determine a first influencing parameter value of the first data source on the prediction result, and determine a second influencing parameter value of the second data source on the prediction result; use the first influencing parameter value as the weight of the first data source, and use the second influencing parameter value as the weight of the second data source; based on the weight of the first data source and the weight of the second data source, perform feature fusion on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature.

[0154] In some embodiments, the target task is for a target object; the task prediction module 4554 is also used to predict the target task based on the fused data features, and obtain the basic object features of the target object before obtaining the prediction result of the target task. The basic object features include at least one of the following features: identification data features that identify the data source, and category data features that classify the data source.

[0155] In some embodiments, the task prediction module 4554 is further used to splice the basic features of the object with the fused data features to obtain spliced ​​data features; based on the spliced ​​data features, predict the target task to obtain a prediction result of the target task.

[0156] In some embodiments, the target task includes m subtasks, where m is an integer greater than 1; the task prediction module 4554 is also used to predict the m subtasks based on the fused data features to obtain subtask prediction results for each of the m subtasks.

[0157] The embodiment of the present application provides a computer program product, which includes computer executable instructions or computer programs, and the computer executable instructions or computer programs are stored in a computer-readable storage medium. The processor of the electronic device reads the computer executable instructions or computer programs from the computer-readable storage medium, and the processor executes the computer executable instructions or computer programs, so that the electronic device performs the task prediction method described above in the embodiment of the present application.

[0158] The present application embodiment provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions or computer programs are stored. When the computer-executable instructions or computer programs are executed by a processor, the processor will be caused to execute the task prediction method provided by the present application embodiment, for example, Figure 3A The task prediction method is shown.

[0159] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0160] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0161] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0162] As an example, computer executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0163] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A task prediction method, characterized in that: The method comprises: Determine a correlation coefficient between a first data feature of a first data source and a first data feature of a second data source; Based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source; Performing feature fusion on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature; Based on the fused data features, the target task is predicted to obtain a prediction result of the target task.

2. The method according to claim 1, characterized in that Before determining the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source, the method further includes: Acquire time series characteristics of the first data source and time series characteristics of the second data source; Time series prediction is performed on the time series features of the first data source and the time series features of the second data source to obtain first data features of the first data source and first data features of the second data source.

3. The method according to claim 1, characterized in that Determining the correlation coefficient between the first data feature of the first data source and the first data feature of the second data source includes: The first data feature of the first data source and the first data feature of the second data source are subjected to correlation processing to obtain the correlation coefficient.

4. The method according to claim 3, characterized in that The performing correlation processing on the first data feature of the first data source and the first data feature of the second data source to obtain the correlation coefficient includes: Multiplying a first data feature of the first data source by a first data feature of the second data source to obtain a first product result of the data sources; Adding the first data feature of the first data source and the first data feature of the second data source to obtain a first summation result; multiplying the first data feature of the first data source by the first summation result to obtain a second product result; A ratio of the first multiplication result to the second multiplication result is determined, and the ratio is used as the correlation coefficient.

5. The method according to claim 1, characterized in that The updating, based on the correlation coefficient, of the first data feature of the first data source and the first data feature of the second data source to obtain the second data feature of the first data source and the second data feature of the second data source includes: When the correlation coefficient is lower than the correlation threshold, based on the correlation coefficient, the first data feature of the first data source and the first data feature of the second data source are updated to obtain the second data feature of the first data source and the second data feature of the second data source; The method further comprises: When the correlation coefficient is not lower than the correlation threshold, select one of the first data source and the second data source as a third data source, and determine a target correlation coefficient between the first data feature of the third data source and the first data feature of the fourth data source; When the target correlation coefficient is lower than the correlation threshold, based on the target correlation coefficient, the first data feature of the third data source and the first data feature of the fourth data source are updated to obtain the second data feature of the third data source and the second data feature of the fourth data source; Performing feature fusion on the second data feature of the third data source and the second data feature of the fourth data source to obtain a target fused data feature; Based on the target fusion data features, the target task is predicted to obtain a target prediction result of the target task.

6. The method according to claim 1, characterized in that The updating, based on the correlation coefficient, of the first data feature of the first data source and the first data feature of the second data source to obtain the second data feature of the first data source and the second data feature of the second data source includes: Using the correlation coefficient as the weight of the first data feature of the first data source, performing data processing on the first data feature of the first data source, and obtaining a second data feature of the first data source; The correlation coefficient is used as the weight of the first data feature of the second data source, and data processing is performed on the first data feature of the second data source to obtain the second data feature of the second data source.

7. The method according to claim 1, characterized in that The step of fusing the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature includes: Determine a first influencing parameter value of the first data source on the prediction result, and determine a second influencing parameter value of the second data source on the prediction result; Using the first influencing parameter value as the weight of the first data source, and using the second influencing parameter value as the weight of the second data source; Based on the weight of the first data source and the weight of the second data source, feature fusion is performed on the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature.

8. The method according to claim 1, characterized in that The target task is for a target object, and before predicting the target task based on the fused data features and obtaining the prediction result of the target task, the method further includes: Acquire object basic features of the target object, wherein the object basic features include at least one of the following features: an identification data feature for identifying a data source, and a category data feature for a category data source; The step of predicting the target task based on the fused data features to obtain a prediction result of the target task includes: Splicing the object basic feature with the fused data feature to obtain a spliced ​​data feature; Based on the spliced ​​data features, a target task is predicted to obtain a prediction result of the target task.

9. The method according to claim 1, characterized in that: The target task includes m subtasks, where m is an integer greater than 1; and the target task is predicted based on the fused data features to obtain a prediction result of the target task, including: Based on the fused data features, the m subtasks are predicted to obtain subtask prediction results of each of the m subtasks.

10. A task prediction device, characterized in that: The device comprises: A coefficient determination module, used to determine a correlation coefficient between a first data feature of a first data source and a first data feature of a second data source; a feature updating module, configured to update the first data feature of the first data source and the first data feature of the second data source based on the correlation coefficient to obtain the second data feature of the first data source and the second data feature of the second data source; A feature fusion module, used for fusing the second data feature of the first data source and the second data feature of the second data source to obtain a fused data feature; The task prediction module is used to predict the target task based on the fused data features to obtain the prediction result of the target task.

11. An electronic device, characterized in that: The electronic device comprises: A memory for storing computer executable instructions or computer programs; A processor, configured to implement the task prediction method according to any one of claims 1 to 9 when executing the computer executable instructions or computer programs stored in the memory.

12. A computer-readable storage medium storing computer-executable instructions or a computer program, characterized in that: When the computer executable instructions or computer program are executed by a processor, the task prediction method according to any one of claims 1 to 9 is implemented.

13. A computer program product comprising computer executable instructions or a computer program, characterized in that: When the computer executable instructions or computer program are executed by a processor, the task prediction method according to any one of claims 1 to 9 is implemented.

Images