Data processing method and device, model training method and device, equipment and medium

Abstract

The invention provides a data processing method and device, a model training method and device, equipment and a medium, and relates to the technical field of artificial intelligence, in particular to the field of deep learning and data processing. According to the implementation scheme, an object relation graph is obtained; for a target object in the plurality of first objects, obtaining at least one meta path corresponding to the target object in the object relation graph; for each meta path, executing the following operations: determining a plurality of first attention weights of the target object based on the inherent attribute data of the target object and the inherent attribute data of each of a plurality of second objects on the meta path; acquiring a second representation vector of the target object based on the first representation vector of the target object and the plurality of first attention weights; and obtaining a target index prediction result of the target object at least based on the at least one second representation vector, corresponding to the at least one meta path, of the target object.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to the field of deep learning and data processing, and specifically to a data processing method, a model training method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] Artificial intelligence is a discipline that studies how to use computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.). It includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0003] The methods described in this section are not necessarily methods that have been previously conceived or employed. Unless otherwise indicated, it should not be assumed that any method described in this section is considered to be prior art simply because it is included in this section. Similarly, unless otherwise indicated, the issues mentioned in this section should not be considered to have been recognized in any prior art. Summary of the invention

[0004] The present disclosure provides a data processing method, a model training method, an apparatus, an electronic device, a computer-readable storage medium, and a computer program product.

[0005] According to one aspect of the present disclosure, a data processing method is provided, including: obtaining an object relationship graph, the object relationship graph including a plurality of first nodes corresponding to a plurality of first objects and used to represent association relationship information between the plurality of first objects; for a target object among the plurality of first objects, obtaining at least one meta-path corresponding to the target object in the object relationship graph, each meta-path in the at least one meta-path being a node path in the object relationship graph for connecting the target object with an associated object of the target object, and the target object respectively determines a plurality of association paths with a plurality of second objects other than the target object in the meta-path; for each meta-path in the at least one meta-path, performing the following operations: determining a plurality of first attention weights of the target object relative to the plurality of association paths based on inherent attribute data of the target object and inherent attribute data of each of the plurality of second objects on the meta-path; and obtaining a second representation vector of the target object after integrating the association relationship information represented by the plurality of association paths on the meta-path based on the first representation vector of the target object and the plurality of first attention weights, the first representation vector being determined based on the inherent attribute data of the target object; and obtaining a target indicator prediction result of the target object based at least on at least one second representation vector of the target object corresponding to at least one meta-path.

[0006] According to another aspect of the present disclosure, a model training method is provided, wherein the model includes a first feature extraction network and a prediction network, the first feature extraction network includes a first feature extraction sub-network, a first attention sub-network and a first feature fusion sub-network, and the model training method includes: obtaining an object relationship graph and sample data, the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and used to represent association relationship information between the plurality of first objects, the sample data includes inherent attribute data and target indicator data of the sample object, and the sample object is one of the plurality of first objects; for a sample object among the plurality of first objects, obtaining at least one meta-path corresponding to the sample object in the object relationship graph, each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the sample object with an associated object of the sample object, and the sample object and a plurality of associated paths are respectively determined with a plurality of second objects other than the sample object in the meta-path; for at least one meta-path in For each meta-path, the following operations are performed: the inherent attribute data of the sample object and the inherent attribute data of each of the multiple second objects on the meta-path are input into the first feature extraction network to obtain a first characterization vector of the sample object output by the first feature extraction network and multiple third characterization vectors corresponding to the multiple second objects; based on the first characterization vector and the multiple third characterization vectors, a first attention sub-network is used to determine multiple first attention weights of the sample object relative to the multiple associated paths; and based on the first characterization vector and the multiple first attention weights of the sample object, a first feature fusion sub-network is used to obtain a second characterization vector of the target object after integrating the association relationship information represented by the multiple associated paths on the meta-path; at least one second characterization vector of the sample object corresponding to at least one meta-path is input into the prediction network to obtain a target indicator prediction result of the sample object output by the prediction network; and the parameters of the model are adjusted based on the target indicator prediction result and the target indicator data.

[0007] According to another aspect of the present disclosure, a data processing device is provided, comprising: a first acquisition unit, configured to acquire an object relationship graph, the object relationship graph comprising a plurality of first nodes corresponding to a plurality of first objects and used to represent association relationship information between the plurality of first objects; a second acquisition unit, configured to acquire, for a target object among the plurality of first objects, at least one meta-path corresponding to the target object in the object relationship graph, each meta-path in the at least one meta-path being a node path in the object relationship graph for connecting the target object with an associated object of the target object, and the target object respectively determines a plurality of association paths with a plurality of second objects in the meta-path other than the target object; and an execution unit, configured to, for each meta-path in the at least one meta-path, Execute the operations of the following sub-units, the execution unit including: a determination sub-unit, configured to determine multiple first attention weights of the target object relative to multiple associated paths based on the inherent attribute data of the target object and the inherent attribute data of each of the multiple second objects on the meta-path; and an acquisition sub-unit, configured to acquire a second representation vector of the target object after integrating the association relationship information represented by the multiple associated paths on the meta-path based on the first representation vector of the target object and the multiple first attention weights, the first representation vector being determined based on the inherent attribute data of the target object; and a third acquisition unit, configured to obtain a target indicator prediction result of the target object based at least on at least one second representation vector of the target object corresponding to at least one meta-path.

[0008] According to another aspect of the present disclosure, a model training device is provided, wherein the model includes a first feature extraction network and a prediction network, wherein the first feature extraction network includes a first feature extraction subnetwork, a first attention subnetwork and a first feature fusion subnetwork, and the model training device includes: a first acquisition unit, configured to acquire an object relationship graph and sample data, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and used to represent association relationship information between the plurality of first objects, wherein the sample data includes inherent attribute data and target indicator data of the sample object, and the sample object is one of the plurality of first objects; a second acquisition unit, configured to acquire, for a sample object among the plurality of first objects, at least one meta-path corresponding to the sample object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the sample object with an associated object of the sample object, and the sample object respectively determines a plurality of associated paths with a plurality of second objects other than the sample object in the meta-path; an execution unit, configured to execute the operations of the following sub-units for each meta-path in the at least one meta-path, The execution unit includes: a first acquisition subunit, configured to input the inherent attribute data of the sample object and the inherent attribute data of each of the multiple second objects on the meta-path into a first feature extraction network to obtain a first representation vector of the sample object output by the first feature extraction network and multiple third representation vectors corresponding to the multiple second objects; a determination subunit, configured to determine multiple first attention weights of the sample object relative to multiple associated paths using a first attention subnetwork based on the first representation vector and the multiple third representation vectors; and a second acquisition subunit, configured to obtain a second representation vector of the target object after integrating the association relationship information represented by multiple associated paths on the meta-path using a first feature fusion subnetwork based on the first representation vector and the multiple first attention weights of the sample object; a third acquisition unit, configured to input at least one second representation vector of the sample object corresponding to at least one meta-path into a prediction network to obtain a target indicator prediction result of the sample object output by the prediction network; and an adjustment unit, configured to adjust the parameters of the model based on the target indicator prediction result and the target indicator data.

[0009] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned data processing method or model training method.

[0010] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the above-mentioned data processing method or model training method.

[0011] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program implements the above-mentioned data processing method or model training method when executed by a processor.

[0012] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings exemplarily illustrate the embodiments and constitute a part of the specification, and together with the text description of the specification, are used to explain the exemplary implementation of the embodiments. The embodiments shown are for illustrative purposes only and do not limit the scope of the claims. In all drawings, the same reference numerals refer to similar but not necessarily identical elements.

[0014] Figure 1 A schematic diagram showing an exemplary system in which the various methods described herein may be implemented according to an embodiment of the present disclosure;

[0015] Figure 2 A flow chart of a data processing method according to an embodiment of the present disclosure is shown;

[0016] Figure 3 A schematic diagram of the architecture of a model for target indicator prediction of a target object according to an exemplary embodiment of the present disclosure is shown;

[0017] Figure 4 A flowchart of determining a first attention weight according to an embodiment of the present disclosure is shown;

[0018] Figure 5 A flowchart of a method for obtaining a target indicator prediction result of a target object according to an embodiment of the present disclosure is shown;

[0019] Figure 6 A flow chart of a model training method according to an embodiment of the present disclosure is shown;

[0020] Figure 7 A structural block diagram of a data processing device according to an embodiment of the present disclosure is shown;

[0021] Figure 8 A structural block diagram of a model training device according to an embodiment of the present disclosure is shown;

[0022] Fig. 9 A structural block diagram of an exemplary electronic device that can be used to implement the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0023] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0024] In the present disclosure, unless otherwise specified, the use of the terms "first", "second", etc. to describe various elements is not intended to limit the positional relationship, temporal relationship, or importance relationship of these elements, and such terms are only used to distinguish one element from another element. In some examples, the first element and the second element may refer to the same instance of the element, and in some cases, based on the description of the context, they may also refer to different instances.

[0025] The terms used in the description of various examples in this disclosure are only for the purpose of describing specific examples and are not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element can be one or more. In addition, the term "and / or" used in this disclosure covers any one of the listed items and all possible combinations.

[0026] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0027] Figure 1 FIG. 1 is a schematic diagram of an exemplary system 100 in which various methods and apparatuses described herein may be implemented according to an embodiment of the present disclosure. Figure 1 , the system 100 includes one or more client devices 101, 102, 103, 104, 105, and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105, and 106 may be configured to execute one or more applications.

[0028] In an embodiment of the present disclosure, the server 120 may run one or more services or software applications that enable execution of the data processing method or model training method of the present disclosure.

[0029] In some embodiments, server 120 may also provide other services or software applications, which may include non-virtualized environments and virtualized environments. In some embodiments, these services may be provided as web-based services or cloud services, such as provided to users of client devices 101, 102, 103, 104, 105, and / or 106 under a software as a service (SaaS) model.

[0030] exist Figure 1 In the configuration shown, the server 120 may include one or more components that implement the functions performed by the server 120. These components may include software components, hardware components, or a combination thereof that can be executed by one or more processors. Users operating client devices 101, 102, 103, 104, 105, and / or 106 may in turn utilize one or more client applications to interact with the server 120 to utilize the services provided by these components. It should be understood that a variety of different system configurations are possible, which may differ from the system 100. Therefore, Figure 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

[0031] The user may use client devices 101, 102, 103, 104, 105 and / or 106 to obtain the prediction result of the target indicator of the target object. The client device may provide an interface that enables the user of the client device to interact with the client device. The client device may also output information to the user via the interface. Figure 1 Only six client devices are depicted, but one skilled in the art will appreciate that the present disclosure may support any number of client devices.

[0032] Client devices 101, 102, 103, 104, 105 and / or 106 may include various types of computer devices, such as portable handheld devices, general-purpose computers (such as personal computers and laptop computers), workstation computers, wearable devices, smart screen devices, self-service terminal devices, service robots, game systems, thin clients, various messaging devices, sensors or other sensing devices, etc. These computer devices may run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as GOOGLE Chrome OS); or include various mobile operating systems, such as MICROSOFT Windows Mobile OS, iOS, Windows Phone, Android. Portable handheld devices may include cellular phones, smart phones, tablet computers, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays (such as smart glasses) and other devices. Game systems may include various handheld game devices, Internet-enabled game devices, etc. Client devices are capable of executing various different applications, such as various Internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and may use various communication protocols.

[0033] The network 110 may be any type of network known to those skilled in the art that may support data communications using any of a variety of available protocols, including but not limited to TCP / IP, SNA, IPX, etc. By way of example only, the one or more networks 110 may be a local area network (LAN), an Ethernet-based network, a token ring, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a blockchain network, a public switched telephone network (PSTN), an infrared network, a wireless network (e.g., Bluetooth, WIFI), and / or any combination of these and / or other networks.

[0034] Server 120 may include one or more general purpose computers, dedicated server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and / or combination. Server 120 may include one or more virtual machines running virtual operating systems, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that may be virtualized to maintain a server's virtual storage device). In various embodiments, server 120 may run one or more services or software applications that provide the functionality described below.

[0035] The computing units in the server 120 may run one or more operating systems including any of the above operating systems and any commercially available server operating systems. The server 120 may also run any of a variety of additional server applications and / or middle-tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0036] In some implementations, server 120 may include one or more applications to analyze and consolidate data feeds and / or event updates received from users of client devices 101, 102, 103, 104, 105, and / or 106. Server 120 may also include one or more applications to display data feeds and / or real-time events via one or more display devices of client devices 101, 102, 103, 104, 105, and / or 106.

[0037] In some embodiments, the server 120 may be a server of a distributed system, or a server combined with a blockchain. The server 120 may also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. A cloud server is a host product in a cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and virtual private servers (VPS) services.

[0038] The system 100 may also include one or more databases 130. In some embodiments, these databases may be used to store data and other information. For example, one or more of the databases 130 may be used to store information such as audio files and video files. The databases 130 may reside in various locations. For example, the database used by the server 120 may be local to the server 120, or may be remote from the server 120 and may communicate with the server 120 via a network-based or dedicated connection. The databases 130 may be of different types. In some embodiments, the databases used by the server 120 may be, for example, relational databases. One or more of these databases may store, update, and retrieve data to and from the databases in response to commands.

[0039] In some embodiments, one or more of the databases 130 may also be used by applications to store application data. The databases used by the applications may be different types of databases, such as a key-value store, an object store, or a conventional store backed by a file system.

[0040] Figure 1 The system 100 may be configured and operated in various ways to enable the application of various methods and apparatuses described in the present disclosure.

[0041] According to the embodiments of the present disclosure, Figure 2 As shown, a data processing method is provided, comprising:

[0042] Step S201: Acquire an object relationship graph, where the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects;

[0043] Step S202: for a target object among the multiple first objects, obtaining at least one meta-path corresponding to the target object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the target object and an associated object of the target object, and the target object and multiple associated paths are respectively determined with multiple second objects other than the target object in the meta-path;

[0044] For each meta-path in at least one meta-path, perform the following operations:

[0045] Step S203, determining a plurality of first attention weights of the target object relative to the plurality of associated paths based on the inherent attribute data of the target object and the inherent attribute data of each of the plurality of second objects on the meta-path; and

[0046] Step S204, based on the first representation vector of the target object and the plurality of first attention weights, obtaining a second representation vector of the target object after integrating the association relationship information represented by the plurality of association paths on the meta-path, wherein the first representation vector is determined based on the inherent attribute data of the target object; and

[0047] Step S205: Obtain a target indicator prediction result of the target object based on at least one second characterization vector of the target object corresponding to at least one meta-path.

[0048] Therefore, by constructing an object relationship network, the association relationship information between each first object is recorded (for example, including at least one of the association relationship between funds, the social relationship between fund managers, and the management relationship of fund managers to funds); at the same time, by obtaining the attention weights of each association path existing on the meta-path for each meta-path involved in the target fund, and performing feature fusion based on the attention weights to obtain the first representation vector, and then predict the target indicator, so that the association relationship information contained in each meta-path can be integrated into the representation vector of the target object, thereby improving the prediction accuracy of the target indicator of the target object.

[0049] In some scenarios, when predicting the target indicators of a target object, if only the inherent data of the target object itself is considered, some objective connections between things are usually ignored, which leads to the prediction accuracy failing to meet expectations.

[0050] For example, in financial scenarios, the measurement and evaluation of fund performance usually only considers the explicit information characteristics of the fund itself, such as fund return rate, fund size and popularity, as well as the influence of a single fund manager on the management of the fund, and uses traditional statistical calculation methods to evaluate fund performance.

[0051] In actual scenarios, there are also certain intrinsic relationships between funds (e.g., similar relationships between funds). In the exemplary embodiments of the present disclosure, by introducing the above intrinsic relationships into the prediction of the target indicators of the target object (e.g., the performance of the fund) in a certain way, the accuracy of the prediction can be improved.

[0052] In some exemplary embodiments, the target object may be a target fund, and the target indicator may be the performance of the target fund (for example, it may be expressed as the rate of return of the fund). Multiple funds on the market may be used as multiple first objects, and each first object may be expressed as a first node in an object relationship graph, and in response to the existence of a similar relationship between two funds, the first nodes corresponding to the two funds are connected through a node path to form an object relationship graph between the multiple funds.

[0053] In some exemplary embodiments, the similar relationship between two funds can be determined by calculating the similarity of the two funds' holdings based on the proportion of each stock held by the two funds (for example, if the similarity is greater than a first similarity threshold, it is determined that there is an association relationship between the two funds).

[0054] In some exemplary embodiments, based on the target object (target fund), all associated objects (associated funds) that have an association relationship with it in the object relationship graph can be determined, that is, the first object in the object relationship graph that can be indirectly connected to the target object through a node path; then, each path through which the target object connects to each associated object can be determined as at least one meta-path of the target object.

[0055] In some exemplary embodiments, each meta-path includes a target object and a plurality of second objects other than the target object. On the meta-path, the target object and each second object respectively form an associated path.

[0056] Figure 3 A schematic diagram of the architecture of a model for predicting a target indicator of a target object according to an exemplary embodiment of the present disclosure is shown.

[0057] In some exemplary embodiments, Figure 3 The model 300 shown may be a fund performance prediction model, which may include a first feature extraction network 310 and a prediction network 320. The first feature extraction network 310 may include a first feature extraction sub-network 311, a first attention sub-network 312, and a first feature fusion sub-network 313.

[0058] In some exemplary embodiments, for each meta-path in at least one meta-path, intrinsic attribute data of the target object and the second object on the meta-path may be first obtained, wherein the intrinsic attribute data of each object may be represented as an intrinsic attribute vector h j . Then, the intrinsic attribute vector h of each object can be j Input to a feature extraction network (e.g. Figure 3In the first feature extraction subnetwork 311 in the first feature extraction network 310 shown in the figure, the first characterization vector of the target object and the characterization vector of each second object are obtained. The above process can be expressed as the following formula:

[0059]

[0060] in, is the feature transformation matrix (i.e., the parameter matrix of the feature extraction network), h′ j A representation vector representing each object.

[0061] In some exemplary embodiments, the inherent attribute data may include, but is not limited to, at least one of the fund's establishment year, the fund's size at the time of establishment, and the fund's share at the time of establishment.

[0062] In some exemplary embodiments, the representation vectors of the two funds on the meta-path may be concatenated to obtain the representation vectors of all associated paths on the meta-path, and the first attention weight of each associated path representation vector may be obtained based on the attention mechanism (for example, the representation vectors of all associated paths are input into Figure 3 The first attention sub-network 312 in the first feature extraction network 310 shown in the figure is used to obtain the first attention weight of each associated path representation vector output by the first attention sub-network 312); then, the representation vector of the target object can be weighted based on the first attention weights corresponding to the multiple associated paths related to the target object (for example, applying Figure 3 The first feature fusion subnetwork 313 in the first feature extraction network 310 shown in the figure is used to obtain the expression of the target object on the meta-path (ie, the second representation vector).

[0063] Subsequently, the second representation vector of the target object corresponding to each meta-path can be input into a prediction network (e.g. Figure 3 The prediction network 320 shown in the figure) is used to obtain the target indicator prediction result (the prediction result of the fund performance) of the target object output by the prediction network.

[0064] According to some embodiments, the multiple first objects include multiple management subjects and multiple managed objects, the target object and each of the multiple second objects is a managed object, the multiple association paths include multiple association types, and the multiple association types include at least two of the following: two managed objects are directly associated, two managed objects are indirectly associated based on at least one management subject, two managed objects are indirectly associated based on at least one managed object, and two managed objects are indirectly associated based on at least one management subject and at least one managed object.

[0065] And, if Figure 4As shown, based on the inherent attribute data of the target object and the inherent attribute data of each of the multiple second objects on the meta-path, determining multiple first attention weights of the target object relative to multiple association paths includes: for each association type among multiple association types, performing the following operations: step S401, determining at least one association path corresponding to the association type and at least one second object corresponding to at least one association path among multiple association paths; and step S402, based on the first representation vector of the target object and the third representation vector corresponding to each of the at least one second object, determining at least one second attention weight of the target object corresponding to the association type relative to at least one association path, the third representation vector being determined based on the inherent attribute data of the corresponding second object; and step S403, determining multiple first attention weights based on at least one second attention weight corresponding to each of the multiple association types.

[0066] In this way, the associated paths can be further divided into multiple path types, and by calculating the attention weights of each path in each type respectively, the associated type information can be further integrated into the representation vector, thereby improving the prediction accuracy of the target indicator.

[0067] In some exemplary embodiments, in the fund performance forecasting scenario, on the basis of the above, it is also possible to further introduce management relationships between fund managers and funds, association relationships between fund managers and fund managers (for example, two fund managers are employed by the same company at the same time), and other association relationships. By mining the above-mentioned intrinsic association relationships in the relevant data, more abundant feature information is provided for the prediction of the target indicators of the target object, thereby further improving the accuracy of the prediction.

[0068] In some exemplary embodiments, each meta-path may include a target fund and one or more associated funds (i.e., managed objects), and in some cases, may also include one or more fund managers (i.e., management entities). In response to a fund manager managing a fund, there is an association relationship between the fund manager and the fund, connected by a node path; in response to an association relationship between fund managers (e.g., two fund managers belong to the same company), the two fund manager nodes may be connected by a node path.

[0069] In some exemplary embodiments, on each meta-path, there may be multiple relationships between the target fund and its associated funds, including direct relationships (i.e., direct connection through a node path) and indirect relationships (i.e., indirect connection via one or more other nodes), wherein the indirect relationships may include indirect relationships based on at least one fund, indirect relationships based on at least one fund manager, and indirect relationships based on at least one fund manager and at least one fund.

[0070] In some exemplary embodiments, the management entity may further include a company that employs a fund manager, and the object relationship diagram may further include an employment relationship between the company and the fund manager it employs, as well as an association relationship between the two companies (e.g., a cooperative relationship, affiliation, equity relationship, etc.) In response to the above relationship, two nodes may be connected through a node path.

[0071] Therefore, we can further introduce the employment relationship between the company and the fund manager, the relationship between companies, etc., so as to provide richer feature information for the prediction of the target indicators of the target object and further improve the accuracy of its prediction by mining the above-mentioned intrinsic relationships in the relevant data.

[0072] In some exemplary embodiments, the above object relationship diagram can be represented as Figure 3 The node relationship network G is shown.

[0073] In some exemplary embodiments, for each meta-path Φ, the characterization vector h′ of the target fund j on the meta-path Φ may be obtained by: j The characterization vector h′ of the associated fund i i Concatenate to obtain the representation vector of each association path.

[0074] In some exemplary embodiments, the above-mentioned multiple association paths may include multiple association types. When obtaining the first attention weight of each association path, the representation vectors of all association paths related to the target fund of the same association type can be input together. Figure 3 The first attention sub-network 312 in the first feature extraction network 310 shown is used to obtain a second attention weight for each association path representation vector of the association type output by the first attention sub-network 312.

[0075] In some exemplary embodiments, the processing of the first attention sub-network 312 can be expressed by the following formula:

[0076]

[0077]

[0078] in, represents the Φ composed of target fund j and associated fund i k The second attention weight corresponding to the association path of the association type, is the parameter vector, The characterization vector corresponding to each associated path can be represented, where k∈[1,2,…,K t ].

[0079] According to some embodiments, based on the first representation vector and multiple first attention weights of the target object, obtaining a second representation vector of the target object after integrating the association relationship information of multiple association path representations on the meta-path includes: for each association type among multiple association types, based on the first representation vector of the target fund and at least one second attention weight corresponding to the association type among the multiple first attention weights, obtaining a fourth representation vector corresponding to the association type; and splicing multiple fourth representation vectors corresponding to the multiple association types to obtain a second representation vector.

[0080] Therefore, in the feature fusion stage, feature fusion is performed for each association type based on the attention weight information corresponding to the type, and then the representation vectors corresponding to each association type are concatenated, so as to further integrate the association type information into the representation vector of the target object, thereby improving the prediction accuracy of the target indicator of the target object.

[0081] In some exemplary embodiments, the first characterization vector h′ may be j And the corresponding second attention weights of each association path of each association type The first feature fusion sub-network 313 is input together to obtain the second representation vector of the target fund on the meta-path.

[0082] In some exemplary embodiments, the processing of the first feature fusion sub-network 313 can be expressed by the following formula:

[0083]

[0084] in, Represents the second characterization vector of the target fund j on the meta-path Φ.

[0085] According to some embodiments, the number of at least one meta-path is multiple, and Figure 5 As shown, obtaining the target indicator prediction result of the target object based on at least one second characterization vector corresponding to at least one meta-path of the target object may include: step S501, for each meta-path in at least one meta-path, based on the second characterization vector corresponding to each managed object on the meta-path, determining the path characterization vector of the meta-path; step S502, based on each corresponding path characterization vector in at least one meta-path, determining at least one third attention weight corresponding to at least one meta-path; step S503, based on at least one second characterization vector and at least one third attention weight, obtaining the fifth characterization vector of the target object; and step S504, obtaining the target indicator prediction result based at least on the fifth characterization vector.

[0086] Therefore, for multiple meta-paths related to the target object, the representation vector of the path is determined based on the representation vector corresponding to each object on each path; then, based on the representation vector of each path, the attention weight of each path is obtained, and the target-specific representation vector corresponding to each path is weighted based on the path attention weight, so as to further integrate the path information into the representation vector, thereby improving the prediction accuracy of the target indicator of the target object.

[0087] In some exemplary embodiments, for example, in a fund performance prediction scenario, for a target fund j, it may have multiple meta-paths Φ p (where p is a positive integer and p∈[1,P], P represents the number of meta-paths corresponding to the target fund j), where multiple meta-paths Φ p The set of meta-paths Φ P .

[0088] In some exemplary embodiments, for each associated fund on each meta-path, each associated fund may be used as a target fund, and the second representation vector of the associated fund on the meta-path may be obtained by the above method.

[0089] In some exemplary embodiments, Figure 3 In the illustrated model 300 , the prediction network 320 may include a second feature fusion subnetwork 321 , a second attention subnetwork 322 , a third feature fusion subnetwork 323 , and a first prediction subnetwork 324 .

[0090] In some exemplary embodiments, the second representation vectors corresponding to all fund objects on a meta-path may be input into a feature fusion network (e.g. Figure 3 The second feature fusion subnetwork 321 in the prediction network 320 shown in the figure is used to obtain the path representation vector corresponding to the meta-path.

[0091] In some exemplary embodiments, the processing of the second feature fusion sub-network 321 can be expressed by the following formula:

[0092]

[0093] in, Represents the meta-path Φ p The path representation vector of Represents the meta-path Φ p The second characterization vector of fund j on Represents the meta-path Φ p The total number of funds included in the above, d, W and b are all model parameters.

[0094] In some exemplary embodiments, the path representation vector corresponding to each meta-path described above can be input into an attention network (e.g. Figure 3 In the second attention subnetwork 322 in the prediction network 320 shown, a third attention weight corresponding to each meta-path is obtained.

[0095] In some exemplary embodiments, the processing of the second attention sub-network 322 can be expressed by the following formula:

[0096]

[0097] in, is the meta-path Φ p The third attention weight of .

[0098] Then, the second representation vector of the target fund on each meta-path and the corresponding third attention weight can be input into a feature fusion network (e.g. Figure 3 In the third feature fusion subnetwork 323 in the prediction network 320 shown, a fifth representation vector of the target fund that integrates the information of each meta-path is obtained.

[0099] In some exemplary embodiments, the processing of the third feature fusion sub-network 323 can be expressed by the following formula:

[0100]

[0101] in, The fifth characterization vector representing the target fund j, Indicates that the target fund j is in the meta-path Φ p The second characterization vector on .

[0102] Then, the fifth characterization vector can be input as Figure 3 In the first prediction sub-network 324 in the prediction network 320 shown, the performance prediction result of the target fund is obtained.

[0103] In some exemplary embodiments, the object relationship network may also include a meta-path that directly connects the target fund and the associated funds. For this type of meta-path, the representation vector corresponding to each fund therein may be the first representation vector corresponding to the fund obtained by the above method.

[0104] In some exemplary embodiments, the path representation vector of the meta-path can be obtained in a manner similar to that described above, and introduced into the calculation of the path attention weight and the subsequent feature fusion process, thereby further introducing richer path information and improving the accuracy of subsequent predictions.

[0105] According to some embodiments, each of a plurality of managed objects is composed of a plurality of sub-objects in proportion, and the above-mentioned data processing method may further include: obtaining tensor data corresponding to the object relationship graph, the tensor data being used to record the proportion data of each sub-object corresponding to each managed object in the object relationship graph; and performing tensor decomposition on the tensor data to obtain a seventh characterization vector corresponding to each managed object in the object relationship graph, the seventh characterization vector being used to characterize the component characteristics of the corresponding managed object; and wherein, at least based on the sixth characterization vector, obtaining the target indicator prediction result includes: performing feature fusion on the sixth characterization vector and the seventh characterization vector of the target object to obtain an eighth characterization vector of the target object; and obtaining the target indicator prediction result at least based on the eighth characterization vector.

[0106] In this way, the intrinsic component characteristics of the target object can be further integrated into the representation vector of the target object, thereby improving the prediction accuracy of the target indicator of the target object.

[0107] In some exemplary embodiments, such as in a fund performance prediction scenario, the intrinsic component characteristics of the target object may be investment characteristics of the target fund for stocks, such as the holding ratio of the target fund for each stock it holds.

[0108] In some exemplary embodiments, the above object relationship diagram may further include stock nodes. If a fund holds a certain stock, the two nodes may be connected through a node path.

[0109] In some exemplary embodiments, the intrinsic representation of each fund can also be obtained in the following manner: First, based on the above object relationship diagram, the management-investment tensor T can be obtained, where each element t ijk ∈T represents the proportion of stock k held by fund j managed by fund manager i. If fund j managed by fund manager i does not hold stock k, then t ijk Assign a value of 0.

[0110] In some exemplary embodiments, the management-investment tensor T can be decomposed through a tensor decomposition network to obtain three implicit vectors U, V, and Z of the tensor, which respectively represent the fund manager's preference, the fund investment representation, and the stock's intrinsic characteristics.

[0111] In some exemplary embodiments, the tensor decomposition network can measure the loss between the ground truth value and the predicted value of each element in the management investment tensor by using the mean square error (MSE), and train the tensor decomposition network based on the loss. It can be calculated by the following formula:

[0112]

[0113] Manages the loss between the ground truth and the predicted value for each element in the investment tensor It can be expressed by the following formula:

[0114]

[0115] Through the above method, the intrinsic representation V of the fund can be obtained.

[0116] In some exemplary embodiments, Figure 3 The model 300 shown may further include a tensor decomposition network 330, and the first prediction sub-network 324 may include a third feature fusion sub-network 324-1 and a second prediction sub-network 324-2.

[0117] In some exemplary embodiments, the tensor decomposition described above may be implemented using a tensor decomposition network 330 .

[0118] In some exemplary embodiments, after obtaining the fund investment representation V, the vector corresponding to the target fund can be obtained as the investment representation of the target fund (that is, the sixth representation vector). Subsequently, the fifth representation vector and the sixth representation vector of the target fund can be input into the third feature fusion sub-network 324-1 together to obtain the intrinsic representation vector of the target fund (that is, the seventh representation vector). This vector is a representation vector after integrating the above-mentioned dimensional information of the target fund. Subsequently, the representation vector is input into the second prediction sub-network 324-2 to perform fund performance prediction, so as to obtain more accurate prediction results.

[0119] In some exemplary embodiments, the processing of the third feature fusion sub-network 324-1 can be expressed by the following formula:

[0120] e f =e g +e t

[0121] Among them, e g Represents the fifth representation vector of the fund, e t Represents the sixth characterization vector of the fund, e f Represents the seventh characterization vector of the fund.

[0122] According to some embodiments, the target indicator prediction result includes the target indicator prediction value of the target object on the target date, and the above-mentioned data processing method may also include: for each first date of at least one first date before the target date, performing the following operations: based on the target dynamic attribute data of the first date, determining a preset number of third objects from multiple managed objects, and the difference between the dynamic attribute data of the third object and the target dynamic attribute data is less than the difference between the dynamic attribute data of other managed objects other than the third object in the multiple managed objects and the target dynamic attribute data; and performing feature fusion on the eighth characterization vector corresponding to each third object in the preset number of third objects to obtain the first dynamic feature characterization vector corresponding to the first date; and wherein, at least based on the eighth characterization vector, obtaining the target indicator prediction result includes: obtaining the target indicator prediction value of the target object on the target date based on at least the eighth characterization vector and the first dynamic feature characterization vector of each first date of at least one first date.

[0123] In this way, some dynamic feature information (such as the dynamic change information of the overall market every day) can be further integrated into the representation vector of the target object, thereby improving the prediction accuracy of the target indicator of the target object.

[0124] In some exemplary embodiments, in a fund performance prediction scenario, fund market factors (such as dynamic fund market trends) may be further introduced into the prediction process of the target fund performance.

[0125] In some exemplary embodiments, the fund market characteristics can be approximately represented based on the intrinsic characteristics of the fund. For example, the fund market characteristics can be the K most closely related to the fund market changes in each fund. f The average of the intrinsic representation of the funds.

[0126] In some exemplary embodiments, the K that is closest to the fund market trend on the first date can be determined by comparing the fund market return rate on the first date with the difference between the fund return rates of various funds. f A fund.

[0127] In some exemplary embodiments, the rate of return on the first day can be determined by the cumulative net value of the day and the cumulative net value of the previous day. For example, the cumulative net value of fund j on day t is Values t , the cumulative net value on day t-1 is Values t-1 , then the fund's rate of return on day t can be calculated using the following formula:

[0128]

[0129] Market return on day t It can be obtained based on a similar method based on the cumulative market net value on the t-th day and the cumulative market net value on the t-1 day.

[0130] The fund market return rate on day t and the fund yield of fund j The difference between can be expressed as:

[0131]

[0132] Then, we can get the K with the smallest difference f Only funds, and based on this K f The intrinsic characterization vector of the fund is used to obtain the dynamic characteristic characterization vector of the fund market on day t:

[0133]

[0134] In some exemplary embodiments, the seventh characterization vector of the target fund and the first dynamic feature characterization vector of each first date before the target date may be input into the second prediction subnetwork 324 - 2 to obtain the fund performance prediction result of the target fund on the target date.

[0135] In some exemplary embodiments, Figure 3 The model 300 shown may also include a feature fusion network 340 .

[0136] In some exemplary embodiments, considering the lag of market impact, the dynamic feature characterization vector of the fund market on the tth day can be obtained by the above method, and at the same time, the dynamic feature characterization vector of the fund market on the previous n trading days can be obtained. The fund dynamic features of the t-th day and the previous n trading days can be further integrated through the feature fusion network 340, so as to obtain the dynamic feature representation vector of the fund market on the t-th day after considering the market lag, as the first dynamic feature representation vector of the t-th day.

[0137] In some exemplary embodiments, the feature fusion network 340 may be a recurrent neural network (Gate Recurrent Unit, GRU). The processing of the feature fusion network 340 may be represented by the following formula:

[0138]

[0139] in, It represents the dynamic characteristic vector of the fund market on the tth day after considering the market lag, and n is a positive integer.

[0140] In some exemplary embodiments, the first dynamic feature characterization vector of the first date after considering market lag and the seventh characterization vector of the target fund may be input into the second prediction subnetwork 324 - 2 to obtain a fund performance prediction result of the target fund on the target date.

[0141] In some exemplary embodiments, the first dynamic feature characterization vector and the seventh characterization vector of the target fund may be firstly fused to obtain the macro characteristics of the target fund, and the macro characteristics of the target fund are input into the second prediction subnetwork 324-2 to obtain the fund performance prediction result of the target fund on the target date. The above feature fusion may be expressed as:

[0142]

[0143] in, represents the macro characteristics of the target fund, e f The seventh characterization vector representing the target fund, Represents the first dynamic feature representation vector.

[0144] According to some embodiments, the above-mentioned data processing method may also include: obtaining a second dynamic feature characterization vector of the target object on the target date, the second dynamic feature characterization vector being determined based on the dynamic attribute data of the target object on each second date in at least one second date before the target date; and wherein, based at least on the eighth characterization vector and the first dynamic feature characterization vector of each first date in at least one first date, obtaining the target indicator prediction value of the target object on the target date includes: obtaining the target indicator prediction value of the target object on the target date based on the eighth characterization vector, the first dynamic feature characterization vector of each first date in at least one first date, and the second dynamic feature characterization vector.

[0145] In this way, the dynamic feature information of the target object itself can be further integrated into the representation vector of the target object, thereby improving the prediction accuracy of the target indicator of the target object.

[0146] In some exemplary embodiments, in a fund performance forecasting scenario, the dynamic characteristic information of the target fund itself may be further introduced into the fund performance forecast of the target fund, thereby further improving the accuracy of the fund performance forecast.

[0147] In some exemplary embodiments, the dynamic characteristic information of the target fund itself can be determined by the dynamic attribute data of the fund. The dynamic attribute data can include at least one of the cumulative net value and the unit net value of the target fund on each of at least one second date before the target date.

[0148] In some exemplary embodiments, Figure 3The model 300 shown may also include a second feature extraction network 350 .

[0149] In some exemplary embodiments, the dynamic attribute data of each of the at least one second date of the target fund may be input into the second feature extraction network 350 to obtain a second dynamic feature representation vector of the target fund corresponding to the target date. Afterwards, you can

[0150] In some exemplary embodiments, the eighth characterization vector, the first dynamic feature characterization vector of each first date in at least one first date, and the second dynamic feature characterization vector may be input into the second prediction subnetwork 324-2 to obtain the fund performance prediction result of the target fund on the target date.

[0151] In some exemplary embodiments, the second prediction sub-network 324-2 may be a multi-layer perceptron (MLP), and the processing of the second prediction sub-network 324-2 may be expressed by the following formula:

[0152]

[0153] in, is the performance forecast value of the target fund on the target date (day t+1), The macro characteristics of the target fund.

[0154] In some exemplary embodiments, the above information of multiple target funds for multiple target dates can be obtained simultaneously and sorted accordingly, and the above data can be simultaneously input into the model as described above, so that the model can obtain the performance forecast values ​​of multiple target dates for multiple target funds after being processed as above.

[0155] According to some embodiments, Figure 6 As shown, a model training method is also provided, wherein the model includes a first feature extraction network and a prediction network, the first feature extraction network includes a first feature extraction sub-network, a first attention sub-network and a first feature fusion sub-network, and the model training method includes:

[0156] Step S601: obtaining an object relationship graph and sample data, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects, and the sample data includes inherent attribute data and target indicator data of the sample object, and the sample object is one of the plurality of first objects;

[0157] Step S602: for a sample object among the multiple first objects, obtaining at least one meta-path corresponding to the sample object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the sample object with an associated object of the sample object, and the sample object and multiple associated paths are respectively determined with multiple second objects other than the sample object in the meta-path;

[0158] For each meta-path in at least one meta-path, perform the following operations:

[0159] Step S603: input the intrinsic attribute data of the sample object and the intrinsic attribute data of each of the plurality of second objects on the meta-path into the first feature extraction network to obtain a first characterization vector of the sample object output by the first feature extraction network and a plurality of third characterization vectors corresponding to the plurality of second objects;

[0160] Step S604: Based on the first representation vector and the multiple third representation vectors, use the first attention sub-network to determine multiple first attention weights of the sample object relative to the multiple association paths; and Step S605: Based on the first representation vector and the multiple first attention weights of the sample object, use the first feature fusion sub-network to obtain a second representation vector of the target object after fusing the association relationship information represented by the multiple association paths on the meta-path;

[0161] Step S606, inputting at least one second representation vector of the sample object corresponding to at least one meta-path into the prediction network to obtain a target indicator prediction result of the sample object output by the prediction network; and

[0162] Step S607: Adjust the parameters of the model based on the target indicator prediction results and the target indicator data.

[0163] Therefore, the model trained by the above method can record the association relationship information between each first object (for example, including the association relationship between funds, the social relationship between fund managers, and at least one of the management relationship of fund managers to funds) by constructing an object relationship network; at the same time, by obtaining the attention weights of each association path existing on the meta-path for each meta-path involved in the target fund, and performing feature fusion based on the attention weights to obtain the first representation vector, and then predict the target indicator, so that the association relationship information contained in each meta-path can be integrated into the representation vector of the target object, thereby improving the prediction accuracy of the target indicator of the target object.

[0164] According to some embodiments, the plurality of first objects include a plurality of management subjects and a plurality of managed objects, each of the sample object and the plurality of second objects is a managed object, the plurality of association paths include a plurality of association types, the plurality of association types include at least two of the following: direct association between two managed objects, indirect association between two managed objects based on at least one management subject, indirect association between two managed objects based on at least one managed object, and indirect association between two managed objects based on at least one management subject and at least one managed object, and, based on the first characterization vector and the plurality of third characterization vectors, determining a plurality of first attention weights of the sample object relative to the plurality of association paths using the first attention subnetwork includes: for each association type in the plurality of association types, performing the following operations: determining, in the plurality of association paths, at least one association path corresponding to the association type and at least one second object corresponding to the at least one association path; and determining, based on the first characterization vector of the sample object and each corresponding third characterization vector of at least one second object, at least one second attention weight of the sample object corresponding to the association type relative to at least one association path; and determining a plurality of first attention weights based on at least one second attention weight corresponding to each association type in the plurality of association types.

[0165] According to some embodiments, based on the first representation vector and multiple first attention weights of the sample object, a second representation vector of the target object after fusing the association relationship information of multiple association path representations on the meta-path using the first feature fusion subnetwork includes: for each association type among multiple association types, based on the first representation vector of the target object and at least one second attention weight corresponding to the association type among the multiple first attention weights, obtaining a fourth representation vector corresponding to the association type; and splicing multiple fourth representation vectors corresponding to the multiple association types to obtain a second representation vector.

[0166] According to some embodiments, the number of at least one meta-path is multiple, the prediction network includes a second feature fusion subnetwork, a second attention subnetwork, a third feature fusion subnetwork and a first prediction subnetwork, and at least one second representation vector of the sample object corresponding to at least one meta-path is input into the prediction network to obtain a target indicator prediction result of the sample object output by the prediction network, including: for each meta-path in at least one meta-path, the second representation vector corresponding to each managed object on the meta-path is input into the second feature fusion subnetwork to obtain a path representation vector of the meta-path output by the second feature fusion subnetwork; based on each corresponding path representation vector in at least one meta-path, at least one third attention weight corresponding to at least one meta-path is determined using the second attention subnetwork; based on at least one second representation vector and at least one third attention weight, a fifth representation vector of the target object is obtained using the third feature fusion subnetwork; and at least the fifth representation vector is input into the first prediction subnetwork to obtain a target indicator prediction result output by the first prediction subnetwork.

[0167] According to some embodiments, each of a plurality of managed objects is composed of a plurality of sub-objects in proportion, the above-mentioned model may further include a tensor decomposition network, the first prediction sub-network includes a third feature fusion sub-network and a second prediction sub-network, and the above-mentioned model training method may further include: obtaining tensor data corresponding to the object relationship graph, the tensor data being used to record the proportion data of each sub-object corresponding to each managed object in the object relationship graph; and inputting the tensor data into the tensor decomposition network to obtain a seventh representation vector corresponding to each managed object in the object relationship graph output by the tensor decomposition network, the seventh representation vector being used to represent the component characteristics of the corresponding managed object; and wherein, at least inputting the sixth representation vector into the prediction sub-network to obtain the target indicator prediction result output by the first prediction sub-network includes: inputting the sixth representation vector and the seventh representation vector of the sample object into the third feature fusion sub-network to obtain the eighth representation vector of the sample object output by the third feature fusion sub-network; and at least inputting the eighth representation vector into the second prediction sub-network to obtain the target indicator prediction result output by the second prediction sub-network.

[0168] According to some embodiments, the target indicator data includes the target indicator value of the sample object on the sample date, the target indicator prediction result includes the target indicator prediction value of the sample object on the sample date, the above-mentioned model may also include a feature fusion network, and the above-mentioned model training method may also include: for each first date of at least one first date before the sample date, perform the following operations: based on the target dynamic attribute data of the first date, determine a preset number of third objects from multiple managed objects, and the difference between the dynamic attribute data of the third object and the target dynamic attribute data is less than the difference between the dynamic attribute data of other managed objects other than the third object in the multiple managed objects and the target dynamic attribute data; and input the eighth characterization vector corresponding to each third object in the preset number of third objects into the feature fusion network to obtain the first dynamic feature characterization vector corresponding to the first date output by the feature fusion network; and wherein, at least the eighth characterization vector is input into the second prediction subnetwork to obtain the target indicator prediction result output by the second prediction subnetwork includes: at least the eighth characterization vector and the first dynamic feature characterization vector of each first date of at least one first date are input into the second prediction subnetwork to obtain the target indicator prediction value of the sample object on the sample date output by the second prediction subnetwork.

[0169] According to some embodiments, the sample data may also include dynamic attribute data of the sample object for each of at least one second date before the sample date, the above-mentioned model may also include a second feature extraction network, and the above-mentioned model training method may also include: inputting the dynamic attribute data of the sample object for each of at least one second date into the second feature extraction network to obtain a second dynamic feature representation vector of the sample object for the sample date; and wherein, at least the eighth representation vector and the first dynamic feature representation vector of each of at least one first date are input into the second prediction subnetwork to obtain the target indicator prediction value of the sample object for the sample date output by the second prediction subnetwork includes: inputting at least the eighth representation vector, the first dynamic feature representation vector of each of at least one first date, and the second dynamic feature representation vector into the second prediction subnetwork to obtain the target indicator prediction value of the sample object for the sample date output by the second prediction subnetwork.

[0170] In some exemplary embodiments, the above model can be Figure 3 Model 300 shown.

[0171] In some exemplary embodiments, the loss function of the above model is It can be expressed as:

[0172]

[0173]

[0174] in, The difference between the predicted value and the actual value of the target indicator for at least one target date of at least one target object; is the loss of the tensor decomposition network; Θ is all the parameters of the model, λ T and λ Θ is the weighting coefficient.

[0175] According to some embodiments, Figure 7 As shown, a data processing device 700 is provided, including:

[0176] A first acquisition unit 710 is configured to acquire an object relationship graph, where the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects;

[0177] The second acquisition unit 720 is configured to acquire, for a target object among the multiple first objects, at least one meta-path corresponding to the target object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the target object and an associated object of the target object, and the target object and multiple second objects other than the target object in the meta-path respectively determine multiple associated paths;

[0178] The execution unit 730 is configured to execute the operations of the following sub-units for each meta-path in at least one meta-path, and the execution unit 730 includes:

[0179] A determination subunit 731, configured to determine a plurality of first attention weights of the target object relative to the plurality of associated paths based on the inherent attribute data of the target object and the inherent attribute data of each of the plurality of second objects on the meta-path; and

[0180] an acquisition subunit 732, configured to acquire, based on a first representation vector of the target object and a plurality of first attention weights, a second representation vector of the target object after integrating the association relationship information represented by the plurality of association paths on the meta-path, wherein the first representation vector is determined based on the inherent attribute data of the target object; and

[0181] The third acquisition unit 740 is configured to obtain a target indicator prediction result of the target object based on at least one second representation vector of the target object corresponding to at least one meta-path.

[0182] The operations of the above-mentioned units 710-740 and sub-units 731-732 of the data processing device 700 are similar to the operations of steps S201-S205 in the above-mentioned data processing method, and are not described in detail here.

[0183] According to some embodiments, Figure 8As shown, a model training device 800 is provided, the model includes a first feature extraction network and a prediction network, the first feature extraction network includes a first feature extraction sub-network, a first attention sub-network and a first feature fusion sub-network, and the model training device 800 includes:

[0184] A first acquisition unit 810 is configured to acquire an object relationship graph and sample data, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects, and the sample data includes inherent attribute data and target indicator data of the sample object, and the sample object is one of the plurality of first objects;

[0185] The second acquisition unit 820 is configured to acquire, for a sample object among the multiple first objects, at least one meta-path corresponding to the sample object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the sample object with an associated object of the sample object, and the sample object respectively determines multiple associated paths with multiple second objects other than the sample object in the meta-path;

[0186] The execution unit 830 is configured to execute the operations of the following sub-units for each meta-path in at least one meta-path, and the execution unit 830 includes:

[0187] A first acquisition subunit 831 is configured to input the intrinsic attribute data of the sample object and the intrinsic attribute data of each of the plurality of second objects on the meta-path into the first feature extraction network to obtain a first characterization vector of the sample object output by the first feature extraction network and a plurality of third characterization vectors corresponding to the plurality of second objects;

[0188] A determination subunit 832 is configured to determine, based on the first characterization vector and the plurality of third characterization vectors, a plurality of first attention weights of the sample object relative to the plurality of associated paths using the first attention subnetwork; and

[0189] The second acquisition subunit 833 is configured to use the first feature fusion subnetwork to acquire, based on the first representation vector of the sample object and the plurality of first attention weights, a second representation vector of the target object after fusion of the association relationship information represented by the plurality of association paths on the meta-path;

[0190] A third acquisition unit 840 is configured to input at least one second representation vector of the sample object corresponding to at least one meta-path into the prediction network to obtain a target indicator prediction result of the sample object output by the prediction network; and

[0191] The adjustment unit 850 is configured to adjust the parameters of the model based on the target indicator prediction result and the target indicator data.

[0192] Among them, the operations of the above-mentioned units 810~850 and sub-units 831~833 of the model training device 800 are respectively similar to the operations of steps S601~S607 in the above-mentioned model training method, and will not be repeated here.

[0193] According to an embodiment of the present disclosure, an electronic device, a readable storage medium and a computer program product are also provided.

[0194] refer to Fig. 9 , a block diagram of an electronic device 900 that can be used as a server or client of the present disclosure will now be described, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0195] like Fig. 9 As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the electronic device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0196] Multiple components in the electronic device 900 are connected to the I / O interface 905, including: an input unit 906, an output unit 907, a storage unit 908, and a communication unit 909. The input unit 906 can be any type of device that can input information to the electronic device 900. The input unit 906 can receive input digital or character information and generate key signal input related to user settings and / or function control of the electronic device, and can include but is not limited to a mouse, a keyboard, a touch screen, a track pad, a track ball, a joystick, a microphone, and / or a remote controller. The output unit 907 can be any type of device that can present information, and can include but is not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 908 can include but is not limited to a disk, an optical disk. The communication unit 909 allows the electronic device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include but is not limited to a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0197] The computing unit 901 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods and processes described above, such as the above-mentioned data processing method or model training method. For example, in some embodiments, the above-mentioned data processing method or model training method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the above-mentioned data processing method or model training method may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to execute the above-mentioned data processing method or model training method in any other appropriate manner (for example, by means of firmware).

[0198] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0199] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0200] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0201] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0202] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0203] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0204] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0205] Although the embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above-mentioned methods, systems and devices are merely exemplary embodiments or examples, and the scope of the present invention is not limited by these embodiments or examples, but only by the claims after authorization and their equivalent scope. Various elements in the embodiments or examples can be omitted or replaced by their equivalent elements. In addition, each step can be performed in an order different from that described in the present disclosure. Further, the various elements in the embodiments or examples can be combined in various ways. It is important that with the evolution of technology, many elements described herein can be replaced by equivalent elements that appear after the present disclosure.

Claims

1. A data processing method, comprising: Acquire an object relationship graph, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects; For a target object among the multiple first objects, obtaining at least one first meta-path corresponding to the target object in the object relationship graph, wherein each first meta-path among the at least one first meta-path is a node path in the object relationship graph for connecting the target object with an associated object of the target object, and the target object and multiple associated paths are respectively determined with multiple second objects other than the target object in the first meta-path; For each first meta-path in the at least one first meta-path, perform the following operations: Determining a plurality of first attention weights of the target object relative to the plurality of associated paths based on the intrinsic attribute data of the target object and the intrinsic attribute data of each of the plurality of second objects on the first meta-path; as well as Based on the first representation vector of the target object and the multiple first attention weights, obtaining a second representation vector of the target object after integrating the association relationship information represented by the multiple association paths on the first element path, wherein the first representation vector is determined based on the inherent attribute data of the target object; as well as A target indicator prediction result of the target object is obtained based on at least one second characterization vector of the target object corresponding to the at least one first element path.

2. The method according to claim 1, wherein: The multiple first objects include multiple management subjects and multiple managed objects, the target object and each of the multiple second objects is a managed object, the multiple association paths include multiple association types, and the multiple association types include at least two of the following: two managed objects are directly associated, two managed objects are indirectly associated based on at least one management subject, two managed objects are indirectly associated based on at least one managed object, and two managed objects are indirectly associated based on at least one management subject and at least one managed object, and, The determining, based on the intrinsic attribute data of the target object and the intrinsic attribute data of each of the plurality of second objects on the first meta-path, a plurality of first attention weights of the target object relative to the plurality of associated paths comprises: For each of the multiple association types, perform the following operations: Determine, among the plurality of association paths, at least one association path corresponding to the association type and at least one second object corresponding to the at least one association path; and Determining at least one second attention weight of the target object corresponding to the association type relative to the at least one association path based on the first representation vector of the target object and each corresponding third representation vector of the at least one second object, wherein the third representation vector is determined based on intrinsic attribute data of the corresponding second object; and The plurality of first attention weights are determined based on at least one second attention weight corresponding to each of the plurality of association types.

3. The method according to claim 2, wherein: The step of obtaining, based on the first representation vector of the target object and the plurality of first attention weights, a second representation vector of the target object after integrating the association relationship information of the plurality of association path representations on the first element path comprises: For each of the plurality of association types, based on the first representation vector of the target object and at least one second attention weight corresponding to the association type among the plurality of first attention weights, obtaining a fourth representation vector corresponding to the association type; and The plurality of fourth characterization vectors corresponding to the plurality of association types are concatenated to obtain the second characterization vector.

4. The method according to claim 2 or 3, wherein: The number of the at least one meta-path is multiple, and obtaining the target indicator prediction result of the target object based on at least one second representation vector of the target object corresponding to the at least one meta-path includes: For each meta-path in the at least one meta-path, determining a path characterization vector of the meta-path based on a second characterization vector corresponding to each managed object on the meta-path; Determining at least one third attention weight corresponding to the at least one meta-path based on each corresponding path representation vector in the at least one meta-path; Based on the at least one second representation vector and the at least one third attention weight, obtaining a fifth representation vector of the target object; and The target indicator prediction result is obtained based at least on the fifth characterization vector.

5. The method according to claim 4, wherein: Each of the multiple managed objects is composed of multiple sub-objects in proportion, and the method further includes: Acquire tensor data corresponding to the object relationship graph, where the tensor data is used to record proportion data of each sub-object corresponding to each managed object in the object relationship graph; and The tensor data is subjected to tensor decomposition to obtain a sixth characterization vector corresponding to each managed object in the object relationship graph, wherein the sixth characterization vector is used to characterize component features of the corresponding managed object; and wherein, The obtaining the target indicator prediction result based at least on the fifth characterization vector includes: performing feature fusion on the fifth representation vector and the sixth representation vector of the target object to obtain a seventh representation vector of the target object; and The target indicator prediction result is obtained based at least on the seventh characterization vector.

6. The method according to claim 5, wherein the target indicator prediction result comprises a target indicator prediction value of the target object on a target date, and the method further comprises: For each first date of at least one first date before the target date, perform the following operations: Based on the target dynamic attribute data of the first date, determining a preset number of third objects from the multiple managed objects, the difference between the dynamic attribute data of the third objects and the target dynamic attribute data being smaller than the difference between the dynamic attribute data of other managed objects other than the third object from the multiple managed objects and the target dynamic attribute data; as well as Performing feature fusion on the seventh characterization vector corresponding to each third object among the preset number of third objects to obtain a first dynamic feature characterization vector corresponding to the first date; And among them, The obtaining of the target indicator prediction result based at least on the seventh characterization vector includes: Based at least on the seventh characterization vector and the first dynamic feature characterization vector of each first date of the at least one first date, a target indicator prediction value of the target object on the target date is obtained.

7. The method according to claim 6, further comprising: Acquire a second dynamic feature characterization vector of the target object on the target date, wherein the second dynamic feature characterization vector is determined based on dynamic attribute data of the target object on each second date of at least one second date before the target date; and wherein, The obtaining the target indicator prediction value of the target object on the target date based on at least the seventh characterization vector and the first dynamic feature characterization vector of each first date of the at least one first date comprises: Based on the seventh characterization vector, the first dynamic feature characterization vector of each first date of the at least one first date, and the second dynamic feature characterization vector, a target indicator prediction value of the target object on the target date is obtained.

8. A model training method, wherein the model includes a first feature extraction network and a prediction network, wherein the first feature extraction network includes a first feature extraction subnetwork, a first attention subnetwork and a first feature fusion subnetwork, and the method includes: Acquire an object relationship graph and sample data, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects, and the sample data includes inherent attribute data and target indicator data of the sample object, and the sample object is one of the plurality of first objects; For a sample object among the plurality of first objects, obtaining at least one meta-path corresponding to the sample object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the sample object with an associated object of the sample object, and the sample object and a plurality of associated paths are respectively determined with a plurality of second objects other than the sample object in the meta-path; For each meta-path in the at least one meta-path, perform the following operations: Inputting the intrinsic attribute data of the sample object and the intrinsic attribute data of each of the plurality of second objects on the meta-path into the first feature extraction network to obtain a first characterization vector of the sample object output by the first feature extraction network and a plurality of third characterization vectors corresponding to the plurality of second objects; Determining, using the first attention subnetwork, a plurality of first attention weights of the sample object relative to the plurality of associated paths based on the first representation vector and the plurality of third representation vectors; as well as Based on the first representation vector of the sample object and the multiple first attention weights, using the first feature fusion subnetwork to obtain a second representation vector of the target object after fusing the association relationship information represented by the multiple association paths on the meta-path; Inputting at least one second representation vector of the sample object corresponding to the at least one meta-path into the prediction network to obtain a target indicator prediction result of the sample object output by the prediction network; as well as Based on the target indicator prediction result and the target indicator data, the parameters of the model are adjusted.

9. The method according to claim 8, wherein: The multiple first objects include multiple management subjects and multiple managed objects, each of the sample object and the multiple second objects is a managed object, the multiple association paths include multiple association types, and the multiple association types include at least two of the following: two managed objects are directly associated, two managed objects are indirectly associated based on at least one management subject, two managed objects are indirectly associated based on at least one managed object, and two managed objects are indirectly associated based on at least one management subject and at least one managed object, and, The determining, based on the first representation vector and the plurality of third representation vectors, using the first attention sub-network to determine a plurality of first attention weights of the sample object relative to the plurality of associated paths comprises: For each of the multiple association types, perform the following operations: Determine, among the plurality of association paths, at least one association path corresponding to the association type and at least one second object corresponding to the at least one association path; and Determining at least one second attention weight of the sample object corresponding to the association type relative to the at least one association path based on the first representation vector of the sample object and each corresponding third representation vector of the at least one second object; and The plurality of first attention weights are determined based on at least one second attention weight corresponding to each of the plurality of association types.

10. The method according to claim 9, wherein: The method of using the first feature fusion subnetwork to obtain the second representation vector of the target object after fusing the association relationship information of the multiple association path representations on the meta-path based on the first representation vector of the sample object and the multiple first attention weights includes: For each of the plurality of association types, based on the first representation vector of the target object and at least one second attention weight corresponding to the association type among the plurality of first attention weights, obtaining a fourth representation vector corresponding to the association type; and The plurality of fourth characterization vectors corresponding to the plurality of association types are concatenated to obtain the second characterization vector.

11. The method according to claim 9 or 10, wherein: The number of the at least one meta-path is multiple, the prediction network includes a second feature fusion sub-network, a second attention sub-network, a third feature fusion sub-network and a first prediction sub-network, and the at least one second representation vector of the sample object corresponding to the at least one meta-path is input into the prediction network to obtain the target indicator prediction result of the sample object output by the prediction network, including: For each meta-path in the at least one meta-path, inputting a second representation vector corresponding to each managed object on the meta-path into the second feature fusion sub-network to obtain a path representation vector of the meta-path output by the second feature fusion sub-network; Determine, using the second attention sub-network, at least one third attention weight corresponding to the at least one meta-path based on each corresponding path representation vector in the at least one meta-path; Based on the at least one second representation vector and the at least one third attention weight, using the third feature fusion subnetwork to obtain a fifth representation vector of the target object; and At least the fifth representation vector is input into the first prediction sub-network to obtain the target indicator prediction result output by the first prediction sub-network.

12. The method according to claim 11, wherein: Each of the multiple managed objects is composed of multiple sub-objects in proportion, the model further includes a tensor decomposition network, the first prediction sub-network includes a fourth feature fusion sub-network and a second prediction sub-network, and the method further includes: Acquire tensor data corresponding to the object relationship graph, where the tensor data is used to record proportion data of each sub-object corresponding to each managed object in the object relationship graph; and The tensor data is input into the tensor decomposition network to obtain a sixth characterization vector corresponding to each managed object in the object relationship graph output by the tensor decomposition network, wherein the sixth characterization vector is used to characterize the component characteristics of the corresponding managed object; and wherein, The step of at least inputting the fifth representation vector into the first prediction sub-network to obtain the target indicator prediction result output by the first prediction sub-network includes: Inputting the fifth representation vector and the sixth representation vector of the sample object into the fourth feature fusion subnetwork to obtain a seventh representation vector of the sample object output by the fourth feature fusion subnetwork; and At least the seventh representation vector is input into the second prediction sub-network to obtain the target indicator prediction result output by the second prediction sub-network.

13. The method according to claim 12, wherein the target indicator data comprises the target indicator value of the sample object on the sample date, the target indicator prediction result comprises the target indicator prediction value of the sample object on the sample date, the model further comprises a feature fusion network, and the method further comprises: For each first date of at least one first date before the sample date, perform the following operations: Based on the target dynamic attribute data of the first date, determining a preset number of third objects from the multiple managed objects, the difference between the dynamic attribute data of the third objects and the target dynamic attribute data being smaller than the difference between the dynamic attribute data of other managed objects other than the third object from the multiple managed objects and the target dynamic attribute data; as well as The seventh characterization vector corresponding to each third object in the preset number of third objects is input into the feature fusion network to obtain the first dynamic feature characterization vector corresponding to the first date output by the feature fusion network; and wherein, The step of at least inputting the seventh representation vector into the second prediction sub-network to obtain the target indicator prediction result output by the second prediction sub-network includes: At least the seventh characterization vector and the first dynamic feature characterization vector of each first date in the at least one first date are input into the second prediction subnetwork to obtain the target indicator prediction value of the sample object on the sample date output by the second prediction subnetwork.

14. The method according to claim 13, wherein the sample data further comprises dynamic attribute data of the sample object on each of at least one second date before the sample date, the model further comprises a second feature extraction network, and the method further comprises: The dynamic attribute data of the sample object of each second date in the at least one second date is input into the second feature extraction network to obtain a second dynamic feature representation vector of the sample object of the sample date; and wherein, The step of inputting at least the seventh characterization vector and the first dynamic feature characterization vector of each first date in the at least one first date into the second prediction subnetwork to obtain the target indicator prediction value of the sample object on the sample date output by the second prediction subnetwork comprises: At least the seventh characterization vector, the first dynamic feature characterization vector of each first date in the at least one first date, and the second dynamic feature characterization vector are input into the second prediction sub-network to obtain the target indicator prediction value of the sample object on the sample date output by the second prediction sub-network.

15. A data processing device, comprising: A first acquisition unit is configured to acquire an object relationship graph, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects; A second acquisition unit is configured to acquire, for a target object among the plurality of first objects, at least one meta-path corresponding to the target object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the target object with an associated object of the target object, and the target object determines a plurality of associated paths with a plurality of second objects other than the target object in the meta-path, respectively; An execution unit is configured to execute the operations of the following sub-units for each meta-path in the at least one meta-path, the execution unit comprising: a determining subunit configured to determine a plurality of first attention weights of the target object relative to the plurality of associated paths based on the intrinsic attribute data of the target object and the intrinsic attribute data of each of the plurality of second objects on the meta-path; and an acquisition subunit, configured to acquire, based on the first representation vector of the target object and the plurality of first attention weights, a second representation vector of the target object after integrating the association relationship information represented by the plurality of association paths on the meta-path, wherein the first representation vector is determined based on the inherent attribute data of the target object; and The third acquisition unit is configured to obtain a target indicator prediction result of the target object based on at least one second representation vector of the target object corresponding to the at least one meta-path.

16. A model training device, the model comprising a first feature extraction network and a prediction network, the first feature extraction network comprising a first feature extraction subnetwork, a first attention subnetwork and a first feature fusion subnetwork, the device comprising: A first acquisition unit is configured to acquire an object relationship graph and sample data, wherein the object relationship graph includes a plurality of first nodes corresponding to a plurality of first objects and is used to represent association relationship information between the plurality of first objects, and the sample data includes inherent attribute data and target indicator data of the sample object, and the sample object is one of the plurality of first objects; a second acquisition unit configured to acquire, for a sample object among the plurality of first objects, at least one meta-path corresponding to the sample object in the object relationship graph, wherein each meta-path in the at least one meta-path is a node path in the object relationship graph for connecting the sample object with an associated object of the sample object, and the sample object respectively determines a plurality of associated paths with a plurality of second objects other than the sample object in the meta-path; An execution unit is configured to execute the operations of the following sub-units for each meta-path in the at least one meta-path, the execution unit comprising: A first acquisition subunit is configured to input the intrinsic attribute data of the sample object and the intrinsic attribute data of each of the plurality of second objects on the meta-path into the first feature extraction network to obtain a first representation vector of the sample object output by the first feature extraction network and a plurality of third representation vectors corresponding to the plurality of second objects; a determining subunit configured to determine, based on the first representation vector and the plurality of third representation vectors, a plurality of first attention weights of the sample object relative to the plurality of associated paths using the first attention subnetwork; and A second acquisition subunit is configured to use the first feature fusion subnetwork to acquire, based on the first representation vector of the sample object and the plurality of first attention weights, a second representation vector of the target object after fusion of the association relationship information represented by the plurality of association paths on the meta-path; A third acquisition unit is configured to input at least one second representation vector of the sample object corresponding to the at least one meta-path into the prediction network to obtain a target indicator prediction result of the sample object output by the prediction network; and An adjustment unit is configured to adjust the parameters of the model based on the target indicator prediction result and the target indicator data.

17. An electronic device comprising: at least one processor; as well as a memory communicatively coupled to the at least one processor; in The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 14.

18. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-14.

19. A computer program product comprising a computer program, wherein: The computer program implements the method according to any one of claims 1 to 14 when executed by a processor.

Images