Object recommendation model training method, recommended object determination method and device

Abstract

Embodiments of the present specification provide an object recommendation model training method, a recommended object determination method and device. In each round of model training, a user feature embedding representation of a user feature, an object feature embedding representation of a target object and a user interaction object are determined. A user interest embedding representation and a user conformity embedding representation are decoupled from the object feature embedding representation of the user interaction object; a target object representation, a user interest representation and a user conformity representation are determined according to the coupling result between the object feature embedding representation of the target object and the user feature embedding representation, and the user interest embedding representation and the user conformity embedding representation, respectively; and model parameters of an object recommendation model are adjusted according to a user interest prediction result and a user conformity prediction result determined based on the user interest representation, the user conformity representation and the target object representation.

Description

Technical Field

[0001] The embodiments in this specification generally relate to the field of object recommendation, and in particular to object recommendation model training methods, recommendation object determination methods and apparatus. Background Technology

[0002] With the development of artificial intelligence technology, object recommendation schemes based on object recommendation models are increasingly being applied to object recommendation scenarios. Object recommendation models are a type of machine learning model. They can typically be modeled based on sequential behavior or decoupling learning. Common sequential behavior-based modeling schemes usually only learn embedding representations driven solely by user interests based on historical behavior, without distinguishing potential behavioral intentions. Common decoupling learning-based modeling schemes struggle to explicitly include user intentions such as herd mentality and risk preferences in object recommendation. Furthermore, the lack of labeled data to explicitly constrain these user intentions makes it difficult to accurately represent them in the model. These shortcomings of sequential behavior-based and decoupling learning-based modeling schemes make them unsuitable for certain object recommendation scenarios, such as intelligent product recommendation and intelligent fund recommendation. Summary of the Invention

[0003] This specification provides a method for training an object recommendation model, a method for determining recommended objects, and an apparatus. Using this object recommendation model training scheme, object recommendation models can be built based on user interests, user herd mentality, and user risk preferences, thereby improving the user understanding of the object recommendation model.

[0004] According to one aspect of an embodiment of this specification, a method for training an object recommendation model is provided, the object recommendation model being used to determine whether to recommend a target object to a user, the method comprising: cyclically executing the following model training process until a training termination condition is met: providing user features, object features of user interaction objects, and object features of target objects to an embedding layer of the object recommendation model, respectively, to obtain user feature embedding representations of the user features, object feature embedding representations of the user interaction objects, and object feature embedding representations of the target objects; and decoupling user interest embedding representations corresponding to user interests and user conformity psychology from the object feature embedding representations of the user interaction objects via a decoupling layer of the object recommendation model. The user conformity embedding representation is fused with the user interest embedding representation and the user conformity embedding representation respectively, and the fused result and the object feature embedding representation of the target object are respectively provided to the representation layer of the object recommendation model to obtain user interest representation, user conformity representation and target object representation; user interest prediction result is predicted based on the user interest representation and the target object representation, and user conformity prediction result is predicted based on the user conformity representation and the target object representation; and in response to the failure to meet the training termination condition, a total loss function is determined based on the user interest prediction result and the user conformity prediction result, and the model parameters of the object recommendation model are adjusted according to the total loss function.

[0005] Optionally, in one example of the above aspects, the method may further include: providing the object feature embedding representations of the target object and the user interaction object to the graph neural network of the object recommendation model to perform graph augmentation processing based on the object knowledge graph, so as to obtain the graph augmented object feature embedding representations of the target object and the user interaction object.

[0006] Optionally, in one example of the above aspects, the user interaction object includes at least two user interaction objects. Decoupling user interest embedding representations and user conformity embedding representations corresponding to user interests and user herd mentality respectively from the object feature embedding representations of the user interaction objects via the decoupling layer of the object recommendation model may include: decoupling object feature embedding representation components corresponding to user interests and user herd mentality respectively from the object feature embedding representations of each user interaction object via the decoupling layer of the object recommendation model, and concatenating them to obtain user interest embedding representations and user conformity embedding representations.

[0007] Optionally, in one example of the above aspects, decoupling user interest embedding representations and user conformity embedding representations corresponding to user interests and user herd mentality respectively from the object feature embedding representations of the user interaction object via the decoupling layer of the object recommendation model may include: decoupling user interest embedding representations and user conformity embedding representations corresponding to user interests and user herd mentality respectively from the object feature embedding representations of the user interaction object via the decoupling layer of the object recommendation model based on an unsupervised mechanism.

[0008] Optionally, in one example of the above aspects, the decoupling layer includes a self-attention network. Decoupling user interest embeddings and user conformity embeddings corresponding to user interests and user herd mentality, respectively, from the object feature embeddings of the user-interacting object via the decoupling layer of the object recommendation model may include: providing the object feature embeddings of the user-interacting object to the self-attention network to decouple the user interest embeddings and user conformity embeddings corresponding to user interests and user herd mentality, respectively, from the object feature embeddings of the user-interacting object.

[0009] Optionally, in one example of the above aspects, determining the total loss function based on the user interest prediction result and the user conformity prediction result may include: determining a user interest loss term based on the user interest prediction result; determining a user conformity loss term based on the user conformity prediction result; and determining the total loss function based on the user interest loss term and the user conformity loss term.

[0010] Optionally, in one example of the above aspects, the target object has an object popularity determined based on the number of interactions in the object interaction history, the user interest loss term is determined based on the user interest prediction result and the object popularity of the target object, and the user conformity loss term is determined based on the user conformity prediction result and the object popularity of the target object.

[0011] Optionally, in one example of the above aspects, providing user features, object features of user interaction objects, and object features of target objects to the embedding layer of the object recommendation model to obtain user feature embedding representations of user features, object feature embedding representations of user interaction objects, and object feature embedding representations of target objects may include: providing user features, object features of user interaction objects, object features of target objects, and object types of user interaction objects to the embedding layer of the object recommendation model to obtain user feature embedding representations of user features, object feature embedding representations of user interaction objects, object feature embedding representations of target objects, and object type embedding representations of user interaction objects. Decoupling user interest embedding representations and user conformity embedding representations corresponding to user interests and user herd mentality, respectively, from the object feature embedding representations of user interaction objects via the decoupling layer of the object recommendation model may include: decoupling user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations corresponding to user interests, user herd mentality, and user risk preference, respectively, from the object feature embedding representations of user interaction objects via the decoupling layer of the object recommendation model. Providing the obtained fusion result and the object feature embedding representation of the target object to the representation layer of the object recommendation model to obtain user interest representation, user conformity representation, and target object representation may include: providing the obtained fusion result, the object feature embedding representation of the target object, and the object type embedding representation of the user interaction object to the representation layer of the object recommendation model to obtain user interest representation, user conformity representation, target object representation, and user risk preference representation. Determining the total loss function based on the user interest prediction result and the user conformity prediction result may include: determining the total loss function based on the user interest prediction result, the user conformity prediction result, the risk preference embedding representation, and the user risk preference representation.

[0012] Optionally, in one example of the above aspects, determining the total loss function based on the user interest prediction result, the user conformity prediction result, the risk preference embedding representation, and the user risk preference representation may include: determining a user interest loss term based on the user interest prediction result; determining a user conformity loss term based on the user conformity prediction result; determining a user risk preference loss term based on the risk preference embedding representation and the user risk preference representation; and determining the total loss function based on the user interest loss term, the user conformity loss term, and the user risk preference loss term.

[0013] Optionally, in one example of the above aspects, the user risk preference loss term has a weighted value, which is used to define the contribution of the user risk preference loss term to the total loss function. Determining the total loss function based on the user interest loss term, the user conformity loss term, and the user risk preference loss term may include: determining the total loss function based on the user interest loss term, the user conformity loss term, and the weighted user risk preference loss term.

[0014] Alternatively, in one example of the above aspects, the object knowledge graph is created based on the relationship between recommended type objects and associated type objects.

[0015] According to another aspect of the embodiments of this specification, a method for determining a recommended object based on an object recommendation model is provided, comprising: providing user features, object features of a user interaction object, and object features of a target object to an embedding layer of the object recommendation model, respectively, to obtain a user feature embedding representation of the user features, an object feature embedding representation of the user interaction object, and an object feature embedding representation of the target object; decoupling user interest embedding representation and user conformity embedding representation corresponding to user interests and user herd mentality, respectively, from the object feature embedding representation of the user interaction object via a decoupling layer of the object recommendation model; fusing the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation, respectively, and providing the obtained fusion result and the object feature embedding representation of the target object to a representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation; and determining whether to recommend the target object to a user based on the user interest representation, the user conformity representation, and the target object representation.

[0016] Optionally, in one example of the above aspects, determining whether to recommend the target object to the user based on the user interest representation, the user conformity representation, and the target object representation may include: predicting user interest prediction results and user conformity prediction results based on the user interest representation and the target object representation, and based on the user conformity representation and the target object representation, respectively; and determining whether to recommend the target object to the user based on the user interest prediction results and the user conformity prediction results.

[0017] Optionally, in one example of the above aspects, the method may further include: providing the object feature embedding representations of the target object and the user interaction object to the graph neural network of the object recommendation model to perform graph augmentation processing based on the object knowledge graph, so as to obtain the graph augmented object feature embedding representations of the target object and the user interaction object.

[0018] Alternatively, in one example of the above aspects, the object recommendation model is trained according to the method described above.

[0019] According to another aspect of the embodiments of this specification, an apparatus for training an object recommendation model is provided, the object recommendation model being used to determine whether to recommend a target object to a user, the apparatus comprising: a first representation generation unit, which provides user features, object features of user interaction objects, and object features of target objects to an embedding layer of the object recommendation model, respectively, to obtain user feature embedding representations of the user features, object feature embedding representations of the user interaction objects, and object feature embedding representations of the target objects; an embedding representation decoupling unit, which decouples user interest embedding representations and user conformity embedding representations corresponding to user interests and user conformity psychology, respectively, from the object feature embedding representations of the user interaction objects via the decoupling layer of the object recommendation model; and a second representation generation unit, which fuses the user feature embedding representations with the user interest embedding representations and user conformity embedding representations, respectively, and combines the obtained fused representations with the user interest embedding representations. The combined result and the object feature embedding representation of the target object are respectively provided to the representation layer of the object recommendation model to obtain user interest representation, user conformity representation, and target object representation; the target object prediction unit predicts user interest prediction result based on the user interest representation and the target object representation, and predicts user conformity prediction result based on the user conformity representation and the target object representation; the loss function determination unit determines the total loss function based on the user interest prediction result and the user conformity prediction result in response to the failure to meet the training termination condition; and the model parameter adjustment unit adjusts the model parameters of the object recommendation model according to the total loss function, wherein the first representation generation unit, the embedding representation decoupling unit, the second representation generation unit, the target object prediction unit, the loss function determination unit, and the model parameter adjustment unit perform operations in a loop until the training termination condition is met.

[0020] According to another aspect of the embodiments of this specification, an apparatus for determining a recommended object based on an object recommendation model is provided, comprising: a first representation generation unit, which provides user features, object features of a user interaction object, and object features of a target object to an embedding layer of the object recommendation model, respectively, to obtain a user feature embedding representation of the user features, an object feature embedding representation of the user interaction object, and an object feature embedding representation of the target object; an embedding representation decoupling unit, which decouples user interest embedding representation and user conformity embedding representation, respectively, from the object feature embedding representation of the user interaction object via the decoupling layer of the object recommendation model; a second representation generation unit, which fuses the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation, respectively, and provides the fusion result and the object feature embedding representation of the target object to the representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation; and a recommendation decision unit, which determines whether to recommend the target object to a user based on the user interest representation, the user conformity representation, and the target object representation.

[0021] Optionally, in one example of the above aspects, the apparatus may further include: a graph augmentation processing unit, which provides the object feature embedding representations of the target object and the user interaction object to the graph neural network of the object recommendation model to perform graph augmentation processing based on the object knowledge graph, so as to obtain the graph-augmented object feature embedding representations of the target object and the user interaction object.

[0022] According to another aspect of the embodiments of this specification, an object recommendation system is provided, comprising: an interactive object acquisition device for acquiring user interactive objects; an device for training an object recommendation model as described above; and an device for determining recommended objects based on the object recommendation model as described above.

[0023] According to another aspect of the embodiments of this specification, an apparatus for training an object recommendation model is provided, comprising: at least one processor, a memory coupled to the at least one processor, and a computer program stored in the memory, wherein the at least one processor executes the computer program to implement the method for training an object recommendation model as described above.

[0024] According to another aspect of the embodiments of this specification, an apparatus for determining recommended objects based on an object recommendation model is provided, comprising: at least one processor, a memory coupled to the at least one processor, and a computer program stored in the memory, wherein the at least one processor executes the computer program to implement the method for determining recommended objects based on an object recommendation model as described above. Attached Figure Description

[0025] A further understanding of the nature and advantages of this specification can be achieved by referring to the following figures. In the figures, similar components or features may have the same reference numerals.

[0026] Figure 1 A block diagram of an object recommendation system based on an object recommendation model according to an embodiment of this specification is shown.

[0027] Figure 2 An example schematic diagram of an object recommendation model according to an embodiment of this specification is shown.

[0028] Figure 3 An example flowchart of a method for training an object recommendation model according to an embodiment of this specification is shown.

[0029] Figure 4 An example flowchart of the total loss function determination process according to an embodiment of this specification is shown.

[0030] Figure 5 An example schematic diagram of the training process of an object recommendation model according to an embodiment of this specification is shown.

[0031] Figure 6 An example flowchart of a method for determining recommended objects based on an object recommendation model, according to an embodiment of this specification, is shown.

[0032] Figure 7 An example schematic diagram of an object recommendation process according to an embodiment of this specification is shown.

[0033] Figure 8 An example block diagram of an object recommendation model training apparatus according to an embodiment of this specification is shown.

[0034] Figure 9 An example block diagram of a loss function determination unit according to an embodiment of this specification is shown.

[0035] Figure 10 An example block diagram of an object recommendation device according to an embodiment of this specification is shown.

[0036] Figure 11 An example schematic diagram of a computer system-based object recommendation model training apparatus according to an embodiment of this specification is shown.

[0037] Figure 12 An example schematic diagram of an object recommendation apparatus based on a computer system according to an embodiment of this specification is shown. Detailed Implementation

[0038] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the various examples. For example, the described methods may be performed in a different order than described, and steps may be added, omitted, or combined. Furthermore, features described in some examples may be combined in other examples.

[0039] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0040] When recommending objects to users, the recommended objects are typically determined and recommended based on an object recommendation model. Object recommendation models can generally be modeled based on sequential behavior or decoupled learning. However, modeling schemes based on sequential behavior usually only learn embedded representations driven by a single user interest based on historical user behavior, without distinguishing between potential behavioral intentions. Modeling schemes based on decoupled learning also struggle to explicitly include user intentions such as herd mentality or risk preferences in object recommendations.

[0041] In some object recommendation applications, such as intelligent product recommendation and intelligent fund recommendation, user selection is influenced not only by their interest in the chosen object, but also by whether the object is chosen by many other users and whether its risk matches the user's risk preferences. Therefore, when modeling object recommendation, it's necessary to consider factors beyond user interest, such as herd mentality and risk preferences. Consequently, the aforementioned modeling schemes based on sequential behavior and decoupling learning are unsuitable for object recommendation scenarios that require consideration of herd mentality and risk preferences.

[0042] Therefore, embodiments of this specification provide a training scheme for an object recommendation model and an object recommendation scheme based on the object recommendation model. Using this object recommendation model training scheme, user features, target objects, and object features of user interaction objects are provided to the embedding layer of the object recommendation model to obtain user feature embedding representations of user features, and object feature embedding representations of target objects and user interaction objects. Then, through the decoupling layer of the object recommendation model, user interest embedding representations and user conformity embedding representations corresponding to user interests and user herd mentality are decoupled from the object feature embedding representations of user interaction objects, respectively. Subsequently, the decoupled user interest embedding representations and user conformity embedding representations are introduced into the object recommendation model training. This allows for object recommendation modeling based on user interests and user herd mentality, thereby improving the user understanding effect of the object recommendation model and making the trained object recommendation model suitable for the aforementioned object recommendation scenarios that require consideration of user interests and user herd mentality.

[0043] The following describes, with reference to the accompanying drawings, an object recommendation model training method and apparatus, a recommendation object determination method and apparatus, and an object recommendation system according to embodiments of this specification.

[0044] Figure 1 A block diagram of an object recommendation system 100 based on an object recommendation model according to an embodiment of this specification is shown.

[0045] like Figure 1 As shown, the object recommendation system 100 includes an interactive object acquisition device 110, an object recommendation model training device 120, an object recommendation model storage device 130, and an object recommendation device 140. The interactive object acquisition device 110, the object recommendation model training device 120, the object recommendation model storage device 130, and the object recommendation device 140 can communicate with each other via a network 150. In some embodiments, the network 150 can be any one or more of a wired network or a wireless network. Examples of the network 150 may include, but are not limited to, cable networks, fiber optic networks, telecommunications networks, enterprise intranets, the Internet, local area networks (LANs), wide area networks (WANs), wireless local area networks (WLANs), metropolitan area networks (MANs), public switched telephone networks (PSTNs), Bluetooth networks, ZigBee networks, near field communication (NFC), device internal buses, device internal lines, etc., or any combination thereof. In some embodiments, some or all of the devices in the interactive object acquisition device 110, the object recommendation model training device 120, the object recommendation model storage device 130, and the object recommendation device 140 can communicate directly without the need for the network 150.

[0046] The interaction object acquisition device 110 is used to acquire user interaction objects. A user interaction object refers to an object with which the user has historical interaction behavior, that is, an object that the user has previously interacted with. Examples of the term "interaction behavior" may include, but are not limited to, user clicks on objects, user views objects, user browses objects, user purchases objects, user inquires about objects, user operates on objects, etc. The term "object" may refer to an object suitable for recommendation to the user. Examples of objects may include, but are not limited to, goods, investment products, video materials such as fitness programs / learning programs, etc. Each object may have an object type and object characteristics. The object type is used to identify the type attribute of the object. For example, when the object is a good, the object type may include ordinary goods, high-end goods, luxury goods, fragile goods, high-renewal goods, etc. When the object is an investment product, the object type may include equity funds, bond funds, futures funds, equity / bond hybrid funds, etc. The object type may have object risk information used to identify the risk of the object. For example, equity funds, bond funds, futures funds, and equity / bond hybrid funds may identify different investment product risks. Object characteristics refer to features that reflect the properties of an object, distinguishing it from other objects. When the object is a fund product, object characteristics may include the fund company, fund manager, etc.

[0047] In some embodiments, the interaction object acquisition device 110 can collect user interaction objects by monitoring the user's historical interaction behavior. In some embodiments, user interaction objects can be collected via a specific object collection device. The interaction object acquisition device 110 can acquire user interaction objects from the specific object collection device via wireless or wired communication.

[0048] The user interaction objects acquired by the interaction object acquisition device 110 can be provided to the object recommendation model training device 120. The object recommendation model training device 120 can use the user interaction objects and target objects to train the object recommendation model. During object recommendation model training, the target object can be one or more pre-selected object samples used for model training, which have behavioral labels for a specific user (e.g., user u). For example, if there is an interaction between user u and the target object, the label value of the target object is 1. If there is no interaction between user u and the target object, the label value of the target object is 0. The object recommendation model training process of the object recommendation model training device 120 will be described in detail below with reference to the accompanying drawings.

[0049] The object recommendation device 140 can determine whether to recommend a target object based on the trained object recommendation model. In response to determining a target object, the object recommendation device 140 can recommend the target object to the user. In some embodiments, the object recommendation model trained by the object recommendation model training device 120 can be stored in the object recommendation model storage device 130. In this case, when the object recommendation device 140 performs object recommendation, it can retrieve the object recommendation model from the object recommendation model storage device 130 for object recommendation, or communicate with the object recommendation model storage device 130 to provide the necessary information to the object recommendation model in the object recommendation model storage device 130 for object recommendation. In some embodiments, the object recommendation model trained by the object recommendation model training device 120 can be stored in the object recommendation device 140.

[0050] Figure 2 An example schematic diagram of an object recommendation model 200 according to an embodiment of this specification is shown. Figure 2 As shown, the object recommendation model 200 may include an embedding layer 210, a graph neural network layer 220, a decoupling layer 230, a cascaded layer 240, a fusion layer 250, a representation layer 260, and a prediction layer 270.

[0051] Embedding layer 210 is used to map the input user features, target object object features, user interaction object object features, and user interaction object object types to corresponding embedded representations in the representation space. Here, a user interaction object may include one or more user interaction objects. For example, embedding layer 210 can map the input user features to user feature embedded representations in the representation space. Embedding layer 210 can map the input target object object features to target object object feature embedded representations in the representation space. Embedding layer 210 can map the input user interaction object object features to user interaction object object feature embedded representations in the representation space. Embedding layer 210 can map the input user interaction object object types to user interaction object object type embedded representations in the representation space. In some embodiments, embedding layer 210 may have multiple embedding layers, such as embedding layer 1 to embedding layer N. Each embedding layer is responsible for the embedding representation mapping processing of one feature or piece of information. For example, an embedding layer can be assigned to each user feature and each user interaction object object type for embedding representation mapping processing.

[0052] Graph neural network layer 220 is used to perform graph augmentation processing on the object feature embedding representations of the target object and the user-interacting object using an object knowledge graph, thereby obtaining graph-augmented object feature embedding representations of the target object and the user-interacting object. Graph neural network layer 220 may include one or more graph neural networks. For example, a graph neural network may be assigned to each object, which is used to perform graph augmentation processing on the object feature embedding representation of the object based on the object knowledge graph, thereby obtaining graph-augmented object feature embedding representations of the object. The graph neural network can be implemented using any graph neural network in the art. Examples of graph neural networks may include, but are not limited to, graph convolutional networks (GCNs), graph attention networks, graph autoencoders, graph generative networks, and graph spatial-temporal networks.

[0053] In some embodiments, an object knowledge graph can be created based on the relationship between recommended type objects and associated type objects. A recommended type object is an object of the same type as the target object. An associated type object is an object that is associated with a recommended type object. For example, when the object is a fund product, associated type objects may include fund managers, fund companies, the industry to which the fund belongs, top holdings, and market indices. When creating the knowledge graph, various nodes and edges in the fund knowledge graph can be created based on five relationships: fund product -> fund manager, fund product -> fund company, fund product -> fund industry, fund product -> top holdings, and fund product -> market indices, thereby constructing the relationships between all fund products. Using graph neural networks to perform graph convolution operations based on the object knowledge graph allows the introduction of similar objects to a given object to enhance its graph representation.

[0054] The object feature embedding representation of the user interaction object, after graph augmentation processing by the graph neural network layer 220, is provided to the decoupling layer 230. The decoupling layer 230 decouples from the object feature embedding representation of the user interaction object the user interest embedding representation, the user conformity embedding representation, and the user risk preference embedding representation, respectively. In some embodiments, the decoupling layer 230 can be implemented as a self-attention network.

[0055] Optionally, when the user interaction objects include multiple user interaction objects, the object recommendation model 200 may further include a concact layer 240. The feature embedding representation components corresponding to user interests, user conformity, and user risk preferences decoupled from the decoupling layer 230 are respectively provided to the concact layer 240 for concatenation, thereby obtaining user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations. It should be noted that in some embodiments, the concact layer 240 may be integrated into the decoupling layer 230. It should also be noted that the term "concact" can be used interchangeably with "concatenation".

[0056] The user interest embedding representation and user conformity embedding representation obtained by cascading through decoupling layer 230 / cascaded layer 240, as well as the user feature embedding representation output by embedding layer 210, are provided to fusion layer 250. Fusion layer 250 fuses the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation, respectively, and provides the resulting fusion results to representation layer 260. Here, the purpose of fusion layer 250 is to incorporate user features into the user interest embedding representation and the user conformity embedding representation, thereby ensuring that the fusion result reflects the user interest embedding representation and the user conformity embedding representation incorporating user features. In some embodiments, fusion layer 250 can be implemented as a cascaded layer or other applicable fusion techniques.

[0057] The representation layer 260 is used to obtain a user interest representation based on the fusion result of the user feature embedding representation and the user interest embedding representation, and to obtain a user conformity representation based on the fusion result of the user feature embedding representation and the user conformity embedding representation. Furthermore, the representation layer 260 also receives the object type embedding representation of the user interaction object provided by the embedding layer 210, and obtains a user risk preference representation based on the object type embedding representation of the user interaction object. Additionally, the representation layer 260 also receives the object feature embedding representation of the target object provided by the graph neural network layer 220, and obtains a target object representation based on the object feature embedding representation of the target object. In some embodiments, the representation layer 260 can be implemented as, for example, a feedforward network (FFN) or a feedforward neural network (FFNN).

[0058] The user interest representation, user conformity representation, and target object representation obtained by the representation layer 260 are provided to the prediction layer 270. The prediction layer 270 predicts the user interest prediction result based on the user interest representation and the target object representation, and predicts the user conformity prediction result based on the user conformity representation and the target object representation. Based on the user interest prediction result and the user conformity prediction result, the prediction layer 270 predicts whether the target object is a recommended object.

[0059] It should be noted that the above examples are merely exemplary embodiments of this disclosure, and not all features included are essential to this disclosure. For example, although a graph neural network layer 220 is included in the above example, this feature is merely exemplary and not required. Target objects and user interaction objects may not be represented in graph form, and therefore the object recommendation model may not include the graph neural network layer 220. Similarly, although the object type of the user interaction object and the user risk preference representation are included in the above example, these features are also merely exemplary, used to better extract user conformity embedding representations, and are not required. When modeling the object recommendation model, only the user interest embedding representation and the user conformity embedding representation can be used, without the user risk preference representation; correspondingly, the object type of the user interaction object may also be omitted.

[0060] Figure 3 An embodiment for training according to this specification is shown. Figure 2 The example flowchart of method 300 of object recommendation model 200 is shown. Figure 3 The object recommendation model training process shown is a cyclical process.

[0061] like Figure 3 As shown, in each iteration, at step 310, user features, target object features, user interaction object features, and user interaction object type are provided to the embedding layer of the object recommendation model, resulting in user feature embedding representations, target object feature embedding representations, user interaction object feature embedding representations, and user interaction object type embedding representations. In this specification, user interaction objects and target objects belong to the same object type, and the object type contains object risk information. For example, when the object is a fund product, the user interaction object is the user interaction fund product, and the target object is the target fund product. The object type is the fund product type, such as equity fund, bond fund, futures fund, etc. This fund product type reflects the risk level of the fund product, such as low risk, medium risk, high risk, etc.

[0062] In step 320, the object feature embedding representations of the target object and the user interaction object are respectively provided to the graph neural network of the object recommendation model to perform graph augmentation processing based on the object knowledge graph, so as to obtain the object feature embedding representations of the target object and the user interaction object after graph augmentation processing.

[0063] In 330, through the decoupling layer of the object recommendation model, user interest embedding representation, user conformity embedding representation, and user risk preference embedding representation, respectively, are decoupled from the graph-enhanced object feature embedding representation of the user interaction object.

[0064] In some embodiments, the decoupling layer of the object recommendation model can be used to decouple user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations from the graph-enhanced object feature embedding representations of user interaction objects based on an unsupervised mechanism.

[0065] In some embodiments, user interaction objects may include multiple user interaction objects. In this case, feature embedding representation components corresponding to user interests, user conformity, and user risk preferences can be decoupled from the graph-enhanced object feature embedding representations of each user interaction object via the decoupling layer of the object recommendation model, and then concatenated to obtain user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations.

[0066] For example, in some embodiments, the decoupling layer can be implemented as a self-attention network. In this case, the graph-enhanced object feature embedding representations of each user interaction object can be provided to the self-attention network to decouple the feature embedding representation components corresponding to user interests, user herd mentality, and user risk preferences from the graph-enhanced object feature embedding representations of each user interaction object.

[0067] For example, suppose we use a graph neural network to perform graph convolution operations on the object features of a given object based on an object knowledge graph to obtain a set of graph-enhanced object feature embeddings, H. (L) ∈R | ε|×d, where |ε| is the number of objects and d is the size of the graph representation. Given the historical interaction sequence of user u as S={f1,…,f |S|}, can be derived from the object feature embedding representation set H after graph augmentation. (L) Retrieve object behavior sequence representation Representation of the object's behavioral sequence Self-attention mechanisms can be used to extract a vector set {ω} of user interests, user herd mentality, and user risk preferences. I ,ω C ,ω R The following operations are performed to obtain the decoupled object feature embedding representation:

[0068]

[0069] Where σ(·) is a nonlinear function and f(·) is a softmax function.

[0070] Through the self-attention operation described above, a multi-granularity decoupled object feature embedding representation for user u can be obtained based on an unsupervised mechanism.

[0071] In step 340, the user feature embedding representation is fused with the user interest embedding representation and the user conformity embedding representation, respectively, and the two fusion results are provided to the representation layer of the object recommendation model to obtain the user interest representation and the user conformity representation.

[0072] In 350, the object feature embedding representation of the target object is provided to the representation layer of the object recommendation model to obtain the target object representation, and the object type embedding representation of the user interaction object is provided to the representation layer of the object recommendation model to obtain the user risk preference representation.

[0073] At 360, the prediction layer of the object recommendation model predicts user interest based on user interest representations and target object representations, and predicts user conformity based on user conformity representations and target object representations. User interest prediction refers to the prediction of user interests, and user conformity prediction refers to the prediction of user conformity. That is, the prediction of user interests is obtained by scoring the similarity between user interest representations and target object representations, and the prediction of user conformity is obtained by scoring the similarity between user conformity representations and target object representations.

[0074] In some embodiments, user interest prediction results Based on It is determined that, where σ(·) is the sigmoid function, and FFN I (·) represents the forward propagation network corresponding to the user interest representation. Embedding representations for user features Embedding representations for user interests, || for cascading operations, (·) T For the transpose operation, and x f To represent the target object.

[0075] In some embodiments, user conformity prediction results Based on It is determined that, where σ(·) is the sigmoid function, and FFN C (·) represents the forward propagation network corresponding to the user conformity representation. Embedding representations for user features For user conformity embedding representation, || is for cascading operations, (·) T For the transpose operation, and x f To represent the target object.

[0076] At step 370, determine whether the model training process of the object recommendation model has ended. For example, this can be determined by whether the model training termination condition is met. Examples of model training termination conditions include, but are not limited to, reaching a preset number of training epochs, or the trained model achieving a predetermined accuracy. When the model training termination condition is that the trained model has achieved a predetermined accuracy, the prediction result for the target object can be determined based on the user interest prediction result and the user conformity prediction result, and the training model accuracy can be determined based on the prediction result for the target object.

[0077] If the training termination condition is met, the model training process ends. If the training termination condition is not met, then at 380, the total loss function is determined based on the user interest prediction results, user conformity prediction results, risk preference embedding representation, and user risk preference representation.

[0078] Figure 4 An example flowchart of a total loss function determination process 400 according to an embodiment of this specification is shown.

[0079] like Figure 4 As shown, at 410, the user interest loss term is determined based on the user interest prediction result, and at 420, the user conformity loss term is determined based on the user conformity prediction result.

[0080] In some embodiments, the target object may have object popularity determined based on the number of interactions in the object's interaction history.

[0081] In some embodiments, the popularity of the target object f can be defined as follows:

[0082]

[0083] Where, γ f For the popularity of target object f, C f C represents the number of interactions of the target object f in historical data. max The maximum number of interactions between each object in the object set in historical data, and C min This represents the minimum number of interactions between the objects in the set of objects in the historical data.

[0084] In this scenario, the user interest loss term can be determined based on the user interest prediction results and the object popularity of the target object, and the user conformity loss term can be determined based on the user conformity prediction results and the object popularity of the target object.

[0085] For example, the user interest loss term L I It can be represented as: in, For user interest prediction results, Let be the action label of user u on target object f, and CE(·) be the cross-entropy loss function.

[0086] User herd mentality loss item L C It can be represented as: in, To predict results based on user conformity, Let be the action label of user u on target object f, and CE(·) be the cross-entropy loss function.

[0087] In 430, the user risk preference loss term is determined based on the risk preference embedding representation and the user risk preference representation.

[0088] Since object type can effectively characterize the risk level of an object, useful prior information can be extracted from the object type sequence in historical behavior to model user risk preferences. For example, suppose a given user u has a historical fund type sequence of... A specific representation can be extracted from it and used as self-supervised information on risk preferences. This specific representation can be defined as follows:

[0089]

[0090] Where Φ(·) is the Embedding operator, g(·) is the Pooling function (e.g., it can be defined as the Concact operation), and FFN(·) is the forward propagation network.

[0091] Based on the above definition, the user risk preference loss term L can be... R The contrastive loss function is defined as follows, which serves as the self-supervised loss function:

[0092]

[0093] Where B represents a specific batch of data, and negative samples can be based on a noisy distribution. The data is obtained from sampling, and τ is a temperature parameter. In some embodiments, the distribution... It is implemented as a uniform distribution. In other embodiments, the distribution... Other biased distributions can also be used. The temperature parameter τ can be used to control the intensity of the penalty applied to different negative samples.

[0094] At 440, the total loss function is determined based on the user interest loss term, the user conformity loss term, and the user risk preference loss term.

[0095] In some embodiments, the total loss function can be obtained by combining the user interest loss term, the user conformity loss term, and the user risk preference loss term, i.e., L = L I+L C +L R In some embodiments, the user risk preference loss term L R It can have a weighted value ε, which defines the contribution of the user risk preference loss term to the total loss function. In this case, the total loss function can be determined based on the user interest loss term, the user conformity loss term, and the weighted user risk preference loss term. For example, L = L I +L C +ε·L R ε is a constant greater than 0. This constant ε is used to control the risk preference term L. R The contribution to the total loss function, while the user interest term L I And user herd mentality item L C The contribution to the total loss function can be determined by the object popularity parameter γ. f Make adaptive adjustments.

[0096] Back Figure 3 After determining the total loss function as described above, at step 390, the model parameters of the object recommendation model are adjusted based on the total loss function. For example, the parameters of one or more layers in the embedding layer, graph neural network layer, representation layer, and prediction layer of the object recommendation model are adjusted. Then, the process returns to step 310 to execute the next iteration. This loop continues until the model training process is complete.

[0097] It should be noted that, Figure 3 and Figure 4 What is shown is only Figure 2 The example shown is an illustration of the training process of the object recommendation model 200. As described above... Figure 2 The above, Figure 2 Some features of the object recommendation model 200 are not essential to this disclosure. Similarly, Figure 3 and Figure 4 Not all steps of the training process shown are necessary for this disclosure.

[0098] As an example, in Figure 4 In the example, user interest embeddings, user conformity embeddings, and user risk preference embeddings are used to model the object recommendation model. In some embodiments, only user interest embeddings and user conformity embeddings can be used to model the object recommendation model. In this case, one can start from... Figure 3 and Figure 4The illustrated training process modifies operations related to the user risk preference embedding representation. For example, in 310, it is not necessary to provide the object type of the user interaction object to the embedding layer of the object recommendation model to obtain the object type embedding representation of the user interaction object. Thus, in 330, through the decoupling layer of the object recommendation model, user interest embedding representations and user conformity embedding representations corresponding to user interests and user herd mentality, respectively, are decoupled from the graph-enhanced object feature embedding representation of the user interaction object. In 350, only the object feature embedding representation of the target object is provided to the representation layer of the object recommendation model to obtain the target object representation. Furthermore, in 380, the total loss function can be determined based on the user interest prediction result and the user conformity prediction result. Specifically, a user interest loss term is determined based on the user interest prediction result, and a user conformity loss term is determined based on the user conformity prediction result. Then, the total loss function is determined based on the user interest loss term and the user conformity loss term.

[0099] In some embodiments, the object recommendation model may not include a graph neural network, so that graph augmentation based on the object knowledge graph is not performed on the object feature embedding representations of the target object and the user-interacting object during the training of the object recommendation model. In some embodiments, only user interest embedding representations and user conformity embedding representations can be used to model the object recommendation model, without performing graph augmentation based on the object knowledge graph on the object feature embedding representations of the target object and the user-interacting object.

[0100] As referred above Figures 1 to 4 The training process of an object recommendation model according to embodiments of this specification is described. The following uses product recommendation as an example to illustrate the above-described object recommendation model training process.

[0101] Figure 5 A schematic diagram illustrating an example of the training process for an object recommendation model according to an embodiment of this specification is shown. Figure 5 In the example, the object is a product, and the interactive objects include interactive product 1 to interactive product n.

[0102] User features are provided to the embedding layer to obtain the user feature embedding representation. The object features of interactive product 1 are provided to the embedding layer to obtain the object feature embedding representation of interactive product 1. This object feature embedding representation is then subjected to graph augmentation processing using a product knowledge graph via a graph neural network, resulting in the graph-augmented object feature embedding representation of interactive product 1. Similarly, the object features of interactive product n are provided to the embedding layer to obtain the object feature embedding representation of interactive product n. This object feature embedding representation is then subjected to graph augmentation processing using a product knowledge graph via a graph neural network, resulting in the graph-augmented object feature embedding representation of interactive product n. Finally, the object features of the target product are provided to the embedding layer to obtain the object feature embedding representation of the interactive product. This object feature embedding representation is then subjected to graph augmentation processing using a product knowledge graph via a graph neural network, resulting in the graph-augmented object feature embedding representation x of the target product. f .

[0103] The object types of interactive goods 1, 2, and n are respectively provided to the embedding layer to obtain the object type embedding representations of interactive goods 1, 2, and n. Subsequently, these representations are provided to the feedforward network (FFN) to obtain the user risk preference representation. The target object's feature embedding representation, after graph augmentation, is provided to the representation layer of the object recommendation model to obtain the target object representation.

[0104] From the graph-enhanced object feature embedding representation of interactive product 1, object feature embedding representation components corresponding to user interests, user conformity, and user risk preferences are decoupled. From the graph-enhanced object feature embedding representation of interactive product 2, object feature embedding representation components corresponding to user interests, user conformity, and user risk preferences are decoupled. From the graph-enhanced object feature embedding representation of interactive product n, object feature embedding representation components corresponding to user interests, user conformity, and user risk preferences are decoupled. Then, the object feature embedding representation components corresponding to user interests decoupled from the graph-enhanced object feature embedding representations of interactive products 1 to n are concatenated to obtain the user interest embedding representation. The object feature embedding representation components corresponding to user conformity psychology, decoupled from the graph-enhanced object feature embedding representations of interactive products 1 to n, are concatenated to obtain the user conformity embedding representation. The object feature embedding representation components corresponding to user risk preferences, decoupled from the graph-enhanced object feature embedding representations of interactive products 1 to n, are concatenated to obtain the user risk preference embedding representation.

[0105] Subsequently, user interests are embedded in the representation. The user interest representation is fused (cascaded) with the user feature embedding representation and fed into a feedforward network. This process incorporates user conformity embedding representation. The user interest representation is fused with the user feature embedding representation, and the fusion result is fed into a feedforward network to obtain the user conformity representation. Then, based on the user interest representation and the target object representation, the user interest prediction result is obtained. And user conformity prediction results obtained based on user conformity representation and target object representation.

[0106] After determining the user interest prediction results and user herd prediction results Then, based on the user interest prediction results User u's behavior tags towards target product f Determine the user interest loss term L I and prediction results based on user herd mentality User u's behavior tags towards target product f Determine the user's herd loss item L C Furthermore, it is also based on the embedded representation of user risk preferences. and user risk preference representation Determine the user's risk preference loss term L R .

[0107] Subsequently, based on the user interest loss term L I User herd mentality loss item L C And user risk preference loss item L R Determine the total loss function L, and adjust the model parameters of the object recommendation model according to the total loss function L.

[0108] Using the above-mentioned object recommendation model training scheme, user feature embeddings and object feature embeddings of target objects and user interaction objects are obtained by providing user features, target objects, and user interaction objects to the embedding layer of the object recommendation model. Then, user interest embeddings and user conformity embeddings corresponding to user interests and user herd mentality are decoupled from the object feature embeddings of user interaction objects through the decoupling layer of the object recommendation model. Subsequently, the decoupled user interest embeddings and user conformity embeddings are introduced into the object recommendation model training. Thus, the object recommendation model can be modeled based on user interests and user herd mentality, thereby improving the user understanding effect of the object recommendation model and making the trained object recommendation model applicable to the above-mentioned object recommendation scenarios that need to consider user interests and user herd mentality.

[0109] By utilizing the above-mentioned object recommendation model training scheme, a graph neural network is deployed in the object recommendation model, and the graph neural network is used to perform graph augmentation processing on the object features of the target object and the user interaction object based on the object knowledge graph. This results in graph-augmented object feature embedding representations of the target object and the user interaction object. In this way, the feature influence of related objects in the object knowledge graph can be introduced into the object feature embedding representations of the target object and the user interaction object, thereby improving the accuracy of the object feature embedding representations of the target object and the user interaction object.

[0110] By utilizing the aforementioned object recommendation model training scheme, a graph neural network is deployed within the object recommendation model. This graph neural network is then used to perform graph augmentation processing on the object features of the target object and the user-interacting object based on the object knowledge graph. This results in graph-augmented object feature embedding representations of the target object and the user-interacting object. Subsequently, user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations, corresponding to user interests, user herd mentality, and user risk preference, are decoupled from the object feature embedding representations of the user-interacting object. These decoupled user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations are then introduced into the object recommendation model training. This allows for object recommendation modeling based on user interests, user herd mentality, and user risk preference, thereby improving the user understanding performance of the object recommendation model and making the trained object recommendation model suitable for the aforementioned object recommendation scenarios that require consideration of user herd mentality and user risk preference.

[0111] By using the above object recommendation model training scheme, and by setting the total loss function to be based on user interest loss, user conformity loss, and user risk preference loss, the training accuracy of the object recommendation model can be further improved.

[0112] By using the above object recommendation model training scheme, the contribution of user interest loss and user conformity loss to the total loss function can be adaptively adjusted by setting object popularity, thereby improving the adaptability of the object recommendation model to various application scenarios.

[0113] By using the above-mentioned object recommendation model training scheme, the contribution of the risk preference term to the total loss function can be controlled by setting a weight value for the risk preference loss term, thereby further improving the adaptability of the object recommendation model to various application scenarios.

[0114] After training the object recommendation model as described above, we can use the object recommendation model to predict whether a target object is a recommended object, thereby realizing object recommendation processing.

[0115] Figure 6 An example flowchart of a method 600 for determining recommended objects based on an object recommendation model, according to an embodiment of this specification, is shown.

[0116] like Figure 6 As shown in Figure 610, user features, target object features, and user interaction object features are provided to the embedding layer of the object recommendation model to obtain user feature embedding representations of user features, and target object and user interaction object feature embedding representations of target object and user interaction object.

[0117] In step 620, the object feature embedding representations of the target object and the user interaction object are respectively provided to the graph neural network of the object recommendation model to perform graph augmentation processing based on the object knowledge graph, so as to obtain the object feature embedding representations of the target object and the user interaction object after graph augmentation processing.

[0118] In 630, through the decoupling layer of the object recommendation model, user interest embedding representation and user conformity embedding representation corresponding to user interests and user herd mentality are decoupled from the graph-enhanced object feature embedding representation of user interaction objects.

[0119] In some embodiments, the user interaction objects include multiple user interaction objects. In this case, feature embedding representation components corresponding to user interests and user conformity can be decoupled from the graph-enhanced object feature embedding representations of each user interaction object via the decoupling layer of the object recommendation model, and then concatenated to obtain user interest embedding representations and user conformity embedding representations.

[0120] In step 640, user feature embedding representations are fused with user interest embedding representations and user conformity embedding representations, respectively, and the resulting fusion results are provided to the representation layer of the object recommendation model to obtain user interest representations and user conformity representations.

[0121] In 650, the object features of the target object are embedded in the representation and provided to the representation layer of the object recommendation model to obtain the target object representation.

[0122] Subsequently, based on user interest representations, user conformity representations, and target object representations, it is determined whether to recommend the target object to the user.

[0123] For example, in some embodiments, at 660, user interest prediction results are predicted based on user interest representations and target object representations, and user conformity prediction results are predicted based on user conformity representations and target object representations.

[0124] In step 670, based on user interest prediction results and user conformity prediction results, it is determined whether to recommend target objects to users. For example, when setting object popularity γ... f In this case, the prediction result y can be determined according to the following formula. u,f : in, For user interest prediction results, and To predict results based on user feedback.

[0125] It should be noted that, Figure 6 The examples shown are merely illustrative embodiments of the object recommendation process. In other embodiments, the object recommendation model may not have a graph neural network, thus not performing graph augmentation based on object knowledge graphs on the object feature embedding representations of the target object and the user-interacting object.

[0126] Figure 7 An example schematic diagram of an object recommendation process according to an embodiment of this specification is shown. Figure 7 The example below uses product recommendations to illustrate this.

[0127] like Figure 7 As shown, user features are provided to the embedding layer to obtain user feature embedding representations. The object features of interactive product 1 are provided to the embedding layer to obtain the object feature embedding representation of interactive product 1. This object feature embedding representation is then processed by a graph neural network using a product knowledge graph to obtain the graph-enhanced object feature embedding representation of interactive product 1. Similarly, the object features of interactive product 2 are provided to the embedding layer to obtain the object feature embedding representation of interactive product 2. This object feature embedding representation is then processed by a graph neural network using a product knowledge graph to obtain the graph-enhanced object feature embedding representation of interactive product 2. Likewise, the object features of interactive product n are provided to the embedding layer to obtain the object feature embedding representation of interactive product n. This object feature embedding representation is then processed by a graph neural network using a product knowledge graph to obtain the graph-enhanced object feature embedding representation x of the target product. f The graph-enhanced feature embedding representation of the target object is provided to the representation layer of the object recommendation model to obtain the target object representation.

[0128] From the graph-enhanced object feature embedding representation of interactive product 1, object feature embedding representation components corresponding to user interests and user conformity psychology are decoupled. From the graph-enhanced object feature embedding representation of interactive product 2, object feature embedding representation components corresponding to user interests and user conformity psychology are decoupled. From the graph-enhanced object feature embedding representation of interactive product n, object feature embedding representation components corresponding to user interests and user conformity psychology are decoupled. Then, the object feature embedding representation components corresponding to user interests decoupled from the object feature embedding representations of interactive products 1 to n are concatenated to obtain the user interest embedding representation. The object feature embedding representation components corresponding to user conformity psychology, decoupled from the object feature embedding representations of interactive products 1 to n, are concatenated to obtain the user conformity embedding representation.

[0129] Subsequently, user interests are embedded in the representation. Embedded representation with user features The data is fused, and the fusion result is fed into a feedforward network to obtain a user interest representation. User conformity is then embedded into the representation. Embedded representation with user features The data is fused together, and the fusion result is fed into a feedforward network to obtain a user conformity representation. Then, based on the user interest representation and the target object representation, a user interest prediction result is obtained. And user conformity prediction results obtained based on user conformity representation and target object representation.

[0130] After determining the user interest prediction results and user herd prediction results Then, based on the user interest prediction results and user herd prediction results Determine whether to recommend the target object to the user.

[0131] The object recommendation model training method and object recommendation method according to embodiments of this specification have been described above with reference to the accompanying drawings.

[0132] Figure 8 An example block diagram of an object recommendation model training apparatus 800 according to an embodiment of this specification is shown. Figure 8 As shown, the object recommendation model training device 800 includes a first representation generation unit 810, a graph enhancement processing unit 820, an embedding representation decoupling unit 830, a second representation generation unit 840, a target object prediction unit 850, a loss function determination unit 860, and a model parameter adjustment unit 870.

[0133] The first representation generation unit 810, the graph enhancement processing unit 820, the embedding representation decoupling unit 830, the second representation generation unit 840, the target object prediction unit 850, the loss function determination unit 860, and the model parameter adjustment unit 870 perform operations in a loop until the training termination condition is met.

[0134] Specifically, in each iteration, the first representation generation unit 810 is configured to provide user features, target object object features, user interaction object object features, and user interaction object object types to the embedding layer of the object recommendation model, respectively, to obtain user feature embedding representations of user features, target object object object features, user interaction object object features, and user interaction object object type embedding representations. Here, the user interaction object and the target object belong to the same type of object, and the object type contains object risk information. The operation of the first representation generation unit 810 can be referenced above. Figure 3 The operation described in 310.

[0135] The graph augmentation processing unit 820 is configured to provide the object feature embedding representations of the target object and the user interaction object to the graph neural network of the object recommendation model to perform graph augmentation processing based on the object knowledge graph, thereby obtaining the graph-augmented object feature embedding representations of the target object and the user interaction object. The operation of the graph augmentation processing unit 820 can be referenced above. Figure 3 The operation described in 320.

[0136] The embedding representation decoupling unit 830 is configured to decouple user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations, respectively, from the graph-augmented object feature embedding representations of user interaction objects via the decoupling layer of the object recommendation model. The operation of the embedding representation decoupling unit 830 can be referenced above. Figure 3 The operation described in 330.

[0137] The second representation generation unit 840 is configured to fuse user feature embedding representations with user interest embedding representations and user conformity embedding representations, respectively, and provide the obtained fusion results, object feature embedding representations of target objects, and object type embedding representations of user interaction objects to the representation layer of the object recommendation model to obtain user interest representations, user conformity representations, target object representations, and user risk preference representations. The operation of the second representation generation unit 840 can be referenced above. Figure 3 The operations described in 340 and 350.

[0138] The target object prediction unit 850 is configured to predict user interest prediction results based on user interest representation and target object representation, and to predict user conformity prediction results based on user conformity representation and target object representation. The operation of the target object prediction unit 850 can be referenced above. Figure 3 The operation described in 360.

[0139] The loss function determination unit 860 is configured to determine the total loss function based on user interest prediction results, user conformity prediction results, risk preference embedding representation, and user risk preference representation in response to the failure to meet the training termination condition. The operation of the loss function determination unit 860 can be referenced above. Figure 3 The operation described in 380.

[0140] The model parameter tuning unit 870 is configured to adjust the model parameters of the object recommendation model according to the total loss function. The operation of the model parameter tuning unit 870 can be referenced above. Figure 3 The operation described in 390.

[0141] Figure 9 An example block diagram of a loss function determination unit 900 according to an embodiment of this specification is shown. Figure 9 As shown, the loss function determination unit 900 includes a user interest loss term determination module 910, a user conformity loss term determination module 920, a user risk preference loss term determination module 930, and a loss function determination module 940.

[0142] The user interest loss term determination module 910 is configured to determine the user interest loss term based on the user interest prediction result. The operation of the user interest loss term determination module 910 can be referenced above. Figure 4 The operation described in 410.

[0143] The user conformity loss term determination module 920 is configured to determine the user conformity loss term based on the user conformity prediction result. The operation of the user conformity loss term determination module 920 can be referenced above. Figure 4 The operation described in 420.

[0144] The user risk preference loss term determination module 930 is configured to determine the user risk preference loss term based on the risk preference embedded representation and the user risk preference representation. The operation of the user risk preference loss term determination module 930 can be referenced above. Figure 4 The operation described in 430.

[0145] The loss function determination module 940 is configured to determine the total loss function based on the user interest loss term, the user conformity loss term, and the user risk preference loss term. The operation of the loss function determination module 940 can be referenced above. Figure 4 The operation described in 440.

[0146] In some embodiments, the embedding representation decoupling unit 830 can decouple user interest embedding representation, user conformity embedding representation, and user risk preference embedding representation from the object feature embedding representation of the user interaction object based on an unsupervised mechanism via the decoupling layer of the object recommendation model.

[0147] In some embodiments, the decoupling layer may include a self-attention network. The embedding representation decoupling unit 830 may provide the object feature embedding representation of the user interaction object to the self-attention network, so as to decouple the user interest embedding representation, the user conformity embedding representation, and the user risk preference embedding representation from the object feature embedding representation of the user interaction object, respectively.

[0148] In some embodiments, user interaction objects may include multiple user interaction objects. The embedding representation decoupling unit 830 is configured to decouple feature embedding representation components corresponding to user interests, user conformity, and user risk preferences from the object feature embedding representations of each user interaction object via the decoupling layer of the object recommendation model, and concatenate them to obtain user interest embedding representation, user conformity embedding representation, and user risk preference embedding representation.

[0149] It should be noted that, Figure 8 The example shown is merely an illustrative embodiment of the object recommendation model training device. Figure 4In the example, user interest embedding representation, user conformity embedding representation, and user risk preference embedding representation are used to model the object recommendation model.

[0150] In some embodiments, object recommendation modeling can be performed using only user interest embeddings and user conformity embeddings. In this case, during model training, a first representation generation unit provides user features, target object features, and user interaction object features to the embedding layer of the object recommendation model, respectively, to obtain user feature embeddings, target object feature embeddings, and user interaction object feature embeddings. An embedding decoupling unit, via the decoupling layer of the object recommendation model, decouples user interest embeddings and user conformity embeddings, respectively, from the graph-enhanced object feature embeddings of the user interaction objects. A second representation generation unit fuses the user feature embeddings with the user interest embeddings and user conformity embeddings, respectively, and provides the resulting fusions and the target object feature embeddings to the representation layer of the object recommendation model to obtain user interest representations, user conformity representations, and target object representations. Subsequently, a loss function determination unit determines the total loss function based on the user interest prediction results and the user conformity prediction results. Specifically, the loss function determination unit determines a user interest loss term based on the user interest prediction results and a user conformity loss term based on the user conformity prediction results. Then, the total loss function is determined based on the user interest loss term and the user conformity loss term.

[0151] In some embodiments, the object recommendation model may not include a graph neural network. Accordingly, the object recommendation model training apparatus may not include a graph augmentation processing unit, thereby not performing object knowledge graph-based graph augmentation processing on the object feature embedding representations of the target object and the user interaction object.

[0152] Figure 10 An example block diagram of an object recommendation device 1000 according to an embodiment of this specification is shown. Figure 10 As shown, the object recommendation device 1000 includes a first representation generation unit 1010, a graph enhancement processing unit 1020, an embedded representation decoupling unit 1030, a second representation generation unit 1040, and a recommendation decision unit 1050.

[0153] The first representation generation unit 1010 is configured to provide user features, target object features, and user interaction object features to the embedding layer of the object recommendation model, thereby obtaining user feature embedding representations of user features and object feature embedding representations of target objects and user interaction objects. The operation of the first representation generation unit 1010 can be referenced above. Figure 6 The operation described in 610.

[0154] The graph augmentation processing unit 1020 is configured to provide the object feature embedding representation of the target object and the object feature embedding representation of the user interaction object to the graph neural network of the object recommendation model for graph augmentation processing based on the object knowledge graph, thereby obtaining the graph augmented object feature embedding representations of the target object and the user interaction object. The operation of the graph augmentation processing unit 1020 can be referenced above. Figure 6 The operation described in 620.

[0155] The embedding representation decoupling unit 1030 is configured, via the decoupling layer of the object recommendation model, to decouple user interest embedding representations and user conformity embedding representations, respectively corresponding to user interests and user herd mentality, from the graph-enhanced object feature embedding representations of user interaction objects. In some embodiments, user interaction objects may include multiple user interaction objects. The embedding representation decoupling unit 1030 is configured, via the decoupling layer of the object recommendation model, to decouple feature embedding representation components, respectively corresponding to user interests, user herd mentality, and user risk preference, from the graph-enhanced object feature embedding representations of each user interaction object, and concatenates them to obtain user interest embedding representations, user conformity embedding representations, and user risk preference embedding representations. The operation of the embedding representation decoupling unit 1030 can be referred to the above reference. Figure 6 The operation described in 630.

[0156] The second representation generation unit 1040 is configured to fuse user feature embedding representations with user interest embedding representations and user conformity embedding representations, respectively, and provide the obtained fusion results and the object feature embedding representation of the target object to the representation layer of the object recommendation model to obtain user interest representations, user conformity representations, and target object representations. The operation of the second representation generation unit 1040 can be referenced above. Figure 6 The operations described in 640 and 650.

[0157] The recommendation decision unit 1050 is configured to determine whether to recommend the target object to the user based on user interest representation, user conformity representation, and target object representation. For example, the recommendation decision unit 1050 can predict user interest prediction results based on user interest representation and target object representation, and predict user conformity prediction results based on user conformity representation and target object representation, and determine whether to recommend the target object to the user based on the user interest prediction results and user conformity prediction results. The operation of the recommendation decision unit 1050 can be referenced above. Figure 6 The operations described in 660 and 670.

[0158] As referred above Figures 1 to 10This specification describes an object recommendation model training method, an object recommendation model training apparatus, an object recommendation method, and an object recommendation apparatus according to embodiments thereof. The object recommendation model training apparatus and object recommendation apparatus described above can be implemented in hardware, or in software, or a combination of hardware and software.

[0159] Figure 11 A schematic diagram illustrating an example of a computer-based object recommendation model training apparatus 1100 implemented according to an embodiment of this specification is shown. Figure 11 As shown, the object recommendation model training apparatus 1100 may include at least one processor 1110, a memory (e.g., non-volatile memory) 1120, a RAM 1130, and a communication interface 1140, and the at least one processor 1110, memory 1120, RAM 1130, and communication interface 1140 are connected together via a bus 1160. At least one processor 1110 executes at least one computer-readable instruction (i.e., the elements implemented in software described above) stored or encoded in the memory.

[0160] In one embodiment, computer-executable instructions are stored in memory, which, when executed, cause at least one processor 1110 to: cyclically execute the following model training process until the training termination condition is met: providing user features, object features of user interaction objects, and object features of target objects to the embedding layer of the object recommendation model, respectively, to obtain user feature embedding representations of user features, object feature embedding representations of user interaction objects, and object feature embedding representations of target objects; and decoupling user interest embedding representations and user conformity embedding representations, respectively, from the object feature embedding representations of user interaction objects via the decoupling layer of the object recommendation model. The process involves: fusing user feature embeddings with user interest embeddings and user conformity embeddings, and providing the fusion results and target object feature embeddings to the representation layer of the object recommendation model to obtain user interest representations, user conformity representations, and target object representations; predicting user interest predictions based on user interest representations and target object representations, and predicting user conformity predictions based on user conformity representations and target object representations; and determining a total loss function based on user interest predictions and user conformity predictions in response to the failure to meet the training termination condition, and adjusting the model parameters of the object recommendation model according to the total loss function.

[0161] It should be understood that the computer-executable instructions stored in memory, when executed, cause at least one processor 1110 to perform the above-described combinations in the various embodiments of this specification. Figures 1-5 as well as Figures 8-9 The description includes various operations and functions.

[0162] Figure 12An example schematic diagram of an object recommendation device 1200 implemented based on a computer system according to an embodiment of this specification is shown. Figure 12 As shown, the object recommendation device 1200 may include at least one processor 1210, a memory (e.g., non-volatile memory) 1220, a RAM 1230, and a communication interface 1240, and the at least one processor 1210, memory 1220, RAM 1230, and communication interface 1240 are connected together via a bus 1260. At least one processor 1210 executes at least one computer-readable instruction (i.e., the elements implemented in software described above) stored or encoded in the memory.

[0163] In one embodiment, computer-executable instructions are stored in memory, which, when executed, cause at least one processor 1210 to: provide user features, object features of user interaction objects, and object features of target objects to the embedding layer of an object recommendation model, respectively, to obtain user feature embedding representations of user features, object feature embedding representations of user interaction objects, and object feature embedding representations of target objects; decouple user interest embedding representations and user conformity embedding representations, respectively, from the object feature embedding representations of user interaction objects via the decoupling layer of the object recommendation model; fuse the user feature embedding representations with the user interest embedding representations and user conformity embedding representations, respectively, and provide the fusion result and the object feature embedding representations of target objects to the representation layer of the object recommendation model to obtain user interest representations, user conformity representations, and target object representations; and determine whether to recommend a target object to the user based on the user interest representations, user conformity representations, and target object representations.

[0164] It should be understood that the computer-executable instructions stored in the memory, when executed, cause at least one processor 1210 to perform the above-described combinations in the various embodiments of this specification. Figures 6-7 as well as Figure 10 The description includes various operations and functions.

[0165] According to one embodiment, a program product, such as a machine-readable medium (e.g., a non-transitory machine-readable medium), is provided. The machine-readable medium may have instructions (i.e., the elements implemented in software as described above), which, when executed by a machine, cause the machine to perform the above-described combinations of the various embodiments of this specification. Figures 1-10 The various operations and functions described. Specifically, a system or apparatus equipped with a readable storage medium storing software program code that implements the functions of any of the embodiments described above, and enabling the computer or processor of the system or apparatus to read and execute the instructions stored in the readable storage medium.

[0166] In this case, the program code itself, which can be read from a readable medium, can perform the functions of any of the above embodiments. Therefore, the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of the present invention.

[0167] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer or the cloud via a communication network.

[0168] According to one embodiment, a computer program product is provided, the computer program product including a computer program, which, when executed by a processor, causes the processor to perform the above-described combinations of the various embodiments of this specification. Figures 1-10 The description includes various operations and functions.

[0169] Those skilled in the art will understand that the various embodiments disclosed above can be modified and varied without departing from the spirit of the invention. Therefore, the scope of protection of this invention should be defined by the appended claims.

[0170] It should be noted that not all steps and units in the above process and system structure diagrams are mandatory; some steps or units can be omitted according to actual needs. The execution order of each step is not fixed and can be determined as needed. The device structure described in the above embodiments can be a physical structure or a logical structure; that is, some units may be implemented by the same physical entity, or some units may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0171] In the above embodiments, the hardware units or modules can be implemented mechanically or electrically. For example, a hardware unit, module, or processor may include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to perform the corresponding operation. The hardware unit or processor may also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily configured by software to perform the corresponding operation. The specific implementation method (mechanical, dedicated permanent circuitry, or temporarily configured circuitry) can be determined based on cost and time considerations.

[0172] The specific embodiments described above with reference to the accompanying drawings are exemplary embodiments, but do not represent all embodiments that can be implemented or fall within the scope of the claims. The term "exemplary" as used throughout this specification means "serving as an example, instance, or illustration" and does not imply that it is "preferred" or "advantageous" compared to other embodiments. Specific details are included to provide an understanding of the described techniques. However, these techniques can be practiced without these specific details. In some instances, well-known structures and apparatuses are shown in block diagram form to avoid obscuring the concepts of the described embodiments.

[0173] The foregoing description of this disclosure is provided to enable any person skilled in the art to implement or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein can be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but is consistent with the widest scope of the principles and novel features disclosed herein.

Claims

1. A method for training an object recommendation model for determining whether to recommend a target object to a user, the method comprising: cyclically performing a model training process until a training end condition is met, the model training process comprising: providing a user feature, an object feature of a user-interacted object, and an object feature of the target object to an embedding layer of the object recommendation model respectively to obtain a user feature embedding representation of the user feature, an object feature embedding representation of the user-interacted object, and an object feature embedding representation of the target object; decoupling, via a decoupling layer of the object recommendation model, a user interest embedding representation and a user conformity embedding representation corresponding to a user interest and a user conformity respectively from the object feature embedding representation of the user-interacted object; fusing the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation respectively, and providing the fused results and the object feature embedding representation of the target object to a representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation respectively; predicting a user interest prediction result based on the user interest representation and the target object representation, and predicting a user conformity prediction result based on the user conformity representation and the target object representation; and in response to the training end condition not being met, determining a total loss function based on the user interest prediction result and the user conformity prediction result, and adjusting model parameters of the object recommendation model according to the total loss function.

2. The method of claim 1, further comprising: providing the object feature embedding representation of the target object and the user-interacted object to a graph neural network of the object recommendation model for graph enhancement processing based on an object knowledge graph to obtain the object feature embedding representation of the target object and the user-interacted object after the graph enhancement processing. the user-interacted object comprises at least two user-interacted objects, 3. The method of claim 1, wherein, decoupling, via the decoupling layer of the object recommendation model, the user interest embedding representation and the user conformity embedding representation corresponding to the user interest and the user conformity respectively from the object feature embedding representation of the user-interacted object comprises: decoupling, via the decoupling layer of the object recommendation model, object feature embedding representation components corresponding to the user interest and the user conformity respectively from the object feature embedding representation of each user-interacted object and concatenating the components to obtain the user interest embedding representation and the user conformity embedding representation. decoupling, via the decoupling layer of the object recommendation model, the user interest embedding representation and the user conformity embedding representation corresponding to the user interest and the user conformity respectively from the object feature embedding representation of the user-interacted object comprises:

4. The method of claim 1, wherein, decoupling, via the decoupling layer of the object recommendation model, the user interest embedding representation and the user conformity embedding representation corresponding to the user interest and the user conformity respectively from the object feature embedding representation of the user-interacted object based on an unsupervised mechanism. ​ 5. The method of claim 4, wherein, The decoupling layer comprises a self-attention network, and the user interest embedding representation and the user conformity embedding representation corresponding to the user interest and the user conformity psychology are decoupled from the object feature embedding representation of the user interactive object via the decoupling layer of the object recommendation model, including: The object feature embedding representation of the user interactive object is provided to the self-attention network to decouple the user interest embedding representation and the user conformity embedding representation corresponding to the user interest and the user conformity psychology from the object feature embedding representation of the user interactive object.

6. The method of any one of claims 1 to 5, wherein, The total loss function is determined based on the user interest prediction result and the user conformity prediction result, including: A user interest loss term is determined based on the user interest prediction result; A user conformity loss term is determined based on the user conformity prediction result; and The total loss function is determined according to the user interest loss term and the user conformity loss term.

7. The method of claim 6, wherein, The target object has an object popularity determined based on the number of interactions in the object interaction historical behavior, the user interest loss term is determined based on the user interest prediction result and the object popularity of the target object, and the user conformity loss term is determined based on the user conformity prediction result and the object popularity of the target object.

8. The method of any one of claims 1 to 5, wherein, The user feature, the object feature of the user interactive object and the object feature of the target object are respectively provided to the embedding layer of the object recommendation model to obtain the user feature embedding representation of the user feature, the object feature embedding representation of the user interactive object and the object feature embedding representation of the target object, including: The user feature, the object feature of the user interactive object, the object feature of the target object and the object type of the user interactive object are respectively provided to the embedding layer of the object recommendation model to obtain the user feature embedding representation of the user feature, the object feature embedding representation of the user interactive object, the object feature embedding representation of the target object and the object type embedding representation of the user interactive object, The user interest embedding representation, the user conformity embedding representation and the user risk preference embedding representation corresponding to the user interest, the user conformity psychology and the user risk preference are decoupled from the object feature embedding representation of the user interactive object via the decoupling layer of the object recommendation model, including: The user interest embedding representation, the user conformity embedding representation and the user risk preference embedding representation corresponding to the user interest, the user conformity psychology and the user risk preference are decoupled from the object feature embedding representation of the user interactive object via the decoupling layer of the object recommendation model, including: The obtained fusion result and the object feature embedding representation of the target object are respectively provided to the representation layer of the object recommendation model to obtain the user interest representation, the user conformity representation and the target object representation, including: The obtained fusion result, the object feature embedding representation of the target object and the object type embedding representation of the user interactive object are respectively provided to the representation layer of the object recommendation model to obtain the user interest representation, the user conformity representation, the target object representation and the user risk preference representation, including: determining a total loss function based on the user interest prediction result and the user conformity prediction result comprises: determining a total loss function based on the user interest prediction result, the user conformity prediction result, the risk preference embedding representation, and the user risk preference representation.

9. The method of claim 8, wherein, determining a total loss function based on the user interest prediction result, the user conformity prediction result, the risk preference embedding representation, and the user risk preference representation comprises: determining a user interest loss term based on the user interest prediction result; determining a user conformity loss term based on the user conformity prediction result; determining a user risk preference loss term based on the risk preference embedding representation and the user risk preference representation; and determining the total loss function according to the user interest loss term, the user conformity loss term, and the user risk preference loss term.

10. The method of claim 9, wherein, the user risk preference loss term has a weighting value for defining a contribution degree of the user risk preference loss term to the total loss function, determining the total loss function according to the user interest loss term, the user conformity loss term, and the user risk preference loss term comprises: determining the total loss function according to the user interest loss term, the user conformity loss term, and the weighted user risk preference loss term.

11. The method of claim 2, wherein, the object knowledge graph is created based on a relationship between a recommendation type object and an associated type object.

12. A method for determining a recommended object based on an object recommendation model, comprising: providing a user feature, an object feature of a user interaction object, and an object feature of a target object to an embedding layer of the object recommendation model respectively to obtain a user feature embedding representation of the user feature, an object feature embedding representation of the user interaction object, and an object feature embedding representation of the target object; decoupling, via a decoupling layer of the object recommendation model, a user interest embedding representation and a user conformity embedding representation corresponding to a user interest and a user conformity psychology respectively from the object feature embedding representation of the user interaction object; fusing the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation respectively, and providing the obtained fusion result and the object feature embedding representation of the target object to a representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation; and determining whether to recommend the target object to a user based on the user interest representation, the user conformity representation, and the target object representation.

13. The method of claim 12, wherein, determining whether to recommend the target object to a user based on the user interest representation, the user conformity representation, and the target object representation comprises: predicting a user interest prediction result and a user conformity prediction result based on the user interest representation and the target object representation respectively and based on the user conformity representation and the target object representation respectively; and determining whether to recommend the target object to a user according to the user interest prediction result and the user conformity prediction result.

14. The method of claim 12, further comprising: The object feature embedding representations of the target object and the user interaction object are respectively provided to a graph neural network of the object recommendation model for graph enhancement processing based on an object knowledge graph, to obtain the object feature embedding representations of the target object and the user interaction object after graph enhancement processing.

15. The method of claim 12, wherein, The object recommendation model is trained according to the method in any one of claims 1 to 11.

16. An apparatus for training an object recommendation model for determining whether to recommend a target object to a user, the apparatus comprising: a first representation generation unit configured to provide a user feature, an object feature of a user interaction object, and an object feature of a target object to an embedding layer of the object recommendation model, to obtain a user feature embedding representation of the user feature, an object feature embedding representation of the user interaction object, and an object feature embedding representation of the target object; an embedding representation decoupling unit configured to decouple, via a decoupling layer of the object recommendation model, a user interest embedding representation and a user conformity embedding representation corresponding to a user interest and a user conformity psychology respectively from the object feature embedding representation of the user interaction object; a second representation generation unit configured to fuse the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation respectively, and to provide the fusion results and the object feature embedding representation of the target object to a representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation; a target object prediction unit configured to predict a user interest prediction result based on the user interest representation and the target object representation, and to predict a user conformity prediction result based on the user conformity representation and the target object representation; a loss function determination unit configured to determine a total loss function based on the user interest prediction result and the user conformity prediction result in response to a training end condition not being met; and a model parameter adjustment unit configured to adjust model parameters of the object recommendation model according to the total loss function, wherein the first representation generation unit, the embedding representation decoupling unit, the second representation generation unit, the target object prediction unit, the loss function determination unit, and the model parameter adjustment unit perform operations cyclically until the training end condition is met.

17. An apparatus for determining a recommended object based on an object recommendation model, comprising: a first representation generation unit configured to provide a user feature, an object feature of a user interaction object, and an object feature of a target object to an embedding layer of the object recommendation model, to obtain a user feature embedding representation of the user feature, an object feature embedding representation of the user interaction object, and an object feature embedding representation of the target object; an embedding representation decoupling unit configured to decouple, via a decoupling layer of the object recommendation model, a user interest embedding representation and a user conformity embedding representation corresponding to a user interest and a user conformity psychology respectively from the object feature embedding representation of the user interaction object; and a second representation generation unit configured to fuse the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation respectively, and to provide the fusion results and the object feature embedding representation of the target object to a representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation. a second representation generation unit that fuses the user feature embedding representation with the user interest embedding representation and the user conformity embedding representation, respectively, and provides the resulting fusion results and the object feature embedding representation of the target object to a representation layer of the object recommendation model to obtain a user interest representation, a user conformity representation, and a target object representation; and a recommendation decision unit that determines whether to recommend the target object to the user based on the user interest representation, the user conformity representation, and the target object representation.

18. The apparatus of claim 17, further comprising: a graph enhancement processing unit that provides the object feature embedding representation of the target object and the user interactive object to a graph neural network of the object recommendation model to perform graph enhancement processing based on an object knowledge graph to obtain the object feature embedding representation of the target object and the user interactive object after graph enhancement processing.

19. An object recommendation system, comprising: an interactive object acquisition apparatus that acquires a user interactive object; the apparatus for training an object recommendation model of claim 16; and the apparatus for determining a recommended object based on an object recommendation model of claim 17 or 18.

20. An apparatus for training an object recommendation model, comprising: at least one processor, a memory coupled with the at least one processor, and a computer program stored in the memory, the at least one processor executing the computer program to implement the method for training an object recommendation model of any one of claims 1 to 11.

21. An apparatus for determining a recommended object based on an object recommendation model, comprising: at least one processor, a memory coupled with the at least one processor, and a computer program stored in the memory, the at least one processor executing the computer program to implement the method for determining a recommended object based on an object recommendation model of any one of claims 12 to 15.

Images