Interaction information prediction method and device, equipment and storage medium

Abstract

The application belongs to the technical field of artificial intelligence, and particularly relates to an interaction information prediction method and device, equipment and a storage medium. The method comprises the following steps: obtaining target recommendation scene features, target account features and display object features; performing feature cross processing on the target account features, the target recommendation scene features and the display object features to obtain first cross features; performing cutting processing on the first cross features based on a preset cutting dimension to obtain at least two sub-cross features corresponding to the first cross features; and predicting interaction information between a target account and a to-be-recommended display object based on the at least two sub-cross features to obtain predicted interaction information. Different types of features are mixed together through feature cross, and the cross features are cut into multiple sub-cross features according to a preset cutting scale, so that feature information in different business scenarios can be fully expressed, and accurate prediction of a recommendation effect is realized.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to an interactive information prediction method, apparatus, device, and storage medium. Background Technology

[0002] With the continuous development of internet technology, the internet is able to provide more and more online services to users. On internet platforms, target accounts can use search functions to search for online information of interest. Furthermore, to facilitate target accounts' access to this information, internet platforms can proactively recommend potentially interesting online information to them based on recommendation systems, such as articles, videos, music, images, products, and advertisements.

[0003] Currently, before recommending network information to a target account, a trained multi-scenario recommendation prediction model is typically used to predict the recommendation effect of the information to be recommended in each scenario. However, a common problem encountered in multi-scenario modeling is that in some business scenarios, data is abundant, requiring a larger feature representation scale to fully express the feature information. Conversely, in other business scenarios, data is scarce, necessitating a smaller feature representation scale to avoid model overfitting. This makes it difficult to balance the feature information representation under different business scenarios during multi-scenario modeling, thus limiting the application of multi-scenario recommendation prediction models and preventing accurate prediction of recommendation results. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides an interactive information prediction method, apparatus, device, and storage medium. By mixing different types of features through feature crossing and dividing the crossing features into multiple sub-crossing features according to a preset segmentation scale, feature information under different business scenarios can be fully expressed, thereby achieving accurate prediction of the recommendation effect.

[0005] On the one hand, this application proposes an interactive information prediction method, the method comprising:

[0006] Obtain the target recommendation scenario features, the target account features corresponding to the target account, and the display object features corresponding to the object to be recommended; the target recommendation scenario features are the features corresponding to the scenario of displaying the object to be recommended to the target account; the target account is the account that will receive the object to be recommended.

[0007] The first cross feature is obtained by performing feature cross processing on the target account features, target recommendation scenario features, and display object features;

[0008] The first cross feature is segmented based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature.

[0009] Based on at least two sub-cross features, the interaction information between the target account and the object to be recommended is predicted to obtain the predicted interaction information.

[0010] On the other hand, embodiments of this application provide an interactive information prediction device, the device comprising:

[0011] The feature acquisition module is used to acquire the target recommendation scenario features, the target account features corresponding to the target account, and the display object features corresponding to the object to be recommended; the target recommendation scenario features are the features corresponding to the scenario of displaying the object to be recommended to the target account; the target account is the account that will receive the object to be recommended.

[0012] The first feature cross module is used to perform feature cross processing on the target account features, target recommendation scenario features and display object features to obtain the first cross feature;

[0013] The feature segmentation module is used to segment the first cross feature based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature.

[0014] The prediction module is used to predict the interaction information between the target account and the object to be recommended based on at least two sub-cross features, and obtain the predicted interaction information.

[0015] On the other hand, this application proposes an electronic device for predicting interactive information, the electronic device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the processor to implement the interactive information prediction method as described above.

[0016] On the other hand, this application proposes a computer-readable storage medium storing at least one instruction or at least one program, which is loaded and executed by a processor to implement the interactive information prediction method as described above.

[0017] On the other hand, this application proposes a computer program product that, when executed by a processor, implements the interactive information prediction method as described above.

[0018] The interaction information prediction method, apparatus, electronic device, and storage medium proposed in this application perform feature cross-processing on the acquired target account features, target recommendation scenario features, and display object features to obtain a first cross feature, thereby allowing various feature information to be mixed together. By segmenting the first cross feature based on a preset segmentation dimension, at least two sub-cross features corresponding to the first cross feature are obtained. Then, based on the at least two sub-cross features, the interaction information between the target account and the object to be recommended is predicted to obtain predicted interaction information. This allows feature information under different business scenarios to be fully expressed, taking into account both the independence and sharing between different business scenarios, realizing recommendation effect prediction in multiple business scenarios, and thus improving the recommendation effect of the object to be recommended. Attached Figure Description

[0019] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram illustrating the implementation environment of an interactive information prediction method according to an exemplary embodiment.

[0021] Figure 2 This is a flowchart illustrating an interactive information prediction method according to an exemplary embodiment.

[0022] Figure 3 This is a schematic diagram illustrating the workflow of a gating submodule according to an exemplary embodiment.

[0023] Figure 4 This is a schematic diagram illustrating a layer normalization principle according to an exemplary embodiment.

[0024] Figure 5 This is a schematic diagram illustrating the structure of a target interactive information prediction model according to an exemplary embodiment.

[0025] Figure 6 This is a schematic diagram illustrating the workflow of a correction module according to an exemplary embodiment.

[0026] Figure 7 This is a block diagram illustrating an interactive information prediction device according to an exemplary embodiment.

[0027] Figure 8 This is a hardware structure block diagram of a server for an interactive information prediction method provided according to an exemplary embodiment. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the present application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0030] To make the objectives, technical solutions, and advantages disclosed in the embodiments of this application clearer, the embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of this application and are not intended to limit the embodiments of this application.

[0031] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "a plurality of" means two or more.

[0032] This application relates to artificial intelligence (AI) and machine learning (ML) technologies.

[0033] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0034] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, interactive operating systems, and mechatronics. Software AI technologies typically include computer vision, natural language processing, and machine learning / deep learning. With the development and advancement of AI, it is being researched and applied in numerous fields, such as smart homes, intelligent customer service, virtual assistants, smart speakers, intelligent marketing, autonomous driving, robotics, and smart healthcare. It is believed that with further technological advancements, AI will be applied in even more areas, playing an increasingly important role.

[0035] Machine learning is a multidisciplinary field that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance.

[0036] Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence. Deep learning, in turn, is the core of machine learning and a technology for realizing machine learning. Machine learning typically includes techniques such as deep learning, reinforcement learning, transfer learning, and inductive learning. Deep learning includes techniques such as MobileNet, Convolutional Neural Networks (CNNs), Deep Belief Networks, Recurrent Neural Networks, Autoencoders, and Generative Adversarial Networks.

[0037] To facilitate understanding of the technical solutions described above and their resulting technical effects in the embodiments of this application, the embodiments of this application first explain the relevant technical terms:

[0038] CTR (Click Through Rate): Also known as click-through rate, it is calculated as the number of clicks divided by the number of impressions. CTR is an internet advertising term that refers to the actual number of clicks on an ad divided by the number of ad impressions.

[0039] Embedding: Dense vectors in deep learning models.

[0040] Normalization: In multi-scenario recommendation models, due to the significant differences in data distribution across different business scenarios, without optimization, activation functions in different network layers may become ineffective during model training, leading to gradient vanishing. To address this issue, multi-scenario recommendation prediction models can employ normalization to ensure consistent data distribution across network layers, thus promoting faster model convergence. Common normalization methods include LayerNorm and BatchNorm. However, in joint modeling of multiple business scenarios, using BatchNorm weakens the magnitude relationship between features, reducing the information gain from the magnitude relationship between features in different business scenarios.

[0041] Currently, most applications (APPs) encompass multiple business scenarios, and the recommendation results for target accounts vary across these different scenarios. To achieve better recommendation performance, object recommendations need to be tailored to different scenarios, recommending objects that the target account is more likely to be interested in.

[0042] For recommendation systems, modeling each scenario individually consumes significant resources and requires substantial maintenance costs. When the number of business scenarios reaches hundreds, this becomes an excessive burden. Furthermore, since the target account receiving recommendations is the same across different business scenarios, there are commonalities. Information sharing between different scenarios facilitates knowledge transfer, which is beneficial for model learning. Moreover, the amount of data in a single business scenario is often limited, especially in small scenarios; modeling them individually increases the difficulty of model training. Therefore, recommendation systems can employ multi-scenario joint modeling to reduce resource consumption and improve model training efficiency. However, a challenge in implementing multi-scenario joint modeling is that, due to the different data volumes and distributions across different business scenarios, it is difficult to account for the differences in data across different scenarios when representing features, thus affecting the overall representation of data performance in the recommendation prediction model.

[0043] In view of this, embodiments of this application provide an interaction information prediction method, apparatus, electronic device, and storage medium. By performing feature cross-processing on the acquired target account features, target recommendation scenario features, and display object features, a first cross feature is obtained, thereby allowing various feature information to be mixed together. By segmenting the first cross feature based on a preset segmentation dimension, at least two sub-cross features corresponding to the first cross feature are obtained. Then, based on the at least two sub-cross features, the interaction information between the target account and the object to be recommended is predicted to obtain predicted interaction information. This allows feature information under different business scenarios to be fully expressed, taking into account both the independence and sharing between different business scenarios, realizing recommendation effect prediction in multiple business scenarios, and thus improving the recommendation effect of the object to be recommended.

[0044] Figure 1 This is a schematic diagram illustrating the implementation environment of an interactive information prediction method according to an exemplary embodiment. For example... Figure 1 As shown, the implementation environment may include at least a client 01 and a server 02. The client 01 and the server 02 may be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0045] In this embodiment, the client 01 can be used to receive display objects recommended by the server. Optionally, the client 01 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smart voice interaction device, smart home appliance, smartwatch, vehicle terminal, aircraft, etc., but is not limited to these. This embodiment can be applied to various scenarios, including but not limited to rendering and displaying interactive information involved in various scenarios such as cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0046] In this embodiment, server 02 can be used to acquire target recommendation scene features, target account features corresponding to the target account, and display object features corresponding to the object to be recommended. It can also perform feature cross-processing on the target account features, target recommendation scene features, and display object features to obtain a first cross feature. Furthermore, it can segment the first cross feature based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature. Based on the at least two sub-cross features, it can predict the interaction information between the target account and the object to be recommended, obtaining predicted interaction information. Finally, it can recommend information to the target account based on the predicted interaction information. Optionally, server 02 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0047] It should be noted that, Figure 1 This is just one example. Other implementation environments may also be included in other scenarios.

[0048] In the specific implementation of this application, data related to target recommendation scenario features, target account features corresponding to target accounts, and display object features corresponding to objects to be recommended and displayed are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0049] Figure 2 This is a flowchart illustrating an interactive information prediction method according to an exemplary embodiment. The method can be used for... Figure 1 In the implementation environment described in this specification, the interactive information prediction method is provided with operational steps as illustrated in the embodiments or flowcharts. However, based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or server product execution, the method can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment) as shown in the embodiments or accompanying drawings. Specifically, as shown in the embodiments or accompanying drawings... Figure 2 As shown, the method may include:

[0050] S101: Obtain the target recommendation scenario features, the target account features corresponding to the target account, and the display object features corresponding to the object to be recommended.

[0051] In this embodiment, the target account is the account that will receive the recommended display object. When recommending information to the target account, the recommendation system can first predict the effect after recommending the display object to the target account. That is, predict whether the target account will interact with the display object after it is recommended to the target account, or predict the probability of the target account interacting with the display object.

[0052] In this embodiment, during interaction prediction, prediction is based on obtaining the target recommendation scenario features, the target account features corresponding to the target account, and the display object features corresponding to the object to be recommended. The target recommendation scenario features are the features corresponding to the scenario in which the object to be recommended is displayed to the target account. Optionally, the target recommendation scenario features include application features, scenario type, account scenario features corresponding to the target account, and display object scenario features corresponding to the object to be recommended. Application scenario features include application type and application version. Optionally, application type can include, but is not limited to, social, content publishing, music, video, and shopping applications. The scenario type can be set according to the business scenarios contained in the application. For example, in a video application, the application's homepage can be a business scenario, the video playback interface can be a business scenario, and the video search results interface can be a business scenario; the type corresponding to these business scenarios is the scenario type. The account scenario features corresponding to the target account refer to the characteristics of the target account in a certain business scenario, such as the target account's account interests and the target account's behavior in that business scenario. The display object scenario features corresponding to the object to be recommended refer to the features of the scenario in which the object to be recommended is displayed, such as the scenario category and scenario identifier. As an example, in video applications, the scene category could be "Information Feed," and the scene identifier could be "Horizontal Immersive Video Stream." The target account characteristics could include age group, gender, and other attributes; the collection and use of this information requires full authorization and permission. The display object characteristics could include the object's identifier, type, and promoter. For example, if the object to be recommended is an advertisement, the display object characteristics could include the advertisement identifier, advertisement size, advertisement category, and advertiser.

[0053] S103: Perform feature cross processing on the target account features, target recommendation scenario features, and display object features to obtain the first cross feature.

[0054] Understandably, in multi-scenario modeling, it is crucial to fully represent the acquired features. In this embodiment, when representing the acquired target account features, target recommendation scenario features, and display object features, since the density of these different categories of features varies, these features are first cross-referenced to obtain the first cross-feature. Optionally, the method for cross-referencing the target account features, target recommendation scenario features, and display object features includes, but is not limited to, Factorization Machines (FM) and their derivatives, Multiple Layer Perceptrons (MLP), etc.

[0055] S105: The first cross feature is segmented based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature.

[0056] In this application embodiment, in multi-business scenarios, some business scenarios have relatively abundant data, and a larger vector dimension (embedding size) is usually chosen when representing features to ensure that feature information can be fully expressed. However, some business scenarios have very scarce data, and a smaller embedding size is usually chosen when representing features in these scenarios to avoid model overfitting. Therefore, when modeling multiple scenarios, a single embedding size is difficult to take into account the data sparsity of different business scenarios during joint modeling, thus affecting the overall expression of data performance in the recommendation prediction model.

[0057] In this embodiment, when performing multi-scenario modeling, various embedding sizes are set to represent the first cross-feature after cross-referencing. The number of embeddings can be set based on the performance requirements of the recommendation prediction model for the business scenario. For example, the number of business scenarios covered by the model can be calculated by performing a logarithmic operation to base 2, and then compared with 2, taking the larger value as the number of embeddings, i.e., Max(log2(number of scenarios), 2). The dimension (size) of each embedding can be set based on the feature dimension of the business scenario with the largest data volume. As an example, the maximum embedding size can be obtained by taking the square root of the feature dimension with the highest coverage of the business scenario with the largest data volume. For example, taking the fourth power of the feature dimension with the highest coverage determines the maximum embedding size. The feature with the highest coverage can be the identifier of the object to be recommended, the content of the object to be recommended, etc. For example, when the object to be recommended is an advertisement, the maximum embedding size can be determined based on the feature dimension of the advertisement identifier. Assuming the feature dimension of the advertisement identifier is in the tens of millions, the maximum embedding size can be set to 64 dimensions. The sizes of other embeddings can be reduced from the maximum embedding size. For example, assuming there are three embeddings and the maximum embedding size is 64 dimensions, the other embedding sizes can be set to 16 dimensions or 32 dimensions. Similarly, assuming there are three embeddings and the maximum embedding size is 60 dimensions, the other embedding sizes can be set to 10 dimensions or 30 dimensions. In some embodiments, the minimum embedding size can also be obtained by taking the square root of the feature dimension that best covers the smallest data volume of the business scenario. In other embodiments, if the data volume corresponding to the business scenarios targeted by the model is not significantly different, all embedding dimensions can be set to the same dimension.

[0058] In this embodiment, the first cross feature obtained after feature crossing can be segmented according to each embedding size to obtain two or more sub-cross features. The sum of the dimensions of the multiple sub-cross features obtained after segmentation is equal to the dimension of the first cross feature. As an example, the dimension of the first cross feature is a preset dimension; at least two sub-cross features include the first sub-cross feature and the second sub-cross feature. The first cross feature is segmented based on the preset segmentation dimension to obtain a first sub-cross feature with a first dimension and a second sub-cross feature with a second dimension; the preset dimension is equal to the sum of the first and second dimensions. By segmenting the first cross feature obtained after crossing, multiple sub-cross features are obtained, thus enabling the features corresponding to different data volume business scenarios to be fully expressed, thereby improving the accuracy of subsequent recommendation effect prediction.

[0059] S107: Based on at least two sub-cross features, predict the interaction information between the target account and the object to be recommended and displayed, and obtain the predicted interaction information.

[0060] In this embodiment, after segmenting the first cross feature to obtain at least two sub-cross features, interaction information can be predicted based on these at least two sub-cross features to obtain predicted interaction information. This predicted interaction information is used to characterize the interaction behavior between the target account and the object to be recommended, or the interaction probability between the target account and the object to be recommended. Based on the predicted interaction information, the recommendation effect after recommending a target account to the object to be recommended can be obtained.

[0061] In this embodiment, the business scenarios contained in any application are not static and may be affected by policies, trending events, etc., leading to changes in the form of business scenarios. This results in changes in the account feedback corresponding to the business scenario, and consequently, changes in the data distribution and data volume of the displayed objects in the business scenario. Therefore, flexibly and adaptively adapting to possible changes in business scenario data can greatly improve work efficiency in practical applications. For possible changes in business scenario data, this embodiment employs a multi-gating adaptive adjustment method to adjust the weights of different business scenarios. Specifically, when performing multi-scenario modeling, a gating module can be set to assign feature weight data to different business scenarios. The gating module may include one or more gating sub-modules, each corresponding to one or more similar business scenarios. For a gating sub-module corresponding to a business scenario, the gating sub-module sets a feature weight data for each sub-intersecting feature obtained after feature expression, and performs weighted calculations on each sub-intersecting feature to adapt to changes in business scenario data.

[0062] In this embodiment, each gating submodule can set weight data based on the target recommendation scene features and the display object features. Specifically, Figure 3 This is a schematic diagram illustrating the workflow of a gating submodule according to an exemplary embodiment, such as... Figure 3 As shown, the target recommendation scenario features include the account scenario features corresponding to the target account and the display object scenario features corresponding to the object to be recommended. A second cross feature is obtained by performing feature cross processing on the account scenario features, display object scenario features, and display object features; the dimension of the second cross feature is equal to the number of sub-cross features. Then, the second cross feature is normalized to obtain the normalized results of the second cross feature in different dimensions. Based on the normalized results of the second cross feature in different dimensions, feature weight data corresponding to at least two sub-cross features is generated; the feature weight data corresponding to at least two sub-cross features is used to determine the model input features. By performing feature cross processing on the account scenario features, display object scenario features, and display object features to obtain the second cross feature, and then normalizing the second cross feature to obtain the feature weight data corresponding to each sub-cross feature, adaptive adjustment of business scenario weights is achieved. This allows the model to adaptively adapt to changes in business scenario data, improve the prediction accuracy of recommendation effects, and reduce the resource consumption of model adjustment caused by data changes.

[0063] As an example, combined Figure 3 Taking advertisements as an example, the workflow of the gating submodule is explained. Account scene features, display object scene features, and display object features are input into the gating submodule. Optionally, account scene features include account scene behavior and account scene interests, display object scene features include scene identifiers and scene categories, and display object features include advertisement identifiers, advertisement specifications, advertisement categories, and advertisers. The input account scene features, display object scene features, and display object features are cross-calculated to obtain the second cross feature. Optionally, the feature cross-calculation method includes, but is not limited to, factorization machines and their derivatives, multilayer perceptrons, etc. The second cross feature is a vector with a preset dimension, the dimension of which is equal to the number of sub-cross features. Then, the second cross feature is normalized (softmax) and calculated. Each position of the vector yields a weight data, which is the feature weight data corresponding to the sub-cross features. For the weight data obtained at each position, a position-based matching method can be used to match it with the sub-cross features. For example, the first sub-cross feature obtained from the first cross feature segmentation is matched with the first weight data calculated by softmax.

[0064] In this embodiment, when predicting the interaction information between a target account and a recommended display object based on at least two sub-cross features, the model input features are determined by acquiring the feature weight data corresponding to each of the at least two sub-cross features and then concatenating the corresponding sub-cross features with the feature weight data. The model input features are obtained by weighting the corresponding sub-cross features based on the feature weight data and then concatenating them. By inputting the model input features into the target interaction information prediction model, the interaction information between the target account and the recommended display object is predicted, thus obtaining the predicted interaction information. The target interaction information prediction model is obtained by training an initial interaction information prediction model based on training sample data. By obtaining the model input features based on the feature weight data corresponding to each sub-cross feature and each sub-cross feature, and then predicting interaction information based on the model input features, the model can adapt to data changes in the business scenario, thereby improving the accuracy of recommendation effect prediction.

[0065] In this embodiment, the target interaction information prediction model includes at least two preset scenario types, each corresponding to an interaction information prediction sub-model. For a model input feature corresponding to a business scenario, it needs to be input into the corresponding interaction information prediction sub-model for prediction. Specifically, when inputting the model input features into the target interaction information prediction model to predict the interaction information between the target account and the object to be recommended, a target interaction information prediction sub-model corresponding to the target recommendation scenario is determined within the target interaction information prediction model based on the scenario type of the target recommendation scenario. The target recommendation scenario features include the target recommendation scenario, and the scenario type of the target recommendation scenario is one of at least two preset scenario types. By inputting the model input features into the target interaction information prediction sub-model to predict the interaction information between the target account and the object to be recommended, predicted interaction information can be obtained. By determining the target interaction information prediction sub-model corresponding to the target recommendation scenario within the target interaction information prediction model and inputting the model input features into the target interaction information prediction sub-model for interaction information prediction, the process of obtaining predicted interaction information is simplified, and the obtained predicted interaction information has high accuracy and reliability.

[0066] In this embodiment, when performing multi-scenario modeling, due to the significant differences in data distribution across different business scenarios, if the input model data is not optimized, activation functions in different network layers may fail during model training due to data distribution differences, leading to gradient vanishing. To address this issue, normalization can be used to ensure consistent data distribution across network layers, promoting faster model training convergence. Commonly used normalization methods include BatchNorm and LayerNorm. However, for multi-scenario joint modeling, BatchNorm weakens the size relationships between features, reducing the information gain from these relationships across different business scenarios. Therefore, LayerNorm can be used for normalization. Figure 4 This is a schematic diagram illustrating a layer normalization principle according to an exemplary embodiment, such as... Figure 4 As shown, LayerNorm works by setting independent scaling and offset coefficients for each business scenario. Based on these coefficients, all dimensional features of the corresponding scenario are normalized, thus ensuring the personalization of each business scenario. The specific calculation formula for LayerNorm is as follows:

[0067]

[0068] in, The result of the representation layer normalization, a l Indicates the input data, μ l This represents the mean of the input data, (σ) l ) 2 γ represents the variance of the input data. p β represents the scaling factor. p This represents the offset coefficient, where ∈ is a very small constant used to avoid the case where the denominator is zero due to zero variance.

[0069] In this embodiment, the model input features are input into the target interaction information prediction sub-model. The processing performed by the target interaction information prediction sub-model on the model input features includes: inputting the model input features into the target interaction information prediction sub-model to perform layer normalization processing on the model input features to obtain a layer normalization result; and then performing activation processing on the layer normalization result based on a preset activation function to obtain an activation result. Based on the activation result, the predicted interaction information between the target account and the object to be recommended can be determined. By performing layer normalization processing on the model input features, the differences in data distribution across different scenarios can be reduced, and the relative magnitudes of different features can be guaranteed. Activation processing and interaction information prediction based on the layer normalization result can improve the accuracy of predicting interaction information.

[0070] In the embodiments of this application, Figure 5This is a schematic diagram illustrating the structure of a target interaction information prediction model according to an exemplary embodiment, such as... Figure 5 As shown, the target interaction information prediction model includes a feature representation module, a gating module, and a prediction module. The feature representation module transforms the acquired target recommendation scenario features, target account features corresponding to the target account, and display object features corresponding to the object to be recommended, resulting in multiple vector-like sub-cross features. The gating module includes multiple gating sub-modules, which determine the feature weight data corresponding to each sub-cross feature, weight each sub-cross feature based on the feature weight data, and concatenate the weighted results to obtain the model input features. Optionally, the number of gating sub-modules can be the same as the amount of data in the business scenario to which the target interaction information prediction model is applicable; that is, each business scenario corresponds to one gating sub-module. Alternatively, the number of gating sub-modules can differ from the number of business scenarios; in this case, multiple similar business scenarios can correspond to the same gating sub-module. The prediction module includes multiple interaction information prediction sub-models, each corresponding to a business scenario. The model input features obtained after processing by the gating module are input into the corresponding interaction information prediction sub-model for prediction processing to obtain the predicted interaction information.

[0071] In the embodiments of this application, such as Figure 5 As shown, when the input features of the model are used to predict interaction information in the sub-model, they first undergo a normalization layer for normalization, resulting in a normalized result. This normalized result is then input into an activation function layer for activation, yielding the activation result. Optionally, the activation function can be a ReLU function. This activation result is then passed through another normalization layer for normalization, resulting in a normalized result. Finally, this normalized result is input into another activation function layer for activation, yielding the activation result. Optionally, the activation function here can be a Sigmoid function. The activation result obtained through the Sigmoid function can then be used as the output for predicting interaction information.

[0072] In some embodiments, using the activation result obtained from the Sigmoid function as the predicted interaction information in certain business scenarios may have some errors. For example, in ad recommendation, the model's output estimate directly affects ad billing, especially in Optimized Cost Per X (oCPX) mode, where both the order and value of the estimate are required to be accurate. Therefore, in multi-scenario modeling, the activation result obtained from the Sigmoid function can be corrected using corrected data to obtain more accurate predicted interaction information. Specifically, when determining the predicted interaction information between the target account and the object to be recommended, corrected data is first obtained, and then the activation result is corrected based on the corrected data to obtain the corrected result. This corrected result is then used as the predicted interaction information between the target account and the object to be recommended. By correcting the activation result based on the corrected data, the final output predicted interaction information has higher accuracy.

[0073] In this embodiment of the application, when performing multi-scene modeling, correction data can be determined by setting a correction module. The correction data can be determined based on the characteristics of the target recommended scene and the characteristics of the displayed object. Figure 6 This is a schematic diagram illustrating the workflow of a correction module according to an exemplary embodiment, such as... Figure 6 As shown, the target recommendation scenario features include the account scenario features corresponding to the target account. By performing feature cross processing on the account scenario features and the display object scenario features, a third cross feature is obtained, and then corrected data is determined based on this third cross feature. Specifically, the third cross feature is processed using the Sigmoid function to obtain the corrected data. By performing feature cross processing on the account scenario features and the display object scenario features, and determining the corrected data based on the feature cross result, the recommendation effect under different business scenarios can be captured, thereby improving the accuracy of the predicted interaction information obtained through correction.

[0074] As an example, combined Figure 6Taking advertisements as an example, this paper explains the workflow of the correction module. The principle of the correction module is to strengthen the intersection of basic account attributes and business scenario features, supplemented by the basic features of the advertisements to be recommended, to capture the overall advertising effect under different business scenarios. The overall advertising effect can be CTR (Click-Through Rate). Specifically, the account scenario features and the display object scenario features are input into the correction module. Optionally, the account scenario features are a combination of basic account features and scenario features, such as the combination of the target account's age group and scenario identifier, or the combination of gender attributes and scenario identifier. Display object features include the advertisement's identifier, advertisement specifications, advertisement category, advertiser, etc. The account scenario features and display object scenario features input into the correction module are cross-calculated to obtain a third cross feature. Optionally, the feature cross-calculation method includes, but is not limited to, factorization machines and their derivatives, multilayer perceptrons, etc. Then, the feature cross-calculation result is subjected to a nonlinear transformation to obtain the corrected data. The nonlinear transformation can include fully connected processing and Sigmoid function processing. The corrected data can be added to the activation result output by the interaction information prediction sub-model to obtain the final output predicted interaction information.

[0075] In this embodiment, after obtaining the predicted interaction information, the predicted interaction information is judged to determine whether it meets preset conditions. If the predicted interaction information meets the preset conditions, a recommended display object is recommended to the target account. Specifically, the predicted interaction information can be the probability that the target account will interact with the recommended display object. When the predicted interaction information is greater than or equal to a preset probability value, such as 60%, the recommended display object can be recommended to the target account. In some embodiments, the predicted interaction information can also be used to indicate whether the target account will interact with the recommended display object. When the predicted interaction information indicates that the target account will interact with the recommended display object, the recommended display object can be recommended to the target account. By determining whether the predicted interaction information meets the preset conditions, the decision on whether to recommend the recommended display object to the target account can be made, thereby improving the efficiency of information recommendation and reducing the recommendation cost.

[0076] In this embodiment, the target interaction information prediction model for predicting interaction information can be obtained through model training. Specifically, training sample data is first acquired. The training sample data includes sample recommendation scene features, sample account features corresponding to sample accounts, sample display object features corresponding to sample display objects, and sample interaction information. The sample recommendation scene features are the features corresponding to the scene of displaying sample display objects to sample accounts. The sample account is the account receiving the sample display object. The sample interaction information is the interaction information between the sample account and the sample display object. Then, feature cross processing is performed on the sample account features, sample display object features, and sample recommendation scene features to obtain sample cross features. By segmenting the sample cross features based on a preset segmentation dimension, at least two sample sub-cross features are obtained. Then, based on at least two sample sub-cross features and sample interaction information, the initial interaction information prediction model is trained to obtain the target interaction information prediction model. By acquiring training sample data and training the initial interaction information prediction model, the target interaction information prediction model can be obtained. Thus, the target interaction information prediction model can be used to predict interaction information, improving the accuracy and reliability of predicting interaction information.

[0077] In this embodiment of the application, reference is made to Figure 5 As shown, the structure of the initial interaction information prediction model is the same as that of the target interaction information prediction model. The parameters of the initial interaction information prediction model are trained using a training sample dataset to obtain the target interaction information prediction model. During model training, the training sample dataset includes multiple training sample data points. These training sample data points can include data corresponding to all business scenarios. For any training sample data point corresponding to a business scenario, the data is processed through a feature representation module and a gating module. The processed sample model input features are then input into the initial interaction information prediction sub-model corresponding to that business scenario for prediction. The prediction result output by the initial interaction information prediction sub-model is corrected based on the sample correction data to obtain the sample predicted interaction information. Then, the loss data is determined based on the sample predicted interaction information and the sample interaction information, and the model parameters are adjusted based on the loss data. Optionally, the loss data can be determined using the cross-entropy loss function. Through multiple iterations of training, the target interaction information prediction model is obtained when the loss data corresponding to the output result of the initial interaction information prediction sub-model for each business scenario is less than a preset loss value.

[0078] In this embodiment, to address potential new business scenarios during business development, a multi-scenario compatible interaction information prediction sub-model can be set up as a fallback to improve work efficiency. Specifically, referring to Figure 5, an initial multi-scenario compatible interaction information prediction sub-model can be set up within the initial interaction information prediction model. During model training, this multi-scenario compatible initial interaction information prediction sub-model is trained based on all training sample data in the training sample dataset, thereby enabling the trained target multi-scenario compatible interaction information prediction sub-model to be compatible with interaction information prediction for various business scenarios. This target multi-scenario compatible interaction information prediction sub-model can subsequently be task-splitting according to actual business needs, and the split interaction information prediction sub-models can be trained based on the accumulated data to improve their prediction accuracy.

[0079] The interactive information prediction method described in this application enables data sharing between different business scenarios, resulting in more comprehensive information expression. It is also compatible with differences in distribution and data volume across different business scenarios, leading to better information recommendation performance. Furthermore, it can adapt to business changes during dynamic business development, achieving end-to-end learning and improving the maintenance efficiency of the interactive information prediction model.

[0080] This application also provides an interactive information prediction device. Figure 7 This is a block diagram illustrating an interactive information prediction device according to an exemplary embodiment. Figure 7 As shown, the device may include at least:

[0081] The feature acquisition module 201 is used to acquire the target recommendation scenario features, the target account features corresponding to the target account, and the display object features corresponding to the object to be recommended; the target recommendation scenario features are the features corresponding to the scenario of displaying the object to be recommended to the target account; the target account is the account that will receive the object to be recommended.

[0082] The first feature cross module 203 is used to perform feature cross processing on the target account features, target recommendation scenario features and display object features to obtain the first cross feature;

[0083] The feature segmentation module 205 is used to segment the first cross feature based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature.

[0084] The prediction module 207 is used to predict the interaction information between the target account and the object to be recommended based on at least two sub-cross features, so as to obtain the predicted interaction information.

[0085] In some optional embodiments, the dimension of the first cross feature is a preset dimension; at least two sub-cross features include the first sub-cross feature and the second sub-cross feature; the feature segmentation module includes:

[0086] The feature segmentation unit is used to segment the first cross feature based on a preset segmentation dimension to obtain a first sub-cross feature with a first dimension and a second sub-cross feature with a second dimension; the preset dimension is equal to the sum of the first dimension and the second dimension.

[0087] In some optional embodiments, the prediction module includes:

[0088] The feature weight data acquisition submodule is used to acquire the feature weight data corresponding to at least two sub-intersecting features.

[0089] The model input feature determination submodule is used to determine the model input features based on the feature weight data corresponding to each sub-cross feature and each sub-cross feature;

[0090] The prediction submodule is used to input the model input features into the target interaction information prediction model to predict the interaction information between the target account and the object to be recommended and displayed, and obtain the predicted interaction information. The target interaction information prediction model is obtained by training the initial interaction information prediction model based on the training sample data.

[0091] In some optional embodiments, the target recommendation scenario features include account scenario features corresponding to the target account and display object scenario features corresponding to the object to be recommended; the device further includes:

[0092] The second feature cross module is used to perform feature cross processing on account scenario features, display object scenario features, and display object features to obtain the second cross feature; the dimension of the second cross feature is equal to the number of sub-cross features;

[0093] The normalization module is used to normalize the second cross feature to obtain the normalization results of the second cross feature in different dimensions.

[0094] The feature weight data generation module is used to generate feature weight data corresponding to at least two sub-cross features based on the normalization processing results of the second cross features in different dimensions; the feature weight data corresponding to at least two sub-cross features is used to determine the input features of the model.

[0095] In some optional embodiments, the target interaction information prediction model includes at least two preset scene types, each corresponding to an interaction information prediction sub-model; the target recommendation scene features include target recommendation scenes, and the scene type of the target recommendation scene is one of at least two preset scene types; the prediction sub-module includes:

[0096] The target interaction information prediction sub-model determination unit is used to determine the target interaction information prediction sub-model corresponding to the target recommendation scenario in the target interaction information prediction model based on the scenario type of the target recommendation scenario.

[0097] The prediction unit is used to input the model input features into the target interaction information prediction sub-model to predict the interaction information between the target account and the object to be recommended and displayed, and obtain the predicted interaction information.

[0098] In some optional embodiments, the prediction unit includes:

[0099] The prediction sub-unit is used to input the model input features into the target interaction information prediction sub-model, so as to perform layer normalization processing on the model input features to obtain the layer normalization result; and to perform activation processing on the layer normalization result based on the preset activation function to obtain the activation result; and to determine the predicted interaction information between the target account and the object to be recommended based on the activation result.

[0100] In some optional embodiments, the prediction subunit includes:

[0101] The correction data acquisition branch unit is used to acquire correction data;

[0102] The correction branch unit is used to correct the activation results based on the correction data to obtain the correction result.

[0103] The predictive interaction information determination branch unit is used to determine the corrective processing result as the predicted interaction information between the target account and the object to be recommended and displayed.

[0104] In some optional embodiments, the target recommendation scenario features include account scenario features corresponding to the target account; the device further includes:

[0105] The third feature cross module is used to perform feature cross processing on account scenario features and display object scenario features to obtain the third cross feature;

[0106] The correction data determination module is used to determine correction data based on the third cross feature; the correction data is used to correct the activation results to obtain the correction result.

[0107] In some optional embodiments, after predicting the interaction information between the target account and the object to be recommended based on at least two sub-cross features, the apparatus further includes:

[0108] The recommendation module is used to recommend objects to be displayed to the target account when the predicted interaction information meets preset conditions.

[0109] In some optional embodiments, the apparatus further includes a target interaction information prediction model training module, the target interaction information prediction model training module comprising:

[0110] The training sample data acquisition submodule is used to acquire training sample data. The training sample data includes sample recommendation scenario features, sample account features corresponding to sample accounts, sample display object features corresponding to sample display objects, and sample interaction information. The sample recommendation scenario features are the features corresponding to the scenario of displaying sample display objects to sample accounts. The sample account is the account that receives the sample display object. The sample interaction information is the interaction information between the sample account and the sample display object.

[0111] The sample feature cross submodule is used to perform feature cross processing on sample account features, sample display object features, and sample recommendation scenario features to obtain sample cross features;

[0112] The sample feature segmentation module is used to segment sample cross features based on a preset segmentation dimension to obtain at least two sample sub-cross features;

[0113] The model training submodule is used to train the initial interaction information prediction model based on at least two sample sub-cross features and sample interaction information to obtain the target interaction information prediction model.

[0114] It should be noted that the interactive information prediction device embodiment provided in this application is based on the same inventive concept as the interactive information prediction method embodiment described above.

[0115] This application also provides an electronic device for predicting interactive information. The electronic device includes a processor and a memory. The memory stores at least one instruction or at least one program. The processor loads and executes the at least one instruction or at least one program to implement the interactive information prediction method provided in any of the above embodiments.

[0116] Embodiments of this application also provide a computer-readable storage medium that can be disposed in a terminal to store at least one instruction or at least one program for implementing an interactive information prediction method in the method embodiments. The at least one instruction or at least one program is loaded and executed by a processor to implement the interactive information prediction method provided in the above method embodiments.

[0117] Optionally, in the embodiments of this specification, the storage medium may be located at at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0118] The memory described in this specification can be used to store software programs and modules. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system, applications required for functions, etc.; the data storage area may store data created based on the use of the device, etc. Furthermore, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide the processor with access to the memory.

[0119] This application also provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the interactive information prediction method provided in the above-described method embodiments.

[0120] The interactive information prediction method embodiments provided in this application can be executed on a terminal, computer terminal, server, or similar computing device. Taking running on a server as an example, Figure 8 This is a hardware structure block diagram of a server for an interactive information prediction method provided according to an exemplary embodiment. For example... Figure 8As shown, the server 300 can vary significantly due to different configurations or performance. It may include one or more Central Processing Units (CPUs) 310 (CPUs 310 may include, but are not limited to, microprocessors (MCUs) or programmable logic devices (FPGAs), a memory 330 for storing data, and one or more storage media 320 (e.g., one or more mass storage devices) for storing application programs 323 or data 322. The memory 330 and storage media 320 may be temporary or persistent storage. The program stored in the storage media 320 may include one or more modules, each module may include a series of instruction operations on the server. Furthermore, the CPU 310 may be configured to communicate with the storage media 320 and execute the series of instruction operations stored in the storage media 320 on the server 300. Server 300 may also include one or more power supplies 360, one or more wired or wireless network interfaces 350, one or more input / output interfaces 340, and / or one or more operating systems 321, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0121] The input / output interface 340 can be used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of server 300. In one example, the input / output interface 340 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the input / output interface 340 may be a radio frequency (RF) module for wireless communication with the Internet.

[0122] Those skilled in the art will understand that Figure 8 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, server 300 may also include... Figure 8 The more or fewer components shown, or having the same Figure 8 The different configurations shown.

[0123] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0124] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and server embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0125] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0126] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An interactive information prediction method, characterized in that, The method includes: The system acquires target recommendation scenario features, target account features corresponding to the target account, and display object features corresponding to the object to be recommended. The target recommendation scenario features are the features corresponding to the scenario in which the object to be recommended is displayed to the target account, and the target recommendation scenario features include scenario type. The target account is the account that will receive the object to be recommended. The target account features, the target recommendation scenario features, and the display object features are subjected to feature cross-processing to obtain the first cross feature; The first cross feature is segmented based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature. Obtain the feature weight data corresponding to at least two of the sub-intersection features; Based on the feature weight data corresponding to each sub-cross feature and each sub-cross feature, the model input features are determined; The input features of the model are input into the target interaction information prediction model to predict the interaction information between the target account and the object to be recommended and displayed, thereby obtaining the predicted interaction information; the target interaction information prediction model is obtained by training an initial interaction information prediction model based on training sample data; The target recommendation scenario features include the account scenario features corresponding to the target account and the display object scenario features corresponding to the object to be recommended. The account scenario features corresponding to the target account are the features of the target account corresponding to the business scenario. The account scenario features include at least one of account scenario behavior and account scenario interest. The display object scenario features are the features of the scenario to which the object to be recommended is displayed. The method further includes: The account scene features, the display object scene features, and the display object features are subjected to feature cross-processing to obtain a second cross feature; the dimension of the second cross feature is equal to the number of the sub-cross features; The second cross feature is normalized to obtain the normalization results of the second cross feature in different dimensions; Based on the normalization results of the second cross feature in different dimensions, feature weight data corresponding to at least two of the sub-cross features are generated; the feature weight data corresponding to at least two of the sub-cross features are used to determine the model input features.

2. The method according to claim 1, characterized in that, The dimension of the first cross feature is a preset dimension; at least two of the sub-cross features include a first sub-cross feature and a second sub-cross feature; The first cross feature is segmented based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature, including: The first cross feature is segmented based on a preset segmentation dimension to obtain a first sub-cross feature with a first dimension and a second sub-cross feature with a second dimension; the preset dimension is equal to the sum of the first dimension and the second dimension.

3. The method according to claim 1, characterized in that, The sum of the dimensions of the at least two sub-cross features obtained after segmentation is equal to the dimension of the first cross feature.

4. The method according to claim 3, characterized in that, The predicted interaction information is used to characterize the interaction behavior between the target account and the object to be recommended, or the interaction probability between the target account and the object to be recommended.

5. The method according to claim 3, characterized in that, The target interaction information prediction model includes at least two preset scene types, each corresponding to an interaction information prediction sub-model; the target recommendation scene features include target recommendation scenes, and the scene type of the target recommendation scene is one of the at least two preset scene types; the step of inputting the model input features into the target interaction information prediction model to predict the interaction information between the target account and the object to be recommended, and obtaining the predicted interaction information, includes: Based on the scenario type of the target recommendation scenario, a target interaction information prediction sub-model corresponding to the target recommendation scenario is determined in the target interaction information prediction model; The input features of the model are input into the target interaction information prediction sub-model to predict the interaction information between the target account and the object to be recommended and displayed, thereby obtaining the predicted interaction information.

6. The method according to claim 5, characterized in that, The step of inputting the model input features into the target interaction information prediction sub-model to predict the interaction information between the target account and the object to be recommended for display, and obtaining the predicted interaction information, includes: The model input features are input into the target interaction information prediction sub-model to perform layer normalization processing on the model input features to obtain a layer normalization result; and based on a preset activation function, the layer normalization result is activated to obtain an activation result; and based on the activation result, the predicted interaction information between the target account and the object to be recommended is determined.

7. The method according to claim 6, characterized in that, The step of determining the predicted interaction information between the target account and the object to be recommended and displayed based on the activation result includes: Obtain correction data; The activation result is corrected based on the correction data to obtain the correction result. The correction result is determined as the predicted interaction information between the target account and the object to be recommended.

8. The method according to claim 7, characterized in that, The target recommendation scenario features include the account scenario features corresponding to the target account; The method further includes: The account scenario features and the display object scenario features are subjected to feature cross-processing to obtain a third cross feature; The correction data is determined based on the third cross feature; the correction data is used to correct the activation result to obtain the correction result.

9. The method according to claim 1, characterized in that, After obtaining the predicted interaction information, the method further includes: If the predicted interaction information meets the preset conditions, the object to be recommended is recommended to the target account.

10. The method according to any one of claims 3 to 8, characterized in that, The training method for the target interaction information prediction model includes: Acquire training sample data; the training sample data includes sample recommendation scenario features, sample account features corresponding to sample accounts, sample display object features corresponding to sample display objects, and sample interaction information; the sample recommendation scenario features are the features corresponding to the scenario of displaying the sample display object to the sample account; the sample account is the account that receives the sample display object; the sample interaction information is the interaction information between the sample account and the sample display object; The sample account features, the sample display object features, and the sample recommendation scenario features are subjected to feature cross processing to obtain sample cross features; The sample cross features are segmented based on a preset segmentation dimension to obtain at least two sample sub-cross features; Based on at least two sample sub-cross features and the sample interaction information, the initial interaction information prediction model is trained to obtain the target interaction information prediction model.

11. An interactive information prediction device, characterized in that, The device includes: The feature acquisition module is used to acquire target recommendation scene features, target account features corresponding to the target account, and display object features corresponding to the object to be recommended; the target recommendation scene features are the features corresponding to the scene in which the object to be recommended is displayed to the target account, and the target recommendation scene features include scene type; the target account is the account that will receive the object to be recommended. The first feature cross module is used to perform feature cross processing on the target account features, the target recommendation scenario features and the display object features to obtain the first cross feature; The feature segmentation module is used to segment the first cross feature based on a preset segmentation dimension to obtain at least two sub-cross features corresponding to the first cross feature. The prediction module is used to predict the interaction information between the target account and the object to be recommended based on at least two of the sub-cross features, so as to obtain the predicted interaction information. The prediction module includes: The feature weight data acquisition submodule is used to acquire the feature weight data corresponding to at least two sub-intersecting features. The model input feature determination submodule is used to determine the model input features based on the feature weight data corresponding to each sub-cross feature and each sub-cross feature; The prediction submodule is used to input the model input features into the target interaction information prediction model to predict the interaction information between the target account and the object to be recommended and displayed, and obtain the predicted interaction information; the target interaction information prediction model is obtained by training the initial interaction information prediction model based on the training sample data; The target recommendation scenario features include the account scenario features corresponding to the target account and the display object scenario features corresponding to the object to be recommended. The account scenario features corresponding to the target account are the features of the target account corresponding to the business scenario. The account scenario features include at least one of account scenario behavior and account scenario interest. The display object scenario features are the features of the scenario to which the object to be recommended is displayed. The device further includes: The second feature cross module is used to perform feature cross processing on account scenario features, display object scenario features, and display object features to obtain the second cross feature; the dimension of the second cross feature is equal to the number of sub-cross features; The normalization module is used to normalize the second cross feature to obtain the normalization results of the second cross feature in different dimensions. The feature weight data generation module is used to generate feature weight data corresponding to at least two sub-cross features based on the normalization processing results of the second cross features in different dimensions; the feature weight data corresponding to at least two sub-cross features is used to determine the input features of the model.

12. An electronic device for interactive information prediction, characterized in that, The device includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor as described in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction or at least one program, which is loaded and executed by a processor to implement the interactive information prediction method as described in any one of claims 1-10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the interactive information prediction method as described in any one of claims 1-10.

Images