Information processing method and device, equipment and storage medium
By setting multiple sizes of coding features and cross-character features of multiple preset dimensions, and combining with expert networks for feature extraction, the problems of transformation link depth and sparse samples are solved, the efficiency and accuracy of the recommendation system are improved, and the Higher value resource push and user information acquisition efficiency.
Patent Information
- Application Number
- CN202310558382.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-16
- Publication Date
- 2024-11-22
AI Technical Summary
In the prior art, the conversion link is deep and the conversion samples are sparse, which leads to a great test on the business data estimation model, and it is impossible to accurately predict the prediction scores of different users relative to different recommendation information, and it takes a lot of time to search the parameter, resulting in Poor recommendations and low model efficiency.
By setting encoding features of multiple sizes, avoiding time consumption in parameter search and improving prediction accuracy, using cross-feature features and expert networks of multiple preset dimensions to enhance the generalization ability of the model to targets at different levels .
Save parameter search time, improve model efficiency and prediction accuracy, can more accurately estimate conversion rate and click-through rate, and improve recommendation effect and input-output ratio.
Smart Images

Figure CN119005373A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to an information processing method, device, equipment and storage medium. Background Art
[0002] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence technology is widely used in recommendation systems. For example, the multi-objective ranking model of the recommendation system is used to process information of interest to appropriate users, and recommendations are optimized based on multiple business goals to obtain recommendation effects that cannot be obtained by single business goal modeling, thereby improving the input-output ratio of promotion resources. Among them, business prediction data under multiple business goals can include predicted click-through rate, conversion rate, and click-through conversion rate.
[0003] Some current related technologies have deep conversion links and sparse conversion samples, which puts a greater test on the business data estimation model and makes it impossible to accurately predict the prediction scores of different users relative to different recommended information, which further leads to poor recommendation effects. Other related technologies require given candidate feature dimensions and then screen the feature dimensions, so a lot of time is spent on parameter search, resulting in low model efficiency. Summary of the invention
[0004] In order to solve the above technical problems, the present application provides an information processing method, device, equipment and storage medium. The present application sets encoding features of various sizes to meet the needs of different prediction targets, which can not only avoid a lot of time spent on parameter search, but also improve the accuracy of prediction.
[0005] On the one hand, the present application proposes an information processing method, the method comprising:
[0006] Encoding and vectorizing each feature information in the multiple feature information to obtain a coding feature group for each feature information; the coding feature group includes coding features of multiple preset dimensions; the multiple feature information includes feature information of the target object and feature information of the information to be recommended;
[0007] Determine cross-features based on a coded feature group of multiple feature information and multiple preset dimensions;
[0008] The cross-features are processed for information prediction to obtain recommendation prediction results; the recommendation prediction results are used to characterize the predicted recommendation effects of multiple business goals associated with the recommendation system corresponding to the target object after the information to be recommended is processed to the target object.
[0009] Optionally, cross features are determined based on a coding feature group of multiple feature information and multiple preset dimensions, including: determining a feature group of the same dimension based on coding features with consistent dimensions in the coding feature group of multiple feature information; performing feature cross processing on any two coding features in the feature group of the same dimension to obtain cross features of each preset dimension.
[0010] Optionally, information prediction processing is performed on the cross-features to obtain recommended prediction results, including: based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among multiple business targets, feature extraction processing is performed on the cross-features of each preset dimension to obtain network prediction features for each business target; the feature extraction network and the preset dimension correspond one-to-one; the weight network and the business target correspond one-to-one; based on the information prediction model corresponding to each business target, information prediction processing is performed on the network prediction features of each business target to obtain recommended prediction results.
[0011] Optionally, based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among the multiple business targets, feature extraction processing is performed on the cross features of each preset dimension to obtain the network prediction features of each business target, including: based on the feature extraction network corresponding to each preset dimension, feature extraction processing is performed on the cross features of each preset dimension to obtain the network output features corresponding to each preset dimension; based on the weight network corresponding to each business target, the cross features are processed to obtain the network weight of each business target; the network weight includes multiple sub-weights corresponding to multiple feature extraction networks, and the multiple sub-weights correspond to the multiple feature extraction networks one-to-one; based on the network weight of each business target, the network output features corresponding to each preset dimension are weighted summed to obtain the network prediction features of each business target.
[0012] Optionally, information prediction processing is performed on the network prediction features of each business target based on the information prediction model corresponding to each business target to obtain a recommended prediction result, including: based on the information prediction model corresponding to each business target, information prediction processing is performed on the network prediction features of each business target to obtain initial prediction information of multiple business targets; the information prediction model and the business target correspond one-to-one; the initial prediction information is normalized to obtain a recommended prediction result.
[0013] Optionally, the method also includes: obtaining a training coding feature group of multiple feature information; the training coding feature group carries annotation information, the annotation information includes sample prediction results, and the sample prediction results are used to characterize the recommendation effects of multiple business objectives after the information to be recommended is processed to the target object; through a preset machine learning model, based on the training coding feature group of multiple feature information and multiple preset dimensions, the training cross-features are determined; information prediction processing is performed on the training cross-features to obtain training prediction results; based on the loss value between the training prediction results and the sample prediction results, the preset machine learning model is trained until a preset end condition is met to obtain an information processing model; wherein, the training combination to be cross-linked includes coding features of the same dimension in the training coding feature group of any two feature information; the information processing model is used to determine the combination to be cross-linked until the recommendation prediction result is obtained.
[0014] Optionally, after performing information prediction processing on the cross-features of multiple preset dimensions and obtaining a recommendation prediction result, the method also includes: sending a recommendation instruction to the recommendation system corresponding to the target object when the recommendation prediction result meets the recommendation condition; the recommendation instruction is used to instruct the recommendation system corresponding to the target object to recommend the information to be recommended to the target object.
[0015] On the other hand, an embodiment of the present application provides an information processing device, the device comprising:
[0016] The encoding module is used to encode and vectorize each feature information of the multiple feature information to obtain a coding feature group of each feature information; the coding feature group includes coding features of multiple preset dimensions; the multiple feature information includes feature information of the target object and feature information of the information to be recommended;
[0017] A cross module, used for determining cross features based on a coding feature group of multiple feature information and multiple preset dimensions;
[0018] The prediction module is used to perform information prediction processing on the cross-features to obtain the recommendation prediction results; the recommendation prediction results are used to characterize the predicted recommendation effects of multiple business objectives associated with the recommendation system corresponding to the target object after the information to be recommended is processed to the target object.
[0019] Optionally, the cross module is used to: determine the same dimension feature group based on the coding features with consistent dimensions in the coding feature groups of multiple feature information; perform feature cross processing on any two coding features in the same dimension feature group to obtain cross features of each preset dimension.
[0020] Optionally, the prediction module is used to: based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among multiple business targets, perform feature extraction processing on the cross-features of each preset dimension to obtain the network prediction features of each business target; the feature extraction network and the preset dimensions correspond one-to-one; the weight network and the business targets correspond one-to-one; based on the information prediction model corresponding to each business target, perform information prediction processing on the network prediction features of each business target to obtain a recommended prediction result.
[0021] Optionally, the prediction module is used to, in the process of performing feature extraction processing on the cross features of each preset dimension based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among multiple business targets to obtain the network prediction features of each business target: based on the feature extraction network corresponding to each preset dimension, performing feature extraction processing on the cross features of each preset dimension to obtain the network output features corresponding to each preset dimension; based on the weight network corresponding to each business target, processing the cross features to obtain the network weight of each business target; the network weight includes multiple sub-weights corresponding to multiple feature extraction networks, and the multiple sub-weights correspond to multiple feature extraction networks one-to-one; based on the network weight of each business target, weighted summing processing is performed on the network output features corresponding to each preset dimension to obtain the network prediction features of each business target.
[0022] Optionally, the prediction module is used to, in the process of performing information prediction processing on the network prediction features of each business target based on the information prediction model corresponding to each business target to obtain recommended prediction results: based on the information prediction model corresponding to each business target, perform information prediction processing on the network prediction features of each business target to obtain initial prediction information for multiple business targets; the information prediction models and business targets correspond one-to-one; the initial prediction information is normalized to obtain recommended prediction results.
[0023] Optionally, the device also includes a training module, which is used to: obtain a training coding feature group of multiple feature information; the training coding feature group carries annotation information, the annotation information includes a sample prediction result, and the sample prediction result is used to characterize the recommendation effect of multiple business objectives after the information to be recommended is processed to the target object; through a preset machine learning model, through the preset machine learning model, determine the training cross-features based on any two training coding feature groups in the training coding feature group and the multiple preset dimensions; perform information prediction processing on the training cross-features to obtain a training prediction result; based on the loss value between the training prediction result and the sample prediction result, train the preset machine learning model until the preset end condition is met to obtain the information processing model; wherein, the training combination to be cross-linked includes the coding features of the same dimension in the training coding feature groups of any two feature information; the information processing model is used to determine the combination to be cross-linked until the recommendation prediction result is obtained.
[0024] Optionally, the device also includes a recommendation module, which is used to send a recommendation instruction to the recommendation system corresponding to the target object when the recommendation prediction result meets the recommendation condition; the recommendation instruction is used to instruct the recommendation system corresponding to the target object to recommend the information to be recommended to the target object.
[0025] On the other hand, the present application proposes an electronic device for information processing, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the information processing method as described above.
[0026] On the other hand, the present application proposes a computer-readable storage medium, which stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by a processor to implement the information processing method as described above.
[0027] The information processing method, device, electronic device and storage medium proposed in the embodiment of the present application obtain embedding vectors of several preset dimensions through feature query, which can greatly save the time consumed in parameter search and improve the model efficiency. Moreover, multiple expert networks and multiple gated networks are used for feature extraction, wherein the expert network is used to extract different features, and the gated network is equivalent to a classifier for assigning the weight of each expert network, so that the recommendation effect of each business goal can be predicted in a targeted manner; at the same time, by inputting cross-features of multiple preset dimensions and establishing different expert networks according to the cross-features of different preset dimensions, the expert network is used to process the cross-features corresponding to the same preset dimension as the expert network, so as to improve the generalization ability of the model for different levels of targets and enhance the model prediction effect. At the same time, the embodiment of the present application more accurately estimates the conversion rate, click-through rate and other data, so that the push resources or recommended information with higher value are expanded to the recall results, so as to expose and display high-value resources on the client and other delivery terminals, which can not only improve the input-output ratio of the delivery party, but also improve the information acquisition efficiency of users at the terminal. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art description are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.
[0029] Figure 1 is a schematic diagram of an implementation environment of an information processing method according to an exemplary embodiment;
[0030] Figure 2 is a flowchart of an information processing method according to an exemplary embodiment;
[0031] Figure 3 is a flow chart of a method for determining cross-features of an information processing method according to an exemplary embodiment;
[0032] Figure 4 is a flowchart of an information prediction processing method of an information processing method according to an exemplary embodiment;
[0033] Figure 5 is a flow chart of a feature extraction processing method of an information processing method according to an exemplary embodiment;
[0034] Figure 6is a flowchart of an information prediction processing method of an information processing method according to an exemplary embodiment;
[0035] Figure 7 is a flowchart of a model training method of an information processing method according to an exemplary embodiment;
[0036] Figure 8 is a structural schematic diagram of an information prediction model of an information processing method according to an exemplary embodiment;
[0037] Fig. 9 The invention is a block diagram of an information processing device according to an exemplary embodiment.
[0038] Fig.10 It is a hardware structure block diagram of a server of an information processing method provided according to an exemplary embodiment. DETAILED DESCRIPTION
[0039] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.
[0040] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes image semantic understanding technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0041] Specifically, the information processing method described in the embodiments of the present application involves AIMachine Learning (ML) in the field. Natural language processing is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language used by people in daily life, so it is closely related to the study of linguistics. Machine learning is a multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning by teaching.
[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0043] It should be noted that the terms "first", "second", etc. in the specification and claims of the embodiments of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.
[0044] In order to make the purpose, technical solution and advantages disclosed in the embodiments of the present application more clearly understood, the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present application and are not used to limit the embodiments of the present application.
[0045] In the following, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this embodiment, unless otherwise specified, "multiple" means two or more. In order to facilitate the understanding of the above-mentioned technical solutions and the technical effects produced by the embodiments of the present application, the embodiments of the present application first explain the relevant professional terms:
[0046] Shallow conversion: generally refers to the conversion behavior that occurs within the advertising domain and is the advertiser’s target.
[0047] Post-advertising link conversion data: generally refers to user behaviors outside the advertising domain, which are closer to the advertiser's business goals, such as placing orders, making payments, visiting stores, etc.
[0048] Post-advertising link conversion effect: Post-advertising link effect generally refers to the effect that occurs outside the advertising domain and is close to the advertiser's business goals, such as order rate, store visit rate, etc.
[0049] AutoML: Automatic machine learning. From the perspective of machine learning, AutoML can be seen as a system with very strong learning and generalization capabilities on given data and tasks. However, it emphasizes that it must be very easy to use. From the perspective of automation, AutoML can be seen as designing a series of advanced control systems to operate machine learning models, so that the models can automatically learn the appropriate parameters and configurations without human intervention.
[0050] Overfitting: Overfitting means that the gap between the training error and the test error is too large. In other words, the model complexity is higher than the actual problem. The model performs well on the training set, but performs poorly on the test set. The model "memorizes" the training set by rote, remembering the properties or characteristics of the training set that are not applicable to the test set, without understanding the rules behind the data, and has poor generalization ability.
[0051] MMoE (Multi-gate Mixture-of-Experts): is a method proposed by Google to model the relationship between different tasks in multi-task learning. The MMoE model uses a mixed expert sub-network with multiple gate controls. The parameters of the mixed expert sub-network are shared between different tasks, but each task controls the importance of different expert sub-networks through different gates, thereby alleviating the conflict between different learning tasks.
[0052] In the field of artificial intelligence, the post-link conversion effect of resource delivery and information recommendation can be estimated through data modeling. Information delivery based on the post-link conversion effect can improve the recommendation effect, thereby increasing the input-output ratio of the publisher and improving the user experience. The current pain points of the post-link algorithm include: the publisher, such as advertisers, usually only sends back tens of thousands of deep post-link samples, which not only has a long conversion link, but also sparse conversion samples, which puts a great test on the model; the overfitting phenomenon of deep conversion is relatively serious; a lot of time is spent on parameter search, resulting in high training time.
[0053] In view of this, the embodiments of the present application provide an information processing method, device, electronic device and storage medium, which realize efficient dimension screening and network parameter training by adopting coding features of multiple preset dimensions, thereby avoiding a large amount of time consumed on parameter search and making the model online less time-consuming; and, since different prediction targets have different requirements for feature dimensions, the embodiments of the present application set coding features of multiple sizes to meet the needs of different prediction targets, which can not only avoid a large amount of time consumed on parameter search, but also improve the generalization ability of the model for multiple targets, thereby improving the accuracy of the prediction.
[0054] Figure 1 FIG. 1 is a schematic diagram of an implementation environment of an information processing method according to an exemplary embodiment. 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, and this application does not impose any limitation thereto.
[0055] Specifically, the client 01 can be used to receive input of text to be recognized, display search results, etc. Optionally, the client 01 can include physical devices such as smart phones, desktop computers, tablet computers, laptop computers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, intelligent voice interaction devices, smart home appliances, smart wearable devices, vehicle-mounted terminal devices, etc., and can also include software running in physical devices, such as applications, etc.
[0056] Specifically, the server 02 can be used to process the back-link data, including encoding and vectorizing each feature information, and then performing feature cross-talk and recommendation prediction, so as to obtain the prediction and recommendation results of multiple business objectives. The server 02 can also be used to train the information processing model and send recommendation instructions to the recommendation system corresponding to the target object, wherein the recommendation instruction can be used to instruct the recommendation system to recommend information. 。Optionally, server 02 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. Specifically, the servers involved above may include physical devices, which may specifically include network communication submodules, processors and memories, etc., and may also include software running in physical devices, which may specifically include applications, etc.
[0057] It should be noted that Figure 1 This is just an example. In other scenarios, other implementation environments may also be included.
[0058] Figure 2 is a flowchart of an information processing method according to an exemplary embodiment. The method can be used for Figure 1 In the implementation environment of the embodiment. This specification provides the information processing method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on routine or non-creative work. The order of steps listed in the embodiment is only one way of executing the steps among many steps, and does not represent the only execution order. When the actual system or server product is executed, it can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment) according to the method shown in the embodiment or the accompanying drawings. Specifically, Figure 2 As shown, the method may include:
[0059] S101: Encode and vectorize each feature information among a plurality of feature information to obtain a coding feature group for each feature information.
[0060] In an optional embodiment, the coding feature group may include coding features of multiple preset dimensions. As an example, the preset dimensions may include 16 dimensions, 32 dimensions, and 64 dimensions. It should be noted that the present application does not limit the number of preset dimensions and specific dimension values. In other optional embodiments, the preset dimensions may also be other numbers and other dimensions, for example, 64 dimensions and 128 dimensions.
[0061] Optionally, the feature information may include continuous features and discrete features. In the case where the parameter feature includes a continuous feature, step S101 may also include: discretizing the continuous feature to obtain discrete features of the continuous feature, and then encoding and vectorizing the discrete features of the continuous feature and the discrete features to obtain a coded feature group for each feature information.
[0062] In step S101, by encoding each feature information in the multiple feature information, each feature information of the target object and the information to be recommended corresponds to a feature code, and then querying multiple preset dimension embedding vectors for each feature code, a coding feature group of each feature information can be obtained. Taking an embodiment in which the preset dimensions include 16 dimensions, 32 dimensions, and 64 dimensions as an example, in step S101, a 16-dimensional embedding vector, a 32-dimensional embedding vector, and a 64-dimensional embedding vector corresponding to each feature information can be obtained through feature query.
[0063] In an optional implementation, the feature information may include the feature information of the target object and the feature information of the information to be recommended. The feature information of the target object may include the object features of the target object, and may specifically include: basic attribute features of the target object, such as age, gender, occupation, education level, region, consumption level, etc.; portrait features, such as hobbies, browsing, clicking, collection, purchase and other behavior data; environmental features of the recommended information, such as recommendation time, recommendation scene, etc. The feature information of the information to be recommended may include the information features of the information to be recommended, and may specifically include: information tags, information categories, information sources, and information content of the information to be recommended. Optionally, the feature information may also include the interaction features of the target object and the information to be recommended, wherein the interaction features refer to the intersection of information features and object features, such as the consumption of the information to be recommended by users of different ages and genders, and the consumption features of the users of different user ages and genders for the recommended information. The consumption features are the interaction features between the information to be recommended and the users.
[0064] The current solution usually requires first giving candidate embedding vectors of different dimensions, then screening them through AutoML, and then using the screened embedding dimensions for subsequent model training and model processing, which consumes a lot of time on parameter search, resulting in high time consumption. In the embodiment of the present application, embedding vectors of several preset dimensions are obtained through feature query, which can greatly save the time spent on parameter search and improve model efficiency.
[0065] S102: Determine cross features based on a plurality of encoding feature groups of feature information and a plurality of preset dimensions.
[0066] In an optional implementation, the NFwFFM (Neural Field-weighted Field-aware Factorization Machines) structure can be used to perform feature crossover on the coding features in any two coding feature groups, and the dimension of the new vector obtained after the crossover is still the preset dimension. For example, in an embodiment where the coding features in each coding feature group are 16-dimensional, 32-dimensional, or 64-dimensional, the dimension of the crossover feature obtained by feature crossover is also 16-dimensional, 32-dimensional, or 64-dimensional.
[0067] Optionally, cross-features may be determined based on the coding feature group of the target object and the coding feature group of the information to be recommended, respectively. Specifically, feature cross-features may be performed between the coding features of the target object and feature cross-features may be performed between the coding features of the information to be recommended.
[0068] The following is based on Figure 3 The exemplary process of step S102 is further introduced.
[0069] Figure 3 FIG. 1 is a flow chart of a method for determining cross-features of an information processing method according to an exemplary embodiment. Specifically, Figure 3 As shown in the figure, the cross-features are determined based on the coded feature groups of multiple feature information and multiple preset dimensions, including:
[0070] S1021: Determine a feature group with the same dimension based on the coding features with the same dimension in the coding feature groups of the plurality of feature information.
[0071] Optionally, feature crossover can be performed based on coding features of the same dimension, so firstly, coding features of the same dimension can be determined from coding features corresponding to all feature information, and the coding features of the same dimension can be determined as a feature group of the same dimension, wherein the dimensions of the coding features in the feature group of the same dimension are consistent.
[0072] As an example, taking the embodiment in which each coding feature group includes 16-dimensional features, 32-dimensional features, and 64-dimensional features, a 16-dimensional feature group of the same dimension can be formed by the 16-dimensional features in the coding feature groups of all feature information, a 32-dimensional feature group of the same dimension can be formed by the 32-dimensional features in the coding feature groups of all feature information, a 64-dimensional feature group of the same dimension can be formed by the 64-dimensional features in the coding feature groups of all feature information, and a 16-dimensional feature group of the same dimension, a 32-dimensional feature group of the same dimension, and a 64-dimensional feature group of the same dimension can be obtained. Among them, the 16-dimensional feature group of the same dimension includes the 16-dimensional coding features in all coding feature groups, the 32-dimensional feature group of the same dimension includes the 32-dimensional coding features in all coding feature groups, and the 64-dimensional feature group of the same dimension includes the 64-dimensional coding features in all coding feature groups.
[0073] S1022: Perform feature cross processing on any two coding features in the same dimensional feature group to obtain cross features of each preset dimension.
[0074] Optionally, feature cross processing can be performed on any two coding features in the same dimensional feature group to obtain cross features of each preset dimension. The feature cross process can include processing two coding features by point multiplication, that is, multiplying any two coding features. Among them, the relevant information of each feature encoding can also be introduced in the point multiplication, that is, in the process of multiplying two coding features, the first coding feature, the relevant information of the first coding feature, the second coding feature, and the relevant information of the second coding feature can be multiplied to obtain the cross feature.
[0075] Next, we will continue based on Figure 3 To elaborate:
[0076] S103: Perform information prediction processing on the cross features to obtain a recommendation prediction result.
[0077] Optionally, information prediction processing can be performed on the cross-features to obtain recommendation prediction results. The recommendation prediction results can be used to characterize the predicted recommendation effects of multiple business goals associated with the recommendation system corresponding to the target object after the information to be recommended is processed to the target object. The business goal can include at least one of clicks, conversions, duration, interactions, etc. The recommendation prediction results can be the predicted results, including click-through rate, conversion rate, click-through conversion rate, etc.
[0078] As an example, the MMoE network can be used to perform information prediction processing to obtain recommendation prediction results. Figure 4 An exemplary process of step S103 is introduced. Figure 4 FIG. 1 is a flow chart of an information prediction processing method according to an exemplary embodiment. Figure 4 , perform information prediction processing on the cross features to obtain the recommended prediction results, including:
[0079] S1031: Based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among the multiple business targets, feature extraction processing is performed on the cross features of each preset dimension to obtain the network prediction features of each business target.
[0080] Optionally, the feature extraction network corresponds to the preset dimension one by one, and the weight network corresponds to the business goal one by one.
[0081] As an example, the MMoE network can be used to extract features from cross features to obtain network prediction features. The MMoE network can dynamically adjust the weights of multi-dimensional expert networks under different optimization objectives for different optimization tasks. Specifically, different weights are used to combine expert models of different dimensions. Figure 5 The exemplary process of step S1031 in this embodiment is further introduced. Figure 5 FIG. 1 is a flow chart of a feature extraction method of an information processing method according to an exemplary embodiment. Figure 5 Based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among the multiple business targets, feature extraction processing is performed on the cross features of each preset dimension to obtain the network prediction features of each business target, including:
[0082] S10311: Based on the feature extraction network corresponding to each preset dimension, feature extraction processing is performed on the cross features of each preset dimension to obtain the network output features corresponding to each preset dimension.
[0083] Optionally, based on the feature extraction network corresponding to each preset dimension, feature extraction processing can be performed on the cross features of each preset dimension to obtain the network output features corresponding to each preset dimension. The feature extraction network can be an expert network, and each expert network can correspond to a preset dimension.
[0084] As an example, in an embodiment where the preset dimensions include 16 dimensions, 32 dimensions, and 64 dimensions, the feature extraction network may include an expert network corresponding to 16 dimensions, an expert network corresponding to 32 dimensions, and an expert network corresponding to 64 dimensions, and the above-mentioned expert networks are used to process 16-dimensional cross features, 32-dimensional cross features, and 64-dimensional cross features, respectively. Optionally, in step S10311, the expert network corresponding to 16 dimensions can perform feature extraction processing on the expert network corresponding to 16 dimensions to obtain 16-dimensional network output features; the expert network corresponding to 32 dimensions can perform feature extraction processing on the expert network corresponding to 32 dimensions to obtain 32-dimensional network output features; the expert network corresponding to 64 dimensions can perform feature extraction processing on the expert network corresponding to 64 dimensions to obtain 64-dimensional network output features. It should be noted that the above embodiment is only an example. In some other optional implementations, as long as the feature extraction network and the preset dimensions correspond one-to-one and are the same in number, the feature extraction network can also be other numbers, the preset dimensions can also be other numbers, and the preset dimensions can also include other dimensions.
[0085] S10312: Based on the weight network corresponding to each business objective, the cross-features are processed to obtain the network weight of each business objective.
[0086] Optionally, the cross-features of each business objective may be processed based on the weight network corresponding to each business objective to obtain the network weight of each business objective. The network weight includes multiple sub-weights corresponding to multiple feature extraction networks, and the multiple sub-weights correspond to the multiple feature extraction networks one-to-one. Optionally, the weight network is a gated network, and each gated network may correspond to one business objective.
[0087] As an example, in an embodiment where the business goals include clicks and conversions, the weight network may include a gating network corresponding to the business goal-click and a gating network corresponding to the business goal-conversion. Optionally, in step S10312, the cross-features may be processed based on the gating network corresponding to the business goal-click to obtain the network weight of the business goal-click; the cross-features may be processed based on the gating network corresponding to the business goal-conversion to obtain the network weight of the business goal-conversion. It should be noted that the above embodiment is only an example. In other optional implementations, as long as the business goals and weight networks correspond one-to-one and are the same in number, the weight networks may also be other numbers, the business goals may also be other numbers, and the business goals may also include other businesses.
[0088] S10313: Perform weighted sum processing on the network output features corresponding to each preset dimension based on the network weight of each business target to obtain the network prediction features of each business target.
[0089] Optionally, the network output features of different feature extraction network outputs can be mapped to 64 dimensions through a fully connected network, and then each network output feature mapped to 64 dimensions is weighted and summed based on the network weight of each business target to obtain the network prediction feature of each business target.
[0090] As an example, there is a network weight for each business goal, and the network weight includes the sub-weight of each feature extraction network, that is, each self-weight represents the weight of the output of the feature extraction network corresponding to the sub-weight in the network prediction feature. In the embodiment where the preset dimensions include 16 dimensions, 32 dimensions and 64 dimensions, and the business goals include clicks and conversions, the network output features of the feature extraction network corresponding to the 16 dimensions after full connection mapping, the network output features of the feature extraction network corresponding to the 32 dimensions after full connection mapping, and the network output features of the feature extraction network corresponding to the 64 dimensions after full connection mapping can be weighted and summed based on the network weight corresponding to the business goal-click to obtain the network prediction feature of the business goal-click; based on the network weight corresponding to the business goal-conversion, the network output features of the feature extraction network corresponding to the 16 dimensions after full connection mapping, the network output features of the feature extraction network corresponding to the 32 dimensions after full connection mapping, and the network output features of the feature extraction network corresponding to the 64 dimensions after full connection mapping can be weighted and summed to obtain the network prediction feature of the business goal-conversion. It should be noted that the above embodiment is only an example. In some other optional implementations, as long as the feature extraction network and the preset dimensions correspond one-to-one and are the same in number, the feature extraction network can also be other numbers, the preset dimensions can also be other numbers, and the preset dimensions can also include other dimensions; as long as the business objectives and the weight networks correspond one-to-one and are the same in number, the weight network can also be other numbers, the business objectives can also be other numbers, and the business objectives can also include other businesses.
[0091] In an optional implementation, the network prediction feature of the kth service target can be expressed as:
[0092]
[0093] Among them, n is the number of feature extraction networks;
[0094] FC 64 A fully connected network for mapping the network output features of the feature extraction network output to a 64-dimensional network;
[0095] f i (x) is the network output feature output by the feature extraction network;
[0096] g k (x) i It is the sub-weight corresponding to the i-th feature extraction network among multiple sub-weights of the network weight.
[0097] In an embodiment of the present application, multiple expert networks and multiple gating networks can be used for feature extraction, wherein the expert network is used to extract different features, and the gating network is equivalent to a classifier for assigning weights to each expert network, so as to predict the recommendation effect of each business goal in a targeted manner; at the same time, by inputting cross-features of multiple preset dimensions and establishing different expert networks according to the cross-features of different preset dimensions, the expert network can be used to process the cross-features corresponding to the same preset dimension as the expert network, so as to improve the generalization ability of the model for targets at different levels and enhance the prediction effect of the model.
[0098] Next, we will continue based on Figure 4 To elaborate:
[0099] S1032: Perform information prediction processing on the network prediction features of each business target based on the information prediction model corresponding to each business target to obtain a recommended prediction result.
[0100] Optionally, information prediction processing may be performed on the network prediction features of each business objective based on the information prediction model corresponding to each business objective to obtain a recommended prediction result.
[0101] As an example, the network prediction features of each business goal can be input into the exclusive tower corresponding to each business goal for learning, and then normalized to obtain the recommended prediction results for each business goal. Figure 6 An exemplary process of step S1032 is introduced. Figure 6 FIG. 1 is a flow chart of an information prediction processing method according to an exemplary embodiment. Specifically, Figure 6 As shown in the figure, based on the information prediction model corresponding to each business goal, the network prediction features of each business goal are processed for information prediction to obtain recommended prediction results, including:
[0102] S10321: Based on the information prediction model corresponding to each business goal, information prediction processing is performed on the network prediction features of each business goal to obtain initial prediction information of multiple business goals.
[0103] Optionally, based on the dedicated tower corresponding to each business target, information prediction processing can be performed on the network prediction characteristics of each business target to obtain initial prediction information of multiple business targets. The dedicated towers can correspond to the business targets one by one.
[0104] In an optional implementation, the information prediction model may include a fully connected network, a multi-layer fully connected neural network, etc.
[0105] S10322: Normalize the initial prediction information to obtain a recommended prediction result.
[0106] Optionally, the initial prediction information may be normalized to obtain a recommended prediction result. In an optional implementation, the initial prediction information of each business target may be processed by a softmax function to obtain a recommended prediction result.
[0107] As an example, in an embodiment where the business goals include clicks and conversions, the initial prediction information corresponding to the business goal-clicks and the business goal-conversion can be normalized by executing step S10322 to obtain the click rate and conversion rate, respectively.
[0108] In an embodiment of the present application, by establishing different expert networks according to the cross-features of different preset dimensions, and predicting the exclusive tower corresponding to each business goal to obtain the recommended prediction data for each business, a model with stronger generalization ability can be used to achieve more accurate information prediction and obtain more accurate recommendation prediction results.
[0109] Next, we will continue based on Figure 2 To elaborate, such as Figure 2 As shown, the method may further include:
[0110] S104: When the recommendation prediction result meets the recommendation condition, a recommendation instruction is sent to the recommendation system corresponding to the target object.
[0111] Optionally, when the recommendation prediction result meets the recommendation condition, a recommendation instruction is sent to the recommendation system corresponding to the target object. The recommendation instruction is used to instruct the recommendation system corresponding to the target object to recommend the information to be recommended to the target object.
[0112] As an example, the recommendation prediction result may include a click-through rate estimate, and the recommendation condition may be that the click-through rate is greater than a preset click-through rate threshold. In step S104, if the click-through rate estimate is greater than the preset click-through rate threshold, a recommendation instruction may be sent to the recommendation system corresponding to the target object.
[0113] In some optional implementations, for different recommendation systems, the recommendation conditions are different, and step S104 may further include: determining the recommendation conditions according to the recommendation system corresponding to the target object; and sending a recommendation instruction to the recommendation system corresponding to the target object when the recommendation prediction result meets the recommendation conditions. The recommendation instruction is used to instruct the recommendation system corresponding to the target object to recommend the information to be recommended to the target object.
[0114] In the embodiment of the present application, by utilizing a model with stronger generalization ability to achieve more accurate information prediction, data such as conversion rate and click-through rate can be more accurately estimated, so that push resources or recommended information with higher value can be expanded to the recall results, so as to expose and display high-value resources on delivery terminals such as clients, which can not only improve the input-output ratio of the delivery party, but also improve the information acquisition efficiency of users on the terminal.
[0115] The following is based on Figure 7 An information processing method according to an embodiment of the present application is further introduced. Figure 7 1 is a flow chart of a model training method of an information processing method according to an exemplary embodiment. In an optional implementation, an information processing method of the present application may further include:
[0116] S201: Obtain a training coding feature group of multiple feature information.
[0117] In an optional implementation, the training coding feature group may carry annotation information, and the annotation information may include sample prediction results, wherein the sample prediction results are used to characterize the recommendation effects of multiple business objectives after the information to be recommended is processed to the target object.
[0118] In an optional implementation, the training coding feature group may include training coding features of multiple preset dimensions. As an example, the preset dimensions may include 16 dimensions, 32 dimensions, and 64 dimensions. It should be noted that the present application does not limit the number of preset dimensions and specific dimension values. In other optional implementations, the preset dimensions may also be other numbers and other dimensions, for example, 64 dimensions and 128 dimensions.
[0119] The specific process of obtaining the training coding features can refer to the relevant description of encoding and vectorizing the feature information in step S101, which will not be repeated here.
[0120] S202: Determine training cross features based on a training coding feature group of multiple feature information and multiple preset dimensions through a preset machine learning model; perform information prediction processing on the training cross features to obtain training prediction results.
[0121] Optionally, by using a preset machine learning model, the training cross-features are determined based on any two training coding feature groups in the training coding feature group and the multiple preset dimensions, including: determining the same dimension feature group of each preset dimension in multiple preset dimensions based on any two training coding feature groups; performing feature cross-processing on any two coding features in the same dimension feature group of each preset dimension to obtain the training cross-features of each preset dimension. Specifically, the specific process of obtaining the training cross-features based on the training coding feature group in this step can refer to the relevant description of obtaining the cross-features based on the coding feature group in step S102 above, which will not be repeated here.
[0122] Optionally, information prediction processing is performed on the training cross features to obtain training prediction results, including: based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among multiple business targets, feature extraction processing is performed on the cross features of each preset dimension to obtain network prediction features for each business target; based on the information prediction model corresponding to each business target, information prediction processing is performed on the network prediction features of each business target to obtain recommended prediction results. Specifically, in this step, the specific process of performing information prediction processing on the training cross features to obtain training prediction results can refer to the relevant description of performing information prediction processing on the cross features to obtain recommended prediction results in step S103 above, which will not be repeated here.
[0123] S203: Based on the loss value between the training prediction result and the sample prediction result, the preset machine learning model is trained until the preset end condition is met to obtain the information prediction model.
[0124] In an optional implementation, the preset machine learning model can be trained based on the loss value between the training prediction result and the sample prediction result until the preset end condition is met to obtain the information prediction model. The training cross-combination includes the coding features of the same dimension in the training coding feature group of any two feature information. The information prediction model is used to determine the cross-combination until the recommended prediction result is obtained.
[0125] The following is based on Figure 8 The information prediction model in the embodiments of the present application is introduced. Figure 8 FIG. 1 is a schematic diagram showing a structure of an information prediction model of an information processing method according to an exemplary embodiment. Figure 8 As shown in the figure, the information prediction model may include a feature layer 810, a feature cross layer, a feature extraction layer, and an information prediction layer. The feature cross layer includes a first cross layer 821, a second cross layer 822, and a third cross layer 823; the feature extraction layer includes a first extraction layer 831, a second extraction layer 832, and a third extraction layer 833; and the information prediction layer includes a first prediction layer 841 and a second prediction layer 842. Figure 8 In the illustrated embodiment, there are two business objectives and three preset dimensions. Figure 8 This is only an optional example. In other optional implementations, as long as the number of cross-layers of the feature cross-layer, the number of extraction layers of the feature extraction layer and the number of preset dimensions are consistent, and the number of prediction layers of the information prediction layer and the number of business objectives are consistent, the number of layers can be other numbers depending on the preset dimensions and the number of business objectives.
[0126] In an optional implementation, step S101 described above can be performed at the feature layer 810 to obtain a feature coding group for each feature information. Optionally, a feature query method can be used to obtain feature codes of different dimensions in the feature coding group. The specific process can refer to the description of step S101 above, which will not be repeated here.
[0127] Optionally, step S102 may be performed at a feature cross-layer, and feature cross-layers of three preset dimensions are performed at a first cross-layer 821, a second cross-layer 822, and a third cross-layer 823, respectively, to obtain cross-features of three preset dimensions. Optionally, the feature cross-layer may include a NFwFFM network, and the specific process may refer to the description of step S102 above, which will not be repeated here.
[0128] Step S103 can be performed in the feature extraction layer and the information prediction layer, and feature extraction is performed on the cross features of the three preset dimensions in the first extraction layer 831, the second extraction layer 832, and the third extraction layer 833, respectively, to obtain network prediction features of the three preset dimensions; the network prediction features are input into the first prediction layer 841 and the second prediction layer 842 according to the network weights of the two business targets output by the gating network, and then information prediction is performed in the first prediction layer 841 and the second prediction layer 842, respectively, and the initial prediction results are output respectively, and then the initial prediction results are normalized to obtain the information prediction results. Optionally, the feature extraction layer may include an MMoE network, and the information prediction layer may include a fully connected layer. The first extraction layer 831, the second extraction layer 832, and the third extraction layer 833 may be expert networks established based on different dimensions, and multiple gating networks are used to determine the sub-weights output by each expert network under each business target, so as to obtain the network prediction features of each business target output by the feature extraction layer and input into the information prediction layer for information prediction. The specific process can refer to the above description of step S103, which will not be repeated here.
[0129] The present application also provides an information processing device, Fig. 9 FIG. 1 is a block diagram of an information processing device according to an exemplary embodiment. Fig. 9 As shown, the information processing device 900 may include at least:
[0130] The encoding module 901 is used to encode and vectorize each feature information in the multiple feature information to obtain a coding feature group for each feature information; the coding feature group includes coding features of multiple preset dimensions; the multiple feature information includes feature information of the target object and feature information of the information to be recommended;
[0131] A cross module 902, configured to determine cross features based on a plurality of encoding feature groups of feature information and a plurality of preset dimensions;
[0132] The prediction module 903 is used to perform information prediction processing on the cross-features to obtain a recommendation prediction result; the recommendation prediction result is used to characterize the predicted recommendation effects of multiple business goals associated with the recommendation system corresponding to the target object after the information to be recommended is processed to the target object.
[0133] Optionally, the cross module 902 is used to: determine the same dimension feature group based on the coding features with consistent dimensions in the coding feature groups of multiple feature information; perform feature cross processing on any two coding features in the same dimension feature group to obtain cross features of each preset dimension.
[0134] Optionally, the prediction module 903 is used to: based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among multiple business targets, perform feature extraction processing on the cross-features of each preset dimension to obtain the network prediction features of each business target; the feature extraction network and the preset dimension correspond one-to-one; the weight network and the business target correspond one-to-one; based on the information prediction model corresponding to each business target, perform information prediction processing on the network prediction features of each business target to obtain a recommended prediction result.
[0135] Optionally, the prediction module 903 is used to, in the process of performing feature extraction processing on the cross-features of each preset dimension based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among multiple business targets to obtain the network prediction features of each business target: based on the feature extraction network corresponding to each preset dimension, perform feature extraction processing on the cross-features of each preset dimension to obtain the network output features corresponding to each preset dimension; based on the weight network corresponding to each business target, process the cross-features to obtain the network weight of each business target; the network weight includes multiple sub-weights corresponding to multiple feature extraction networks, and the multiple sub-weights correspond to the multiple feature extraction networks one-to-one; based on the network weight of each business target, perform weighted sum processing on the network output features corresponding to each preset dimension to obtain the network prediction features of each business target.
[0136] Optionally, the prediction module 903 is used to perform information prediction processing on the network prediction features of each business target based on the information prediction model corresponding to each business target to obtain recommended prediction results: based on the information prediction model corresponding to each business target, perform information prediction processing on the network prediction features of each business target to obtain initial prediction information for multiple business targets; the information prediction models and business targets correspond one-to-one; the initial prediction information is normalized to obtain recommended prediction results.
[0137] Optionally, the device also includes a training module 903, which is used to: obtain a training coding feature group of multiple feature information; the training coding feature group carries annotation information, the annotation information includes sample prediction results, and the sample prediction results are used to characterize the recommendation effects of multiple business goals after the information to be recommended is processed to the target object; through a preset machine learning model, determine the training cross-features based on the training coding feature group of multiple feature information and multiple preset dimensions; perform information prediction processing on the training cross-features to obtain training prediction results; based on the loss value between the training prediction results and the sample prediction results, train the preset machine learning model until the preset end condition is met to obtain the information processing model; wherein, the training combination to be cross-combined includes the coding features of the same dimension in the training coding feature group of any two feature information; the information processing model is used to determine the combination to be cross-combined until the recommendation prediction result is obtained.
[0138] Optionally, the device further includes a recommendation module 903, which is used to send a recommendation instruction to the recommendation system corresponding to the target object when the recommendation prediction result meets the recommendation condition; the recommendation instruction is used to instruct the recommendation system corresponding to the target object to recommend the information to be recommended to the target object.
[0139] It should be noted that the information processing device embodiment provided in the embodiments of the present application and the above-mentioned information processing method embodiment are based on the same inventive concept.
[0140] An embodiment of the present application also provides an electronic device for information processing, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement an information processing method provided in any of the above embodiments.
[0141] An embodiment of the present application also provides a computer-readable storage medium, which can be set in a terminal to store at least one instruction or at least one program for implementing an information processing method in a method embodiment, and the at least one instruction or at least one program is loaded and executed by a processor to implement the information processing method provided in the above method embodiment.
[0142] Optionally, in the embodiment of this specification, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the above storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.
[0143] The memory of the embodiment of this specification can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, applications required for functions, etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a 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.
[0144] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the information processing method provided by the above method embodiment.
[0145] The information processing method embodiment provided in the embodiment of the present application can be executed in a terminal, a computer terminal, a server or a similar computing device. Taking running on a server as an example, Fig.10 FIG. 1 is a hardware structure diagram of a server of an information processing method provided according to an exemplary embodiment. Fig.10As shown, the server 1000 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1010 (the central processing unit 1010 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 1030 for storing data, and one or more storage media 1020 (such as one or more mass storage devices) for storing application programs 1023 or data 1022. Among them, the memory 1030 and the storage medium 1020 can be short-term storage or permanent storage. The program stored in the storage medium 1020 may include one or more modules, each of which may include a series of instruction operations on the server. Furthermore, the central processing unit 1010 can be configured to communicate with the storage medium 1020 and execute a series of instruction operations in the storage medium 1020 on the server 1000. The server 1000 may also include one or more power supplies 1060, one or more wired or wireless network interfaces 1050, one or more input and output interfaces 1040, and / or one or more operating systems 1021, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.
[0146] The input / output interface 1040 may be used to receive or send data via a network. The specific example of the network may include a wireless network provided by a communication provider of the server 1000. In one example, the input / output interface 1040 includes a network adapter (Network Interface Controller, NIC), which may be connected to other network devices through a base station so as to communicate with the Internet. In one example, the input / output interface 1040 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0147] It can be understood by those skilled in the art that Fig.10 The structure shown is only for illustration and does not limit the structure of the above electronic device. Fig.10 More or fewer components as shown, or with Fig.10 Different configurations are shown.
[0148] It should be noted that the above-mentioned sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above-mentioned specific embodiments of this specification are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0149] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and server embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.
[0150] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.
[0151] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. An information processing method, characterized in that: The method comprises: Encoding and vectorizing each feature information of the plurality of feature information to obtain a coding feature group of each feature information; the coding feature group includes coding features of a plurality of preset dimensions; the plurality of feature information includes feature information of the target object and feature information of the information to be recommended; Determining a cross feature based on the coded feature groups of the plurality of feature information and the plurality of preset dimensions; The cross-features are processed for information prediction to obtain a recommendation prediction result; the recommendation prediction result is used to characterize the predicted recommendation effects of multiple business objectives associated with the recommendation system corresponding to the target object after the information to be recommended is processed to the target object.
2. An information processing method according to claim 1, characterized in that: The determining of the cross-features based on the coded feature groups of the plurality of feature information and the plurality of preset dimensions comprises: Determine a feature group with the same dimension based on the coded features with the same dimension in the coded feature groups of the plurality of feature information; Perform feature cross processing on any two coding features in the same dimensional feature group to obtain cross features of each preset dimension.
3. An information processing method according to claim 2, characterized in that: The performing information prediction processing on the cross-features to obtain a recommendation prediction result includes: Based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among the multiple business targets, feature extraction processing is performed on the cross features of each preset dimension to obtain the network prediction features of each business target; the feature extraction network corresponds to the preset dimension one-to-one; the weight network corresponds to the business target one-to-one; Based on the information prediction model corresponding to each business goal, information prediction processing is performed on the network prediction characteristics of each business goal to obtain the recommended prediction result.
4. An information processing method according to claim 3, characterized in that: The step of performing feature extraction processing on the cross features of each preset dimension based on the feature extraction network corresponding to each preset dimension and the weight network corresponding to each business target among the multiple business targets to obtain the network prediction features of each business target includes: Based on the feature extraction network corresponding to each preset dimension, feature extraction processing is performed on the cross features of each preset dimension to obtain the network output features corresponding to each preset dimension; Based on the weight network corresponding to each business target, the cross-features are processed to obtain a network weight for each business target; the network weight includes a plurality of sub-weights corresponding to a plurality of feature extraction networks, and the plurality of sub-weights correspond to the plurality of feature extraction networks one by one; Based on the network weight of each business objective, a weighted sum process is performed on the network output features corresponding to each preset dimension to obtain the network prediction features of each business objective.
5. An information processing method according to claim 3, characterized in that: The performing information prediction processing on the network prediction feature of each business target based on the information prediction model corresponding to each business target to obtain the recommended prediction result includes: Based on the information prediction model corresponding to each business target, information prediction processing is performed on the network prediction characteristics of each business target to obtain initial prediction information of the multiple business targets; the information prediction model and the business target have a one-to-one correspondence; The initial prediction information is normalized to obtain the recommended prediction result.
6. An information processing method according to any one of claims 1 to 5, characterized in that: The method further comprises: Acquire a plurality of training coding feature groups of the feature information; the training coding feature groups carry annotation information, the annotation information includes sample prediction results, and the sample prediction results are used to characterize the recommendation effects of the plurality of business objectives after the information to be recommended is processed to the target object; Determine the training cross-features based on the training coding feature groups of the plurality of feature information and the plurality of preset dimensions through a preset machine learning model; perform information prediction processing on the training cross-features to obtain training prediction results; Based on the loss value between the training prediction result and the sample prediction result, the preset machine learning model is trained until a preset end condition is met to obtain an information processing model; The training to be cross-combined includes the coding features of the same dimension in the training coding feature group of the arbitrary two feature information; The information processing model is used to determine the combination to be crossed until the recommended prediction result is obtained.
7. An information processing method according to any one of claims 1 to 5, characterized in that: After performing information prediction processing on the cross features of the multiple preset dimensions to obtain a recommendation prediction result, the method further includes: In the case where the recommendation prediction result meets the recommendation condition, a recommendation instruction is sent to the recommendation system corresponding to the target object; the recommendation instruction is used to instruct the recommendation system corresponding to the target object to recommend the information to be recommended to the target object.
8. An information processing device, characterized in that: The device comprises: An encoding module, used to encode and vectorize each feature information of the multiple feature information to obtain a coding feature group of each feature information; the coding feature group includes coding features of multiple preset dimensions; the multiple feature information includes feature information of the target object and feature information of the information to be recommended; A cross module, configured to determine cross features based on the coded feature groups of the plurality of feature information and the plurality of preset dimensions; A prediction module is used to perform information prediction processing on the cross-features to obtain a recommendation prediction result; the recommendation prediction result is used to characterize the predicted recommendation effect of multiple business objectives associated with the recommendation system corresponding to the target object after the information to be recommended is processed to the target object.
9. An electronic device for information processing, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded by the processor and executes the information processing method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the information processing method according to any one of claims 1 to 7.
