Object recognition, model training method, device, medium and equipment

Abstract

The present application discloses a method, device, medium and equipment for object recognition and model training, the method comprising: obtaining object feature information of an object to be recognized and resource feature information of all resources; inputting the object feature information and resource feature information into a first object recognition model in a multi-task cascade model, obtaining a first behavior prediction result of the object to be recognized and determining the first intermediate feature information; inputting the first intermediate feature information into a second object recognition model, obtaining a second behavior prediction result of the object to be recognized and determining the second intermediate feature information; inputting the second intermediate feature information into a third object recognition model, obtaining a third behavior prediction result of the object to be recognized; and determining the target recognition result of the object to be recognized based on the first behavior prediction result, the second behavior prediction result and the third behavior prediction result. The present application can estimate the potential of an object at different growth stages and can effectively identify objects with growth potential.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and specifically to methods, devices, media and equipment for object recognition and model training. Background Art

[0002] Artificial Intelligence (AI) is a comprehensive technology in computer science. By studying the design principles and implementation methods of various intelligent machines, it enables machines to have the functions of perception, reasoning and decision-making. Artificial Intelligence technology is a comprehensive discipline that covers a wide range of fields, such as natural language processing, machine learning, deep learning and other major directions. With the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0003] As the penetration rate of mobile Internet gradually reaches saturation, it is becoming increasingly difficult to grow the business scale by relying on industry development trends. In related technologies, deep learning models are used to mine new users, new hot spots, new trends and other objects, but the examination of the growth potential of the objects is still not comprehensive and detailed, which makes it difficult to maintain sustainable and long-term growth. Summary of the invention

[0004] In order to more effectively identify potential objects, this application provides methods, devices, media and equipment for object identification and model training. The technical solution is as follows:

[0005] In a first aspect, the present application provides an object recognition method, the method comprising:

[0006] Obtaining object feature information of each object to be identified and resource feature information of all resources; the object to be identified is a potential object;

[0007] Inputting the object feature information and the resource feature information into a first object recognition model in a trained multi-task cascade model, obtaining a first behavior prediction result corresponding to each of the objects to be recognized, and determining first intermediate feature information output by a first intermediate layer of the first object recognition model; the first behavior prediction result indicates the possibility of each of the objects to be recognized growing into a cognitive object;

[0008] Inputting the first intermediate feature information into the second object recognition model in the multi-task cascade model to obtain a second behavior prediction result of each of the objects to be recognized, and determining the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second behavior prediction result indicates the possibility of each of the objects to be recognized growing into an object of interest;

[0009] Inputting the second intermediate feature information into the third object recognition model in the multi-task cascade model to obtain a third behavior prediction result of each of the objects to be recognized; the third behavior prediction result represents the possibility of each of the objects to be recognized growing into a transformation object;

[0010] Determine a target recognition result of each of the objects to be recognized according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result of each of the objects to be recognized; the target recognition result represents an estimation of the growth potential of the objects to be recognized;

[0011] Determining a target object from among the objects to be identified according to the target identification results of the objects to be identified;

[0012] The potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in sequence.

[0013] In a second aspect, the present application provides a method for training an object recognition model, the method comprising:

[0014] Acquire a training data set, the training data set including object feature information of a sample object, resource feature information of a sample resource, and first behavior feature information, second behavior feature information, and third behavior feature information for the sample resource;

[0015] Acquire an initialized multi-task cascade model, wherein the multi-task cascade model includes a first object recognition model, a second object recognition model, and a third object recognition model;

[0016] The first object recognition model is trained according to the object feature information, the resource feature information and the first behavior feature information to obtain the trained first object recognition model, and first intermediate feature information output by a first intermediate layer of the first object recognition model is determined; the first object recognition model is used to predict the possibility of growing from a potential object to a cognitive object;

[0017] The second object recognition model is trained according to the first intermediate feature information and the second behavior feature information to obtain the trained second object recognition model, and the second intermediate feature information output by the second intermediate layer of the second object recognition model is determined; the second object recognition model is used to predict the possibility of growing into an interest-type object;

[0018] The third object recognition model is trained according to the second intermediate feature information and the third behavior feature information to obtain the trained third object recognition model; the third object recognition model is used to predict the possibility of growing into a transformation object;

[0019] Among them, the first behavior characteristic information, the second behavior characteristic information and the third behavior characteristic information correspond to different behavior levels in turn; the potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in turn.

[0020] In a third aspect, the present application provides an object recognition device, the device comprising:

[0021] An information acquisition module, used to acquire object feature information of each object to be identified and resource feature information of all resources; the object to be identified is a potential object;

[0022] a first recognition module, configured to input the object feature information and the resource feature information into a first object recognition model in a trained multi-task cascade model, obtain a first behavior prediction result corresponding to each of the objects to be recognized, and determine first intermediate feature information output by a first intermediate layer of the first object recognition model; the first behavior prediction result indicates the possibility of each of the objects to be recognized growing into a cognitive object;

[0023] A second recognition module is used to input the first intermediate feature information into a second object recognition model in the multi-task cascade model to obtain a second behavior prediction result of each of the objects to be recognized, and determine the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second behavior prediction result indicates the possibility of each of the objects to be recognized growing into an object of interest;

[0024] A third recognition module, used for inputting the second intermediate feature information into a third object recognition model in the multi-task cascade model to obtain a third behavior prediction result of each of the objects to be recognized; the third behavior prediction result represents the possibility of each of the objects to be recognized growing into a transformation object;

[0025] a target recognition result determination module, configured to determine a target recognition result of each of the objects to be recognized based on the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result of each of the objects to be recognized; the target recognition result represents an estimation of the growth potential of the objects to be recognized;

[0026] A target object determination module, used to determine the target object from each of the objects to be identified according to the target recognition results of each of the objects to be identified;

[0027] The potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in sequence.

[0028] In a fourth aspect, the present application provides a training device for an object recognition model, the device comprising:

[0029] A training data acquisition module, used to acquire a training data set, wherein the training data set includes object feature information of a sample object, resource feature information of a sample resource, and first behavior feature information, second behavior feature information, and third behavior feature information for the sample resource;

[0030] A model acquisition module, used to acquire an initialized multi-task cascade model, wherein the multi-task cascade model includes a first object recognition model, a second object recognition model, and a third object recognition model;

[0031] a first training recognition module, configured to train the first object recognition model according to the object feature information, the resource feature information and the first behavior feature information to obtain the trained first object recognition model, and determine first intermediate feature information output by a first intermediate layer of the first object recognition model; the first object recognition model is used to predict the possibility of growing from a potential object to a cognitive object;

[0032] a second training recognition module, configured to train the second object recognition model according to the first intermediate feature information and the second behavior feature information to obtain the trained second object recognition model, and to determine the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second object recognition model is used to predict the possibility of growing into an interest-type object;

[0033] A third training recognition module is used to train the third object recognition model according to the second intermediate feature information and the third behavior feature information to obtain the trained third object recognition model; the third object recognition model is used to predict the possibility of growing into a conversion-type object;

[0034] Among them, the first behavior characteristic information, the second behavior characteristic information and the third behavior characteristic information correspond to different behavior levels in turn; the potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in turn.

[0035] In a fifth aspect, the present application provides a computer-readable storage medium, which 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 a processor to implement an object recognition method as described in the first aspect or a training method for an object recognition model as described in the second aspect.

[0036] In a sixth aspect, the present application provides a computer device, comprising 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 object recognition method as described in the first aspect or a training method for an object recognition model as described in the second aspect.

[0037] In a seventh aspect, the present application provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, they implement an object recognition method as described in the first aspect or a training method for an object recognition model as described in the second aspect.

[0038] The object recognition and model training methods, devices, media, and equipment provided in this application have the following technical effects:

[0039] The solution provided in the present application further divides the growth stages of objects into potential objects, cognitive objects, interest objects and conversion objects. At the same time, the behaviors implemented for resources are divided into first behaviors, second behaviors and third behaviors, which are used as indicators for object level upgrades. The solution provided in the present application constructs and trains a multi-task cascade model based on the above-mentioned segmentation. For objects to be identified and all resources, the object feature information of the objects to be identified and the resource feature information of the all resources are input into the first object recognition model in the cascade model to obtain the first behavior prediction result corresponding to each object to be identified, and the first behavior prediction result indicates the possibility of each object to be identified growing into a cognitive object; at the same time, the first intermediate feature information output by the first intermediate layer of the first object recognition model is input into the second object recognition model to obtain the second behavior prediction result corresponding to each object to be identified, and the second behavior prediction result indicates the possibility of each object to be identified growing into an interest object; at the same time, the first intermediate feature information output by the first intermediate layer of the first object recognition model is input into the second object recognition model to obtain the second behavior prediction result corresponding to each object to be identified, and the second behavior prediction result indicates the possibility of each object to be identified growing into an interest object; at the same time, the first intermediate feature information output by the first intermediate layer of the first object recognition model is input into the second object recognition model to obtain the second behavior prediction result corresponding to each object to be identified, and the second behavior prediction result indicates the possibility of each object to be identified growing into an interest object. The second intermediate feature information output by the second intermediate layer of the two-object recognition model is input into the third object recognition model to obtain the third behavior prediction result corresponding to each object to be recognized, and the third behavior prediction result indicates the possibility of each object to be recognized growing into a transformation object; then, the target recognition result can be obtained according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result, and the target recognition result can comprehensively measure the growth potential of the object to be recognized. Compared with simply estimating the possibility of transformation, the solution provided in the present application measures and estimates the potential of multiple different growth stages, which can more effectively identify potential objects, thereby obtaining long-term and sustainable growth and development of objects.

[0040] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 It is a schematic diagram of an implementation environment of an object recognition method provided in an embodiment of the present application;

[0043] Figure 2 It is a flowchart of a training method for an object recognition model provided in an embodiment of the present application;

[0044] Figure 3 It is a schematic diagram of a training process of a multi-task cascade model provided in an embodiment of the present application;

[0045] Figure 4 It is a flowchart of an object recognition method provided in an embodiment of the present application;

[0046] Figure 5 It is a schematic diagram of the application process of a multi-task cascade model provided in an embodiment of the present application;

[0047] Figure 6 is a schematic diagram of a training device for an object recognition model provided in an embodiment of the present application;

[0048] Figure 7 is a schematic diagram of an object recognition device provided in an embodiment of the present application;

[0049] Figure 8 It is a schematic diagram of the hardware structure of a device for implementing a training method for an object recognition model or an object recognition method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] 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. Artificial intelligence technology is an interdisciplinary subject that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics.

[0051] The solution provided in the embodiments of the present application involves technologies such as deep learning (DL) of artificial intelligence.

[0052] Deep learning (DL) is a major research direction in the field of machine learning (ML). It is introduced into machine learning to make it closer to its original goal - artificial intelligence. Deep learning is the inherent laws and representation levels of learning sample data. The information obtained in the learning process is of great help in the interpretation of data such as text, images and sounds. Its ultimate goal is to enable machines to have analytical learning capabilities like humans and to recognize data such as text, images and sounds. Deep learning is a complex machine learning algorithm that has achieved results in speech and image recognition that far exceed previous related technologies. Deep learning has achieved many results in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech, recommendation and personalization technology, and other related fields. Deep learning enables machines to imitate human activities such as seeing, hearing and thinking, solves many complex pattern recognition problems, and makes great progress in artificial intelligence related technologies.

[0053] The solution provided in the embodiment of the present application can be deployed in the cloud, which also involves cloud technology, etc.

[0054] Cloud technology: refers to a hosting technology that unifies hardware, software, network and other resources in a wide area network or local area network to achieve data computing, storage, processing and sharing. It can also be understood as a general term for network technology, information technology, integration technology, management platform technology and application technology based on the cloud computing business model application, which can form a resource pool and be used on demand, flexible and convenient. The back-end services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, each item may have its own identification mark in the future, and all need to be transmitted to the back-end system for logical processing. Data of different levels will be processed separately. All kinds of industry data need strong system backing support, so cloud technology needs to be supported by cloud computing. Cloud computing is a computing model that distributes computing tasks on a resource pool composed of a large number of computers, so that various application systems can obtain computing power, storage space and information services as needed. The network that provides resources is called "cloud". The resources in the "cloud" are infinitely expandable in the eyes of users, and can be obtained at any time, used on demand, expanded at any time, and paid for by use. As a cloud computing basic capability provider, a cloud computing resource pool platform will be established, referred to as a cloud platform, generally referred to as Infrastructure as a Service (IaaS), and various types of virtual resources will be deployed in the resource pool for external customers to choose to use. The cloud computing resource pool mainly includes: computing devices (which can be virtualized machines, including operating systems), storage devices, and network devices.

[0055] In order to more effectively identify potential objects, embodiments of the present application provide methods, devices, media and equipment for object recognition and model training. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings 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 creative work are within the scope of protection of this application. Examples of the embodiments are shown in the drawings, where the same or similar numbers throughout represent the same or similar elements or elements with the same or similar functions.

[0056] It should be noted that the terms "first", "second", etc. in the specification and claims 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 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 that includes 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.

[0057] It is understandable that in the specific implementation of this application, related data such as user information is 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 relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0058] See also Figure 1 , which is a schematic diagram of an implementation environment of an object recognition method provided in an embodiment of the present application, such as Figure 1 As shown, the implementation environment may include at least a client 01 and a server 02 .

[0059] Specifically, the client 01 may include devices such as smart phones, desktop computers, tablet computers, laptops, vehicle terminals, digital assistants, smart wearable devices, and voice interaction devices, and may also include software running in the device, such as web pages provided by some service providers to users, or applications provided by these service providers to users. Specifically, the client 01 may be used to obtain object feature information of the object to be identified; and the client 01 may also receive and display the target resources pushed by the server 02 when the corresponding user to be identified is the target object.

[0060] Specifically, the server 02 can 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, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server 02 may include a network communication unit, a processor, and a memory, etc. The terminal and the server may be directly or indirectly connected via wired or wireless communication, and this application is not limited here. Specifically, the server 02 may be used to input the object feature information of the object to be identified and the resource feature information of the full amount of resources into the trained multi-task cascade model, and obtain the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result corresponding to the object to be identified, respectively, so as to obtain a target recognition result that can measure the growth potential of the user to be identified as a whole, and filter out the target object according to the target recognition result, and push the target resource for the target object.

[0061] The embodiments of the present application can also be implemented in combination with cloud technology. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software and network in a wide area network or a local area network to realize the calculation, storage, processing and sharing of data. It can also be understood as a general term for network technology, information technology, integration technology, management platform technology and application technology based on the cloud computing business model application. Cloud technology needs to be supported by cloud computing. Cloud computing is a computing model that distributes computing tasks on a resource pool composed of a large number of computers, so that various application systems can obtain computing power, storage space and information services as needed. The network that provides resources is called a "cloud". Specifically, the server 02 and the database are located in the cloud, and the server 02 can be a physical machine or a virtualized machine.

[0062] The following introduces a training method for an object recognition model provided by this application. Figure 2 It is a flowchart of a method for training an object recognition model provided in an embodiment of the present application. The present application provides method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. 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. Please refer to Figure 2 A training method for an object recognition model provided in an embodiment of the present application may include the following steps:

[0063] S210: Acquire a training data set, where the training data set includes object feature information of a sample object, resource feature information of a sample resource, and first behavior feature information, second behavior feature information, and third behavior feature information for the sample resource.

[0064] The embodiments of the present application can be applied to object growth and growth business. The most important link in object growth is to attract new users. For example, potential objects can be encouraged to grow into conversion objects through advertising, diversion, and push.

[0065] In the embodiment of the present application, in the above business scenario, the growth stage of the object is further divided, and specifically, it can be divided into potential objects, cognitive objects, interest objects and conversion objects step by step, and optionally, loyalty objects can also be included. At the same time, the behavior implemented for the resource is divided into levels, which can specifically include the first behavior, the second behavior, and the third behavior. The behavior feature information corresponding to different behavior levels may include different behavior types, behavior frequencies, behavior contexts and other features, which can be used to characterize the behavior implementation indication information in the present application, that is, whether a certain type of behavior has been implemented. Exemplarily, in a specific product, the potential object is an object that has not yet had any click-to-view behavior in the product, the cognitive object is an object that has only a certain number of click-to-view behaviors in the product, the interest object is an object that has a certain number of click-to-view behaviors and message discussion behaviors in the product, and the conversion object is an object that has purchased the product in the product, that is, the first behavior can be a click-to-view behavior, the second behavior can be a message discussion behavior, and the third behavior can be a purchase behavior. It is understandable that the first behavior, the second behavior and the third behavior can be specifically set according to different business types and product functions to serve as the basis for dividing the object's growth stage and the indicator for upgrading. For example, the second behavior can also include one or more of the search behavior, reading behavior or comment behavior, and the third behavior can also include one or more of the item holding behavior, item giving behavior, dynamic publishing behavior, etc.

[0066] In an embodiment of the present application, during the model training process, the sample objects may include objects at different growth stages, and the object feature information of the sample objects may represent various attributes of the objects, such as identity identification, category identification, etc. If the sample object is a potential object, its first behavior feature information, second behavior feature information, and third behavior feature information may be empty. If the sample object is a cognitive object, it has first behavior feature information, while the second behavior feature information and third behavior feature information may be empty. During the model training process, the sample resources may be historical resources recommended to the sample objects or historical push resources associated with the sample objects. The type of resources may be set according to the business type or product function, such as advertisements, commodities, financial products, articles, videos, etc. The resource feature information may represent the category of the resource, the text or pictures contained in the resource, the price of the resource, and other attributes.

[0067] S230: Obtain an initialized multi-task cascade model, where the multi-task cascade model includes a first object recognition model, a second object recognition model, and a third object recognition model.

[0068] In an embodiment of the present application, based on the subdivision of the object growth stage and the corresponding behavioral indicators when upgrading at different growth stages, a multi-task cascade model is constructed to uniformly measure the overall growth potential of the object. The multi-task cascade model includes a first object recognition model, a second object recognition model and a third object recognition model, wherein the first object recognition model is used to predict the possibility of growing into a cognitive object, which can be the probability of estimating the sample object to implement the first behavior on the sample resource during the potential object period or the probability of the sample object being implemented with the first behavior during the potential object period; the second object recognition model is used to predict the possibility of growing into an interest object, which can be the probability of estimating the sample object to implement the second behavior on the sample resource during the cognitive object period or the probability of the sample object being implemented with the second behavior during the cognitive object period; the third object recognition model is used to predict the possibility of growing into a conversion object, which can be the probability of estimating the sample object to implement the third behavior on the sample resource during the interest object period or the probability of the sample object being implemented with the third behavior during the interest object period.

[0069] Figure 3 A schematic diagram of the training process of a multi-task cascade model is shown, such as Figure 3 As shown in the figure, the architecture of the multi-task cascade model is progressive, that is, the intermediate feature output of the previous level object recognition model serves as the input of the next level object recognition model.

[0070] S250: Train the first object recognition model according to the object feature information, the resource feature information and the first behavior feature information to obtain a trained first object recognition model, and determine first intermediate feature information output by a first intermediate layer of the first object recognition model.

[0071] In one embodiment of the present application, the first object recognition model is used to predict the possibility of a potential object growing into a cognitive object, which can be represented by the probability of implementing the first behavior on the sample resource.

[0072] In a feasible implementation, specifically, training the first object recognition model may include the following steps:

[0073] S251: Inputting object feature information, resource feature information and first behavior feature information into a first object recognition model, performing feature splicing, and obtaining a first training sample and a first sample label corresponding to the first training sample.

[0074] It is feasible that the initialized first object recognition model can adopt an architecture based on the predicted click-through rate model (PredictClick-Through Rate, pCTR). In the feature processing stage, various types of feature information are embedded and finally concatenated together. The constructed first training samples can be divided into positive samples and negative samples according to the first sample labels. In one application embodiment, the positive sample can represent that a certain sample object has implemented the first behavior on a certain sample resource that has been pushed, and the negative sample represents that a certain sample object has not implemented the first behavior on a certain sample resource that has been pushed. In another application embodiment, the positive and negative samples can also represent whether a certain sample resource corresponding to a certain sample object (such as a hot topic or trend) has been implemented with the first behavior.

[0075] S253: The first object recognition model determines a corresponding first behavior prediction result according to the first training sample, where the first behavior prediction result indicates the possibility of the sample object growing from a potential object to a cognitive object.

[0076] Optionally, the first behavior prediction result may also characterize the probability that the sample resource of the sample object will perform the first behavior when it is a potential object. According to the subdivision of the object growth stages and the division of the behavior levels in the embodiments of the present application, the first behavior prediction result may also be used to indicate the possibility of the sample object growing from a potential object to a cognitive object.

[0077] S255: Calculate first loss data according to the first behavior prediction result and the first sample label.

[0078] It is understandable that the first behavior prediction result is an estimated value, the first sample label is a true value, and the first loss data is calculated based on the deviation between the estimated value and the true value. Optionally, the first loss data can be calculated based on a cross entropy loss function or an exponential loss function.

[0079] S257: Adjust the first object recognition model based on the first loss data to complete the training of the first object recognition model.

[0080] Specifically, the network weights in the first object recognition model are adjusted according to the first loss data, thereby obtaining the trained first object recognition model. The first loss data represents the deviation between the first behavior prediction result and the first sample label. The smaller the first loss data, the smaller the estimated loss cost. Therefore, the connection weights between neurons in the first object recognition model and the output threshold of each functional neuron can be reversely adjusted through the gradient descent algorithm, and the training of the first object recognition model is completed when the first loss data reaches the minimum or extremely small value.

[0081] In a feasible implementation, specifically, determining the first intermediate feature information output by the first intermediate layer of the first object recognition model may include the following steps:

[0082] S252: When determining the first behavior prediction result, determine the first intermediate layer of the first object recognition model; the first intermediate layer is any one or more layers of the network of the first object recognition model.

[0083] S254: Obtain first intermediate feature information output by the first intermediate layer, or combine outputs of each layer network in the first intermediate layer to obtain first intermediate feature information.

[0084] If the first intermediate layer is any layer of the network, the output of this layer of the network is directly used as the first intermediate feature information; if the first intermediate layer includes a multi-layer network in the first object recognition model, the outputs of each layer of the network can be combined according to the corresponding weights to obtain the first intermediate feature information.

[0085] In general, in addition to the input layer and the output layer, the neural network model also includes a hidden layer in the middle. The hidden layer abstracts the features of the input data to another dimensional space to show its more abstract features so that these features can be better linearly divided. Multiple hidden layers are actually multi-level abstractions of the input features, which can ultimately better linearly divide different types of data. Hidden layers can include convolutional layers, pooling layers, activation layers, and fully connected layers.

[0086] Exemplarily, the first intermediate layer may be an input layer, and the corresponding first intermediate feature information may be a feature vector obtained by concatenating and embedding the object feature information, the resource feature information, and the first behavior feature information; the first intermediate layer may also be a fully connected layer, and the corresponding first intermediate feature information may be a feature vector obtained by concatenating, embedding, convolution, pooling, activating, and fully connecting the object feature information, the resource feature information, and the first behavior feature information; the first intermediate layer may also include multiple activation layers in the first object recognition model, and the outputs of the multiple activation layers are combined to obtain the above-mentioned first intermediate feature information. It should be noted that when combining multi-layer networks to obtain the first intermediate feature information, the outputs of the combined multi-layer networks are consistent in dimension.

[0087] The first intermediate feature information includes first object intermediate feature information and first resource intermediate feature information.

[0088] S270: Train the second object recognition model according to the first intermediate feature information and the second behavior feature information to obtain a trained second object recognition model, and determine second intermediate feature information output by the second intermediate layer of the second object recognition model.

[0089] In the embodiment of the present application, the second object recognition model is used to predict the possibility of growing into an interest-type object. Further, the possibility of growing from a cognitive object to an interest-type object can be predicted under the condition that the potential object grows into a cognitive object.

[0090] In a feasible implementation, specifically, the training of the second object recognition model may include the following steps:

[0091] S271: Input the first intermediate feature information and the second behavior feature information into the second object recognition model, perform feature concatenation, and obtain a second training sample and a second sample label corresponding to the second training sample.

[0092] It is feasible that the initialized second object recognition model can adopt an architecture based on the predicted click-through rate model (PredictClick-Through Rate, pCTR). In the feature processing stage, the first intermediate feature information and the second behavior feature information are spliced, and the constructed second training samples can be divided into positive samples and negative samples according to the corresponding second sample labels. In one application embodiment, the positive sample can represent that a certain sample object has implemented the second behavior on a certain sample resource pushed, and the negative sample represents that a certain sample object has not implemented the second behavior on a certain sample resource pushed. In another application embodiment, the positive and negative samples can also represent whether a certain sample resource corresponding to a certain sample object (such as a hot topic or trend) has been implemented with a second behavior. Exemplarily, the second behavior corresponding to the positive sample may include a click behavior that is not less than a preset number of times threshold, and the click behavior may be a search, read an article, view details, add to favorites, and other behaviors. The preset number of times threshold can be determined based on the statistical distribution results of the operation data of users in the product.

[0093] S273: The second object recognition model determines a corresponding second behavior prediction result based on the second training sample, where the second behavior prediction result indicates the possibility of the sample object growing from a cognitive object to an interest object.

[0094] Optionally, the second behavior prediction result can characterize the probability that the sample object will perform the second behavior on the sample resource during the cognitive object period. According to the subdivision of the object growth stages and the division of the behavior levels in the embodiments of the present application, the second behavior prediction result can also be used to indicate the possibility of the sample object growing from a cognitive object to an interest-type object.

[0095] S275: Calculate second loss data according to the second behavior prediction result and the second sample label.

[0096] It is understandable that the second behavior prediction result is an estimated value, the second sample label is a true value, and the second loss data is calculated based on the deviation between the estimated value and the true value. Optionally, the first loss data can be calculated based on a cross entropy loss function or an exponential loss function.

[0097] S277: Adjust the second object recognition model based on the second loss data to complete the training of the second object recognition model.

[0098] Specifically, the network weights in the second object recognition model are adjusted according to the second loss data, thereby obtaining a trained second object recognition model. The second loss data represents the deviation between the second behavior prediction result and the second sample label. The smaller the second loss data, the smaller the estimated loss cost. Therefore, the connection weights between neurons in the second object recognition model and the output threshold of each functional neuron can be reversely adjusted to complete the training of the second object recognition model when the second loss data reaches a minimum or extremely small value.

[0099] In a feasible implementation manner, specifically, determining the second intermediate feature information output by the second intermediate layer of the second object recognition model includes:

[0100] S272: When determining the second behavior prediction result, determine the second intermediate layer of the second object recognition model; the second intermediate layer is any one or more layers of the network of the second object recognition model.

[0101] S274: Obtain second intermediate feature information output by the second intermediate layer, or combine outputs of each layer network in the second intermediate layer to obtain second intermediate feature information.

[0102] If the second intermediate layer is any layer of the network, the output of this layer of the network is directly used as the second intermediate feature information; if the second intermediate layer includes a multi-layer network in the second object recognition model, the outputs of each layer of the network can be combined according to the corresponding weights to obtain the second intermediate feature information.

[0103] In general, in addition to the input layer and the output layer, the neural network model also includes a hidden layer in the middle. The hidden layer abstracts the features of the input data into another dimensional space to show its more abstract features so that these features can be better linearly divided. Multiple hidden layers are actually multi-level abstractions of the input features. Hidden layers can include convolutional layers, pooling layers, activation layers, and fully connected layers.

[0104] Exemplarily, the second intermediate layer may be an input layer, and the corresponding second intermediate feature information may be a feature vector obtained by embedding and concatenating the first intermediate feature information and the second behavior feature information; the second intermediate layer may also be a fully connected layer, and the corresponding second intermediate feature information may be a feature vector obtained by performing convolution, pooling, activation, and full connection operations on the output of the input layer; the second intermediate layer may also include multiple activation layers in the second object recognition model, and the outputs of the multiple activation layers are combined to obtain the above-mentioned second intermediate feature information. It should be noted that when the multi-layer networks are combined to obtain the second intermediate feature information, the outputs of the combined multi-layer networks are consistent in dimension.

[0105] The second intermediate characteristic information includes second object intermediate characteristic information and second resource intermediate characteristic information.

[0106] The second object intermediate feature information may be an update of the first object intermediate feature information output in the first object recognition model, and the second resource intermediate feature information may be an update of the first resource intermediate feature information.

[0107] S290: Training a third object recognition model according to the second intermediate feature information and the third behavior feature information to obtain a trained third object recognition model.

[0108] In the embodiment of the present application, the third object recognition model is used to predict the possibility of growing into a conversion type object. Further, it can be predicted that the possibility of growing from an interest type object to a conversion type object is under the condition that the potential type object grows into a cognitive type object and the cognitive type object grows into an interest type object.

[0109] It is understandable that the first object recognition model, the second object recognition model, and the third object recognition model can use the same basic model structure, or can select basic models of different structures for training according to economic costs and business needs, but they are all predictions of the probability of behavior occurrence, and essentially solve the same type of problem. According to the division of the growth stages of the object in this application and the division of the behavior categories and behavior times corresponding to different growth stages, training samples and training labels corresponding to different growth stages are constructed to train the three models, so that the three models can realize specific different application functions.

[0110] In a feasible implementation, specifically, the training of the third object recognition model may include the following steps:

[0111] S291: Input the second intermediate feature information and the third behavior feature information into a third object recognition model to perform feature concatenation to obtain a third training sample and a third sample label corresponding to the third training sample.

[0112] It is feasible that the initialized third object recognition model can adopt an architecture based on the predicted conversion rate model (PredictConversion Rate, pCVR). In the feature processing stage, the second intermediate feature information and the third behavior feature information are spliced, and the constructed third training samples can be divided into positive samples and negative samples according to the corresponding third sample labels.

[0113] S293: The third object recognition model determines a corresponding third behavior prediction result based on the third training sample; the third behavior prediction result indicates the possibility of the sample object growing from an interest-type object to a conversion-type object.

[0114] Optionally, the third behavior prediction result can represent the probability that the sample object will perform the third behavior on the sample resource during the interest-type object period. According to the subdivision of the object growth stages and the division of the behavior levels in the embodiments of the present application, the third behavior prediction result can also be used to indicate the possibility of the sample object growing from an interest-type object to a conversion-type object.

[0115] S295: Calculate third loss data according to the third behavior prediction result and the third sample label.

[0116] It is understandable that the third behavior prediction result is an estimated value, the third sample label is a true value, and the third loss data is calculated based on the deviation between the estimated value and the true value. Optionally, the first loss data can be calculated based on a cross entropy loss function or an exponential loss function.

[0117] S297: Adjust the third object recognition model based on the third loss data to complete the training of the third object recognition model.

[0118] Specifically, the network weights in the third object recognition model are adjusted according to the third loss data, thereby obtaining a trained third object recognition model. The third loss data represents the deviation between the third behavior prediction result and the third sample label. The smaller the third loss data, the smaller the estimated loss cost. Therefore, the connection weights between neurons in the third object recognition model and the output threshold of each functional neuron can be reversely adjusted through the gradient descent algorithm, and the training of the third object recognition model is completed when the third loss data reaches a minimum or extremely small value.

[0119] In another feasible implementation, the first intermediate feature information can also be introduced from the first intermediate layer of the first object recognition model as the input of the third object recognition model to be spliced ​​with the third behavior feature information. The obtained third behavior prediction result specifically represents the probability of the sample object performing the third behavior on the sample resources during the cognitive object period, that is, it can be used to indicate the probability of the sample object growing directly from a cognitive object to a transformation object.

[0120] In a model training embodiment for resource recommendation business provided in the present application, the resources are items, and the operations that can be performed on the items are divided into three types of behaviors: click to view, add to shopping list, and place an order to purchase. If a potential object (i.e., an object that has not performed any operation on any recommended item) performs a click to view operation on a recommended item, then the potential object can be considered to have grown into a cognitive object; if the cognitive object adds the recommended item to the shopping list based on the aforementioned operation, then the cognitive object can be considered to have become an interested object; if the interested object places an order to purchase the recommended item based on the aforementioned operation, then the interested object can be considered to have grown into a converted object. In order to identify objects with growth potential and promote the interaction between objects and items, a system is constructed based on the above-mentioned division of operation behaviors and growth stages. Figure 3 Cascade model shown.

[0121] In the training process, first obtain the object feature information U = {u1, u2..., u n}, n is a positive integer, where u i The object characteristic information of the i-th sample object includes the identity, age, gender, etc. of the sample object; and the resource characteristic information S={s1, s2..., s m}, m is a positive integer, where s j The resource feature information of the jth sample resource includes the resource identifier, category, price, sales volume, etc. of the sample resource; in addition, the first line feature information A for the sample resource needs to be obtained during the training phase. 11 , A 12 .....A 1m ; A 21 , A 22 .....A 2m ;......;A n1 , A n2 .....A nm}, the second behavior characteristic information B = {B 11 , B 12 .....B 1m ; B 21 , B 22 .....B 2m ;......;B n1 , B n2 .....B nm} and the third behavior characteristic information C={C 11 , C 12 .....C 1m ; C 21 , C 22 .....C2m ;......;C n1 , C n2 .....C nm}, where A ij , B ij , C ij Indicates whether the i-th sample object has clicked to view, added to the shopping list, or placed an order to purchase the j-th sample resource.

[0122] Secondly, the object feature information U, the resource feature information S and the first behavior feature information A are input into the first object recognition model in the cascade model to train the first object recognition model, and at the same time obtain the first intermediate feature information for input into the second object recognition model. Taking the Deep Crossing model as an example, the first object recognition model includes an Embedding layer, a Stacking layer, a Multiple Residual Units layer and a Scoring layer. Specifically, the object feature information U and the resource feature information S are input into the Embedding layer for embedding representation, and the sparse category features are converted into dense Embedding vectors. The Embedding layer is composed of a single-layer neural network. Then, the Embedding vectors corresponding to each feature information are input into the Stacking layer for feature concatenation (Concat) to obtain the first training sample. The Stacking layer is also called the connection layer. The first training sample can be represented by a feature vector Feature1. The number of first training samples is nm, and the first sample label Lable1 corresponding to the first training sample is obtained according to the first behavior feature information A. The sample with Lable1 being 1 indicates that an object has clicked to view a resource, which is also a positive sample. The sample with Lable1 being 0 indicates that an object has not clicked to view a resource, which is also a negative sample. Feature1 is then input to the Multiple Residual Units layer (multi-layer residual network). The multi-layer residual network fully cross-combines the dimensions of the feature vector Feature1, so that the model can capture more nonlinear features and combined features. The single-layer residual network can be composed of residual units. The residual unit passes the input feature vector Feature1 through two layers of fully connected layers with ReLU as the activation function to obtain an intermediate output vector. Then, the input feature vector Feature1 is directly element-added with the intermediate output vector to obtain the final output data Feature. 1E Finally, the Feature in the output 1EThe data is input to the Scoring layer. The Scoring layer can use a softmax-based logistic regression model to estimate the probability of the click-to-view behavior. The Scoring layer outputs the estimated value P1 corresponding to the first training sample, and the value of P1 is 0-1. Based on the log loss function, the loss value is calculated according to the estimated value P1 and the first sample label, and the gradient back propagation method is used to continuously adjust the network weights in the first object recognition model, such as the weights and biases in the ReLU function, and the first object recognition model is iteratively trained until the training end conditions are met. In addition, the Feature 1E As the first intermediate feature information to be input into the second object recognition model.

[0123] Next, the first intermediate feature information Feature 1E The second behavior feature information B is input into the second object recognition model in the cascade model to train the second object recognition model, and at the same time, the second intermediate feature information for input into the third object recognition model is obtained. The second object recognition model takes the FNN (Factorization Machine supported Neural Network, a neural network based on a factorization machine) model as an example. The FNN model includes an input layer, an embedding (Dense) layer, a hidden layer 1, a hidden layer 2, and an output layer. Specifically, the second behavior feature information characterizes whether a sample object implements the behavior of adding to the shopping list under the condition that a sample object implements the click-to-view behavior on a sample resource. The second behavior feature information may specifically include behavior context feature information (such as implementation time, implementation scenario, implementation results, etc.). At the input layer, the context feature information in the second behavior feature information is represented as a feature vector based on one-hot encoding corresponding to different feature categories. In the Dense layer, the feature vectors corresponding to different feature categories are multiplied by the latent vectors obtained after pre-training of the embedding layer and horizontally spliced. The spliced ​​result is then spliced ​​with the first intermediate feature information to obtain the input Feature2 of the hidden layer (also the fully connected layer). Feature2 represents the second training sample. According to the second behavioral feature information, the corresponding second sample label Lable2 can be constructed, and the value is 1 or 0. It can be understood that the first intermediate feature information is already a high-order feature representation and no embedding representation processing is required. In the hidden layer, Feature2 is input to the fully connected layer based on the tanh function as the activation function to obtain the output vector Feature 2E In the output layer, the sigmoid function is used to calculate the feature 2E(that is, the behavior probability estimate P2 corresponding to the second training sample), the value of P2 is 0-1. Finally, based on the log loss function, the loss value is calculated according to the estimate P2 and the second sample label, and the gradient back propagation method is used to continuously adjust the network weights in the second object recognition model, and the second object recognition model is iteratively trained until the training end condition is met. In addition, the feature vector Feature2 output by the embedding layer in the second object recognition model is used as the second intermediate feature information to be input into the third object recognition model.

[0124] Finally, the second intermediate feature information Feature2 and the third behavior feature information C are input into the third object recognition model in the cascade model to train the third object recognition model. The third object recognition model can also be based on the FNN model, including an input layer, an embedding (Dense) layer, a hidden layer 1, a hidden layer 2 and an output layer. Specifically, the third behavior feature information characterizes whether a sample object implements the behavior of clicking to view a sample resource and adding it to the shopping list, and then implements the behavior of placing an order to purchase. The third behavior feature information can specifically include behavior context feature information (such as implementation time, implementation scenario, purchase amount, etc.). The processes such as feature processing and loss calculation in the model training process are similar to the above steps and will not be repeated here.

[0125] The above process is an implementation process under a specific application scenario provided by this application. The implementation process will be different for different business types, different object growth stage divisions, different behavior level divisions, and different model selections.

[0126] In a training method for an object recognition model provided in an embodiment of the present application, a multi-task cascade model is constructed and trained based on the subdivision of the object's growth stage and the subdivision of the behavior implemented on the resources, wherein the first object recognition model is used to predict the possibility of growing into a cognitive object, the second object recognition model is used to predict the possibility of growing into an interest-type object, and the third object recognition model is used to predict the possibility of growing into a conversion-type object. The cascade model can be used to uniformly measure the overall growth potential of the object.

[0127] The following introduces an object recognition method provided by this application. Figure 4 This is a flowchart of an object recognition method provided by an embodiment of the present application. The present application provides method operation steps as described in the embodiment or flowchart, and may also include more or fewer operation steps. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or server product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, in a parallel processor or multi-threaded processing environment). Please refer to Figure 4, an object recognition method provided by an embodiment of the present application may include the following steps:

[0128] S410: Obtaining object feature information of each object to be identified and resource feature information of all resources; the object to be identified is a potential object.

[0129] The embodiments of the present application can be applied to the object growth and growth business. The most important part of object growth is to attract new users. For example, potential objects can be promoted to transform into conversion objects through advertising, traffic diversion, and push. Objects can include but are not limited to user accounts, hot topics, and trends. For example, objects can be user entities that perform actions on resources, or topics and product categories corresponding to resources.

[0130] In the embodiment of the present application, in the above business scenario, the growth stage of the object is further divided, and specifically, it can be divided into potential objects, cognitive objects, interest objects and conversion objects step by step, and optionally, loyalty objects can also be included. At the same time, the behavior implemented for the resources is divided into levels, and specifically, it can include the first behavior, the second behavior, and the third behavior. The behavior feature information corresponding to different behavior levels can characterize different behavior types, behavior frequencies, resources corresponding to the implementation of the behavior, and other characteristics. Exemplarily, in a specific product, the potential object is an object that has not yet had any click-to-view behavior in the product, the cognitive object is an object that has only a certain number of click-to-view behaviors in the product, the interest object is an object that has a certain number of click-to-view behaviors and message discussion behaviors in the product, and the conversion object is an object that has purchased the product. That is, the first behavior can be a click-to-view behavior, the second behavior can be a message discussion behavior, and the third behavior can be a purchase behavior. It is understandable that the first behavior, the second behavior and the third behavior can be specifically set according to different business types and product functions to serve as the basis for dividing the object's growth stage and the indicator for upgrading. For example, the second behavior can also include one or more of the search behavior, reading behavior or comment behavior, and the third behavior can also include one or more of the item holding behavior, item giving behavior, dynamic publishing behavior, etc.

[0131] The embodiments of the present application can more comprehensively identify potential targets through a more detailed division of growth stages and corresponding feature examinations, thereby bringing long-term and sustainable growth and development to the business.

[0132] S420: Input the object feature information and resource feature information into the first object recognition model in the trained multi-task cascade model to obtain the first behavior prediction result corresponding to each object to be recognized, and determine the first intermediate feature information output by the first intermediate layer of the first object recognition model.

[0133] In an embodiment of the present application, based on the subdivision of the object growth stage (divided into potential objects, cognitive objects, interest objects and conversion objects, corresponding to different object levels in turn) and the corresponding behavioral indicators when upgrading at different growth stages, a multi-task cascade model is constructed to uniformly measure the overall growth potential of the object. The above-mentioned multi-task cascade model includes a first object recognition model, a second object recognition model and a third object recognition model, wherein the first object recognition model is used to predict the possibility of growing into a cognitive object, and an example thereof may be the probability of estimating the probability that a certain resource of the object to be recognized is implemented with a first behavior during the period of a potential object; the second object recognition model is used to predict the possibility of growing into an interest object, and an example thereof may be the probability that a certain resource of the object to be recognized is implemented with a second behavior during the period of a cognitive object; the third object recognition model is used to predict the possibility of growing into a conversion object, and an example thereof may be the probability that a certain resource of the object to be recognized is implemented with a third behavior during the period of an interest object.

[0134] Figure 5 A schematic diagram of the application process of a multi-task cascade model is shown, Figure 5 As shown, the architecture of the multi-task cascade model is progressive, that is, the intermediate feature output of the previous level object recognition model is used as the input of the next level object recognition model. The architecture and principle of each object recognition model can refer to the training method of an object recognition model provided in the embodiment of the present application, which will not be repeated here.

[0135] Specifically, the object feature information of the object to be identified can represent various attributes of the object, such as identity identification, category identification, etc. The resources can be all the resources in the product or all the resources initially recalled. The type of resources can be set according to the business type or product function, such as advertisements, commodities, financial products, articles, videos, etc. The resource feature information can represent the category of the resource, the text or pictures contained in the resource, the price of the resource and other attributes.

[0136] Specifically, the first behavior prediction result can also represent the probability that the first behavior will be implemented on each resource of the object to be identified when it is a potential object. According to the subdivision of the object growth stages and the division of behavior levels in the embodiments of the present application, the first behavior prediction result can also be used to indicate the possibility of the object to be identified growing from a potential object to a cognitive object.

[0137] Specifically, the first intermediate layer can be any one or more layers of the network of the first object recognition model. If the first intermediate layer is any one layer of the network, the output of the network layer can be directly used as the first intermediate feature information; if the first intermediate layer includes multiple layers of the network in the first object recognition model, the outputs of each layer of the network can be combined according to the corresponding weights to obtain the first intermediate feature information. The above-mentioned first intermediate feature information can include the first object intermediate feature information and the first resource intermediate feature information.

[0138] S430: Input the first intermediate feature information to the second object recognition model in the multi-task cascade model to obtain a second behavior prediction result of each object to be recognized, and determine the second intermediate feature information output by the second intermediate layer of the second object recognition model.

[0139] In one embodiment of the present application, the second behavior prediction result indicates the possibility of each to-be-recognized object growing into an interest-type object. Further, the possibility of a cognitive object growing into an interest-type object may be predicted under the condition that a potential object grows into a cognitive object.

[0140] Optionally, the second behavior prediction result can represent the probability that the object to be identified will perform the second behavior on each resource during the cognitive object period. According to the subdivision of the object growth stages and the division of the behavior levels in the embodiments of the present application, the second behavior prediction result can also be used to indicate the possibility of the object to be identified growing from a cognitive object to an interest object.

[0141] Specifically, the second intermediate layer is any one or more layers of the network of the second object recognition model; if the second intermediate layer is any one layer of the network, the output of the layer of network can be directly used as the second intermediate feature information; if the second intermediate layer includes multiple layers of the network in the second object recognition model, the output of each layer of the network can be combined according to the corresponding weights to obtain the second intermediate feature information. The above second intermediate feature information can include the second object intermediate feature information and the second resource intermediate feature information.

[0142] S440: Input the second intermediate feature information into the third object recognition model in the multi-task cascade model to obtain a third behavior prediction result of each object to be recognized.

[0143] In one embodiment of the present application, the third behavior prediction result characterizes the possibility of each to-be-recognized object growing into a conversion object. Further, the possibility of growing from an interest-type object into a conversion-type object can be predicted under the condition that the potential object grows into a cognitive object and the cognitive object grows into an interest-type object.

[0144] Optionally, the third behavior prediction result can represent the probability that the object to be identified will perform the third behavior on each resource during the interest-type object period. According to the subdivision of the object growth stages and the division of the behavior levels in the embodiments of the present application, the third behavior prediction result can also be used to indicate the possibility of the object to be identified growing from an interest-type object to a conversion-type object.

[0145] S450: Determine a target recognition result of each object to be recognized according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result of each object to be recognized.

[0146] In the embodiment of the present application, the target recognition result represents the overall estimation of the growth potential of the object to be recognized.

[0147] In a feasible implementation manner, specifically, step S450 may include the following steps:

[0148] S451: Determine, based on the object growth data in the historical period, a first weight value, a second weight value, and a third weight value corresponding to the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result, respectively.

[0149] That is, based on a periodic offline data statistics, the weight values ​​of the potential estimation results at different growth stages in the overall potential estimation results are determined. For example, if an average of 10 objects to be identified grew into cognitive objects one month ago, the first weight value may be 0.1.

[0150] S453: Determine the target recognition result of each object to be recognized according to the first behavior prediction result, the second behavior prediction result, the third behavior prediction result, and the first weight value, the second weight value, and the third weight value.

[0151] Exemplarily, by adding the product of the first behavior prediction result and the first weight value, the product of the second behavior prediction result and the second weight value, and the product of the third behavior prediction result and the third weight value, the target recognition results of each object to be recognized can be obtained.

[0152] For example, as shown in formula (1), the target recognition result Score(x) of the object to be recognized x can be expressed as:

[0153] Score(x)=w1*cognitive pCTR(x)+w2*interest pCTR(x)+conversion PCVR; (1)

[0154] Among them, cognitive pCTR(x) represents the first behavior prediction result for the object to be identified x output by the first object recognition model, interest pCTR(x) represents the second behavior prediction result for the object to be identified x output by the second object recognition model, conversion PCVR represents the third behavior prediction result for the object to be identified x output by the third object recognition model, w1 and w2 are the first weight value and the second weight value respectively, and the third weight value is 1.

[0155] S460: Determine a target object from among the objects to be identified according to the target identification results of the objects to be identified.

[0156] In the embodiment of the present application, an object that meets a preset potential condition is determined from among the objects to be identified through the predicted target recognition result as a target object.

[0157] In a feasible implementation manner, specifically, step S460 may include the following steps:

[0158] S461: Determine the number of objects that meet the resource recommendation business requirement.

[0159] S463: Determine a recognition result threshold according to the number of objects and the target recognition result of each object to be recognized.

[0160] S465: Determine the target object from the objects to be identified based on the recognition result threshold and the target recognition results of the objects to be identified.

[0161] Furthermore, the highest score in the target recognition results of the target object may be determined, and the target resource corresponding to the highest score may be recommended to the target object.

[0162] In the resource recommendation business, specific resources can be recommended to objects with growth potential to meet the object's product needs and promote resource conversion.

[0163] In a model recognition embodiment applied to a resource recommendation business provided in the present application, the resources are items, and the operations that can be performed on the items are divided into three types of behaviors: click to view, add to shopping list, and place an order to purchase. If a potential object (i.e., an object that has not performed any operation on any recommended item) performs a click to view operation on a recommended item, then the potential object can be considered to have grown into a cognitive object; if the cognitive object adds the recommended item to the shopping list based on the aforementioned operation, then the cognitive object can be considered to have become an interested object; if the interested object places an order to purchase the recommended item based on the aforementioned operation, then the interested object can be considered to have grown into a converted object. In order to identify objects with growth potential from potential objects, a system is constructed based on the above-mentioned division of operation behaviors and growth stages. Figure 3The cascade model shown is trained based on a model training embodiment of an application resource recommendation service provided above.

[0164] In the recognition process, first obtain the object feature information U = {u1, u2..., u n}, n is a positive integer, where u i The object feature information of the i-th object to be identified includes the identity, age, gender, etc. of the object to be identified. All objects to be identified are potential objects; and the resource feature information S = {s1, s2..., s m}, m is a positive integer, where s j Represents the resource feature information of the jth resource, including the resource identifier, category, price, sales volume, etc. of the resource.

[0165] Secondly, the object feature information U and the resource feature information are input into the first object recognition model in the cascade model to obtain the first behavior prediction result P1 corresponding to the object to be recognized = {P 11 , P 12 ,......,P 1n}, where P 1i It represents the prediction result of the first behavior corresponding to the i-th object to be identified, which contains m components, component P 1ij represents the probability that the i-th object to be identified clicks to view the j-th resource, that is, P 1ij The possibility that the i-th object to be identified can grow into a cognitive object by implementing a click-to-view behavior on the j-th resource. At the same time, the final output of the fully connected layer in the first object recognition model is used as the first intermediate feature information, and the first intermediate feature information is input into the second object recognition model to obtain the second behavior prediction result P2 corresponding to the object to be identified = {P 21 , P 22 ,......,P 2n}, where P 2i It represents the prediction result of the second behavior corresponding to the i-th object to be identified, which contains m components, component P 2ij represents the probability that the i-th object to be identified adds the j-th resource to the shopping list, that is, P 2ij It can indicate the possibility that the i-th object to be identified grows into an object of interest by adding the j-th resource to the shopping list. At the same time, the feature vector output from the embedding layer of the second object recognition model is used as the second intermediate feature information, and the second intermediate feature information is input into the third object recognition model to obtain the third behavior prediction result P3 corresponding to the object to be identified = {P 31 , P 32 ,......,P 3n}, where P 3i It represents the prediction result of the second behavior corresponding to the i-th object to be identified, which contains m components, component P 3ij represents the probability that the i-th object to be identified places an order to purchase the j-th resource, that is, P 3ij It can indicate the possibility that the i-th object to be identified grows into a conversion-type object by placing an order for the j-th resource.

[0166] Then, the target recognition result of each object to be recognized can be determined according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result and the first weight value w1, the second weight value w2 and the third weight value w3. The first weight value, the second weight value and the third weight value can be obtained by statistical calculation based on periodic offline data. Specifically, for the i-th object to be recognized, its score Score(i) = {Score(i1), Score(i2), ..., Score(i m )}, where Score(i j )=w1*P 1ij +w2*P 2ij +w3*P 3ij , used to represent the overall growth potential that the i-th object to be identified may show when recommending the j-th resource to the i-th object to be identified. If there is a component in Score(i) that is higher than the preset threshold, the i-th object to be identified is taken as the target object to be identified, and the resources corresponding to the components in Score(i) that are higher than the preset threshold are recommended to the target object as the resources to be recommended.

[0167] The above process is an implementation process in a specific application scenario provided by this application. The above process does not limit the actual business type, the division of object growth stages, the division of behavior levels, the model type, etc.

[0168] An object recognition method provided by an embodiment of the present application inputs object feature information of the object to be recognized and resource feature information of the full amount of resources into a first object recognition model in the cascade model for the object to be recognized and the full amount of resources, and obtains a first behavior prediction result corresponding to each object to be recognized, and the first behavior prediction result indicates the possibility of each object to be recognized growing into a cognitive object; at the same time, inputs the first intermediate feature information output by the first intermediate layer of the first object recognition model into a second object recognition model, and obtains a second behavior prediction result corresponding to each object to be recognized, and the second behavior prediction result indicates the possibility of each object to be recognized growing into an interest-type object; at the same time, The second intermediate feature information output by the second intermediate layer of the type is input into the third object recognition model to obtain the third behavior prediction result corresponding to each object to be recognized, and the third behavior prediction result indicates the possibility of each object to be recognized growing into a transformation object; then the target recognition result can be obtained according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result, and the target recognition result can comprehensively measure the growth potential of the object to be recognized. Compared with simply estimating the possibility of transformation, the solution provided in the present application measures and estimates the potential of multiple different growth stages, which can more effectively identify potential objects, thereby obtaining the growth and development of objects in a long-term and sustainable manner.

[0169] The present application also provides a training device 600 for an object recognition model. Figure 6 As shown, the device 600 may include:

[0170] A training data acquisition module 610 is used to acquire a training data set, wherein the training data set includes object feature information of a sample object, resource feature information of a sample resource, and first behavior feature information, second behavior feature information, and third behavior feature information for the sample resource;

[0171] A model acquisition module 620, configured to acquire an initialized multi-task cascade model, wherein the multi-task cascade model includes a first object recognition model, a second object recognition model, and a third object recognition model;

[0172] A first training recognition module 630 is used to train the first object recognition model according to the object feature information, the resource feature information and the first behavior feature information to obtain the trained first object recognition model, and determine the first intermediate feature information output by the first intermediate layer of the first object recognition model; the first object recognition model is used to predict the possibility of growing from a potential object to a cognitive object;

[0173] A second training recognition module 640 is used to train the second object recognition model according to the first intermediate feature information and the second behavior feature information to obtain the trained second object recognition model, and to determine the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second object recognition model is used to predict the possibility of growing into an interest-type object;

[0174] A third training recognition module 650 is used to train the third object recognition model according to the second intermediate feature information and the third behavior feature information to obtain the trained third object recognition model; the third object recognition model is used to predict the possibility of growing into a conversion-type object;

[0175] Among them, the first behavior characteristic information, the second behavior characteristic information and the third behavior characteristic information correspond to different behavior levels in turn; the potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in turn.

[0176] In one embodiment of the present application, the first training identification module 630 may include:

[0177] A first feature splicing unit, configured to input the object feature information, the resource feature information, and the first behavior feature information into the first object recognition model, perform feature splicing, and obtain a first training sample and a first sample label corresponding to the first training sample;

[0178] A first training prediction unit, configured for the first object recognition model to determine a corresponding first behavior prediction result according to the first training sample, wherein the first behavior prediction result indicates a possibility that the sample object grows from the potential object to the cognitive object;

[0179] A first loss calculation unit, configured to calculate first loss data according to the first behavior prediction result and the first sample label;

[0180] The first model adjustment unit is used to adjust the first object recognition model based on the first loss data to complete the training of the first object recognition model.

[0181] In one embodiment of the present application, the first training identification module 630 may further include:

[0182] A first intermediate layer determining unit, configured to determine a first intermediate layer of the first object recognition model when determining the first behavior prediction result; the first intermediate layer is any one or more layers of the network of the first object recognition model;

[0183] a first intermediate feature information determining unit, configured to obtain the first intermediate feature information output by the first intermediate layer, or to combine the outputs of the networks of each layer in the first intermediate layer to obtain the first intermediate feature information;

[0184] The first intermediate characteristic information includes first object intermediate characteristic information and first resource intermediate characteristic information.

[0185] In one embodiment of the present application, the second training identification module 640 may include:

[0186] A second feature splicing unit, used for inputting the first intermediate feature information and the second behavior feature information into the second object recognition model, performing feature splicing, and obtaining a second training sample and a second sample label corresponding to the second training sample;

[0187] A second training prediction unit, configured for the second object recognition model to determine a corresponding second behavior prediction result according to the second training sample, wherein the second behavior prediction result indicates a possibility that the sample object grows from the cognitive object to the interest object;

[0188] A second loss calculation unit, used for calculating second loss data according to the second behavior prediction result and the second sample label;

[0189] The second model adjustment unit is used to adjust the second object recognition model based on the second loss data to complete the training of the second object recognition model.

[0190] In one embodiment of the present application, the second training identification module 640 may further include:

[0191] A second intermediate layer determination unit, configured to determine a second intermediate layer of the second object recognition model when determining the second behavior prediction result; the second intermediate layer is any one or more layers of the network of the second object recognition model;

[0192] a second intermediate feature information determining unit, configured to obtain the second intermediate feature information output by the second intermediate layer, or to combine the outputs of the networks of each layer in the second intermediate layer to obtain the second intermediate feature information;

[0193] The second intermediate characteristic information includes second object intermediate characteristic information and second resource intermediate characteristic information.

[0194] In one embodiment of the present application, the third training identification module 650 may include:

[0195] A third feature splicing unit, used for inputting the second intermediate feature information and the third behavior feature information into the third object recognition model, performing feature splicing, and obtaining a third training sample and a third sample label corresponding to the third training sample;

[0196] A third training prediction unit, configured for the third object recognition model to determine a corresponding third behavior prediction result according to the third training sample; the third behavior prediction result indicates the possibility of the sample object growing from the interest-type object to the conversion-type object;

[0197] A third loss calculation unit, used for calculating third loss data according to the third behavior prediction result and the third sample label;

[0198] The third model adjustment unit is used to adjust the third object recognition model based on the third loss data to complete the training of the third object recognition model.

[0199] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0200] The present application embodiment also provides an object recognition device 700, such as Figure 7 As shown, the device 700 may include:

[0201] The information acquisition module 710 is used to acquire object feature information of each object to be identified and resource feature information of all resources; the object to be identified is a potential object;

[0202] The first recognition module 720 is used to input the object feature information and the resource feature information into the first object recognition model in the trained multi-task cascade model, obtain the first behavior prediction result corresponding to each of the objects to be recognized, and determine the first intermediate feature information output by the first intermediate layer of the first object recognition model; the first behavior prediction result indicates the possibility of each of the objects to be recognized growing into a cognitive object;

[0203] The second recognition module 730 is used to input the first intermediate feature information into the second object recognition model in the multi-task cascade model, obtain the second behavior prediction result of each of the objects to be recognized, and determine the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second behavior prediction result indicates the possibility of each of the objects to be recognized growing into an object of interest;

[0204] A third recognition module 740 is used to input the second intermediate feature information into a third object recognition model in the multi-task cascade model to obtain a third behavior prediction result of each of the objects to be recognized; the third behavior prediction result represents the possibility of each of the objects to be recognized growing into a transformation object;

[0205] The target recognition result determination module 750 is used to determine the target recognition result of each of the objects to be recognized according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result of each of the objects to be recognized; the target recognition result represents the estimation of the growth potential of the objects to be recognized;

[0206] A target object determination module 760 is used to determine a target object from each of the objects to be identified according to the target recognition results of each of the objects to be identified;

[0207] The potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in sequence.

[0208] In one embodiment of the present application, the target recognition result determination module 750 may include:

[0209] a weight determination unit, configured to determine, based on the object growth data in a historical period, a first weight value, a second weight value, and a third weight value corresponding to the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result, respectively;

[0210] A target recognition result determination unit is used to determine the target recognition result of each object to be identified based on the first behavior prediction result, the second behavior prediction result and the third behavior prediction result and the first weight value, the second weight value and the third weight value.

[0211] In one embodiment of the present application, the target object determination module 760 may include:

[0212] An object number determination unit, used to determine the number of objects that meet the resource recommendation business requirements;

[0213] A threshold determination unit, used to determine a recognition result threshold according to the number of objects and the target recognition result of each object to be recognized;

[0214] The target object determining unit is used to determine the target object from the objects to be identified based on the recognition result threshold and the target recognition results of the objects to be identified.

[0215] In one embodiment of the present application, the device 700 may further include:

[0216] a highest score determination unit, used to determine the highest score among the target recognition results of the target object;

[0217] The resource recommendation unit is used to recommend the target resource corresponding to the highest score to the target object.

[0218] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0219] An embodiment of the present application provides a computer device, 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 the at least one program is loaded and executed by the processor to implement a training method for an object recognition model or an object recognition method as provided in the above method embodiment.

[0220] Figure 8 A hardware structure diagram of a device for implementing a training method for an object recognition model or an object recognition method provided in an embodiment of the present application is shown, and the device may participate in or include the apparatus or system provided in an embodiment of the present application. Figure 8 As shown, the device 10 may include one or more (1002a, 1002b, ..., 1002n are used to illustrate) processors 1002 (the processor 1002 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 1004 for storing data, and a transmission device 1006 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 8 The structure shown is for illustration only and does not limit the structure of the electronic device. Figure 8 More or fewer components as shown, or with Figure 8 Different configurations are shown.

[0221] It should be noted that the one or more processors 1002 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the device 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0222] The memory 1004 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the methods described in the embodiments of the present application. The processor 1002 executes various functional applications and data processing by running the software programs and modules stored in the memory 1004, that is, to implement the above-mentioned training method of an object recognition model or an object recognition method. The memory 1004 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1004 may further include a memory remotely arranged relative to the processor 1002, and these remote memories may be connected to the device 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0223] The transmission device 1006 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the device 10. In one example, the transmission device 1006 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 1006 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0224] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of device 10 (or mobile device).

[0225] An embodiment of the present application also provides a computer-readable storage medium, which can be set in a server to store at least one instruction or at least one program related to a training method for an object recognition model or an object recognition method in a method embodiment, and the at least one instruction or the at least one program is loaded and executed by the processor to implement a training method for an object recognition model or an object recognition method provided in the above method embodiment.

[0226] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as 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.

[0227] The embodiment of the present invention further provides a computer program product or a computer program, wherein the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes a training method for an object recognition model or an object recognition method provided in the above various optional implementations.

[0228] It can be seen from the embodiments of the object recognition and model training methods, devices, media and equipment provided by the present application above that

[0229] The solution provided in the present application further divides the growth stages of objects into potential objects, cognitive objects, interest objects and conversion objects. At the same time, the behaviors implemented for resources are divided into first behaviors, second behaviors and third behaviors, which are used as indicators for object level upgrades. The solution provided in the present application constructs and trains a multi-task cascade model based on the above-mentioned segmentation. For objects to be identified and all resources, the object feature information of the objects to be identified and the resource feature information of the all resources are input into the first object recognition model in the cascade model to obtain the first behavior prediction result corresponding to each object to be identified, and the first behavior prediction result indicates the possibility of each object to be identified growing into a cognitive object; at the same time, the first intermediate feature information output by the first intermediate layer of the first object recognition model is input into the second object recognition model to obtain the second behavior prediction result corresponding to each object to be identified, and the second behavior prediction result indicates the possibility of each object to be identified growing into an interest object; at the same time, the first intermediate feature information output by the first intermediate layer of the first object recognition model is input into the second object recognition model to obtain the second behavior prediction result corresponding to each object to be identified, and the second behavior prediction result indicates the possibility of each object to be identified growing into an interest object; at the same time, the first intermediate feature information output by the first intermediate layer of the first object recognition model is input into the second object recognition model to obtain the second behavior prediction result corresponding to each object to be identified, and the second behavior prediction result indicates the possibility of each object to be identified growing into an interest object. The second intermediate feature information output by the second intermediate layer of the two-object recognition model is input into the third object recognition model to obtain the third behavior prediction result corresponding to each object to be recognized, and the third behavior prediction result indicates the possibility of each object to be recognized growing into a transformation object; then, the target recognition result can be obtained according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result, and the target recognition result can comprehensively measure the growth potential of the object to be recognized. Compared with simply estimating the possibility of transformation, the solution provided in the present application measures and estimates the potential of multiple different growth stages, which can more effectively identify potential objects, thereby obtaining long-term and sustainable growth and development of objects.

[0230] 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 the present application are described. Other embodiments are within the scope of the appended 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.

[0231] Each embodiment in this application 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, equipment and storage medium 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 embodiment.

[0232] 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 instructing related hardware through a program, 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.

[0233] The above description is only a preferred embodiment of the present application and is 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 object recognition method, characterized in that: The method comprises: Obtaining object feature information of each object to be identified and resource feature information of all resources; the object to be identified is a potential object; Inputting the object feature information and the resource feature information into a first object recognition model in a trained multi-task cascade model, obtaining a first behavior prediction result corresponding to each of the objects to be recognized, and determining first intermediate feature information output by a first intermediate layer of the first object recognition model; the first behavior prediction result indicates the possibility of each of the objects to be recognized growing into a cognitive object; Inputting the first intermediate feature information into the second object recognition model in the multi-task cascade model to obtain a second behavior prediction result of each of the objects to be recognized, and determining the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second behavior prediction result indicates the possibility of each of the objects to be recognized growing into an object of interest; Inputting the second intermediate feature information into the third object recognition model in the multi-task cascade model to obtain a third behavior prediction result of each of the objects to be recognized; the third behavior prediction result represents the possibility of each of the objects to be recognized growing into a transformation object; Determine a target recognition result of each of the objects to be recognized according to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result of each of the objects to be recognized; the target recognition result represents an estimation of the growth potential of the objects to be recognized; Determining a target object from among the objects to be identified according to the target identification results of the objects to be identified; The potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in sequence.

2. The method according to claim 1, characterized in that: Determining the target recognition result of each of the objects to be recognized according to the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result of each of the objects to be recognized includes: Determine, according to the object growth data in the historical period, a first weight value, a second weight value, and a third weight value corresponding to the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result, respectively; According to the first behavior prediction result, the second behavior prediction result and the third behavior prediction result and the first weight value, the second weight value and the third weight value, a target recognition result of each of the objects to be recognized is determined.

3. The method according to claim 1, characterized in that The step of determining the target object from the objects to be identified according to the target identification results of the objects to be identified includes: Determine the number of objects that meet the business needs of resource recommendation; Determining a recognition result threshold according to the number of objects and the target recognition result of each of the objects to be recognized; Based on the recognition result threshold and the target recognition results of each of the objects to be recognized, the target object is determined from the objects to be recognized.

4. The method according to claim 3, characterized in that: The method further comprises: Determining a highest score among the target recognition results for the target object; The target resource corresponding to the highest score is recommended to the target object.

5. A method for training an object recognition model, characterized in that: The method comprises: Acquire a training data set, the training data set including object feature information of a sample object, resource feature information of a sample resource, and first behavior feature information, second behavior feature information, and third behavior feature information for the sample resource; Acquire an initialized multi-task cascade model, wherein the multi-task cascade model includes a first object recognition model, a second object recognition model, and a third object recognition model; The first object recognition model is trained according to the object feature information, the resource feature information and the first behavior feature information to obtain the trained first object recognition model, and first intermediate feature information output by a first intermediate layer of the first object recognition model is determined; the first object recognition model is used to predict the possibility of growing from a potential object to a cognitive object; The second object recognition model is trained according to the first intermediate feature information and the second behavior feature information to obtain the trained second object recognition model, and the second intermediate feature information output by the second intermediate layer of the second object recognition model is determined; the second object recognition model is used to predict the possibility of growing into an interest-type object; The third object recognition model is trained according to the second intermediate feature information and the third behavior feature information to obtain the trained third object recognition model; the third object recognition model is used to predict the possibility of growing into a transformation type object; Among them, the first behavior characteristic information, the second behavior characteristic information and the third behavior characteristic information correspond to different behavior levels in turn; the potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in turn.

6. The method according to claim 5, characterized in that The step of training the first object recognition model according to the object feature information, the resource feature information, and the first behavior feature information to obtain the trained first object recognition model includes: Inputting the object feature information, the resource feature information and the first behavior feature information into the first object recognition model, performing feature splicing, and obtaining a first training sample and a first sample label corresponding to the first training sample; The first object recognition model determines a corresponding first behavior prediction result according to the first training sample, wherein the first behavior prediction result indicates a possibility that the sample object grows from the potential object to the cognitive object; Calculating first loss data according to the first behavior prediction result and the first sample label; The first object recognition model is adjusted based on the first loss data to complete the training of the first object recognition model.

7. The method according to claim 6, characterized in that The determining of first intermediate feature information output by a first intermediate layer of the first object recognition model includes: When determining the first behavior prediction result, determining a first intermediate layer of the first object recognition model; the first intermediate layer is any one or more layers of the network of the first object recognition model; Obtaining the first intermediate feature information output by the first intermediate layer, or combining the outputs of each layer network in the first intermediate layer to obtain the first intermediate feature information; The first intermediate characteristic information includes first object intermediate characteristic information and first resource intermediate characteristic information.

8. The method according to claim 5, characterized in that The step of training the second object recognition model according to the first intermediate feature information and the second behavior feature information to obtain the trained second object recognition model includes: Inputting the first intermediate feature information and the second behavior feature information into the second object recognition model to perform feature concatenation to obtain a second training sample and a second sample label corresponding to the second training sample; The second object recognition model determines a corresponding second behavior prediction result according to the second training sample, wherein the second behavior prediction result indicates a possibility that the sample object grows from the cognitive object to the interest object; Calculating second loss data according to the second behavior prediction result and the second sample label; The second object recognition model is adjusted based on the second loss data to complete the training of the second object recognition model.

9. The method according to claim 8, characterized in that The determining of second intermediate feature information output by a second intermediate layer of the second object recognition model includes: When determining the second behavior prediction result, determining a second intermediate layer of the second object recognition model; the second intermediate layer is any one or more layers of the network of the second object recognition model; Obtaining the second intermediate feature information output by the second intermediate layer, or combining the outputs of each layer network in the second intermediate layer to obtain the second intermediate feature information; The second intermediate characteristic information includes second object intermediate characteristic information and second resource intermediate characteristic information.

10. The method according to claim 5, characterized in that The step of training the third object recognition model according to the second intermediate feature information and the third behavior feature information to obtain the trained third object recognition model includes: Inputting the second intermediate feature information and the third behavior feature information into the third object recognition model to perform feature concatenation to obtain a third training sample and a third sample label corresponding to the third training sample; The third object recognition model determines a corresponding third behavior prediction result according to the third training sample; the third behavior prediction result indicates the possibility of the sample object growing from the interest-type object to the conversion-type object; Calculating third loss data according to the third behavior prediction result and the third sample label; The third object recognition model is adjusted based on the third loss data to complete the training of the third object recognition model.

11. An object recognition device, characterized in that: The device comprises: An information acquisition module, used to acquire object feature information of each object to be identified and resource feature information of all resources; the object to be identified is a potential object; a first recognition module, configured to input the object feature information and the resource feature information into a first object recognition model in a trained multi-task cascade model, obtain a first behavior prediction result corresponding to each of the objects to be recognized, and determine first intermediate feature information output by a first intermediate layer of the first object recognition model; the first behavior prediction result indicates the possibility of each of the objects to be recognized growing into a cognitive object; A second recognition module is used to input the first intermediate feature information into a second object recognition model in the multi-task cascade model to obtain a second behavior prediction result of each of the objects to be recognized, and determine the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second behavior prediction result indicates the possibility of each of the objects to be recognized growing into an object of interest; A third recognition module, used for inputting the second intermediate feature information into a third object recognition model in the multi-task cascade model to obtain a third behavior prediction result of each of the objects to be recognized; the third behavior prediction result represents the possibility of each of the objects to be recognized growing into a transformation object; a target recognition result determination module, configured to determine a target recognition result of each of the objects to be recognized based on the first behavior prediction result, the second behavior prediction result, and the third behavior prediction result of each of the objects to be recognized; the target recognition result represents an estimation of the growth potential of the objects to be recognized; A target object determination module, used to determine the target object from each of the objects to be identified according to the target recognition results of each of the objects to be identified; The potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in sequence.

12. A training device for an object recognition model, characterized in that: The device comprises: A training data acquisition module, used to acquire a training data set, wherein the training data set includes object feature information of a sample object, resource feature information of a sample resource, and first behavior feature information, second behavior feature information, and third behavior feature information for the sample resource; A model acquisition module, used to acquire an initialized multi-task cascade model, wherein the multi-task cascade model includes a first object recognition model, a second object recognition model, and a third object recognition model; a first training recognition module, configured to train the first object recognition model according to the object feature information, the resource feature information and the first behavior feature information to obtain the trained first object recognition model, and determine first intermediate feature information output by a first intermediate layer of the first object recognition model; the first object recognition model is used to predict the possibility of growing from a potential object to a cognitive object; a second training recognition module, configured to train the second object recognition model according to the first intermediate feature information and the second behavior feature information to obtain the trained second object recognition model, and to determine the second intermediate feature information output by the second intermediate layer of the second object recognition model; the second object recognition model is used to predict the possibility of growing into an interest-type object; A third training recognition module is used to train the third object recognition model according to the second intermediate feature information and the third behavior feature information to obtain the trained third object recognition model; the third object recognition model is used to predict the possibility of growing into a conversion-type object; Among them, the first behavior characteristic information, the second behavior characteristic information and the third behavior characteristic information correspond to different behavior levels in turn; the potential object, the cognitive object, the interest object and the conversion object correspond to different object levels in turn.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium 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 the processor to implement an object recognition method as described in any one of claims 1 to 4 or to implement a training method for an object recognition model as described in any one of claims 5 to 10.

14. A computer device, characterized in that: The computer 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 at least one program is loaded and executed by the processor to implement an object recognition method as described in any one of claims 1 to 4 or to implement a training method for an object recognition model as described in any one of claims 5 to 10.

15. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, it implements an object recognition method as described in any one of claims 1 to 4 or a training method for an object recognition model as described in any one of claims 5 to 10.

Images