Resource recycling strategy acquisition method and device, electronic equipment and storage medium

Abstract

The application provides a resource recycling strategy acquisition method and device, electronic equipment and a storage medium. The method comprises the following steps: obtaining a label of a sample object; obtaining feature data of each stage in a resource retention process of the sample object; training a classification model by taking the feature data of each stage in the resource retention process of the sample object as input and taking the label as a mapping target, to obtain a trained classification model; in response to an update of a stage in which a target object is located in the resource retention process, updating feature data of the stage in which the target object is located in the resource retention process, inputting the updated feature data of the target object into the trained classification model, and obtaining an output result of the trained classification model for the target object update; and updating a resource recycling strategy for the target object according to the output result for the target object update. The embodiments of the application can improve the rationality of the resource recycling strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a resource recycling strategy acquisition method and device, electronic equipment and storage medium. BACKGROUND

[0002] In many fields, resources are usually limited. In order to ensure efficient use of resources or for other considerations, when a certain part of the resource is long-stayed in an object, the resource held by the object needs to be recycled. For example, in a computer system, in order to ensure efficient use of CPU resources, when a certain part of the CPU resources is long-stayed by a process, the CPU resources held by the process need to be recycled.

[0003] At the same time, in many cases, the control right of the resource is in the hands of the object holding the resource. Therefore, in order to successfully recycle the resource held by the object, a resource recycling strategy needs to be developed specifically. The resource recycling strategy developed in the prior art is relatively single and one-sided, and lacks rationality to some extent. SUMMARY

[0004] One purpose of the present application is to provide a resource recycling strategy acquisition method and device, electronic equipment and storage medium, which can improve the rationality of the resource recycling strategy.

[0005] According to an aspect of an embodiment of the present application, a resource recycling strategy acquisition method is disclosed, the method comprising:

[0006] obtaining a label of a sample object, wherein the label is used to describe the stay degree of the resource held by the sample object;

[0007] obtaining feature data of the sample object at each stage in the resource stay process;

[0008] respectively taking the feature data of the sample object at each stage in the resource stay process as input, and taking the label as a mapping target, training a preset classification model until the output result of the classification model is mapped to the label with an accuracy rate meeting a preset condition, and obtaining a trained classification model;

[0009] in response to an update of the stage of a target object in the resource stay process, updating the feature data of the target object in the resource stay process, and inputting the updated feature data of the target object into the trained classification model to obtain an output result of the trained classification model for the target object update;

[0010] updating the resource recycling strategy for the target object according to the output result for the target object update.

[0011] According to an aspect of an example embodiment of the present application, a device for obtaining a resource recycling strategy is disclosed, the device comprising:

[0012] a first obtaining module configured to obtain a label of a sample object, wherein the label is used to describe a retention degree of a resource held by the sample object;

[0013] a second obtaining module configured to obtain feature data of the sample object at each stage in a resource retention process;

[0014] a training module configured to take the feature data of the sample object at each stage in the resource retention process as input and take the label as a mapping target, and train a preset classification model until an output result of the classification model is mapped to the label with an accuracy rate satisfying a preset condition, so as to obtain a trained classification model;

[0015] a first updating module configured to update the feature data of a target object at a stage in a resource retention process in response to an update of the stage of the target object in the resource retention process, and input updated feature data of the target object into the trained classification model to obtain an output result of the trained classification model for the update of the target object;

[0016] a second updating module configured to update a resource recycling strategy for the target object according to the output result for the update of the target object.

[0017] In an example embodiment of the present application, the first obtaining module is configured to:

[0018] obtain, for a preset number of statistical objects, a probability that a resource which has been retained for different days continues to be retained on the next day based on resource retention records of the preset number of statistical objects;

[0019] determine a retention day threshold value for dividing ratings based on a comparison result between the probability and a preset probability threshold value;

[0020] obtain a maximum retention day of the resource held by the sample object;

[0021] generate the label of the sample object based on a comparison result between the maximum retention day of the resource held by the sample object and the retention day threshold value.

[0022] In an example embodiment of the present application, the first obtaining module is configured to:

[0023] determine a first retention day threshold value for dividing ratings based on a comparison result between the probability and a preset probability threshold value;

[0024] obtaining a preset second retention day threshold for dividing ratings, wherein the second retention day threshold is greater than the first retention day threshold;

[0025] if the maximum retention day of the resource held by the sample object is less than or equal to the first retention day threshold, a first rating is taken as the label of the sample object;

[0026] if the maximum retention day of the resource held by the sample object is greater than the first retention day threshold and less than or equal to the second retention day threshold, a second rating is taken as the label of the sample object;

[0027] if the maximum retention day of the resource held by the sample object is greater than the second retention day threshold, a third rating is taken as the label of the sample object.

[0028] In an example embodiment of the present application, the device is configured to:

[0029] the sample objects with the second rating are excluded, and the sample objects with the first rating and the third rating are used to train the classification model.

[0030] In an example embodiment of the present application, the training module is configured to:

[0031] the feature data of each stage of the sample object in the resource retention process is input into the classification model respectively, and the retention probability output by the classification model is obtained, wherein the retention probability is used to describe the probability of successfully recovering the resource held by the sample object within a preset retention day threshold or the probability of the resource held by the sample object continuing to retain after the retention day threshold;

[0032] based on the mapping relationship between the retention probability and the label, the retention probability output by the classification model is converted into a corresponding label to obtain a model classification label;

[0033] based on the deviation between the model classification label and the label of the sample object, the classification model is adjusted in the direction of reducing the deviation.

[0034] In an example embodiment of the present application, the training module is configured to:

[0035] the feature data of each stage of the sample object in the resource retention process is input into the classification model respectively, and the maximum retention day output by the classification model is obtained, wherein the maximum retention day is used to describe the maximum retention day of the resource held by the sample object;

[0036] The maximum retention days output by the classification model are converted into corresponding labels based on a mapping relationship between the maximum retention days and the labels, to obtain model classification labels;

[0037] The classification model is adjusted in a direction of reducing a deviation between the model classification labels and the labels of the sample objects based on the deviation.

[0038] In an example embodiment of the present application, the feature data includes: object basic data for describing basic attribute information of an object, object behavior data for describing historical scheduling behavior of the object for the resource held by the object, and object recovery data for describing recovery measures taken for the object since the resource held by the object is retained.

[0039] In an example embodiment of the present application, the object recovery data includes: recovery behavior data for describing attribute information of recovery behavior taken for the object since the resource held by the object is retained, and recovery text data for describing interaction information between the object and the object in a text form in a recovery resource process since the resource held by the object is retained.

[0040] In an example embodiment of the present application, the device is configured to:

[0041] acquire interaction text between the object and the object in a recovery process since the resource held by the object is retained;

[0042] generate the recovery text data based on the interaction text.

[0043] In an example embodiment of the present application, the device is configured to:

[0044] acquire interaction voice between the object and the object in a recovery process since the resource held by the object is retained;

[0045] generate the recovery text data based on the interaction voice.

[0046] In an example embodiment of the present application, the classification model includes a first sub-model, a second sub-model, and a main model, an output result of the classification model is an output result of the main model, an input of the main model is an output result of the first sub-model and an output result of the second sub-model, an input of the first sub-model is the recovery text data, and an input of the second sub-model is feature data other than the recovery text data.

[0047] In an example embodiment of the present application, the second update module is configured to:

[0048] Based on the updated retention probability for the target object, the rating of the target object is updated to obtain the updated rating of the target object;

[0049] Based on the updated rating of the target object, update the resource recycling strategy for the target object.

[0050] According to one aspect of the embodiments of this application, an electronic device is disclosed, comprising: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement any of the above embodiments.

[0051] According to one aspect of the embodiments of this application, a computer program medium is disclosed, on which computer-readable instructions are stored, which, when executed by a computer's processor, cause the computer to perform any of the above embodiments.

[0052] According to one aspect of the embodiments of this application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations described above.

[0053] In this embodiment, the resource retention process of a sample object is divided into multiple stages, and a preset classification model is trained using the feature data of each stage of the resource retention process as input. This allows the trained classification model to accurately describe the degree of resource retention held by the sample object based on the feature data of each stage. Consequently, the trained classification model can accurately describe the degree of resource retention held by the target object based on the feature data of each stage of the resource retention process. Therefore, the trained classification model enables the resource recovery strategy for the target object to dynamically adapt to the updated stage of the resource retention process, thereby improving the rationality of the resource recovery strategy.

[0054] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0055] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0056] The above and other objectives, features and advantages of this application will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0057] Figure 1 A schematic diagram showing an exemplary system architecture of the provided technical solution according to an embodiment of the present application is shown.

[0058] Figure 2 A flow chart showing a resource recycling strategy obtaining method according to an embodiment of the present application is shown.

[0059] Figure 3 A structural schematic diagram of a classification model according to an embodiment of the present application is shown.

[0060] Figure 4 A training schematic diagram of a classification model according to an embodiment of the present application is shown.

[0061] Figure 5 A mapping relationship between ratings and resource recycling strategies according to an embodiment of the present application is shown.

[0062] Figure 6 A block diagram of a resource recycling strategy obtaining apparatus according to an embodiment of the present application is shown.

[0063] Figure 7 An electronic device hardware diagram according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0064] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art. Like reference numerals refer to like elements throughout the description. Repeated use of illustrations indicates reusability of a drawing figure across one or more examples.

[0065] Moreover, described features, structures, or characteristics can be combined in any suitable manner in one or more example implementations. In the following description, numerous specific details are provided to give a thorough understanding of example implementations. One skilled in relevant art will recognize, however, that the

[0066] Some of the block diagrams shown in the drawings are functional entities, which do not necessarily have to correspond to physically or logically independent entities. These functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0067] The application provides a resource recycling strategy acquisition method, relates to machine learning in the field of artificial intelligence, and is mainly used for determining a corresponding resource recycling strategy for a target object with held resources.

[0068] Artificial intelligence (AI) is the use of digital computers or digital computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react 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.

[0069] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology and machine learning / deep learning and other major directions.

[0070] Machine learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It is a subject that studies how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and example-based learning.

[0071] Figure 1 A schematic diagram of an exemplary system architecture of the technical solution provided by an embodiment of the present application is shown.

[0072] As shown in Figure 1 , the system architecture can include one or more of terminal devices (such as a smartphone 101, a tablet computer 102, and a portable computer 103 as shown in Figure 1 , of course, a desktop computer, etc.), a network 104, and a server 105. The network 104 is a medium for providing a communication link between the terminal devices and the server 105. The network 104 can include various connection types, such as wired communication links, wireless communication links, and the like.

[0073] It should be understood that Figure 1 the number of terminal devices, networks, and servers in is only illustrative. Depending on the implementation needs, there can be any number of terminal devices, networks, and servers. For example, the server 105 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDNs, and basic cloud computing services such as big data and artificial intelligence platforms.

[0074] Among them, the terminal device is mainly used for the server 105 to obtain the feature data of the sample object, or to obtain the feature data of the target object. The server 105 is mainly used to obtain the label of the sample object and the feature data of the sample object at each stage in the resource retention process according to the resource recycling strategy acquisition method provided by the present application, and then train the preset classification model on this basis, and then use the trained classification model to update the resource recycling strategy for the target object.

[0075] It should be noted that this embodiment is only illustratively shows a system architecture to which the present application can be applied, and should not limit the functions and use range of the present application.

[0076] Figure 2 A flowchart of the resource recycling strategy acquisition method proposed by the present application is shown. Illustratively, the execution subject of the method can be a server, and the method includes:

[0077] Step S210, obtaining a label of a sample object, wherein the label is used to describe the retention degree of the resource held by the sample object;

[0078] Step S220, obtaining feature data of the sample object at each stage in the resource retention process;

[0079] Step S230, inputting the feature data of each stage of the sample object in the resource retention process as input, and mapping the label as the target, training the preset classification model until the output result of the classification model is mapped to the label with the accuracy rate meeting the preset condition, and obtaining the trained classification model;

[0080] Step S240, in response to the update of the stage of the target object in the resource retention process, updating the feature data of the stage of the target object in the resource retention process, and inputting the updated feature data of the target object into the trained classification model to obtain the output result of the trained classification model for the target object update;

[0081] Step S250, updating the resource recycling strategy for the target object according to the output result for the target object update.

[0082] In the embodiment of the application, after the sample object is selected, the label of the sample object is obtained. The label is used to describe the retention degree of the resource held by the corresponding sample object. For example, if the sample object holds the resource for a long time, the resource held by the sample object is confirmed as heavy retention, and the label of the sample object is "low segment"; if the sample object holds the resource for a long time, the resource held by the sample object is confirmed as moderate retention, and the label of the sample object is "medium segment"; if the sample object holds the resource for a short time, the resource held by the sample object is confirmed as light retention, and the label of the sample object is "high segment".

[0083] In the embodiment of the application, the retention process is divided into multiple stages in advance according to the chronological order. After the sample object is selected, the feature data of the sample object in each stage of the resource retention process is obtained, and the feature data of each stage is used to describe the related features of the resource holding behavior of the sample object in the corresponding stage. For example, each natural day is divided into a stage in advance according to the chronological order. The first day of retention is the first retention stage, the second day of retention is the second retention stage, the third day of retention is the third retention stage, and so on. Then, after the sample object is selected, the feature data of the sample object in the first retention stage, the feature data of the sample object in the second retention stage, the feature data of the sample object in the third retention stage, and so on are obtained.

[0084] Then, the feature data of the sample object in each stage of the resource retention process is input into the preset classification model to obtain the output result of the classification model. Then, each output result is mapped according to the preset mapping relationship, and the mapping target is the label of the sample object. That is, if a certain output result obtains the label of the sample object after mapping, it means that the output result is accurately mapped to the label of the sample object; if a certain output result does not obtain the label of the sample object after mapping, it means that the output result is not accurately mapped to the label of the sample object.

[0085] The classification model is trained until the output result of the classification model can be mapped to the label of the sample object with an accuracy rate meeting a preset condition, and a trained classification model is obtained. Since the label of the sample object can be accurately mapped according to the output result of the trained classification model, it indicates that the trained classification model can accurately describe the retention degree of the resource held by the sample object according to the feature data of the sample object at each stage in the resource retention process, and thus the trained classification model can also accurately describe the retention degree of the resource held by the target object according to the feature data of the target object at each stage in the resource retention process.

[0086] Further, for the target object to be determined the resource recycling strategy, whenever the stage of the target object in the resource retention process is updated, the feature data of the stage of the target object in the resource retention process is updated accordingly, and then the updated feature data of the target object is input into the trained classification model, and the output result of the trained classification model for the updated target object is obtained. Then, according to the output result for the updated target object, the resource recycling strategy for the target object is updated. For example, when the target object is retained for 1 day and enters the first retention stage, the feature data of the target object in the first retention stage is input into the trained classification model, and then the resource recycling strategy for the target object is updated according to the output result of the trained classification model for the updated target object. When the target object is retained for 2 days and enters the second retention stage, the feature data of the target object in the second retention stage is input into the trained classification model, and then the resource recycling strategy for the target object is updated according to the output result of the trained classification model for the updated target object. And so on.

[0087] As can be seen, in the embodiments of the present application, the resource retention process of the sample object is divided into multiple stages, and the preset classification model is trained with the feature data of the sample object at each stage in the resource retention process as input, so that the trained classification model can accurately describe the retention degree of the resource held by the sample object according to the feature data of the sample object at each stage in the resource retention process, thereby enabling the trained classification model to accurately describe the retention degree of the resource held by the target object according to the feature data of the target object at each stage in the resource retention process. Thus, through the trained classification model, the resource recycling strategy for the target object can dynamically adapt to the update of the stage of the target object in the resource retention process, thereby improving the rationality of the resource recycling strategy.

[0088] In an embodiment, the label of the sample object includes a rating of the sample object. For a preset number of statistical objects, a probability that resources having stayed for different days continue to stay for the next day is obtained based on resource stay records of the preset number of statistical objects. A stay day threshold for dividing ratings is determined based on a comparison result between the probability and a preset probability threshold. A maximum stay day of resources held by the sample object is obtained. The label of the sample object is generated based on a comparison result between the maximum stay day of resources held by the sample object and the stay day threshold.

[0089] Specifically, the design idea of the embodiment can be summarized as follows: obtaining a probability that resources having stayed for different days continue to stay for the next day; then determining a stay day threshold in combination with a preset probability threshold; and then generating a label of a sample object in combination with a maximum stay day of resources held by the sample object.

[0090] The probability that resources having stayed for different days continue to stay for the next day can be described by a proportion of resources having stayed for different days that continue to stay for the next day. For example, among 100 resources having stayed for 1 day, 40 continue to stay for the next day, so the probability that resources having stayed for 1 day continue to stay for the next day is 40 / 100=40%; among 40 resources having stayed for 2 days, 24 continue to stay for the next day, so the probability that resources having stayed for 2 days continue to stay for the next day is 24 / 40=60%, and so on.

[0091] Generally, the longer the stay days are, the greater the probability that resources continue to stay for the next day is. Let P be a preset probability threshold, and the probability that resources having stayed for N days continue to stay for the next day be P N . N satisfying P N ≤P≤P N+1 is determined, and then N or (N+1) is taken as the stay day threshold. Then the maximum stay day of resources held by the sample object is compared with the stay day threshold, and the rating of the sample object is determined based on the comparison result, so as to generate the label of the sample object.

[0092] The embodiment has the advantage that the stay day threshold for dividing ratings is determined based on the probability that resources having stayed for different days continue to stay for the next day, so that the label of the sample object can be generated more objectively.

[0093] In an embodiment, the first retention day threshold for dividing the rating is determined based on a comparison result between the probability and a preset probability threshold. A preset second retention day threshold for dividing the rating is obtained, wherein the second retention day threshold is greater than the first retention day threshold. If the maximum retention day of the resource held by the sample object is less than or equal to the first retention day threshold, the first rating is taken as the label of the sample object. If the maximum retention day of the resource held by the sample object is greater than the first retention day threshold and less than or equal to the second retention day threshold, the second rating is taken as the label of the sample object. If the maximum retention day of the resource held by the sample object is greater than the second retention day threshold, the third rating is taken as the label of the sample object.

[0094] Specifically, the rating of the sample is divided into three ratings (the first rating, the second rating and the third rating) through two retention day thresholds (the first retention day threshold and the second retention day threshold). The first retention day threshold is determined based on a comparison result between the probability that the resource which has been retained for different days continues to be retained the next day and a probability threshold. The second retention day threshold is preset, and the first retention day is less than the second retention day.

[0095] Let the first retention day threshold be T1, the second retention day threshold be T2, and the maximum retention day of the resource held by the sample object be T, T1

[0096] For example, when the object is a borrower and the resource held by the object is a borrowed loan, the maximum retention day of the resource held by the object can be the maximum overdue day of the borrowed loan, and the retention day threshold is the overdue day threshold of the borrowed loan.

[0097] Generally, 30 days is taken as the boundary of whether the object is seriously overdue or not. If the maximum overdue day of the borrowed loan is greater than 30 days, the object is seriously overdue. If the maximum overdue day is less than or equal to 30 days, the object is not seriously overdue. Therefore, the second overdue day threshold can be set to 30.

[0098] If the first overdue day threshold is determined to be 20 based on a comparison result between the probability that the borrowed loan which has been overdue for different days continues to be overdue the next day and a probability threshold, the first overdue day threshold is set to 20.

[0099] If the maximum overdue days of the loan borrowed by the sample object is less than or equal to 20, it indicates that the sample object is a low-risk object, and the label of the sample object is the first rating; if the maximum overdue days of the loan borrowed by the sample object is greater than 20 and less than or equal to 30, it indicates that the sample object is a medium-risk object, and the label of the sample object is the second rating; if the maximum overdue days of the loan borrowed by the sample object is greater than 30, it indicates that the sample object is a high-risk object, and the label of the sample object is the third rating.

[0100] In an embodiment, the sample objects with the second rating are excluded, and the sample objects with the first rating and the third rating are used to train the classification model.

[0101] Specifically, after determining the three levels of sample objects, the sample objects in the middle level are excluded from the modeling samples, and only the sample objects in the other two levels are used to train the classification model.

[0102] The advantage of this embodiment is that by excluding the sample objects in the middle level from the modeling samples and only using the sample objects in the other two levels to train the classification model, the classification model has stronger risk differentiation capability.

[0103] In an embodiment, the feature data of the sample object at each stage in the resource retention process is input into the classification model to obtain the retention probability output by the classification model, wherein the retention probability is used to describe the probability of successfully recovering the resources held by the sample object within a preset retention day threshold or the probability of the resources held by the sample object continuing to be retained after the retention day threshold. Based on the mapping relationship between the retention probability and the label, the retention probability output by the classification model is converted into a corresponding label to obtain a model classification label. Based on the deviation between the model classification label and the label of the sample object, the classification model is adjusted in the direction of reducing the deviation.

[0104] Specifically, the classification model is used to predict the probability P of successfully recovering the resources held by the sample object within a preset retention day threshold, or to predict the probability (1-P) of the resources held by the sample object continuing to be retained after the retention day threshold. Then, according to the mapping relationship between the retention probability and the label, P or (1-P) predicted by the classification model is converted into a corresponding label, and the label obtained by the conversion is a model classification label. Then, based on the deviation between the model classification label and the label of the sample object, the classification model is adjusted so that the deviation continuously decreases, so as to improve the accuracy of the retention probability of the classification model.

[0105] For example, the classification model is used to predict a probability P that a loan borrowed by the subject is successfully recovered within 30 days. If P is greater than or equal to 80%, the model classification label mapped thereby is a first rating; if P is greater than or equal to 60% and less than 80%, the model classification label mapped thereby is a second rating; and if P is less than 60%, the model classification label mapped thereby is a third rating.

[0106] In an embodiment, the feature data of the sample subject at each stage of the resource retention process is respectively input into the classification model to obtain a maximum retention day number output by the classification model, where the maximum retention day number is used to describe a maximum retention day number of the resource held by the sample subject. Based on a mapping relationship between the maximum retention day number and the label, the maximum retention day number output by the classification model is converted into a corresponding label to obtain a model classification label. Based on a deviation between the model classification label and the label of the sample subject, the classification model is adjusted in a direction to reduce the deviation.

[0107] Specifically, the retention probability output by the classification model is used to predict a maximum retention day number T of the resource held by the sample subject. Then, according to a mapping relationship between the maximum retention day number and the label, T predicted by the classification model is converted into a corresponding label, and the label converted thereby is a model classification label. Then, based on a deviation between the model classification label and the label of the sample subject, the classification model is adjusted so that the deviation is continuously reduced, so as to improve the accuracy of the retention probability of the classification model.

[0108] In an embodiment, the feature data includes: object basic data used to describe basic attribute information of the subject, object behavior data used to describe historical scheduling behaviors of the subject with respect to the resource held thereby, and object recovery data used to describe recovery measures taken with respect to the subject since the resource held thereby is retained.

[0109] Specifically, the object basic data includes: data used to describe a responsibility category of the subject, data used to describe a survival duration of the subject, and the like.

[0110] The object behavior data includes: data used to describe historical behaviors of the subject to obtain the resource, data used to describe historical behaviors of the subject to release the resource, and the like.

[0111] The object recovery data includes: data used to describe a recovery frequency taken with respect to the subject since the resource held thereby is retained, data used to describe a recovery mode taken with respect to the subject since the resource held thereby is retained, data used to describe interaction information taken with respect to the subject since the resource held thereby is retained, and the like.

[0112] The embodiment has the advantage that the feature data of the object can more comprehensively reflect the external influence applied to the object by using the dynamically updated object recycling data to construct the feature data of the object, and the expression capability of the feature data is improved.

[0113] In an embodiment, the object recycling data comprises recycling behavior data and recycling text data, wherein the recycling behavior data is used to describe attribute information of recycling behavior performed on the object since the resource held by the object is retained, and the recycling text data is used to describe interaction information between the object and the recycling personnel in a text form since the resource held by the object is retained.

[0114] Specifically, the attribute information of the recycling behavior performed on the object comprises recycling frequency performed on the object, recycling mode performed on the object, and the like.

[0115] It should be noted that the object recycling data can only comprise the recycling behavior data, only comprise the recycling text data, or comprise both the recycling behavior data and the recycling text data.

[0116] In an embodiment, interaction text between the object and the recycling personnel in a recycling process since the resource held by the object is retained is acquired. The recycling text data is generated based on the interaction text.

[0117] Specifically, in the recycling process, the object can be interacted with through short messages, emails, or the like to recycle the resource. In this case, the recycling text data can be generated based on the interaction text contained in the short messages or the emails, and the recycling text data is taken as part of the feature data of the object.

[0118] For example, the target object has been interacted with through short messages to recycle the resource since the resource held by the target object is retained. The short message used to interact with the target object in the first retention stage is TEXT1, and the short message used to interact with the target object in the second retention stage is TEXT2. When the resource held by the target object is retained for 1 day, the target object enters the first retention stage, and thus the feature data of the target object in the first retention stage comprises recycling text data generated based on TEXT1. When the resource held by the target object is retained for 2 days, the target object enters the second retention stage, and thus the feature data of the target object in the second retention stage comprises recycling text data generated based on TEXT1 and TEXT2.

[0119] In an embodiment, interaction voice between the object and the recycling personnel in a recycling process since the resource held by the object is retained is acquired. The recycling text data is generated based on the interaction voice.

[0120] Specifically, during the recovery process, the subject can be interacted with through a phone, a voice message, etc. to recover the resource. In this case, the interactive voice contained in the phone or voice message can be converted into recovery text data through voice-to-text technology, and the recovery text data can be taken as part of the feature data of the subject.

[0121] For example, the subject has been interacted with through a phone to recover the resource since the resource held by the subject is retained. The interactive voice contained in the phone used to interact with the subject in the first retention stage is VOICE1, and the interactive voice contained in the phone used to interact with the subject in the second retention stage is VOICE2. When the resource held by the subject is retained for 1 day, the subject enters the first retention stage, and the feature data of the subject in the first retention stage contains recovery text data generated based on VOICE1. When the resource held by the subject is retained for 2 days, the subject enters the second retention stage, and the feature data of the subject in the second retention stage contains recovery text data generated based on VOICE1 and VOICE2.

[0122] In an embodiment, the classification model includes a first sub-model, a second sub-model, and a main model, the output result of the classification model is the output result of the main model, the input of the main model is the output result of the first sub-model and the output result of the second sub-model, the input of the first sub-model is the recovery text data, and the input of the second sub-model is feature data other than the recovery text data.

[0123] Specifically, referring to Figure 3 A structural diagram of a classification model of an embodiment of the present application is shown. In this embodiment, the data source is divided into two parts: data source 1 containing only recovery text data, and data source 2 not containing recovery text data. The first sub-model can be an LR model or an Xgboost model, the second sub-model can be an Xgboost model, and the main model can be an LR model.

[0124] Data source 1 is converted into Embedding features after using Word2vec or Glove algorithm, and then an LR algorithm or an Xgboost algorithm is used for modeling to obtain the first sub-model. Data source 2 is aggregated to form features after feature engineering, and then an Xgboost algorithm is used for modeling to obtain the second sub-model. Then, the two sub-models are fused through the main model to obtain a complete classification model.

[0125] Figure 4 A training diagram of a classification model of an embodiment of the present application is shown.

[0126] Referring to Figure 4As shown, in this embodiment, before training the classification model, the sample object used to obtain the modeling sample is selected. Specifically, the loan customers in the past period of time are selected as the sample object, and the feature data of the sample object is used as the modeling sample. When selecting the modeling sample, on the one hand, the feature data close to the modeling time point is selected as the modeling sample, and on the other hand, the interference of similar marketing activities, seasons and other factors is avoided.

[0127] From the beginning of the overdue to the pre-collection, the feature data of the sample object is always in the modeling sample, that is, the modeling sample repeatedly contains the feature data of a sample object several times until the sample object no longer overdues.

[0128] The purpose of the classification model is to predict whether the borrowed loan will continue to overdue after 30 days of overdue, and therefore the performance period is 30 days after the borrowed loan begins to overdue. There are generally two methods for label definition. One method is to determine an overdue day threshold value by using business experience, if the maximum overdue day of the borrowed loan of the sample object is less than the overdue day threshold value, the label of the sample object is determined as the first rating, if the maximum overdue day of the borrowed loan of the sample object is between the overdue day threshold value and 30 days, the label of the sample object is determined as the second rating, and if the maximum overdue day of the borrowed loan of the sample object is greater than 30 days, the label of the sample object is determined as the third rating. Another method is to use the method of overdue rolling rate analysis, to determine an overdue day threshold value by analyzing the probability of the borrowed loan of the current overdue N days continuing to overdue the next day, and then to determine the label of the sample object according to the comparison between the maximum overdue day of the sample, the overdue day threshold value and 30 days according to the first method.

[0129] In the process of feature engineering, the object basic data, object behavior data, recycling behavior data and recycling text data are obtained. These feature data are divided into two data sources: data source 1 containing only recycling text data, and data source 2 containing other feature data except recycling text data.

[0130] The data source 1 is converted into Embedding feature after using Word2vec or Glove algorithm, and then the LR algorithm or Xgboost algorithm is used for modeling to obtain the first sub-model; the data source 2 is aggregated after feature engineering to form features, and then the Xgboost algorithm is used for modeling to obtain the second sub-model. Then the two sub-models are fused by the main model to obtain the complete classification model.

[0131] After obtaining the complete classification model, it is trained and evaluated until the evaluation result determines that the classification model meets the online standard, and the classification model is deployed and landed.

[0132] When the classification model is deployed, the feature processing logic of the feature engineering environment should form an explanatory requirement document for the model deployment personnel to consult. After the classification model is deployed, the deployment result needs to be verified to ensure that the deployment logic is consistent with the development logic before going online.

[0133] The classification model outputs a rating for the object. Before the classification model goes online, a mapping relationship between the rating and the recycling strategy can be defined, so that after the classification model goes online, different resource recycling strategies can be determined according to different ratings output by the classification model.

[0134] After the classification model goes online, the model indicators are monitored monthly. On the one hand, it is determined whether the classification model has certain distinguishing ability for good and bad customers in business application, and on the other hand, the distribution of the classification model output and the degree of deviation from the modeling sample are always concerned.

[0135] The distinguishing ability of the classification model mainly focuses on the KS (Kolmogorov-Smirnov) indicator. KS is used to evaluate the risk distinguishing ability of the model, and measures the difference between the cumulative distributions of good and bad samples. The greater the cumulative difference between good and bad samples, the greater the KS indicator, and the stronger the risk distinguishing ability of the classification model.

[0136] The distribution of the classification model output and the degree of deviation from the modeling sample mainly focus on the aPSI (Population Stability Index) indicator. The aPSI indicator reflects the stability of the distribution of the validation sample at each rating and the distribution of the modeling sample. Stability is relative, so there need to be two distributions - actual distribution and expected distribution. During modeling, the training sample is usually taken as the expected distribution, and the validation sample is taken as the actual distribution.

[0137] In an embodiment, the updated output result for the target object includes an updated retention probability for the target object, the updated retention probability for the target object being used to describe a probability of successfully recycling resources held by the target object within a preset retention day threshold or a probability of the resources held by the target object continuing to be retained after the retention day threshold. According to the updated retention probability for the target object, an updated rating of the target object is obtained. According to the updated rating of the target object, a resource recycling strategy for the target object is updated.

[0138] Specifically, the resource recycling strategy for the target object is determined through the rating of the target object. Whenever the stage of the target object in the resource retention process is updated, the updated retention probability for the target object is obtained through the trained classification model, and then the rating of the target object is updated, and then the resource recycling strategy for the target object is updated.

[0139] For example, refer to Figure 5 The mapping relationship between the rating and the resource recovery strategy of an embodiment of the present application is shown in the diagram. If the updated rating of a target object is the first rating, it indicates that the target object is a low-risk object in the current stage, and the resource recovery strategy for the target object can be not reminding the target object to return the resource. If the updated rating of the target object is the second rating, it indicates that the target object is a medium-risk object in the current stage, and the resource recovery strategy for the target object can be reminding the target object to return the resource at a relatively low frequency. If the updated rating of the target object is the third rating, it indicates that the target object is a high-risk object in the current stage, and the resource recovery strategy for the target object can be reminding the target object to return the resource at a relatively high frequency.

[0140] Figure 6 A block diagram of a resource recovery strategy obtaining apparatus according to an embodiment of the present application is shown. The apparatus comprises:

[0141] The first obtaining module 310 is configured to obtain a label of a sample object, wherein the label is used to describe the retention degree of the resource held by the sample object.

[0142] The second obtaining module 320 is configured to obtain feature data of the sample object in each stage of the resource retention process.

[0143] The training module 330 is configured to take the feature data of the sample object in each stage of the resource retention process as input, take the label as a mapping target, and train a preset classification model until the output result of the classification model is mapped to the label with an accuracy rate that meets a preset condition, to obtain a trained classification model.

[0144] The first updating module 340 is configured to update the feature data of the target object in the resource retention process in response to the update of the stage in which the target object is located, and input the updated feature data of the target object into the trained classification model to obtain the output result of the trained classification model for the update of the target object.

[0145] The second updating module 350 is configured to update the resource recovery strategy for the target object according to the output result for the update of the target object.

[0146] In an exemplary embodiment of the present application, the first obtaining module is configured to:

[0147] For a preset number of statistical objects, based on the resource retention records of the preset number of statistical objects, obtain the probability that the resource that has been retained for different days continues to be retained the next day.

[0148] determine a threshold of the number of days of retention for dividing the rating based on a comparison result between the probability and a preset probability threshold;

[0149] obtain a maximum number of days of retention of the resource held by the sample object;

[0150] generate the label of the sample object based on a comparison result between the maximum number of days of retention of the resource held by the sample object and the threshold of the number of days of retention.

[0151] In an example embodiment of the present application, the first obtaining module is configured to:

[0152] determine a first threshold of the number of days of retention for dividing the rating based on a comparison result between the probability and a preset probability threshold;

[0153] obtain a second threshold of the number of days of retention for dividing the rating, wherein the second threshold of the number of days of retention is greater than the first threshold of the number of days of retention;

[0154] if the maximum number of days of retention of the resource held by the sample object is less than or equal to the first threshold of the number of days of retention, take the first rating as the label of the sample object;

[0155] if the maximum number of days of retention of the resource held by the sample object is greater than the first threshold of the number of days of retention and less than or equal to the second threshold of the number of days of retention, take the second rating as the label of the sample object;

[0156] if the maximum number of days of retention of the resource held by the sample object is greater than the second threshold of the number of days of retention, take the third rating as the label of the sample object.

[0157] In an example embodiment of the present application, the device is configured to:

[0158] eliminate the sample object with the second rating as the label, and use the sample object with the first rating and the third rating as the label for training the classification model.

[0159] In an example embodiment of the present application, the training module is configured to:

[0160] input the feature data of each stage of the resource retention process of the sample object into the classification model respectively, to obtain the retention probability output by the classification model, wherein the retention probability is used to describe the probability of successfully recovering the resource held by the sample object within a preset threshold of the number of days of retention or the probability of the resource held by the sample object continuing to be retained after the threshold of the number of days of retention;

[0161] convert the retention probability output by the classification model into a corresponding label based on a mapping relationship between the retention probability and the label, to obtain the model classification label.

[0162] adjust the classification model in a direction of reducing the deviation between the model classification label and the label of the sample object based on a deviation between the model classification label and the label of the sample object.

[0163] In an example embodiment of the present application, the training module is configured to:

[0164] input the feature data of the sample object at each stage of the resource retention process into the classification model respectively to obtain the maximum retention days output by the classification model, wherein the maximum retention days are used to describe the maximum retention days of the resource held by the sample object;

[0165] convert the maximum retention days output by the classification model into a corresponding label based on a mapping relationship between the maximum retention days and the label to obtain a model classification label;

[0166] adjust the classification model in a direction of reducing the deviation between the model classification label and the label of the sample object based on a deviation between the model classification label and the label of the sample object.

[0167] In an example embodiment of the present application, the feature data includes: object basic data used to describe basic attribute information of the object, object behavior data used to describe historical scheduling behavior of the object for the resource held by the object, and object recovery data used to describe recovery measures taken for the object since the resource held by the object is retained.

[0168] In an example embodiment of the present application, the object recovery data includes: recovery behavior data used to describe attribute information of the recovery behavior taken for the object since the resource held by the object is retained, and recovery text data in a text form used to describe interaction information between the object and the object in the process of recovering the resource since the resource held by the object is retained.

[0169] In an example embodiment of the present application, the device is configured to:

[0170] obtain interaction text between the object and the object in the process of recovering the resource since the resource held by the object is retained;

[0171] generate the recovery text data based on the interaction text.

[0172] In an example embodiment of the present application, the device is configured to:

[0173] obtain interaction voice between the object and the object in the process of recovering the resource since the resource held by the object is retained;

[0174] generate the recovery text data based on the interaction voice.

[0175] In an example embodiment of the present application, the classification model comprises a first sub-model, a second sub-model and a main model, an output result of the classification model is an output result of the main model, an input of the main model is an output result of the first sub-model and an output result of the second sub-model, an input of the first sub-model is the recovered text data, and an input of the second sub-model is feature data other than the recovered text data.

[0176] In an example embodiment of the present application, the second updating module is configured to:

[0177] update the rating of the target object according to the updated stay probability of the target object, to obtain an updated rating of the target object;

[0178] update the resource recycling strategy for the target object according to the updated rating of the target object.

[0179] The electronic device 40 according to the embodiments of the present application will be described below with reference to Figure 7 FIG. 1. Figure 7 The electronic device 40 shown is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0180] As shown in Figure 7 FIG. 1, the electronic device 40 is in the form of a general computing device. The components of the electronic device 40 can include, but are not limited to, the at least one processing unit 410 described above, the at least one storage unit 420 described above, and a bus 430 connecting different system components, including the storage unit 420 and the processing unit 410.

[0181] The storage unit stores program codes which can be executed by the processing unit 410, so that the processing unit 410 performs the steps according to various example embodiments of the present application described in the description part of the above example method of the present specification. For example, the processing unit 410 can perform each step as shown in Figure 2 FIG. 1.

[0182] The storage unit 420 can include a readable medium in the form of a volatile storage unit, such as a random access memory (RAM) 4201 and / or a cache memory 4202, and can further include a read-only memory (ROM) 4203.

[0183] The storage unit 420 can further include program / utilities 4204 having a set of (at least one) program modules 4205, such as an operating system, one or more application programs, other program modules, and program data, each of which can include, or be included in, the implementation of a network environment, alone or in combination with some examples.

[0184] Bus 430 can be one or more of several types of bus structures including a memory bus or memory controller, a peripheral bus, a graphics bus, a processor or local bus using any of a variety of bus architectures.

[0185] Electronic device 40 can also communicate with one or more external devices 500 such as a keyboard or a pointing device, a Bluetooth device, etc. ; other devices associated with electronic device 40 ; and / or one or more devices that enable user interaction with electronic device 40 (for example, a display, speakers, a haptic output device, etc. ) ; and / or one or more devices that enable communication of electronic device 40 with other computing devices. Such communication can occur via Input / Output (I / O) interface 450. Input / Output (I / O) interface 450 can also communicate human or machine-generated

[0186] Those skilled in the art will readily understand that the example embodiments described herein can be implemented by software and / or by hardware coupled with software, as described above. Thus, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash disk, a mobile hard disk, etc. ) or a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc. ) to perform the methods according to the embodiments of the present application.

[0187] In the example embodiments of the present application, a computer readable storage medium is also provided, which stores computer readable instructions. When the computer readable instructions are executed by a processor of a computer, the computer is enabled to perform the method described in the method embodiment part.

[0188] According to one embodiment of the present application, there is also provided a program product for implementing the method in the above-mentioned method embodiments, which can adopt a portable compact disc read-only memory (CD-ROM) and include program codes, and can run on a terminal device, such as a personal computer. However, the program product of the present application is not limited to this, and in this document, the readable storage medium can be any tangible medium containing or storing a program, which can be used by or in conjunction with an instruction execution system, apparatus or device.

[0189] The program product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0190] The computer readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, in which readable program codes are borne. Such a propagated data signal can take on many forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The readable signal medium can also be any readable medium that is not a readable storage medium and that can transmit, propagate or transport program for use by or in connection with an instruction execution system, apparatus or device.

[0191] The program codes contained on the readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0192] The program code may be implemented in any of various ways, including procedure-based, narrative-based, object-based, and / or architectural-based versions. In a procedure-based version, the program code is implemented as a set of procedures that are called by application programs and / or other procedures. In a narrative-based version, the program code is implemented as a set of functions that are called by application programs and / or other procedures. In an object-based version, the program code is implemented as a set of objects that are called by application programs and / or other objects. In an architectural-based version, the program code is implemented as a set of modules that are called by application programs and / or other modules. The program code can be implemented in a variety of ways, including by conventional and / or meta-programming techniques, such as using Java.

[0193] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, such a division is not mandatory. Indeed, according to embodiments of the application, the features and functionalities of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functionalities of one module or unit described above can be further divided into embodied by a plurality of modules or units.

[0194] Furthermore, although the various steps of the methods of the present application are described in a particular order in the figures, this is not required or implied as to the order of execution of the steps, nor is it required that all of the steps be executed to achieve the desired result. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, one step can be broken into multiple steps, etc.

[0195] From the above description of embodiments of the application, those skilled in the art will readily perceive that the example embodiments described herein can be implemented by software and / or by software in combination with the necessary hardware. Thus, the technical solutions according to the embodiments of the application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash disk, a mobile hard disk, etc.) or a network, and includes a plurality of instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to perform the methods according to the embodiments of the application.

Claims

1. A resource recovery strategy acquisition method, characterized in that, The method comprises: acquiring a label of a sample object, wherein the label is used to describe a retention degree of a resource held by the sample object; acquiring feature data of the sample object at each stage in a resource retention process; the feature data comprises: object basic data used to describe basic attribute information of the object, object behavior data used to describe historical scheduling behavior of the object for the resource held by the object, and object recovery data used to describe recovery measures taken for the object since the resource held by the object is retained; the object recovery data comprises: recovery behavior data used to describe attribute information of recovery behavior taken for the object since the resource held by the object is retained, and recovery text data in a text form used to describe interaction information between the object and a recovery party in a resource recovery process since the resource held by the object is retained; training a preset classification model by taking the feature data of the sample object at each stage in the resource retention process as input and taking the label as a mapping target until an output result of the classification model is mapped to the label with an accuracy rate meeting a preset condition, to obtain a trained classification model; the classification model comprises a first sub-model, a second sub-model and a main model, the output result of the classification model is an output result of the main model, an input of the main model is an output result of the first sub-model and an output result of the second sub-model, an input of the first sub-model is the recovery text data, and an input of the second sub-model is feature data other than the recovery text data; in response to an update of a stage of a target object in a resource retention process, updating feature data of the target object at the stage in the resource retention process, and inputting the updated feature data of the target object into the trained classification model to obtain an output result of the trained classification model for the target object update; updating a resource recovery strategy for the target object according to the output result for the target object update.

2. The method of claim 1, wherein, The label comprises a rating of the sample object, and acquiring the label of the sample object comprises: for a preset number of statistical objects, acquiring a probability that a resource retained for different days continues to be retained on the next day based on resource retention records of the preset number of statistical objects; determining a retention day threshold for dividing ratings based on a comparison result between the probability and a preset probability threshold; acquiring a maximum retention day of the resource held by the sample object; generating the label of the sample object based on a comparison result between the maximum retention day of the resource held by the sample object and the retention day threshold.

3. The method of claim 2, wherein, Determining a retention day threshold for dividing ratings based on a comparison result between the probability and a preset probability threshold comprises: determining a first retention day threshold for dividing ratings based on a comparison result between the probability and a preset probability threshold; Generating the label of the sample object based on a comparison result between the maximum retention day of the resource held by the sample object and the retention day threshold comprises: obtaining a preset second retention day threshold for dividing ratings, wherein the second retention day threshold is greater than the first retention day threshold; if the maximum retention day of the resource held by the sample object is less than or equal to the first retention day threshold, a first rating is taken as the label of the sample object; if the maximum retention day of the resource held by the sample object is greater than the first retention day threshold and less than or equal to the second retention day threshold, a second rating is taken as the label of the sample object; if the maximum retention day of the resource held by the sample object is greater than the second retention day threshold, a third rating is taken as the label of the sample object.

4. The method of claim 3, wherein, The method further comprises: eliminating sample objects labeled with the second rating, and using sample objects labeled with the first rating and the third rating to train the classification model.

5. The method of claim 1, wherein, respectively inputting feature data of each stage of the sample object in the resource retention process into the classification model, obtaining the retention probability output by the classification model, wherein the retention probability is used to describe the probability of successfully recovering the resource held by the sample object within a preset retention day threshold or the probability of the resource held by the sample object continuing to retain after the retention day threshold; based on the mapping relationship between the retention probability and the label, converting the retention probability output by the classification model into a corresponding label to obtain a model classification label; based on the deviation between the model classification label and the label of the sample object, adjusting the classification model in the direction of reducing the deviation. respectively inputting feature data of each stage of the sample object in the resource retention process into the classification model, obtaining the maximum retention day output by the classification model, wherein the maximum retention day is used to describe the maximum retention day of the resource held by the sample object; 6. The method of claim 1, wherein, based on the mapping relationship between the maximum retention day and the label, converting the maximum retention day output by the classification model into a corresponding label to obtain a model classification label; based on the deviation between the model classification label and the label of the sample object, adjusting the classification model in the direction of reducing the deviation. The method further comprises: obtaining interactive text between the object and the object since the resource held by the object has been retained in the recovery process; 7. The method of claim 1, wherein, generating the recovery text data based on the interactive text. The method further comprises: obtaining interactive speech between the object and the object since the resource held by the object has been retained in the recovery process; 8. The method of claim 1, wherein, generating the recovery text data based on the interactive speech. ​ ​ 9. The method of claim 1, wherein, The output result updated for the target object includes a stay probability of the target object, which is used to describe a probability of successfully recycling resources held by the target object within a preset stay day threshold or a probability of the resources held by the target object continuing to stay after the stay day threshold. According to the output result updated for the target object, a resource recycling strategy for the target object is updated, including: According to the stay probability of the target object, an updated rating of the target object is obtained. According to the updated rating of the target object, the resource recycling strategy for the target object is updated.

10. A resource recovery strategy acquisition apparatus characterized by comprising: The device includes: A first obtaining module configured to obtain a label of a sample object, wherein the label is used to describe a stay degree of resources held by the sample object; A second obtaining module configured to obtain feature data of the sample object at each stage in a resource stay process; the feature data includes object basic data used to describe basic attribute information of the object, object behavior data used to describe historical scheduling behaviors of the object for the resources held by the object, and object recycling data used to describe recycling measures taken for the object since the resources held by the object stay; the object recycling data includes recycling behavior data and recycling text data, wherein the recycling behavior data is used to describe attribute information of recycling behaviors taken for the object since the resources held by the object stay, and the recycling text data is used to describe interactive information between the object and the recycling personnel in a recycling process since the resources held by the object stay in a text form; A training module configured to train a preset classification model by taking the feature data of the sample object at each stage in the resource stay process as input and taking the label as a mapping target until an output result of the classification model is mapped to the label with an accuracy rate meeting a preset condition, to obtain a trained classification model; the classification model includes a first sub-model, a second sub-model, and a main model, the output result of the classification model is an output result of the main model, an input of the main model is an output result of the first sub-model and an output result of the second sub-model, an input of the first sub-model is the recycling text data, and an input of the second sub-model is feature data other than the recycling text data; A first updating module configured to update feature data of a target object at a stage in a resource stay process in response to an update of the stage, input updated feature data of the target object into the trained classification model, and obtain an output result of the trained classification model for the target object; A second updating module configured to update a resource recycling strategy for the target object according to the output result of the target object.

11. An electronic device, comprising: One or more processors; ​ a storage device for storing one or more programs, which when executed by the one or more processors, cause the electronic device to implement the method of any of claims 1-9.

12. A computer-readable storage medium, characterized in that, a computer readable medium having stored thereon computer readable instructions which, when executed by a processor of a computer, cause the computer to perform the method of any of claims 1-9.

Images