CN116821503A

CN116821503A - Object recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN116821503A
Application number: CN202310822537.5A
Authority: CN
Inventors: 张敏; 汪佳茵; 张元�; 李彪; 江鹏
Original assignee: Tsinghua University; Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Tsinghua University; Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2023-07-05
Filing date: 2023-07-05
Publication date: 2023-09-29

Abstract

The disclosure relates to an object recommendation method, an object recommendation device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring the current aging parameters of the candidate object; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; determining recommendation indexes of the candidate objects according to account characteristics of the accounts to be recommended, current aging parameters of the candidate objects and object characteristics of the candidate objects; the recommendation index of the candidate object is positively correlated to the current aging parameter of the candidate object; and recommending the object to the account to be recommended based on the recommendation index of the candidate object. According to the object recommendation method, the influence of the object self timeliness on the recommendation result in the object recommendation process can be enhanced, the exposure probability of the candidate object with high recommendation value probability at the current moment is improved, the final recommended object is guaranteed to have high recommendation value, and the problem of unfair exposure of the object caused by bias such as popularity or recommendation modes excessively depending on user preference can be further solved.

Description

Object recommendation method and device, electronic equipment and storage medium

Technical Field

The disclosure relates to the technical field of internet, and in particular relates to an object recommendation method, an object recommendation device, electronic equipment and a storage medium.

Background

In recent years, bias (bias) problems in recommendation systems raise more and more attention and concerns, because bias (such as popularity bias) often causes a small part of objects to get most of exposure resources in the system, resulting in problems such as long tail phenomenon and martai effect, which can cause degradation of consumer experience, loss of creator, and loss of diversity of system ecology.

Therefore, there is a need to develop a fair exposure mechanism at the object level from the perspective of the recommendation system, and implementing reasonable recommendation resource allocation is a key to improving recommendation performance.

Disclosure of Invention

The disclosure provides an object recommendation method, an object recommendation device, electronic equipment and a storage medium, and the technical scheme of the disclosure is as follows:

according to a first aspect of an embodiment of the present disclosure, there is provided an object recommendation method, including:

acquiring the current aging parameters of the candidate object; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; determining a current aging parameter according to the interaction data of the candidate object in the historical time period and the current basic failure parameter; the current basic failure parameter characterizes the basic probability that the candidate object has recommendation value at the current moment and loses the recommendation value in the instant time after the current moment;

Determining recommendation indexes of the candidate objects according to account characteristics of the accounts to be recommended, current aging parameters of the candidate objects and object characteristics of the candidate objects; the recommendation index characterizes the probability of successful recommendation of the candidate object to the account to be recommended; the recommendation index of the candidate object is positively correlated to the current aging parameter of the candidate object;

and recommending the object to the account to be recommended based on the recommendation index of the candidate object.

In some possible embodiments, obtaining the current aging parameters of the candidate includes:

responding to a recommendation request sent by a client corresponding to an account to be recommended, and determining the current moment according to timestamp information corresponding to the recommendation request;

obtaining aging data of a candidate object; the ageing data comprise a plurality of ageing parameters, the ageing parameters correspond to a plurality of different moments, and each ageing parameter in the ageing parameters represents the probability that the candidate object has recommended value at the moment corresponding to each ageing parameter;

based on the current time instant, a current aging parameter is determined from a plurality of aging parameters.

In some possible embodiments, the historical time period takes the uploading time of the candidate object as the initial time; the method further comprises the steps of:

starting timing from an initial time;

When the timing reaches a first preset time length, acquiring interaction data between the candidate object and the recommended account in the first preset time length; the interaction data includes interaction characteristic values of the candidate object with respect to each of the plurality of interaction behaviors;

obtaining an aging parameter determination model; the aging parameter determining model comprises a trained basic failure parameter determining model and weight coefficients corresponding to each interaction behavior; the base failure parameter determination model is used for determining a base probability that a candidate object has a recommendation value at each of a plurality of different moments and loses the recommendation value in an instant time after each moment;

and determining the aging parameters of the candidate object at each different moment according to the trained basic failure parameter determination model, the weight coefficient corresponding to each interaction behavior and the interaction characteristic value of the candidate object about each interaction behavior.

In some possible embodiments, the current aging parameters are determined from an aging parameter determination model; the training mode of the aging parameter determination model comprises the following steps:

acquiring historical interaction data of each historical object in a plurality of historical objects; the historical interaction data of each historical object comprises first interaction data in a first preset duration from the uploading moment of each historical object and second interaction data in a second preset duration from the middle moment; the middle time is the time when the first preset duration is ended;

Determining tag data of each historical object according to the second interaction data of each historical object; the tag data of each history object represents the history moment when each history object loses recommendation value;

acquiring an initial aging parameter determination model; the initial aging parameter determination model comprises a basic failure parameter determination model to be trained and initial weight coefficients corresponding to each interaction behavior;

performing iterative updating on the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior by using the label data of each historical object and the first interaction data of each historical object;

and obtaining a trained aging parameter determination model until the preset finishing training condition is met.

In some possible embodiments, the second interaction data of each historical object includes a number of exposures of each historical object at each time within a second preset time period and an interaction characteristic value of each historical object at each time with respect to each interaction behavior;

determining tag data for each historical object from the second interaction data for each historical object, comprising:

determining the recommended ranking percentage of each historical object in a plurality of historical objects at each moment according to the interaction characteristic value of each historical object at each moment about each interaction behavior;

At each moment, if the exposure quantity of each historical object at the moment is more than or equal to the preset quantity, determining the recommended degree value of each historical object at each moment based on the recommended ranking percentage of each historical object at the moment;

accumulating the recommended degree value of each historical object at each moment to obtain the accumulated recommended degree value of each historical object at each moment;

for each historical object, when the cumulative recommendation degree value is smaller than or equal to a preset recommendation degree value, determining the moment corresponding to the cumulative recommendation degree value as the historical moment when the historical object loses recommendation value.

In some possible embodiments, the first interaction data for each historical object includes an interaction characteristic value for each historical object for each interaction behavior;

and iteratively updating a basic failure parameter determination model to be trained and an initial weight coefficient corresponding to each interaction behavior by using the label data of each historical object and the first interaction data of each historical object, wherein the method comprises the following steps:

determining a model, an initial weight coefficient corresponding to each interaction behavior and an interaction characteristic value of each history object about each interaction behavior according to failure parameters to be trained, and determining a prediction aging parameter of each history object at each moment; the prediction aging parameter of each historical object at each moment represents the prediction probability of the recommendation value of each historical object at each moment;

According to the historical moment when each historical object loses recommendation value and the prediction aging parameter of each historical object at each moment, the basis failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior are iteratively updated by using a maximum likelihood estimation method, and the trained basis failure parameter determination model and the weight coefficient corresponding to each interaction behavior are obtained.

In some possible embodiments, determining the recommendation index for the candidate object based on the account characteristics of the account to be recommended, the current aging parameters of the candidate object, and the object characteristics of the candidate object includes:

determining a correlation index between the account to be recommended and the candidate object according to the account characteristics of the account to be recommended and the object characteristics of the candidate object;

and adjusting the correlation index by using the current aging parameter of the candidate object to obtain the recommended index of the candidate object.

According to a second aspect of the embodiments of the present disclosure, there is provided an object recommendation apparatus, including:

an acquisition module configured to perform acquiring a current aging parameter of the candidate object; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; determining a current aging parameter according to the interaction data of the candidate object in the historical time period and the current basic failure parameter; the current basic failure parameter characterizes the basic probability that the candidate object has recommendation value at the current moment and loses the recommendation value in the instant time after the current moment;

The determining module is configured to determine a recommendation index of the candidate object according to account characteristics of the account to be recommended, current aging parameters of the candidate object and object characteristics of the candidate object; the recommendation index characterizes the probability of successful recommendation of the candidate object to the account to be recommended; the recommendation index of the candidate object is positively correlated to the current aging parameter of the candidate object;

and the recommending module is configured to execute the recommending index based on the candidate object and conduct object recommending to the account to be recommended.

In some possible embodiments, the obtaining module is further configured to execute a recommendation request sent by a client corresponding to the account to be recommended, and determine the current moment according to timestamp information corresponding to the recommendation request; obtaining aging data of a candidate object; the ageing data comprise a plurality of ageing parameters, the ageing parameters correspond to a plurality of different moments, and each ageing parameter in the ageing parameters represents the probability that the candidate object has recommended value at the moment corresponding to each ageing parameter; based on the current time instant, a current aging parameter is determined from a plurality of aging parameters.

In some possible embodiments, the historical time period takes the uploading time of the candidate object as the initial time; the apparatus further comprises:

An aging parameter determination module configured to perform timing from an initial time; when the timing reaches a first preset time length, acquiring interaction data between the candidate object and the recommended account in the first preset time length; the interaction data includes interaction characteristic values of the candidate object with respect to each of the plurality of interaction behaviors; obtaining an aging parameter determination model; the aging parameter determining model comprises a trained basic failure parameter determining model and weight coefficients corresponding to each interaction behavior; the base failure parameter determination model is used for determining a base probability that a candidate object has a recommendation value at each of a plurality of different moments and loses the recommendation value in an instant time after each moment; and determining the aging parameters of the candidate object at each different moment according to the trained basic failure parameter determination model, the weight coefficient corresponding to each interaction behavior and the interaction characteristic value of the candidate object about each interaction behavior.

In some possible embodiments, the current aging parameters are determined from an aging parameter determination model; the apparatus further includes a training module of the aging parameter determination model configured to perform obtaining historical interaction data for each of a plurality of historical objects; the historical interaction data of each historical object comprises first interaction data in a first preset duration from the uploading moment of each historical object and second interaction data in a second preset duration from the middle moment; the middle time is the time when the first preset duration is ended; determining tag data of each historical object according to the second interaction data of each historical object; the tag data of each history object represents the history moment when each history object loses recommendation value; acquiring an initial aging parameter determination model; the initial aging parameter determination model comprises a basic failure parameter determination model to be trained and initial weight coefficients corresponding to each interaction behavior; performing iterative updating on the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior by using the label data of each historical object and the first interaction data of each historical object; and obtaining a trained aging parameter determination model until the preset finishing training condition is met.

the training module of the aging parameter determination model is further configured to perform determining a recommended ranking percentage of each historical object in the plurality of historical objects at each moment according to the interaction characteristic value of each historical object at each moment in time with respect to each interaction behavior; at each moment, if the exposure quantity of each historical object at the moment is more than or equal to the preset quantity, determining the recommended degree value of each historical object at each moment based on the recommended ranking percentage of each historical object at the moment; accumulating the recommended degree value of each historical object at each moment to obtain the accumulated recommended degree value of each historical object at each moment; for each historical object, when the cumulative recommendation degree value is smaller than or equal to a preset recommendation degree value, determining the moment corresponding to the cumulative recommendation degree value as the historical moment when the historical object loses recommendation value.

The training module of the aging parameter determination model is further configured to execute the initial weight coefficient corresponding to each interaction behavior according to the failure parameter determination model to be trained and the interaction characteristic value of each history object about each interaction behavior, and determine the prediction aging parameter of each history object at each moment; the prediction aging parameter of each historical object at each moment represents the prediction probability of the recommendation value of each historical object at each moment; according to the historical moment when each historical object loses recommendation value and the prediction aging parameter of each historical object at each moment, the basis failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior are iteratively updated by using a maximum likelihood estimation method, and the trained basis failure parameter determination model and the weight coefficient corresponding to each interaction behavior are obtained.

In some possible embodiments, the determining module is configured to perform determining a relevance index between the account to be recommended and the candidate object according to the account characteristics of the account to be recommended and the object characteristics of the candidate object; and adjusting the correlation index by using the current aging parameter of the candidate object to obtain the recommended index of the candidate object.

According to a third aspect of embodiments of the present disclosure, there is provided an electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to execute instructions to implement the object recommendation method of the first aspect of the embodiments of the present disclosure.

According to a fourth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium, which when executed by a processor of an electronic device, causes the electronic device to perform the object recommendation method of the first aspect of embodiments of the present disclosure.

According to a fifth aspect of embodiments of the present disclosure, there is provided a computer program product comprising a computer program stored in a readable storage medium, from which at least one processor of the computer device reads and executes the computer program, causing the computer device to perform the object recommendation method of the first aspect of embodiments of the present disclosure.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:

in the object recommendation process, the server acquires the current aging parameters of the candidate objects; comprehensively considering account characteristics of an account to be recommended, current aging parameters of different candidate objects and object characteristics of different candidate objects to determine recommendation indexes of the candidate objects; the current aging parameters of different candidate objects reflect the probability of the existence of recommendation value of different candidate objects at the current moment, and the probability of the existence of recommendation value at the current moment is higher, so that the influence of the object self-aging property on recommendation results in the object recommendation process can be enhanced, the exposure probability of the candidate object with higher probability of existence of recommendation value at the current moment is improved, the object to be finally recommended is ensured to have higher recommendation value, and the problem of unfair exposure of the object caused by bias such as popularity and the like or recommendation modes which depend on user preference excessively can be further improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.

FIG. 1 is a schematic diagram of an application environment shown in accordance with an exemplary embodiment;

FIG. 2 is a flowchart illustrating an object recommendation method according to an exemplary embodiment;

FIG. 3 is a flowchart illustrating one method of obtaining current aging parameters for a candidate object, according to an example embodiment;

FIG. 4 is a flowchart illustrating one method of determining aging parameters at different times, according to an example embodiment;

FIG. 5 is a schematic diagram showing the trend of change in an aging parameter according to an exemplary embodiment;

FIG. 6 is a flowchart illustrating a training manner of the aging parameter determination model, according to an exemplary embodiment;

FIG. 7 is a schematic diagram of a training process, according to an example embodiment;

FIG. 8 is a flowchart illustrating a method of determining tag data for each historical object according to an example embodiment;

FIG. 9 is a schematic diagram illustrating a trend of cumulative recommendation level values, according to an example embodiment;

FIG. 10 is a flowchart illustrating an iterative update according to an exemplary embodiment;

FIG. 11 is a schematic diagram of a prediction process, according to an example embodiment;

FIG. 12 is a schematic diagram of a recommendation process, according to an example embodiment;

FIG. 13 is a flowchart illustrating a method of determining recommendation indicators for candidate objects in accordance with an exemplary embodiment;

FIG. 14 is a block diagram of an object recommendation device, according to an example embodiment;

FIG. 15 is a block diagram of an electronic device for object recommendation, according to an example embodiment.

Detailed Description

In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar first objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.

The user information (including but not limited to user equipment information, user personal information, etc.) related to the present disclosure is information authorized by the user or sufficiently authorized by each party.

An ideal recommender system would not only meet the different preferences of users, but would also provide fair exposure opportunities for recommended objects. Currently, most recommendation systems mainly focus on modeling user preferences to generate recommendation results that are satisfactory to users, but lack focus on object exposure mechanisms, ignoring object fairness issues.

The lack of a reasonable object exposure mechanism may cause serious problems in a real scene. For example, due to popularity bias, a small portion of the objects often get most of the exposure resources in the system, resulting in phenomena such as long tails and martai effects. Unfair exposure between different objects negatively affects the propagation of new objects in the recommendation system and further affects the quality of the recommendation, eventually compromising user satisfaction and the enthusiasm of the creator.

Based on the above, the embodiment of the disclosure provides an object recommendation method, which provides fair exposure opportunities for different objects based on timeliness of the objects, realizes reasonable recommended resource allocation, and further improves recommendation performance.

Referring to fig. 1, fig. 1 is a schematic diagram of an application environment according to an exemplary embodiment, as shown in fig. 1, including a server 01 and a terminal device 02. Alternatively, the server 01 and the terminal device 02 may be connected through a wireless link, or may be connected through a wired link, which is not limited herein.

Referring to fig. 1, a server 01 may provide a background service, such as an object recommendation service, for a terminal device 02. When the server 01 provides the object recommendation service for the terminal device 02, a plurality of candidate objects can be obtained; for each candidate object, the server 01 can comprehensively determine the recommendation index of the candidate object according to the account characteristics of the account to be recommended corresponding to the terminal device 02, the object characteristics of the candidate object and the current aging parameters of the candidate object; then, the server 01 may push the candidate whose recommendation index satisfies a certain requirement to the terminal device 02 according to the recommendation index of each candidate.

In some possible embodiments, the server 01 may include a stand-alone physical server, a server cluster or a distributed system formed by a plurality of physical servers, and may also be a cloud server for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network ), and basic cloud computing services such as big data and artificial intelligence platforms. Operating systems running on the server may include, but are not limited to, android systems, IOS systems, linux, windows, unix, and the like.

In some possible embodiments, the terminal device 02 described above may include, but is not limited to, a smart phone, a desktop computer, a tablet computer, a notebook computer, a smart speaker, a digital assistant, an augmented reality (augmented reality, AR)/Virtual Reality (VR) device, a smart wearable device, and the like. Or may be software running on the client, such as an application, applet, etc. Alternatively, the operating system running on the client may include, but is not limited to, an android system, an IOS system, linux, windows, unix, and the like.

In addition, it should be noted that the application environment shown in fig. 1 is merely an example. In practical applications, the method for recommending the object in the embodiment of the present disclosure may be executed by the terminal device and the server in cooperation, or the method for recommending the object in the embodiment of the present disclosure may be executed independently by the terminal device or the server, which is not limited to a specific application environment.

FIG. 2 is a flowchart illustrating an object recommendation method, as shown in FIG. 2, that may be applied to a server, according to an exemplary embodiment, including the steps of:

in step S201, a current aging parameter of a candidate object is acquired; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; determining a current aging parameter according to the interaction data of the candidate object in the historical time period and the current basic failure parameter; the current base failure parameter characterizes a base probability that the candidate object has a recommendation value at the current time and loses the recommendation value in an instant in time after the current time.

In the embodiment of the disclosure, the candidate object refers to an object which is prepared by the server to recommend to the terminal equipment, and the object is matched with the actual application scene; in some possible application scenarios, the objects may include music, short videos, merchandise, news information, advertisements, and the like. The server can respond to the object recommendation request of the terminal equipment to acquire a plurality of candidate objects, and then push the objects meeting the requirements to the terminal equipment after a series of operations such as screening, sorting and the like are performed on the plurality of candidate objects. Optionally, the server may collect a plurality of candidate objects in advance, and when receiving an object recommendation request sent by the terminal device, may immediately perform operations such as screening and sorting on the plurality of candidate objects; or, after receiving the object recommendation request sent by the terminal device, the server may take the account currently logged in by the terminal device as an account to be recommended, determine a plurality of candidate objects according to the account to be recommended, and perform operations such as screening and sorting on the plurality of candidate objects, so that candidate objects corresponding to different terminal devices are different; in the former scheme, the server has high response speed, and in the latter scheme, the server can provide more personalized recommendation services for different terminal devices.

As described above, in the process of recommending objects to a terminal device, the server generally needs to perform operations such as screening and sorting on a plurality of candidate objects; in the related technology, the selection and the sorting are performed in long-term dependence on the preference of the account to be recommended, so that various bias problems occur in the recommendation result, and the phenomenon of Martai effect and the like are caused. In particular, the newly uploaded object cannot obtain enough exposure opportunities, which is very unfavorable for timely propagation of the current class object; the popular objects are generally long in uploading time, and the popular objects are frequently recommended for a long time, so that the masses cannot acquire new contents in time; therefore, the unfair exposure among different objects easily causes the problems of reduced user experience, reduced enthusiasm of creator creation, reduced recommendation diversity loss and the like.

Based on this, in the embodiment of the present disclosure, when the server performs object recommendation to the terminal device, a plurality of candidate objects are first obtained, and at the same time, a current aging parameter of each candidate object in the plurality of candidate objects is also obtained, where the current aging parameter characterizes a probability that the candidate object has a recommendation value at the current time. The current time may be an acquisition time, a recommendation time, or a time in response to an object recommendation request of the terminal device. It should be noted that, at any moment, the object has either a recommendation value or no recommendation value; objects with higher probability of having a recommendation value have higher timeliness, e.g., current events objects are generally more timeliness, and the recommendation value of such objects is higher; conversely, objects that have a lower probability of having a recommendation value or that do not have a recommendation value have a lower timeliness, such objects having a lower recommendation value. Therefore, the current aging parameters of each candidate object can be synthesized during subsequent processing of the server, the exposure probability of the candidate object with higher probability of having recommendation value at the current moment is improved, and the object which is finally pushed to the terminal equipment is ensured to have higher recommendation value; here, the recommended value refers to a value measured from the degrees such as user satisfaction, practicality, and the like.

Generally, in the current event object, the closer the uploading time is to the object at the current moment, the higher the timeliness is, and the higher the probability of having recommendation value at the current moment is, so that the newly uploaded current event object can obtain higher recommendation probability by virtue of the higher current timeliness parameter, and indirectly, the recommendation probability of the object which is hot but has lower timeliness can be reduced; therefore, the method and the device for processing the exposure information of the current class object start from the timeliness of the object, ensure that the object with high timeliness can be timely recommended, solve the problem of unfair exposure among different objects, shorten the cold start time of the current class object, and simultaneously avoid pushing the outdated current class object to a user.

The meaning of the "ageing parameters" is further described in detail below.

First, the current aging parameter of the candidate, i.e., the aging parameter of the candidate at the current time. The aging parameters of each candidate object are changed along with time, and the aging parameters of the candidate objects are different at different current moments or recommended moments; that is, the current aging parameters for each candidate are dynamically changing. In general, the aging parameters of different types of objects have different trends over time, and the aging parameters of the same type of object have substantially the same trend over time. For example, the displayed value of the current event object is obviously reduced along with the time loss, and the recommended value is quickly lost after hot spot iteration or new information appears, so that the ageing parameter of the current event object is increased along with the time and is reduced; for example, the reading topic objects share the contents in classical works, the recommendation value of the reading topic objects is not obviously reduced along with the increase of uploading time, and the reading topic objects are not basically influenced by time.

Considering that the objects still have large individual differences under the same type, in the embodiment of the disclosure, the server determines, in advance, for each object, an aging parameter of each object at different moments; therefore, when receiving an object recommendation request of the terminal equipment, the server can directly call and obtain ageing parameters of the candidate object at the current moment;

specifically, in some possible embodiments, the step S201 of obtaining the current aging parameter of the candidate object may include the following steps as shown in fig. 3:

in step S301, in response to a recommendation request sent by a client corresponding to an account to be recommended, the current time is determined according to timestamp information corresponding to the recommendation request.

The terminal equipment is a client corresponding to the account to be recommended. The timestamp information corresponding to the recommendation request may include the time when the server receives the recommendation request or the time when the client carried in the recommendation request sends the recommendation request. Thus, in this step, after receiving the recommendation request, the server may determine the time when the recommendation request is received or the time when the client sends the recommendation request as the current time.

In step S303, aging data of the candidate object is acquired; the aging data comprises a plurality of aging parameters, the plurality of aging parameters correspond to a plurality of different moments, and each aging parameter in the plurality of aging parameters characterizes the probability that the candidate object has recommended value at the moment corresponding to each aging parameter.

Here, the number of candidates is plural; as described above, the plurality of candidate objects may be predetermined by the server before receiving the recommendation request, or may be determined by the server according to the account to be recommended corresponding to the recommendation request; specific ways of determining may refer to the prior art, which is not limited by the present disclosure.

In the step, a server acquires ageing data of each candidate object, wherein the ageing data comprises a plurality of ageing parameters corresponding to a plurality of different moments, and each ageing parameter represents the probability that the candidate object has a recommendation value at the corresponding moment; the time-effect data is pre-calculated by the server, and specific calculation time and calculation mode will be described in detail below, which is not repeated here. A plurality of different times are determined based on the calculated time, the plurality of different times including each time within a future time period after the calculated time; specifically, the time granularity may take minutes, hours, days, weeks, etc.; for example, taking the time granularity of hours as an example, a plurality of different times including each hour within 7 days of the future, the aging data includes aging parameters corresponding to each of 7 x 24 hours; for another example, taking the time granularity as a day, a plurality of different times including a day of 7 days into the future, the aging data includes corresponding aging parameters for each day of 7 days.

In step S305, a current aging parameter is determined from a plurality of aging parameters based on the current time.

In the step, the server determines a time matched with the current time from a plurality of different times based on the current time, and then determines an aging parameter corresponding to the time as the current aging parameter.

For example, the server calculates the aging data of candidate a at 2022, 3, 1, 8, 00, including the aging parameters corresponding to each hour between 2022, 3, 1, 9, 00 and 2022, 3, 8, 9, 00; and when the server receives the recommendation request from the 12:30 days 3 and 3 of 2022, the server can determine that the current moment is 12 points of the 12 days 3 and 3 of 2022, find the aging parameters corresponding to the 12 points of the 3 days and 3 of 2022 from the pre-calculated aging data, and determine the aging parameters as the current aging parameters.

In the above embodiment, the server calculates the aging parameters of each candidate object at different moments in advance, and can match the corresponding aging parameters according to the actual recommendation moment during the object recommendation process; therefore, the calculation work of a large number of parameters is processed offline, so that the time for online real-time calculation can be saved, the storage and calculation resources can be saved, the processing of the subsequent steps can be quickened, and the overall recommendation efficiency can be improved.

The specific meaning of the "aging parameters" is explained further below.

In the embodiment of the disclosure, the current aging parameters of the candidate object are defined by combining the relevant knowledge of survival analysis, and the current aging parameters characterize the probability that the candidate object has recommendation value at the current moment. The survival analysis is mainly used for analyzing and deducing the occurrence time of a given event, and when the survival analysis is applied to a recommendation scene, the given event can be understood as the transition of an object from the presence of a recommendation value to the loss of the recommendation value; thus, the current aging parameter of the object can be expressed by equation (1):

GRV _i (t)＝P(T _i >t) (1)

wherein GRV _i (t) represents an aging parameter of the object i at time t; t (T) _i The specific calculation method will be described in detail below, which indicates the time point when the object i makes a state transition from the presence of the recommended value to the loss of the recommended value.

Thus, in connection with equation (1), the current age parameter of the object can be understood as the probability that the object has not yet lost the recommended value at the current time.

Before describing how to calculate the aging parameters of an object, several basic concepts in survival analysis are first described:

Risk function (Hazard function): h is a _i (t), representing the risk of instantaneous death of subject i at time t, at an instantaneous time t+δ (t) after time t, can be represented by a limit, as shown in formula (2) below:

when applied to the recommendation scenario of the present disclosure, it can be understood that the object i has a recommendation value at the time t, and the probability of losing the recommendation value within the instant time t+δ (t) after the time t.

Through equivalent deduction, the risk function h _i (t) and aging parameter GRV _i (t) the following formula (3):

cumulative risk function (cumulative hazard function): h _i (t), from h _i (t) integral to indicate the cumulative risk of death, which is related to the aging parameter GRV _i The equivalent relationships of (t) are as follows (4) and (5):

GRV _i (t)＝exp(-H _i (t)) (5)

it can be seen that only h of object i at each instant is determined _i (t) obtaining GRV of the object i at the future time t _i (t). Thus, in the practice of the present disclosure, the GRV of i of the object can be calculated _i (t) conversion to h at each time instant of the computation object i _i (t)；

Further, consider P (T _i >t) is affected by a number of factors; in a recommendation scenario, the probability that an object has not lost recommendation value at time t is related to multiple attributes of the object, while h at each time is determined for object i _i (t) in embodiments of the present disclosure, a Cox proportional risk regression model is used to analyze multiple attributes of an object with h _i (t) relationship between (c) and (d). The formula of the Cox proportional hazards regression model is as follows (6):

wherein i is _m An attribute value representing the mth attribute of the object i;representing the average attribute value; sigma (sigma) _i 1 represents the number of attributes; alpha _m A weight coefficient representing an mth attribute; h is a ₀ (t) is a base risk function of the population of objects for determining a base failure parameter, i.e. the base of the object having a recommended value at time t and losing the recommended value at a transient time after time tProbability; h is a ₀ (t) is the same for any one object, so that the difference between different objects is only +.>Is different from the above. Wherein alpha is _m And h ₀ (t) need to be obtained by training.

In the formula (6), the following is usedIntegrally as covariates, h _i (t) is a dependent variable; in other embodiments, as shown in formula (7) below, i can be used _m Replace->I is i _m Is covariate, h _i And (t) is a dependent variable:

h _i (t)＝h ₀ (t)*exp(∑ _m α _m (i _m )) (7)

the advantage of equation (6) over equation (7) is that each attribute value of the object is averaged, i.e., each attribute value is subtracted by the average attribute value, which is beneficial to alpha in the parameter training process _m And h ₀ And (3) fast fitting of (t).

By combining the living analysis concept, the aging parameters of the object are influenced by various attributes of the object, so that the aging parameters of the object at different moments can be predicted by utilizing the various attributes of the object. In some possible embodiments, the plurality of attributes of the object may be determined according to the interaction data of the candidate object in the historical time period, and then the plurality of attributes of the object may include an interaction characteristic value of at least one interaction behavior, such as a play completion rate, a click rate, a praise rate, and the like; then, combining the pre-trained basic risk function and the weight coefficient corresponding to each interaction behavior, and determining h of the object i at a plurality of different moments according to the formula (6) _i (t) obtaining the aging parameters GRV of the object i at a plurality of different moments according to the formulas (4) and (5) _i (t)。

The embodiment introduces a calculation mode of the aging parameter, and the aging parameter needs to be determined according to the interaction data of the object in the historical time period, so that the calculation time of the aging parameter is necessarily after the historical time period after the object is uploaded; accordingly, in some possible embodiments, the historical time period takes the uploading time of the candidate object as the initial time; the object recommendation method of the embodiment of the present disclosure further includes the following steps as shown in fig. 4:

in step S401, the timer starts from the initial time.

In step S403, when the timer reaches the first preset duration, interaction data between the candidate object and the recommended account within the first preset duration is obtained.

In the step, for each candidate object, the server takes the uploading time of the candidate object as the initial time, starts timing from the initial time, and acquires interaction data between the candidate object and the recommended account in the first preset time when the timing reaches the first preset time. The first preset duration may be determined according to actual requirements, for example, the first preset duration may be half a day, 1 day, 2 days, etc.; the interaction data includes candidate objects about each interaction characteristic value of a plurality of interactions, the interactions are described above, including but not limited to at least one of playing, clicking, praying, focusing and commenting, the corresponding interaction characteristic values of playing behaviors can include a play rate, the interaction characteristic values of clicking behaviors can include a click rate, the interaction characteristic values of praying behaviors can include a praying rate, the interaction characteristic values of focusing behaviors can include a focusing rate, the interaction characteristic values of commenting behaviors can include a commenting rate, and the specific calculation mode of the interaction characteristic values of each behavior adopts a calculation method commonly used in the art, which is not repeated in the present disclosure.

In step S405, an aging parameter determination model is acquired; the aging parameter determination model comprises a trained basic failure parameter determination model and weight coefficients corresponding to each interaction behavior.

In this step, the server obtains a model for determining the aging parameters, which is designed to use the interaction characteristics of the object with respect to each interactionThe sign value predicts the aging parameters of the subject at different moments through a series of mathematical operations. The aging parameter determination model needs to train the basic failure parameter determination model and the weight coefficient alpha corresponding to each interaction behavior in advance _m The method comprises the steps of carrying out a first treatment on the surface of the Wherein the basic failure parameter determination model refers to the basic risk function h ₀ (t), a base risk function h ₀ (t) is the same for different objects, and is only time dependent, for determining a base probability that an object has a recommendation value at each of a plurality of different times and loses the recommendation value at an instant in time after each time.

In step S407, according to the trained basic failure parameter determination model, the weight coefficient corresponding to each interaction behavior, and the interaction characteristic value of the candidate object about each interaction behavior, the aging parameters of the candidate object at each different moment in a plurality of different moments are determined.

In the recommendation process, the server inputs the interaction characteristic value of each candidate object about each interaction behavior into the aging parameter determining model according to the steps S405-S407, the model automatically calls the trained basic failure parameter determining model and the weight coefficient corresponding to each interaction behavior, and sequentially calculates according to the formulas (6), (4) and (5), and finally outputs the aging parameters of each candidate object at different moments.

In a specific example, the interaction data of the candidate object a of the current event class in 1 day after uploading is obtained, specifically may include a play completion rate, a click rate, a praise rate and the like of the candidate object a in 1 day, the play completion rate, the click rate and the praise rate of the candidate object a in 1 day are input into an aging parameter determination model, and aging parameters of each hour of the candidate object a in 7 days in the future are predicted; referring to fig. 5, fig. 5 exemplarily shows the trend of the aging parameters of candidate a every hour in the future 7 days.

In the above embodiment, the aging parameter determining model is established by combining with the survival analysis technology, and the server predicts the aging parameters of the candidate object at different moments in the future by collecting the interaction data of the candidate object in a period of time and then utilizing the interaction data, wherein the aging parameters of different objects can reflect the probability of the recommendation value of different objects in the same time dimension, so that the data support of the object layer can be provided for the selection of the recommendation object in the subsequent recommendation process, the influence of the object self-aging on the recommendation result is enhanced, and the problem of unfair exposure of the object caused by bias such as popularity is improved.

The basic failure parameter determining model and the weight coefficient corresponding to each interaction behavior in the aging parameter determining model need to be trained in advance, and the training mode of the aging parameter determining model is described below.

In some possible embodiments, the training method of the aging parameter determination model may include the following steps as shown in fig. 6:

in step S601, history interaction data of each of a plurality of history objects is acquired.

In the training process, the server mainly utilizes the historical interaction data corresponding to a plurality of historical objects to train parameters. Thus, the server firstly acquires the historical interaction data of each historical object, wherein the historical interaction data of each historical object comprises first interaction data in a first preset duration from the uploading moment of each historical object and second interaction data in a second preset duration from the middle moment; the intermediate time is the time when the first preset duration ends. The first preset duration refers to the embodiment of step S403 above, and the second preset duration may be determined according to actual requirements, for example, 5 days, 7 days, etc.

The steps are described in detail below with reference to fig. 7. As shown in fig. 7, three history objects are shown in fig. 7: the history object B, the history object C and the history object D are taken as examples to illustrate the specific content of the history interaction data. From the whole, the history interaction data of the history object B includes: from the time of upload of history object B (in t ₀ Indicated) for a first predetermined time period (indicated by T) ₁ Represented) from a time instant (denoted by t) ₁ Indicated) for a second preset time period (indicated by T) ₂ Representation) of the second interaction number withinAccording to the above; wherein the intermediate time t ₁ For a first preset time period T ₁ Ending time, a second preset time length T ₂ Time of end t ₂ And (3) representing. From a specific dimension, each column is an interaction characteristic value of an object about a certain interaction behavior at different moments, and 4 columns in an exemplary fig. 7 can be respectively a click rate, a play rate, a praise rate and a attention rate; each horizontal line is an interaction characteristic value of the object about different interaction behaviors at a certain moment; thus, the historical interaction data for each historical object can be understood as a matrix of time dimensions and interaction behavior dimensions.

In step S603, tag data of each history object is determined according to the second interaction data of each history object.

Before training the model parameters, the trained tag data needs to be determined. In the step, the server determines label data of each historical object according to second interaction data of each historical object aiming at each historical object; the tag data of each history object represents the history moment when each history object loses recommendation value; it should be noted that, the historical moment when the historical object actually loses the recommendation value needs to occur within the second preset duration, otherwise, effective tag data cannot be obtained, which is not beneficial to training of the model and convergence of model parameters.

The label data can be determined by adopting a manual labeling mode, and can be automatically determined by formulating a corresponding strategy. An embodiment of automatic determination by the corresponding policy is described below.

accordingly, the determining the tag data of each history object according to the second interaction data of each history object may include the following steps as shown in fig. 8:

in step S801, a recommended ranking percentage of each history object in the plurality of history objects at each moment is determined according to the interaction characteristic value of each history object with respect to each interaction behavior at each moment.

In the step, the server considers the recommendation ranking percentage of each history object under different time dimensions, namely different moments; specifically, for each moment, the server determines the ranking order of each historical object in a plurality of historical objects according to the interaction characteristic value of each interaction behavior of each historical object at the moment, and then converts the ranking order into ranking percentage.

This is explained with reference to fig. 7. Taking the history object B as an example, the server performs the same steps on other history objects. The server is directed to T ₂ Inner instant t ₁ To t ₂ Each time in between, determining the recommended ranking percentage of the historical object B at the time, using R _B (t) represents R _B (t)＝Rank(F _B (t),f _I (t)); wherein F is _B (t) represents the interaction characteristic value of each interaction behavior of the historical object B at the moment t; f (f) _I (t) represents a set of interaction characteristic values for each interaction behavior at time t for all historical objects; here, assuming a total of 8 history objects, it can be seen that history object B is at time t ₁ The recommendation ranking percentage is R _B (t ₁ )＝25％.

In step S803, at each time, if the number of exposures of each history object at the time is equal to or greater than the preset number, the recommendation degree value of each history object at each time is determined based on the recommendation ranking percentage of each history object at the time.

In the step, the server compares the exposure quantity of each historical object at different moments with the preset quantity from the time dimension, and if the exposure quantity is larger than or equal to the preset quantity, the server determines the recommendation degree value of each historical object at the moment based on the recommendation ranking percentage of each historical object at the moment; the preset number may be set to 1, or may be set according to actual requirements.

In a specific example, if the number of exposure is greater than or equal to 1, that is, the number of exposure is not 0, at each time, the server may directly determine the recommended ranking percentage of each history object at the time as the recommended degree value of each history object at the time.

Alternatively, in another specific example, the recommended degree value is determined according to the following formula (8):

v _i (t)＝1 _{(Expi(t)＞0)} *(R _i (t)-β _E )+1 _{(Expi(t)＝0)} *(-β _nE ) (8)

wherein v is _i (t) represents a recommended degree value of the object i at the time t; r is R _i (t) represents the recommended ranking percentage of object i at time t; exp (Exp) _i (t) represents the exposure quantity of the object i at the time t; meet Exp _i (t)>At the time of 0, the temperature of the liquid,otherwiseIn the same way, satisfy Exp _i (t) =0>Otherwise->β _E 、β _nE Is super-parameter and can be set according to actual requirements, and is exemplified by beta _E And beta _nE Are all set to 0.5;

in step S805, the recommendation level values of each history object at each time are accumulated, so as to obtain an accumulated recommendation level value of each history object at each time.

In the step, the server accumulates the recommendation degree value of each historical object at each time in the past for each historical object to obtain the accumulated recommendation degree value of each historical object at the current time.

In step S807, when the cumulative recommended degree value is equal to or smaller than the preset recommended degree value for each history object, the time corresponding to the cumulative recommended degree value is determined as the history time at which the history object loses the recommended value.

In the step, a server monitors the cumulative recommendation degree value of each historical object at each moment aiming at each historical object; optionally, the server may perform visualization processing on the monitored data to form a trend chart of cumulative recommendation values as shown in fig. 9, where the cumulative recommendation values of different historical objects are generally gradually decreasing along with the increase of time; the horizontal dotted line in the figure represents a preset recommendation degree value, wherein the preset recommendation degree value refers to the lowest value which can confirm that the object has a recommendation value, and if the object is lower than the preset recommendation degree value, the object loses the recommendation value; when the cumulative recommended degree value of the historical object touches the dotted line, the cumulative recommended degree value is equal to the preset recommended degree value, and the subsequent cumulative recommended degree value may be smaller than the preset recommended degree value, namely the object loses the recommended value; therefore, the time when the cumulative recommendation level value touches the dotted line is the historical time when the historical object loses the recommendation value.

Specifically, the server may determine a historical moment when the historical object loses the recommendation value according to the following formula (9):

wherein T is _i A historical moment of losing recommendation value of the object i is represented;representing t _x A cumulative recommended degree value for time object i; beta _d And representing a preset recommended degree value. Equation (9) may be understood as solving the minimum time when the cumulative recommended degree value is smaller than the preset recommended degree value.

In step S605, an initial aging parameter determination model is acquired; the initial aging parameter determination model comprises a basic failure parameter determination model to be trained and initial weight coefficients corresponding to each interaction behavior.

In the step, the server acquires an initial aging parameter determination model, namely, builds a corresponding mathematical model based on the formulas (4) - (6), and then determines a basic failure parameter determination model to be trained and initial weight coefficients corresponding to each interaction behavior. For example, for four interactions in FIG. 7: clicking, playing, ordering and focusing to set corresponding initial weight coefficients respectively.

In step S607, the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior are iteratively updated by using the tag data of each history object and the first interaction data of each history object.

In step S609, a trained aging parameter determination model is obtained until a preset end training condition is satisfied.

In the step, a server utilizes an initial aging parameter determination model to analyze the first interaction data of each historical object and predict the aging parameter of each historical object at each moment in a second preset time length; performing iterative updating on the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior according to the label data of each historical object and the predicted aging parameter of each historical object until the preset ending training condition is met, so as to obtain a trained aging parameter determination model; the preset end training conditions comprise a basic failure parameter determination model to be trained and initial weight coefficient fitting corresponding to each interaction behavior.

The process of iterative updating is described in detail below.

In some possible embodiments, the first interaction data for each historical object includes an interaction characteristic value for each historical object for each interaction behavior; the iterative updating of the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior by using the tag data of each historical object and the first interaction data of each historical object may include the following steps as shown in fig. 10:

in step S1001, according to the basic failure parameter determination model to be trained, the initial weight coefficient corresponding to each interaction behavior, and the interaction characteristic value of each history object with respect to each interaction behavior, the prediction aging parameter of each history object at each moment is determined.

In the step, a server inputs each interaction characteristic value of each historical object in a first preset time length into an initial aging parameter determining model for each historical object, and predicts the probability of recommendation value of each historical object in a second preset time length based on the basic failure parameter determining model to be trained and an initial weight coefficient corresponding to each interaction behavior to obtain a predicted aging parameter of each time of the historical object in the second preset time length.

As shown in fig. 11, the history object B is displayed for a first preset time period T ₁ Inner instant t ₀ To t ₁ Inputting the interaction characteristic values at each moment between-1 into an initial aging parameter determination model, and based on the basic failure parameter determination model to be trained and an initial weight coefficient corresponding to each interaction behavior, inputting a second preset time length T for the historical object B ₂ Predicting the probability of recommendation value at each time in the time frame, wherein, taking each time frame of the second preset time period as an example of each hour in 7 days, the history object B at t can be obtained ₁ -1 a predicted aging parameter for each hour within 7 days after future;

correspondingly, in the actual recommendation process, the server determines a model and a weight coefficient corresponding to each interaction behavior by using the trained basic failure parameters, and executes a corresponding process in fig. 11 on each interaction characteristic value of the candidate object 1, the candidate object 2 and the candidate object 3 within a first preset duration to finally obtain aging parameters of the candidate object 1, the candidate object 2 and the candidate object 3 at each moment within a second preset duration; the server may present the prediction in a visual form, and the upper right hand corner of fig. 11 shows, by way of example, the trend of the aging parameters for candidate 1, candidate 2 and candidate 3 for each hour over the next 7 days.

In step S1003, according to the historical time when each historical object loses the recommended value and the prediction aging parameter of each historical object at each time, the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior are iteratively updated by using the maximum likelihood estimation method, so as to obtain the trained basic failure parameter determination model and the weight coefficient corresponding to each interaction behavior.

In the step, the server performs iterative update on the basic risk function to be trained and the initial weight coefficient corresponding to each interaction behavior by using a maximum likelihood estimation method according to the label data, namely the historical moment when each historical object loses the recommendation value, and the prediction data, namely the prediction aging parameter of each historical object at each moment, until the fitted basic risk function and the weight coefficient corresponding to each interaction behavior are obtained. The specific iteration process, referring to the iteration method commonly used in the art, is not described in detail in this disclosure.

In the above embodiment, the training mode of the time efficiency parameter determining model, the tag data required for training and the data transfer process in the training process are introduced, the time efficiency parameter determining model is built based on survival analysis and a Cox proportion risk regression model, the server utilizes the historical interaction data of the historical object to fit the unknown parameters in the time efficiency parameter determining model, the trained time efficiency parameter determining model is obtained after the parameter fitting, and the time efficiency parameter determining model is subsequently used for determining the time efficiency parameters of the candidate object; different from the definition of object timeliness in the related art, the embodiment of the disclosure creatively introduces a survival analysis technology and a Cox proportion risk regression model, can analyze the influence of various different factors on the timeliness parameters of the candidate objects at the same time, and can effectively improve the accuracy of the timeliness parameters.

In step S203, a recommendation index of the candidate object is determined according to the account characteristics of the account to be recommended, the current aging parameter of the candidate object, and the object characteristics of the candidate object.

In step S205, object recommendation is performed to the account to be recommended based on the recommendation index of the candidate object.

In the embodiment of the disclosure, the server may comprehensively determine the recommendation index of the candidate object according to the account characteristics of the account to be recommended, the current aging parameter of the candidate object and the object characteristics of the candidate object; the recommendation index characterizes the probability of successful recommendation of the candidate object to the account to be recommended, and the recommendation index of the candidate object is relevant to the current aging parameter of the candidate object; and finally, recommending the object to the account to be recommended based on the recommendation index of each candidate object in the plurality of candidate objects.

As shown in fig. 12, in a specific application scenario, a server responds to a recommendation request of an account to be recommended to obtain a plurality of candidate objects, including candidate object 1, candidate object 2, candidate object 3 and … …, and further obtains aging data corresponding to each candidate object; then, according to the current moment, determining the current aging parameters corresponding to the candidate objects at the current moment from the aging data corresponding to the candidate objects; then, the server performs fusion processing on account characteristics of the account to be recommended, current aging parameters of the candidate objects and object characteristics of the candidate objects, and determines recommendation indexes of the candidate objects.

Further optionally, the server sorts the plurality of candidate objects according to the sequence of the recommendation indexes from large to small, and pushes the candidate objects with the front N digits of the sorting to the client corresponding to the account to be recommended; wherein N is an integer of 1 or more. The recommendation index of the candidate object is positively correlated to the current aging parameter of the candidate object, so that the higher the current aging parameter of the candidate object is, the higher the corresponding recommendation index is, and the probability that the object with high aging is recommended can be improved.

Wherein the account characteristics may include behavioral characteristics and/or attribute characteristics of the account to be recommended, wherein the behavioral characteristics include, but are not limited to, preferred object types, preferred interaction behaviors, and the like; attribute features include, but are not limited to, account gender, age, territory, etc.; the object features of the candidate object are determined according to what is actually displayed by the object, and can be determined according to an object feature extraction algorithm commonly used in the art, and the disclosure is not repeated.

In some possible embodiments, determining the recommendation index of the candidate object according to the account feature of the account to be recommended, the current aging parameter of the candidate object, and the object feature of the candidate object may include the following steps as shown in fig. 13:

In step S1301, a correlation index between the account to be recommended and the candidate object is determined according to the account characteristics of the account to be recommended and the object characteristics of the candidate object.

In this step, the server may determine, according to the account characteristics of the account to be recommended and the object characteristics of the candidate object, a correlation index between the account to be recommended and the candidate object, where the correlation index characterizes the degree of interest of the account to be recommended in the candidate object. Optionally, the server may use a feature fusion algorithm commonly used in the art, for example, a dot product operation, to perform feature fusion processing on the account feature of the account to be recommended and the object feature of the candidate object, so as to obtain a correlation index between the account to be recommended and the candidate object.

In step S1303, the correlation index is adjusted by using the current aging parameter of the candidate object, so as to obtain a recommendation index of the candidate object.

In the step, a server adjusts the correlation index by utilizing the current aging parameter of the candidate object to obtain the recommended index of the candidate object; optionally, the current aging parameter of the candidate object is directly multiplied by the correlation index, and the multiplied result is the recommended index of the candidate object.

In the above embodiment, the server synthesizes account characteristics of the account to be recommended, current aging parameters of the candidate object and object characteristics of the candidate object to determine recommendation indexes of the candidate object; compared with the prior art, the method and the device for ranking the candidate objects directly utilize the correlation between the account to be recommended and the candidate objects to guide the ranking of the plurality of candidate objects finally.

In summary, according to the object recommendation method in the embodiment of the present disclosure, a server considers the difference of timeliness of different candidate objects, so as to comprehensively consider account characteristics of an account to be recommended, current timeliness parameters of different candidate objects, and object characteristics of different candidate objects to determine recommendation indexes of each candidate object; the current aging parameters of different candidate objects reflect the probability of the recommendation value of the different candidate objects at the current moment, and the higher the probability of the recommendation value at the current moment, the higher the recommendation index is, so that in the object recommendation process, the influence of the object self-aging on recommendation results is enhanced, the exposure probability of the candidate objects with the higher probability of the recommendation value at the current moment is improved, the objects which are finally pushed to the terminal equipment are guaranteed to have higher recommendation values, and the problem of unfair exposure of the objects caused by bias such as popularity and the like or recommendation modes which excessively depend on user preference can be solved.

FIG. 14 is a block diagram of an object recommendation device, according to an example embodiment. Referring to fig. 14, the apparatus includes an acquisition module 1401, a determination module 1402, and a recommendation module 1403;

an acquisition module 1401 configured to perform acquiring current aging parameters of the candidate object; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; determining a current aging parameter according to the interaction data of the candidate object in the historical time period and the current basic failure parameter; the current basic failure parameter characterizes the basic probability that the candidate object has recommendation value at the current moment and loses the recommendation value in the instant time after the current moment;

a determining module 1402 configured to perform determining a recommendation index for the candidate object according to account characteristics of the account to be recommended, current aging parameters of the candidate object, and object characteristics of the candidate object; the recommendation index characterizes the probability of successful recommendation of the candidate object to the account to be recommended; the recommendation index of the candidate object is positively correlated to the current aging parameter of the candidate object;

the recommending module 1403 is configured to execute object recommendation to the account to be recommended based on the recommendation index of the candidate object.

In some possible embodiments, the obtaining module 1401 is further configured to execute a recommendation request sent in response to a client corresponding to an account to be recommended, and determine the current time according to timestamp information corresponding to the recommendation request; obtaining aging data of a candidate object; the ageing data comprise a plurality of ageing parameters, the ageing parameters correspond to a plurality of different moments, and each ageing parameter in the ageing parameters represents the probability that the candidate object has recommended value at the moment corresponding to each ageing parameter; based on the current time instant, a current aging parameter is determined from a plurality of aging parameters.

In some possible embodiments, the determining module 1402 is configured to perform determining a relevance indicator between the account to be recommended and the candidate object according to the account characteristics of the account to be recommended and the object characteristics of the candidate object; and adjusting the correlation index by using the current aging parameter of the candidate object to obtain the recommended index of the candidate object.

The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.

Fig. 15 is a block diagram illustrating an electronic device for object recommendation, which may be a terminal, according to an exemplary embodiment, and an internal structure diagram thereof may be as shown in fig. 15. The electronic device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the electronic device is configured to provide computing and control capabilities. The memory of the electronic device includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the electronic device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement an object recommendation method. The display screen of the electronic equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the electronic equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the structure shown in fig. 15 is merely a block diagram of a portion of the structure associated with the disclosed aspects and is not limiting of the electronic device to which the disclosed aspects apply, and that a particular electronic device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In an exemplary embodiment, there is also provided an electronic device including: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement the object recommendation method as in the embodiments of the present disclosure.

In an exemplary embodiment, a computer readable storage medium is also provided, which when executed by a processor of an electronic device, causes the electronic device to perform the object recommendation method in the embodiments of the present disclosure.

In an exemplary embodiment, there is also provided a computer program product containing instructions, the computer program product comprising a computer program stored in a readable storage medium, from which at least one processor of the computer device reads and executes the computer program, causing the computer device to perform the object recommendation method of the embodiments of the present disclosure.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An object recommendation method, comprising:

acquiring the current aging parameters of the candidate object; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; the current aging parameters are determined according to the interaction data of the candidate object in the historical time period and the current basic failure parameters; the current base failure parameter characterizes a base probability that the candidate object has a recommendation value at the current time and loses the recommendation value in an instant time after the current time;

Determining a recommendation index of the candidate object according to account characteristics of an account to be recommended, current aging parameters of the candidate object and object characteristics of the candidate object; the recommendation index characterizes the probability of successful recommendation of the candidate object to the account to be recommended; the recommendation index of the candidate object is positively related to the current aging parameter of the candidate object;

2. The object recommendation method according to claim 1, wherein the obtaining the current aging parameters of the candidate object comprises:

responding to a recommendation request sent by a client corresponding to the account to be recommended, and determining the current moment according to timestamp information corresponding to the recommendation request;

obtaining aging data of the candidate object; the ageing data comprise a plurality of ageing parameters, the ageing parameters correspond to a plurality of different moments, and each ageing parameter in the plurality of ageing parameters represents the probability that the candidate object has recommended value at the moment corresponding to each ageing parameter;

determining the current aging parameter from the plurality of aging parameters based on the current time.

3. The object recommendation method according to claim 2, wherein the history period takes an uploading time of the candidate object as an initial time; the method further comprises the steps of:

starting timing from the initial time;

when the timing reaches a first preset time length, acquiring interaction data between the candidate object and the recommended account in the first preset time length; the interaction data comprises interaction characteristic values of the candidate object about each interaction behavior of a plurality of interaction behaviors;

obtaining an aging parameter determination model; the aging parameter determining model comprises a trained basic failure parameter determining model and weight coefficients corresponding to each interaction behavior; the base failure parameter determination model is configured to determine a base probability that the candidate object has a recommendation value at each of the plurality of different moments and loses the recommendation value within an instant time after each moment;

and determining aging parameters of the candidate object at each different moment according to the trained basic failure parameter determination model, the weight coefficient corresponding to each interaction behavior and the interaction characteristic value of the candidate object relative to each interaction behavior.

4. The object recommendation method according to claim 1, wherein the current aging parameter is determined according to an aging parameter determination model; the training mode of the aging parameter determination model comprises the following steps:

acquiring historical interaction data of each historical object in a plurality of historical objects; the historical interaction data of each historical object comprises first interaction data in the first preset duration from the uploading moment of each historical object and second interaction data in the second preset duration from the middle moment; the intermediate time is the time when the first preset duration ends;

determining the tag data of each historical object according to the second interaction data of each historical object; the tag data of each historical object represents the historical moment when each historical object loses recommendation value;

iteratively updating the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior by using the label data of each history object and the first interaction data of each history object;

5. The object recommendation method according to claim 4, wherein the second interaction data of each history object includes an exposure amount of each history object at each time within the second preset time period and an interaction characteristic value of each history object with respect to each interaction behavior at each time;

the determining the tag data of each historical object according to the second interaction data of each historical object comprises the following steps:

determining a recommended ranking percentage of each historical object in the plurality of historical objects at each moment according to the interaction characteristic value of each historical object at each moment about each interaction behavior;

at each moment, if the exposure quantity of each historical object at the moment is greater than or equal to a preset quantity, determining a recommendation degree value of each historical object at each moment based on a recommendation ranking percentage of each historical object at the moment;

accumulating the recommendation degree value of each historical object at each moment to obtain an accumulated recommendation degree value of each historical object at each moment;

And for each historical object, when the accumulated recommendation degree value is smaller than or equal to a preset recommendation degree value, determining the moment corresponding to the accumulated recommendation degree value as the historical moment when the historical object loses recommendation value.

6. The object recommendation method of claim 4 wherein said first interaction data for each historical object comprises an interaction characteristic value for said each historical object for said each interaction behavior;

the step of iteratively updating the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior by using the label data of each history object and the first interaction data of each history object comprises the following steps:

determining a model, an initial weight coefficient corresponding to each interaction behavior and an interaction characteristic value of each historical object about each interaction behavior according to the failure parameter to be trained, and determining a predicted aging parameter of each historical object at each moment; the prediction aging parameter of each historical object at each moment represents the prediction probability that each historical object has recommendation value at each moment;

And according to the historical moment when each historical object loses the recommended value and the predicted aging parameter of each historical object at each moment, carrying out iterative updating on the basic failure parameter determination model to be trained and the initial weight coefficient corresponding to each interaction behavior by using a maximum likelihood estimation method to obtain the trained basic failure parameter determination model and the weight coefficient corresponding to each interaction behavior.

7. The method for recommending objects according to any one of claims 1 to 6, wherein the determining the recommendation index of the candidate object according to the account feature of the account to be recommended, the current aging parameter of the candidate object, and the object feature of the candidate object comprises:

and adjusting the correlation index by utilizing the current aging parameter of the candidate object to obtain the recommended index of the candidate object.

8. An object recommendation device, characterized by comprising:

an acquisition module configured to perform acquiring a current aging parameter of the candidate object; the current aging parameter characterizes the probability that the candidate object has recommendation value at the current moment; the current aging parameters are determined according to the interaction data of the candidate object in the historical time period and the current basic failure parameters; the current base failure parameter characterizes a base probability that the candidate object has a recommendation value at the current time and loses the recommendation value in an instant time after the current time;

A determining module configured to perform determining a recommendation index for the candidate object according to account characteristics of an account to be recommended, current aging parameters of the candidate object, and object characteristics of the candidate object; the recommendation index characterizes the probability of successful recommendation of the candidate object to the account to be recommended; the recommendation index of the candidate object is positively related to the current aging parameter of the candidate object;

and the recommending module is configured to execute object recommendation to the account to be recommended based on the recommending index of the candidate object.

9. An electronic device, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the object recommendation method of any one of claims 1-7.

10. A computer readable storage medium, characterized in that instructions in the computer readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the object recommendation method of any one of claims 1-7.