CN110971659A

CN110971659A - Recommendation message pushing method and device and storage medium

Info

Publication number: CN110971659A
Application number: CN201910967133.9A
Authority: CN
Inventors: 邢可
Original assignee: Beike Technology Co Ltd
Current assignee: Seashell Housing Beijing Technology Co Ltd
Priority date: 2019-10-11
Filing date: 2019-10-11
Publication date: 2020-04-07

Abstract

The application discloses a recommendation message pushing method, a recommendation message pushing device and a storage medium, and relates to the internet technology. The scheme specifically comprises the following steps: determining a target user meeting a preset screening time condition, and acquiring attribute information of the target user and behavior information of the target user for at least one object to be recommended; for each object to be recommended, determining the attention degree score of a target user to the object to be recommended according to the behavior category and the behavior time of each operation behavior of the object to be recommended; determining the recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended and the attention degree score of each object to be recommended; determining a preset number of target objects to be recommended according to the recommendation degree of the objects to be recommended, and pushing recommendation information generated according to the target objects to be recommended to a target user when a preset recommendation time condition is met. The method and the device can improve the selection probability of the target user for the object to be recommended and improve the transaction rate.

Description

Recommendation message pushing method and device and storage medium

Technical Field

The present application relates to internet technologies, and in particular, to a method and an apparatus for pushing a recommendation message, and a storage medium.

Background

The user needs to go through a certain selection period from browsing, clicking, searching and other operation behaviors of the object to be recommended to selecting the object to be recommended. In the selection period, the user's will to select the recommended object will change constantly.

In the prior art, a recommendation mechanism for an object to be recommended, which pushes a recommendation message to a target user at a proper time, is absent, and the target user may run away at any stage in a selection period, so that the transaction rate is affected.

Disclosure of Invention

In view of the above, a main object of the present application is to provide a method for pushing recommendation messages, which can improve the selection probability of a target user for a to-be-recommended object and improve the transaction rate by pushing the recommendation messages to the target user at a proper time.

In order to achieve the purpose, the technical scheme provided by the application is as follows:

in a first aspect, an embodiment of the present application provides a method for pushing a recommendation message, including the following steps:

determining a target user meeting a preset screening time condition, and acquiring attribute information of the target user and behavior information of the target user for at least one object to be recommended; the behavior information comprises the behavior category and the behavior time of the target user for operating the object to be recommended at least once;

for each object to be recommended, determining the attention degree score of the target user to the object to be recommended according to the behavior category and the behavior time of each operation behavior of the object to be recommended; the attention degree score is used for evaluating the interest degree of the target user on the object to be recommended;

acquiring attribute information of each object to be recommended, and determining the recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended and the attention degree score of each object to be recommended;

and determining a preset number of target objects to be recommended from the objects to be recommended according to the recommendation degree of each object to be recommended, and pushing a recommendation message generated according to the target objects to the target user when a preset recommendation time condition is met.

In a possible implementation manner, the step of determining the target user meeting the preset filtering time condition includes:

acquiring at least one user who has not generated a transaction on a transaction object and has performed an operation action on the transaction object;

and aiming at each user, determining the target user according to the behavior time of the user for operating the transaction object, the current time and the time threshold corresponding to the screening time condition.

In a possible implementation manner, for each object to be recommended, the step of determining the attention degree score of the target user for the object to be recommended according to the behavior category and the behavior time of each operation behavior on the object to be recommended includes:

aiming at each operation behavior of the user on the object to be recommended, determining the score of the operation behavior according to the duration between the behavior time of the operation behavior and the current time;

determining the score weight of the operation behavior according to the behavior category of the operation behavior;

and determining the attention degree score of the recommended object according to the score of each operation behavior and the score weight.

In one possible embodiment, the score for the action is determined using the following formula:

R＝e^-λx

wherein, R is the score value, x is the time difference between the date of the operation behavior and the current date, and lambda is the decay rate.

In a possible implementation manner, the step of determining the recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended, and the attention degree score of each object to be recommended includes:

determining a preset number of objects to be recommended with the maximum attention degree value as alternative objects to be recommended according to the attention degree value of the target user to each object to be recommended;

generating a characteristic vector according to the attribute information of the target user and the attribute information of each alternative object to be recommended;

and inputting the characteristic vector into a pre-trained recommendation degree calculation model to obtain the recommendation degree of each object to be recommended.

In a possible implementation, the step of pre-training the recommendation degree calculation model includes:

acquiring each user performing an operation action on the sample object to be recommended as a sample user;

constructing a training sample set according to the attribute information of each sample user, the attribute information of each sample object to be recommended and the sample object to be recommended actually selected by each sample user;

inputting the training sample set into the initial recommendation calculation model to obtain a loss value;

and updating parameters of the initial recommendation degree calculation model according to the loss value to obtain the recommendation degree calculation model.

In a possible implementation manner, the step of constructing a training sample set according to the attribute information of each sample user, the attribute information of each sample object to be recommended, and the sample object to be recommended actually selected by each sample user includes:

for each sample user, extracting a sample feature vector according to the attribute information of the sample user and the attribute information of the sample object to be recommended of the sample user who performs the operation behavior;

constructing the training sample of the sample user according to the sample feature vector of the sample user and the sample object to be recommended actually selected by the sample user; the training sample set includes the training samples of each of the sample users.

In a possible implementation manner, the step of updating the parameters of the initial recommendation degree calculation model according to the loss value to obtain the recommendation degree calculation model includes:

calculating a residual of a loss function based on the loss value of the initial recommendation calculation model;

fitting the sample characteristic vector and the residual error to obtain a fitting result;

and determining the recommendation degree calculation model according to the initial recommendation degree calculation model and the fitting result.

In one possible embodiment, the sample feature vector and the residual are fitted using the following formula:

wherein, c_jFor the fitting result, R_jIs the region where the fitting result is located, x_iSample feature vector, y, for the ith sample user_iSample to-be-recommended object actually selected by the ith sample user, f (x)_i) Calculate output of model for the ith training sample for initial recommendation, L (y)_i，f(x_i) + c) is a loss function, c being a constant.

In one possible embodiment, the recommendation calculation model is determined using the following formula:

wherein f (x) is a recommendation degree calculation model, c_jIs the fitting result, j is the number of the fitting result, I is the correctness parameter of the initial recommendation degree calculation model, f₀(x) A model is calculated for the initial recommendation.

In a possible implementation manner, after the step of updating the parameters of the initial recommendation degree calculation model according to the loss value and before the step of obtaining the recommendation degree calculation model, the method further includes:

judging whether a preset iteration number threshold value is reached;

when the initial recommendation degree calculation model does not reach a preset iteration number threshold value, updating parameters of the initial recommendation degree calculation model according to the loss value to obtain an intermediate recommendation degree calculation model;

inputting the training sample set into the intermediate recommendation degree calculation model to obtain a loss value;

and updating parameters of the intermediate recommendation degree calculation model according to the loss value, and returning to the step of judging whether the threshold value of the preset iteration times is reached.

In a possible implementation manner, after the step of determining a predetermined number of target objects to be recommended, before the step of pushing a recommendation message generated according to the target objects to the target user, the method further includes:

generating the recommendation message using the attribute information of the target recommendation object.

acquiring a contact person associated with the target recommendation object, and acquiring attribute information of the contact person;

and generating the recommendation message according to the attribute information of the contact.

In a second aspect, based on the same design concept, an embodiment of the present application further provides a recommendation message pushing apparatus, including:

the target user determination module is used for determining target users meeting the preset screening time condition;

the information acquisition module is used for acquiring the attribute information of the target user and the behavior information of the target user aiming at least one object to be recommended; the behavior information comprises the behavior category and the behavior time of the target user for operating the object to be recommended at least once;

the attention degree score module is used for determining the attention degree score of the target user to each object to be recommended according to the behavior category and the behavior time of each operation behavior of the object to be recommended; the attention degree score is used for evaluating the interest degree of the target user on the object to be recommended;

the recommendation degree determining module is used for acquiring attribute information of each object to be recommended and determining the recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended and the attention degree score of each object to be recommended;

the recommendation message generation module is used for determining a preset number of target objects to be recommended from the objects to be recommended according to the recommendation degree of each object to be recommended and generating recommendation messages according to the target objects to be recommended;

and the recommendation message pushing module is used for pushing the recommendation message generated according to the target recommendation object to the target user when a preset recommendation time condition is met.

In one possible embodiment, the target user determination module is configured to:

In one possible embodiment, the attention scoring module is configured to:

In one possible implementation, the attention score module determines the score of the operation behavior by using the following formula:

R＝e^-λx

In a possible implementation manner, the recommendation degree determining module is configured to:

In a possible implementation manner, the recommendation message pushing apparatus further includes a model training module, configured to:

In one possible embodiment, the model training module is configured to:

In a possible embodiment, the model training module is further configured to:

In a possible embodiment, the model training module is further configured to: fitting the sample feature vector and the residual using the following formula:

In a possible embodiment, the model training module is further configured to: determining the recommendation degree calculation model by adopting the following formula:

In a possible embodiment, the model training module is further configured to:

judging whether a preset iteration number threshold value is reached;

In a possible implementation manner, the recommendation message generation module is configured to:

In a possible implementation manner, the recommendation message generating module is further configured to:

In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which can improve recommendation degrees of a target user for an object to be recommended, and improve a transaction rate. The specific scheme is as follows:

a computer readable storage medium storing computer instructions which, when executed by a processor, may implement the steps of any one of the possible embodiments of the first aspect and the first aspect.

In a fourth aspect, the embodiment of the present application further provides an electronic device, which can improve the recommendation degree of a target user for an object to be recommended, and improve a transaction rate. The specific scheme is as follows:

an electronic device comprising the computer-readable storage medium described above, further comprising a processor that can execute the computer-readable storage medium.

In summary, the present application provides a method, an apparatus, and a storage medium for pushing a recommendation message. According to the method and the device, the attention degree score of the target user to the object to be recommended is determined by analyzing the behavior information of the target user to operate the object to be recommended, the interest degree of the target user to each object to be recommended can be determined according to the attention degree score, the attribute information of the target user, the attribute information of each object to be recommended and the attention degree score of each object to be recommended are comprehensively analyzed, the recommendation degree of each object to be recommended is determined, a preset number of objects to be recommended are determined according to the recommendation degree of each object to be recommended, and the selection probability of the target user to the object to be recommended can be improved. In addition, the target users meeting the preset screening time condition are determined, and when the preset recommendation time condition is met, the recommendation message generated according to the target recommendation object is pushed to the target users, so that the recommendation message can be pushed to the target users at proper time, and the transaction rate is improved.

Drawings

Fig. 1 is a schematic flowchart of a method for pushing a recommendation message according to an embodiment of the present application;

fig. 2 is a schematic flowchart of another method for pushing a recommendation message according to an embodiment of the present application;

fig. 3 is a system architecture diagram of another method for recommending message pushing according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of a recommendation message pushing device according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

The terms "first," "second," "third," "fourth," and the like in the description and in the claims of the present application and in the drawings described above, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprising" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, article, or apparatus.

The recommendation message pushing method provided by the embodiment of the application can be applied to a network transaction scene for trading any transaction object, before the network transaction is completed, a user usually conducts at least one operation behavior such as browsing, clicking, searching or collecting on the transaction object on a network transaction platform, the transaction object is known and screened through the operation behaviors, the user usually needs to go through a certain selection period from the operation behavior of browsing, clicking, searching or collecting on the transaction object to the transaction object selected to generate the transaction, and in the selection period, the selection will of the transaction object will change constantly through the knowledge of each transaction object. Without intervention, the target user may be lost at any stage in the selection cycle.

In view of this, in the embodiment of the present application, by pushing the recommendation message to the target user at an appropriate time, the selection probability of the target user for the object to be recommended is improved, and the transaction rate is improved.

To make the objects, technical solutions and advantages of the present application more clear, the present application will be further described in detail with reference to the accompanying drawings and specific embodiments.

Example one

Fig. 1 is a schematic flow chart of a first embodiment of the present application, as shown in fig. 1, the first embodiment mainly includes:

s101: determining a target user meeting a preset screening time condition, and acquiring attribute information of the target user and behavior information of the target user for at least one object to be recommended; the behavior information comprises the behavior category and the behavior time of the target user for operating the object to be recommended at least once.

Here, the object to be recommended is a transaction object in network transaction, and the specific object to be recommended is a transaction object subjected to an operation by a target user in network transaction. In this embodiment, a preset screening time condition is set, and a user meeting the screening time condition is determined as a target user from users who perform an operation on an object to be recommended.

The attribute information of the target user is information describing features of the target user, and the attribute information of the target user is usually static information, and usually does not change for a long time, and specifically may include information such as age, sex, location, and the like of the target user.

The behavior information of the target user for the object to be recommended is generated when the target user performs an operation on the object to be recommended, and the operation on the object to be recommended is generally an operation on a page, an application program or a network platform displaying attribute information of the object to be recommended, and may specifically include a browsing behavior, a clicking behavior, a searching behavior, a collecting behavior, and the like. The operation behavior of the target user on the object to be recommended and the behavior information of the operation behavior can be acquired by adding a data acquisition interface or a buried point in a page, an application program or a network platform for displaying the attribute information of the object to be recommended. The operation behaviors performed by the target user on the object to be recommended generally belong to dynamic information, and the operation behaviors performed by the target user on the object to be recommended are often increased before and after the target user selects the object to be recommended.

The behavior information may specifically include a behavior category and a behavior time when the target user performs an operation on the object to be recommended at least once. Here, the behavior category is a kind of operation behavior, and for example, browsing behavior, clicking behavior, searching behavior, collecting behavior, and the like are different behavior categories. The action time is information on a time point when the operation action is generated, may be information on a date when the operation action is generated, or may be information on a date and hour when the operation action is generated.

S102: for each object to be recommended, determining the attention degree score of the target user to the object to be recommended according to the behavior category and the behavior time of each operation behavior of the object to be recommended; and the attention degree score is used for evaluating the interest degree of the target user on the object to be recommended.

In the implementation process, the interest degree of the target user in the object to be recommended, which generates the over-operation behavior, of the target user gradually fades along with the time, or the interest of the target user in the object to be recommended, which generates the over-operation behavior, of the target user is transferred to a new object to be recommended along with the time, and the interest degree of the target user in the object to be recommended, which generates the over-operation behavior, of the target user is often timeliness, so that the interest degree of the target user in the object to be recommended can be determined according to the behavior time of the target user in each operation behavior of the object to be recommended.

In addition, for the operation behavior generated by the target user for the object to be recommended, different behavior categories represent the interest level of the object to be recommended, for example, compared with the behavior of the target user for browsing the object to be recommended, the behavior of the target user actively searching the object to be recommended represents that the interest level of the target user for the object to be recommended is higher. Therefore, according to the behavior category of each operation behavior of the target user on the object to be recommended, the interest degree of the target user on the object to be recommended can be determined.

By integrating the behavior time and the behavior category of each operation behavior of the target user on the object to be recommended, the attention degree score of the target user on the object to be recommended can be determined, the interest degree of the target user on each object to be recommended is evaluated by the attention degree score, and the interest and preference description of the target user on each object to be recommended can be more specific, more definite and more targeted.

S103: and acquiring attribute information of each object to be recommended, and determining the recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended and the attention degree score of each object to be recommended.

The attribute information of the object to be recommended is information describing characteristics of the object to be recommended, such as information of unit price, size, position, type, environment and the like of the object to be recommended. The attribute information of the object to be recommended is also static information, and usually does not change for a long time. According to the attribute information of the target user, the attribute information of each object to be recommended and the attention degree score of each object to be recommended, the recommendation degree of each object to be recommended of the target user performing the operation behavior can be determined. Here, the recommendation degree is a numerical value representing the estimated probability that the target user may select the object to be recommended to generate a trading action.

S104: and determining a preset number of target objects to be recommended from the objects to be recommended according to the recommendation degree of each object to be recommended, and pushing a recommendation message generated according to the target objects to the target user when a preset recommendation time condition is met.

Generally, the larger the recommendation degree is, the more likely the target user selects the object to be recommended to generate a trading behavior, and therefore, a predetermined number of objects to be recommended with the maximum recommendation degree may be determined as the target object to be recommended. The recommendation message is generated according to the target recommendation object, and the recommendation message is generated according to the object to be recommended with the largest recommendation degree, so that the selection probability of the target user for the object to be recommended can be effectively improved.

Too many pushed recommendation messages are easy to cause objectionability, and too few pushed recommendation messages cannot achieve the recommendation effect, so that the recommendation messages are pushed to the target user when the preset recommendation time condition is met, the recommendation messages are pushed to the target user at proper time, and the transaction rate is effectively improved.

Example two

As shown in fig. 2, another method for pushing a recommendation message provided in the embodiment of the present application includes:

s201: and determining the target users meeting the preset screening time condition.

The target user is determined from at least one user who has never made a transaction to the transaction object and has performed an operational action with respect to the transaction object.

Specifically, the target user meeting the preset screening time condition may be determined according to the following steps 1 and 2:

step 1, at least one user who has not generated a transaction for a transaction object and has performed an operation action for the transaction object is obtained.

For convenience of judgment, a user who has not transacted the trading object within a preset time range may be acquired, for example, when the embodiment of the present application is applied to the house property trading field, an advisory behavior such as a broker operation or an online advisory behavior may be acquired, and a user who has not transacted the trading object within 60 days before the advisory behavior is performed for the trading house property and within 3 days after the advisory behavior is performed for the trading house property. Here, the consulting action of the trading property may refer to a consulting action through an Instant Messenger (IM) or a telephone, for example, a 400 telephone.

And 2, aiming at each user, determining a target user according to the behavior time of the user for operating the transaction object, the current time and a time threshold corresponding to the screening time condition.

A screening time condition may be specified for the current time, the screening time condition corresponding to a time threshold specified based on the current time; different screening time conditions can be defined for different operation behaviors, and each screening time condition corresponds to a time threshold value defined based on the behavior time of the operation behavior.

For example, when the embodiment of the present application is applied to the field of house property trading, different screening time conditions may be defined according to the current time and different operation behaviors, for example, the user does not initiate a consultation behavior within seven days before the current time, the user has more than 40 hours after initiating the last consultation behavior, the user has not responded to a conversation of a consultant until the current time after initiating the consultation behavior, the user has not effectively communicated with the consultant within 7 days after initiating the consultation behavior, the user has not reached a trading behavior of a paired trade house property within 2 days after effectively communicating with the consultant, the user has browsed the trading house property within seven days before the current time, and the user meeting at least one of the above example screening time conditions may be determined as a target user.

In a possible implementation manner, more than two different screening time conditions can be combined for use, corresponding operation behavior scenes are set according to different screening time condition combinations, and users meeting different operation behavior scenes are screened as target users.

In the actual implementation process, repeated users may be screened out when different screening time conditions are met or different operation behavior scenes are met, and at this time, the screened users need to be subjected to duplicate removal operation, and then target users are obtained. Furthermore, priorities can be set for different screening time conditions or operation behavior scenes, the screening time conditions or the operation behavior scenes met by the target user are recorded according to the priorities during deduplication, and the recorded screening time conditions or the recorded operation behavior scenes met by the target user are stored as attribute information of the target user.

In addition, the user who has not performed a transaction on the transaction object and has performed an operation on the transaction object may also be another merchant or a malicious user with a competitive relationship, so that before the target user is determined according to the behavior time of the user performing an operation on the transaction object, the current time and the time threshold corresponding to the screening time condition, the user may be screened in advance based on a preset blacklist, and the target user is determined from the users who are not on the blacklist.

S202: and for each object to be recommended, determining the attention degree score of the target user to the object to be recommended according to the behavior category and the behavior time of each operation behavior of the object to be recommended.

Because the interest degree of the target user in the object to be recommended, on which the target user generates the over-operation behavior, is often timeliness, the score of the operation behavior can be determined according to the duration between the behavior time of the operation behavior and the current time for each operation behavior of the user on the object to be recommended.

The action time of the operation action may be the date of executing the operation action, or may be the specific hour, minute and second of executing the operation action, and correspondingly, the current time may be the current date, or may be the current hour, minute and second. The time length between the action time of the operation action and the current time can be the time difference between the date of executing the operation action and the current date, or the time difference between the specific hour/minute of executing the operation action and the current hour/minute.

For example, when the embodiment of the present application is applied to the real estate transaction field, the score of the operation behavior can be determined according to the time difference between the date and the current date of the operation behavior. For example, the score of the operation behavior may be determined by using the following formula (1).

R＝e^-λx(1)

Wherein, R is a score, x is a time difference between the date of the operation behavior and the current date, and λ is an attenuation rate, which is used for representing the attenuation speed of the interest degree of the object to be recommended, on which the target user generates the operation behavior, along with time. The specific value of lambda can be set or adjusted according to expert experience.

For example, for the target user, for example, for each operation behavior of the target user a, the score of each operation behavior is determined, and the score table shown in table 1 below can be obtained. Wherein, the user identification column (ID) characterizes the target user to which each operation behavior belongs. The property ID column represents the transaction object targeted by each operation behavior, and the transaction object that has been operated by the target user is the object to be recommended, so the property ID also represents the object to be recommended targeted by each operation behavior. The time difference is the time difference between the date of the operation row and the current date. The operation behavior column represents the behavior category of the operation behavior of the target user specifically aiming at the object to be recommended. The attribute column represents attribute information of an object to be recommended, wherein the object to be recommended is a trading property, and therefore the attribute column represents attribute information of the trading property. The score column characterizes the score of the action.

TABLE 1 scoring table for each operation behavior of target user A

In addition, because the behavior categories of different operation behaviors of the target user to be recommended represent the interest degree of the object to be recommended, the score weight of the operation behavior can be determined according to the behavior category of the operation behavior. Here, the score weight is a weight corresponding to the score of each operation behavior described above.

For example, when the embodiment of the present application is applied to the field of house property trading, generally, the target user has more interest preference on actively searched trading house property than on actively clicked trading house property, and the target user has more interest preference on actively clicked trading house property than on passively browsed trading house property. Therefore, according to the behavior category of each operation behavior, the score of the operation behavior is given a score weight.

For example, for each operation behavior of the target user a, the score weights of the browsing behavior B, the clicking behavior C and the searching behavior S of the target user may be set to ω_B＝1、ω_C1.5 and ω_s＝2。

Specifically, the score of each operation behavior may be weighted and summed according to the score weight of each operation behavior, so as to obtain the attention score of the recommendation object.

For example, when the embodiment of the present application is applied to the real estate transaction field, the following formula (2) may be used to determine the attention score of the recommended object.

Wherein, w (k)_j) A focus score, ω, for the recommended object_CIs a score weight, omega, corresponding to the click behavior of the object to be recommended_BA score weight, omega, corresponding to the browsing behavior of the object to be recommended_SThe score weight corresponding to the search behavior of the object to be recommended,

is the score of the click behavior for the object to be recommended,

is the score weight corresponding to the browsing behavior of the object to be recommended,

is the score weight, k, corresponding to the search behavior of the object to be recommended_jIs identified for the recommended object.

Illustratively, for each operation behavior of the target user a, the three operation behaviors in table 1 are all for different objects to be recommended, and therefore, the attention scores of the three objects to be recommended are respectively shown in table 2.

TABLE 2 attention score of target user A object to be recommended

S203: and determining the preset number of objects to be recommended with the maximum attention degree value as alternative objects to be recommended according to the attention degree value of the target user to each object to be recommended.

The target user usually undergoes a long selection process from the operation of the object to be recommended to the selection of the object to be recommended by the target user for trading, in the process, the number of the objects to be recommended for which the target user performs the operation may be large, the reference meaning of the object to be recommended for which the target user performs the operation before the target user is longer is smaller for the object to be recommended for which the target user finally selects for trading, and the attention score is determined according to the behavior category and the behavior time of the operation. Therefore, the objects to be recommended of the operation behavior of the target user can be selected less preliminarily according to the attention degree score of each object to be recommended, and the objects to be recommended with the maximum attention degree score and the preset number of objects to be recommended are determined as the candidate objects to be recommended. And the target object to be recommended is determined from the alternative objects to be recommended, and the target object to be recommended is stronger in pertinence and higher in accuracy than the target object to be recommended is determined from all the objects to be recommended.

S204: and generating a feature vector according to the attribute information of the target user and the attribute information of each candidate object to be recommended.

The attribute information of the target user and the attribute information of each candidate object to be recommended are obtained from the database, for example, when the embodiment of the application is applied to the field of house property transaction, the attribute information of the target user, such as age, sex, registration time and the like, can be obtained. Attribute information such as unit price, area, floor, type of family room, and location of the trading house corresponding to the object to be recommended may also be obtained, and a main infrastructure around the trading house and a distance between the trading house and the main infrastructure may also be calculated as the attribute information by using an Application Programming Interface (API) provided by a map. Here, the primary infrastructure may include hospitals, supermarkets, schools, malls, parks, sports facilities, and the like.

Any One of common vector generation modes such as One-Hot coding (One-Hot), feature description and the like can be utilized to generate feature vectors according to the attribute information of the target user and the attribute information of each candidate object to be recommended. And preprocessing the characteristic vector, extracting abnormal values in the characteristic vector, and normalizing the continuous characteristic vector.

After the feature vector is obtained, the importance of each dimension feature in the feature vector can be obtained by using a feature engineering method or according to expert experience, each dimension feature in the feature vector is screened according to the importance of each dimension feature, the feature with higher importance is reserved, and the feature with higher importance is used for training a recommendation calculation model.

S205: and inputting the characteristic vector into a pre-trained recommendation degree calculation model to obtain the recommendation degree of each object to be recommended.

The attention degree score of each object to be recommended of the target user provided by this embodiment may be used as a basis for recommending the target user, and according to the attention degree score of each object to be recommended, a predetermined number of objects to be recommended before the highest attention degree score are used as the target objects to be recommended, and then a recommendation message is generated according to the target objects to be recommended.

However, the target object to be recommended obtained by the implementation method has too few factors to consider, and is often not accurate enough, so that the purpose of improving the selection probability of the target user for the object to be recommended is difficult to achieve.

Therefore, the recommendation degree of each object to be recommended is determined by the embodiment by adopting a pre-trained recommendation degree calculation model. For example, the recommendation degree calculation model may be trained by using the following steps 1 to 4:

step 1, obtaining each user performing an operation action on the sample object to be recommended as various users.

When the recommendation degree calculation model is trained, all users who perform the operation on the object to be recommended of the sample are preferably adopted as sample users, and are not screened or distinguished.

And 2, constructing a training sample set according to the attribute information of each sample user, the attribute information of each sample object to be recommended and the sample object to be recommended actually selected by each sample user.

Specifically, for each sample user, extracting a sample feature vector according to the attribute information of the sample user and the attribute information of the sample object to be recommended of the sample user who has performed the operation behavior; constructing the training sample of the sample user according to the sample feature vector of the sample user and the sample object to be recommended actually selected by the sample user; the training sample set includes the training samples of each of the sample users. And marking the sample object to be recommended according to the sample object to be recommended actually selected by each sample user, or extracting a result feature vector according to the sample object to be recommended actually selected by each sample user.

Constructing a training sample set T according to the sample feature vector and the actually selected sample object to be recommended, wherein the training sample set is T { (x) exemplarily when the embodiment of the application is applied to the field of house property trading₁，y₁)，(x₂，y₂)，...，(x_m，y_m)}. Wherein, the training sample set T comprises m training samples, each training sample comprises a sample user and an object to be recommended of the sample user after operation, x_iSample feature vectors, y, extracted according to attribute information of sample users of the ith training sample and attribute information of sample objects to be recommended of the sample in which the ith sample user performs operation behaviors_iAnd the sample object to be recommended actually selected by the ith sample user.

And 3, inputting the training sample set into the initial recommendation calculation model to obtain a loss value.

The training sample set T is input into the initial recommendation calculation model, and any common loss function can be used to calculate the loss value. For example, when the embodiment of the present application is applied to the real estate transaction field, the loss function L (y) can be adopted_i，f(x_i) Get the loss value of the initial recommendation calculation model. Wherein y is_iSample to-be-recommended object actually selected by the ith sample user, f (x)_i) Computing a model for an ith training for an initial recommendationAnd (5) outputting the result of the training sample.

And 4, updating parameters of the initial recommendation degree calculation model according to the loss value to obtain the recommendation degree calculation model.

Specifically, when the embodiment of the present application is applied to the real estate transaction field, the residual error of the loss function, or the negative gradient of the loss function, may be calculated based on the loss value of the initial recommendation degree calculation model according to the following formula (3):

wherein, f (x)_i) Calculate output of model for the ith training sample for initial recommendation, L (y)_i，f(x_i) A loss function of the model is calculated for the recommendation. The degree of error of the algorithm model is usually calculated by using the loss value to represent the recommendation degree, and the degree of error of the algorithm model is calculated by using the loss function, so that the loss value can be decreased along the direction of the negative gradient as soon as possible by applying the formula (3), the loss value is minimized, and the convergence of the algorithm model is accelerated.

To minimize the loss value, the sample feature vector and the residual are fitted to obtain a fitting result, for example, x may be used_iAnd r_iFitting is performed with R_jRepresenting the region where the fitting result is located, and estimating the region R where the fitting result is located by linear search_jObtaining a fitting result c_jThereby minimizing the loss value. Specifically, the fitting result c can be calculated by the following formula (4)_j：

Wherein, c_jFor the fitting result, R_jL (y) is the region where the fitting result is located_i，f(x_i) + c) is a loss function, c being a constant. Finally, according to the initial recommendation degree calculation model and the fitting result, determining the recommendation degree calculation model as an initial recommendation degreeThe calculation model is f₀(x) Then, the recommendation degree calculation model f (x) can be expressed by the following formula (5):

wherein, c_jAnd J is the number of the fitting results, and the J better fitting results are obtained by one-time fitting. I is a correctness parameter of the initial recommendation degree calculation model, is determined according to a sample object to be recommended actually selected by a sample user, and is 1 when the sample object to be recommended with the maximum recommendation degree given by the initial recommendation degree calculation model is the sample object to be recommended actually selected by the sample user; on the contrary, when the sample object to be recommended with the maximum recommendation degree given by the initial recommendation degree calculation model is not the sample object to be recommended actually selected by the sample user, I is 0.

In a possible embodiment, the above steps 3 and 4 may also be performed iteratively until a preset threshold number of iterations is reached.

Specifically, whether a preset iteration number threshold value is reached is judged; when the initial recommendation degree calculation model does not reach a preset iteration number threshold value, updating parameters of the initial recommendation degree calculation model according to the loss value to obtain an intermediate recommendation degree calculation model; inputting the training sample set into the intermediate recommendation degree calculation model to obtain a loss value; and updating parameters of the intermediate recommendation degree calculation model according to the loss value, and returning to the step of judging whether the threshold value of the preset iteration times is reached.

And in each iteration process, inputting the training sample set into an intermediate recommendation degree calculation model obtained in the previous iteration, and calculating the residual error of the loss function based on the loss value of the intermediate recommendation degree calculation model. For example, when the intermediate recommendation degree calculation model performs the tth round of iterative training, the residual of the loss function of the ith sample during the tth round of iterative training may be calculated according to the following formula (6):

wherein f (x) f_t-1(x) Thus, f (x)_i) Output results for the t-1 th round intermediate recommendation calculation model for the ith training sample, L (y)_i，f(x_i) A loss function of the model is calculated for the recommendation.

Then, a fitting optimal solution obtained by fitting the residual of the loss function of the ith sample can be calculated by using the following formula (7):

wherein R is_tjIndicates the region in which the fitting result is located, f_t-1(x_i) Output results for the t-1 th round intermediate recommendation calculation model for the ith training sample, L (y)_i，f_t-1(x_i) + c) is a loss function, c being a constant.

Therefore, the intermediate recommendation degree calculation model of the tth round obtained after the tth round training is shown in the following formula (8):

wherein, c_tjAnd J is the number of the fitting results, and the J better fitting results are obtained by one-time fitting. I is determined according to a sample object to be recommended actually selected by a sample user, and when the sample object to be recommended with the largest recommendation degree given by the model is the sample object to be recommended actually selected by the sample user, I is 1; on the contrary, when the sample object to be recommended with the largest recommendation degree given by the model is not the sample object to be recommended actually selected by the sample user, I is 0.

Assuming that a preset iteration threshold is T, performing T-round iterative training on the initial recommendation degree calculation model, and obtaining a recommendation degree calculation model f (x) which can be expressed by the following formula (9):

wherein f is₀(x) Is an initialA recommendation calculation model, T is the number of iterations, T is a preset iteration threshold, c_tjAnd J is the number of the fitting results, and each iteration fitting obtains J better fitting results. I is determined according to a sample object to be recommended actually selected by a sample user, and when the sample object to be recommended with the largest recommendation degree given by the model is the sample object to be recommended actually selected by the sample user, I is 1; on the contrary, when the sample object to be recommended with the largest recommendation degree given by the model is not the sample object to be recommended actually selected by the sample user, I is 0.

The initial recommendation degree calculation model is trained in the iteration mode, so that the accuracy of the task can be greatly improved, and various types of data including continuous values and discrete values can be flexibly processed. Meanwhile, the training method is simple and convenient.

S206: and determining a preset number of target objects to be recommended from each object to be recommended according to the recommendation degree of each object to be recommended.

Specifically, a predetermined number of objects to be recommended with the maximum recommendation degree in each object to be recommended are determined as target objects to be recommended. For example, 1-5 objects to be recommended with the largest recommendation degree are selected to be determined as target objects to be recommended.

S207: and generating a recommendation message according to the target recommendation object.

Specifically, the following two possible embodiments may be adopted to generate the recommendation message according to the target recommendation object:

in one possible implementation, the recommendation message may be generated using the attribute information of the target recommendation object. For example, when the embodiment of the present application is applied to the field of house property trading, a color page, a hovering page, a poster or a push message having an introduction or advertisement function may be generated as a recommendation message using attribute information such as the area, unit price, house type, location, etc. of a traded house property corresponding to a target recommendation object.

In another possible implementation manner, a contact person associated with the target recommendation object is acquired, and attribute information of the contact person is acquired; and generating the recommendation message according to the attribute information of the contact. For example, when the embodiment of the application is applied to the field of house property trading, at least one house property broker associated with a trading house property corresponding to a target recommendation object may be obtained as a contact associated with the target recommendation object, attribute information of each house property broker, such as name, gender, age, working experience, photos, contact way, and the like of the house property broker may be obtained, a recommendation message may be generated according to the attribute information of each house property broker, or one highest score may be determined from each house property broker, and the recommendation message may be generated according to the attribute information of the highest score of the house property broker, for example, a color page, a hovering page, a poster, or a push message having an introduction or advertisement function may be generated from the attribute information of the house property broker, and used as the recommendation message.

In addition, the attribute information of the cell and the business circle where the trading house is located can be obtained according to the cell and the business circle where the trading house is located, and the recommendation message is generated based on the attribute information of the cell and the business circle where the trading house is located.

S208: and judging whether a preset recommended time condition is met.

And setting a recommendation time condition according to a recommendation strategy, so that the push of the recommendation message does not cause much dislike and not too little to achieve the recommendation effect.

Specifically, when the embodiment of the present application is applied to the real estate transaction field, the following recommended time condition may be specified:

(1) if the recommendation message is continuously pushed to a target user for 3 times, but the target user does not click the recommendation message, the recommendation message is not pushed to the target user within 15 days.

(2) If the user performs a contact action or a consultation action with the contact person involved in the recommendation message or the object to be recommended in the recommendation message after the recommendation message is pushed to a target user, the recommendation message for the target user is not generated again after the contact person or the object to be recommended is pushed.

(3) The same target user receives the recommendation message at most once in 24 hours, namely the time interval for pushing the recommendation message to the same user is at least 24 hours at least, and the same target user is pushed with the recommendation message for 2 times at most in one week. No matter what form the recommendation message is pushed through what channel, the recommendation message is recorded as being pushed once, and the counting of the recommendation message is not limited by the recommendation form and the recommendation channel.

S209: and pushing a recommendation message generated according to the target recommendation object to the target user when a preset recommendation time condition is met.

Specifically, the recommendation message can be pushed to the target user by adopting a plurality of different recommendation forms and recommendation channels. When a plurality of different recommendation forms and recommendation channels exist, priorities can be set for the different recommendation forms and recommendation channels.

For example, when the embodiment of the application is applied to the field of house property transaction, the recommendation message can be pushed by displaying a transaction house property detail page of the target recommendation object on a webpage, an APP or a transaction platform, or pushing detail information of the transaction house property of the target recommendation object at a fixed time. The two different pushing manners may be set with priorities, for example, the priority of the manner of displaying the details page of the transaction property may be set to be greater than the priority of pushing the details information at a fixed time, and after the recommendation message is pushed in the manner of displaying the details page of the transaction property, the recommendation message is not pushed in the manner of pushing the details information at the fixed time within a certain time range.

For example, when the embodiment of the present application is applied to the field of house property trading, the system architecture diagram of the embodiment of the present application is as shown in fig. 3, and the operation behaviors that the user has generated on trading house properties at the house property trading platform, such as browsing behaviors, clicking behaviors, searching behaviors, collecting behaviors, contact brokers, and the like, are all stored in the user behavior path library. The method comprises the steps of obtaining at least one user who does not generate a transaction for a transaction object from a user behavior path library, analyzing the user behavior path library aiming at the at least one user who performs an operation on the transaction object, and setting a configurable screening time condition so as to define a target user who performs recommendation message pushing.

Aiming at each target user, the object to be recommended of the operation behavior of the target user can be obtained from the user behavior path library, the behavior portrait of the target user is analyzed, the interest degree of the target user to each object to be recommended is obtained, therefore, a more accurate recommendation strategy is formulated, and the target object to be recommended is determined based on the attribute information of the target user, the behavior information of the operation behavior and the attribute information of the object to be recommended.

And generating a recommendation message according to the attribute information of the target object to be recommended or the attribute information of the contact person associated with the target object to be recommended.

The push master control module can judge whether the preset recommendation time condition is met or not, and when the preset recommendation time condition is met, the recommendation message is pushed to the target user through various push channels and push modes such as a hovering color page of a house property transaction platform or a push message of a house property transaction public number. In the generation and pushing process of the recommendation message, the pushing time, the pushing target, the pushing amount and the pushing effect of the pushing message can be recorded and counted, so that model parameters or characteristic vectors related to the pushing method are improved in a targeted manner.

Therefore, the selection probability of the target user for the object to be recommended is improved, the recommendation message is pushed to the target user at a proper time, and the transaction rate is improved.

Based on the same design concept, the embodiment of the application also provides a device for pushing the recommendation message and a storage medium.

EXAMPLE III

As shown in fig. 4, an apparatus 400 for pushing a recommendation message provided in an embodiment of the present application includes:

a target user determination module 401, configured to determine a target user meeting a preset screening time condition;

an information obtaining module 402, configured to obtain attribute information of the target user and behavior information of the target user for at least one object to be recommended; the behavior information comprises the behavior category and the behavior time of the target user for operating the object to be recommended at least once;

the attention degree score module 403 is configured to determine, for each object to be recommended, an attention degree score of the target user for the object to be recommended according to the behavior category and the behavior time of each operation behavior on the object to be recommended; the attention degree score is used for evaluating the interest degree of the target user on the object to be recommended;

a recommendation degree determining module 404, configured to obtain attribute information of each object to be recommended, and determine a recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended, and the attention degree score of each object to be recommended;

a recommendation message generating module 405, configured to determine, according to the recommendation degree of each object to be recommended, a predetermined number of target objects to be recommended from the objects to be recommended, and generate a recommendation message according to the target objects to be recommended;

and the recommendation message pushing module 406 is configured to push a recommendation message generated according to the target recommendation object to the target user when a preset recommendation time condition is met.

In a possible implementation, the target user determining module 401 is configured to:

In one possible implementation, the attention scoring module 403 is configured to:

In one possible embodiment, the attention score module 403 determines the score of the operation behavior by using the following formula:

R＝e^-λx

In a possible implementation manner, the recommendation degree determining module 404 is configured to:

In a possible implementation manner, the recommendation message pushing apparatus 400 further includes a model training module 407, configured to:

In one possible implementation, the model training module 407 is configured to:

In a possible implementation, the model training module 407 is further configured to:

In a possible implementation, the model training module 407 is further configured to: fitting the sample feature vector and the residual using the following formula:

In a possible implementation, the model training module 407 is further configured to: determining the recommendation degree calculation model by adopting the following formula:

judging whether a preset iteration number threshold value is reached;

In a possible implementation, the recommendation message generating module 405 is configured to:

In a possible implementation, the recommendation message generating module 405 is further configured to:

Example four

A computer readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the steps of the method provided in embodiment one or embodiment two. In practice, the computer readable medium may be RAM, ROM, EPROM, magnetic disk, optical disk, etc., and is not intended to limit the scope of protection of this application.

The method steps described herein may be implemented in hardware, for example, logic gates, switches, Application Specific Integrated Circuits (ASICs), programmable logic controllers, embedded microcontrollers, etc., in addition to data processing programs. Such hardware capable of implementing the methods described herein may also constitute the present application.

EXAMPLE five

The embodiment of the present application further provides an electronic device, which may be a computer or a server, and the apparatus in the third apparatus embodiment of the present application may be integrated therein. As shown in fig. 4, an electronic device 500 according to a third embodiment of the apparatus of the present application is shown.

The electronic device may include a processor 501 of one or more processing cores, one or more computer-readable storage media 502. The electronic device may further include a power supply 503, an input-output unit 504. Those skilled in the art will appreciate that fig. 5 is not limiting of electronic devices and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components.

Wherein:

the processor 501 is a control section of the electronic device, connects the respective sections using various interfaces and lines, and executes the steps of the method provided in the first embodiment or the second embodiment by running or executing a software program stored in the computer-readable storage medium 502.

The computer-readable storage medium 502 may be used to store a software program, i.e., a program involved in the method provided in embodiment one or embodiment two.

The processor 501 executes various functional applications and data processing by executing software programs stored in the computer-readable storage medium 502. The computer-readable storage medium 502 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data or the like used according to the needs of the electronic device. Further, the computer-readable storage medium 502 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the computer-readable storage medium 502 may also include a memory controller to provide the processor 501 access to the computer-readable storage medium 502.

The electronic device further comprises a power supply 503 for supplying power to each component, and preferably, the power supply 503 may be logically connected to the processor 501 through a power management system, so that functions of managing charging, discharging, power consumption, and the like are realized through the power management system. The power supply 503 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.

The server may also include an input-output unit 504, such as may be used to receive entered numeric or character information, and to generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control; such as various graphical user interfaces that may be used to display information entered by or provided to the user, as well as the server, which may be composed of graphics, text, icons, video, and any combination thereof.

In summary, the above description is only a preferred embodiment of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A recommendation message pushing method is characterized by comprising the following steps:

2. The method according to claim 1, wherein the step of determining the target users meeting the preset filtering time condition comprises:

3. The method according to claim 1, wherein the step of determining, for each object to be recommended, the attention score of the target user for the object to be recommended according to the behavior category and the behavior time of each operation behavior on the object to be recommended comprises:

4. The method of claim 3, wherein the score for the operational behavior is determined using the following equation:

R＝e^-λx

5. The method according to claim 1, wherein the step of determining the recommendation degree of each object to be recommended according to the attribute information of the target user, the attribute information of each object to be recommended, and the attention degree score of each object to be recommended comprises:

6. The method of claim 5, wherein the step of pre-training the recommendation calculation model comprises:

7. The method according to claim 6, wherein the step of constructing a training sample set according to the attribute information of each sample user, the attribute information of each sample object to be recommended, and the sample object to be recommended actually selected by each sample user comprises:

8. A recommendation message pushing apparatus, comprising:

9. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 8.

10. An electronic device comprising the computer-readable storage medium of claim 9, further comprising a processor that can execute the computer-readable storage medium.