CN117349509A

CN117349509A - Information pushing method and device

Info

Publication number: CN117349509A
Application number: CN202210784846.3A
Authority: CN
Inventors: 刘艺超; 胡佳椰
Original assignee: Beijing Xiaomi Mobile Software Co Ltd; Beijing Xiaomi Pinecone Electronic Co Ltd
Current assignee: Beijing Xiaomi Mobile Software Co Ltd; Beijing Xiaomi Pinecone Electronic Co Ltd
Priority date: 2022-06-29
Filing date: 2022-06-29
Publication date: 2024-01-05

Abstract

The disclosure relates to an information pushing method and a device thereof, and relates to the field of deep learning by acquiring keywords of information to be pushed; matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library; determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information; based on the target recall strategy, a target pushing object corresponding to the information to be pushed is obtained, and the information to be pushed is pushed to the target pushing object. According to the recall reference information of the information to be pushed, the target recall strategy of the information to be pushed is automatically selected, recall speed and recall accuracy are considered, recall efficiency of the target pushed object is improved, and waste of resources is avoided.

Description

Information pushing method and device

Technical Field

The disclosure relates to the field of deep learning, in particular to an information pushing method and device.

Background

In a product operation scene, operators need to recall tens of millions of target push objects rapidly and accurately aiming at specific commodities, and do relevant operation activities on the target push objects, so that the operation efficiency is improved.

Disclosure of Invention

The disclosure provides an information pushing method and an information pushing device, which at least solve the problem that in the process of determining a target pushing object, balance and compromise are difficult to achieve between recall speed and recall accuracy of the target pushing object.

The technical scheme of the present disclosure is as follows:

according to a first aspect of an embodiment of the present disclosure, there is provided an information pushing method, including: acquiring keywords of information to be pushed; matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library; determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information; based on the target recall strategy, a target pushing object corresponding to the information to be pushed is obtained, and the information to be pushed is pushed to the target pushing object.

According to the recall reference information of the information to be pushed, the target recall strategy of the information to be pushed is automatically selected, recall speed and recall accuracy are considered, recall efficiency of the target pushed object is improved, and waste of resources is avoided.

In some embodiments, based on a target recall policy, obtaining a target push object corresponding to information to be pushed includes: acquiring a first candidate pushing object corresponding to information to be pushed; the method comprises the steps of obtaining a pushing type of information to be pushed, and screening a first candidate pushing object based on the pushing type to obtain a second candidate pushing object; and acquiring the target push object based on the target recall strategy and the second candidate push object.

In some embodiments, based on the target recall policy and the second candidate push object, obtaining the target push object comprises: obtaining pushing optimization parameters of information to be pushed, and determining a target network model under the target recall strategy based on the pushing optimization parameters; acquiring first characteristic information of information to be pushed and second characteristic information of a second candidate pushing object; inputting the first characteristic information and the second characteristic information into a target network model to acquire a first preference value of the second candidate push object to the information to be pushed; and sorting the first preference values from large to small, and acquiring a target pushing object according to the sorted first preference values.

In some embodiments, the recall reference information includes a number of successful pushes of the information to be pushed and a qualification rate of the information to be pushed, and determining a target recall policy of the information to be pushed from the candidate recall policies based on the recall reference information includes: determining a numerical interval corresponding to successful pushing times and good score; and determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the numerical intervals.

In some embodiments, determining a target recall policy for information to be pushed from among candidate recall policies based on the value interval comprises: determining the pre-recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the good score being greater than or equal to the set good score; based on the pre-recall policy, obtaining a target push object corresponding to information to be pushed includes: calling push object information corresponding to the information to be pushed from a database; and determining a target push object from the push objects corresponding to the push object information.

In some embodiments, determining a target recall policy for information to be pushed from among candidate recall policies based on the value interval comprises: determining the model recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the good score being less than the set good score; based on the model recall strategy, obtaining a target push object corresponding to the information to be pushed comprises the following steps: acquiring first real-time characteristic information of information to be pushed; acquiring second real-time characteristic information of a second candidate push object; inputting the first real-time characteristic information and the second real-time characteristic information into a preset model to obtain a second preference value of a second candidate pushing object to be pushed, which is output by the preset model; and sorting the second preference values from large to small, and acquiring a target pushing object according to the sorted second preference values.

In some embodiments, determining a target recall policy for information to be pushed from among candidate recall policies based on the value interval comprises: determining the vector recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being smaller than the set times; the method for acquiring the target push object corresponding to the information to be pushed based on the vector recall strategy comprises the following steps: calling a first target vector of information to be pushed and a second target vector of a second candidate push object from a database; and performing inner product value calculation on the first target vector and the second target vector, sorting the inner product values obtained after calculation according to the sizes from large to small, and determining a target pushing object according to the sorted inner product values.

According to a second aspect of the embodiments of the present disclosure, there is provided an information pushing apparatus, including: the acquisition module is used for acquiring keywords of the information to be pushed; the matching module is used for matching the keywords with the keyword text library so as to acquire recall reference information corresponding to the information to be pushed from the keyword text library; the determining module is used for determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information; and the pushing module is used for acquiring a target pushing object corresponding to the information to be pushed based on the target recall strategy and pushing the information to be pushed to the target pushing object.

In some embodiments, the pushing module is further configured to: acquiring a first candidate pushing object corresponding to information to be pushed; the method comprises the steps of obtaining a pushing type of information to be pushed, and screening a first candidate pushing object based on the pushing type to obtain a second candidate pushing object; and acquiring the target push object based on the target recall strategy and the second candidate push object.

In some embodiments, the pushing module is further configured to: obtaining pushing optimization parameters of information to be pushed, and determining a target network model under the target recall strategy based on the pushing optimization parameters; acquiring first characteristic information of information to be pushed and second characteristic information of a second candidate pushing object; inputting the first characteristic information and the second characteristic information into a target network model to acquire a first preference value of the second candidate push object to the information to be pushed; and sorting the first preference values from large to small, and acquiring a target pushing object according to the sorted first preference values.

In some embodiments, the recall reference information includes a number of successful pushes of the information to be pushed and a score of the information to be pushed, and the determining module is further configured to: determining a numerical interval corresponding to successful pushing times and good score; and determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the numerical intervals.

In some embodiments, the determining module is further to: determining the pre-recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the good score being greater than or equal to the set good score; based on the pre-recall policy, obtaining a target push object corresponding to information to be pushed includes: calling push object information corresponding to the information to be pushed from a database; and determining a target push object from the push objects corresponding to the push object information.

In some embodiments, the determining module is further to: determining the model recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the good score being less than the set good score; based on the model recall strategy, obtaining a target push object corresponding to the information to be pushed comprises the following steps: acquiring first real-time characteristic information of information to be pushed; acquiring second real-time characteristic information of a second candidate push object; inputting the first real-time characteristic information and the second real-time characteristic information into a preset model to obtain a second preference value of a second candidate pushing object to be pushed, which is output by the preset model; and sorting the second preference values from large to small, and acquiring a target pushing object according to the sorted second preference values.

In some embodiments, the determining module is further to: determining the vector recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being smaller than the set times; the method for acquiring the target push object corresponding to the information to be pushed based on the vector recall strategy comprises the following steps: calling a first target vector of information to be pushed and a second target vector of a second candidate push object from a database; and performing inner product value calculation on the first target vector and the second target vector, sorting the inner product values obtained after calculation according to the sizes from large to small, and determining a target pushing object according to the sorted inner product values.

According to a third aspect of embodiments of the present disclosure, there is provided an electronic device, including: at least one processor; and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor to implement an information pushing method according to an embodiment of the first aspect of the present application.

According to a fourth aspect of embodiments of the present disclosure, a non-transitory computer readable storage medium storing computer instructions for implementing an information push method as the embodiments of the first aspect of the present application is presented.

According to a fifth aspect of the embodiments of the present disclosure, a computer program product is presented, comprising a computer program which, when executed by a processor, implements an information pushing method as the embodiments of the first aspect of the present application.

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

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 an exemplary schematic diagram of an information pushing method according to an embodiment of the present disclosure.

Fig. 2 is a schematic diagram of an information pushing method according to an example of the present disclosure in a product operation scenario.

Fig. 3 is an exemplary schematic diagram of an information pushing method according to an embodiment of the present disclosure.

Fig. 4 is a schematic diagram of interface creation in an operational scenario set forth in an embodiment of the present disclosure.

Fig. 5 is an exemplary schematic diagram of an information pushing method according to an embodiment of the present disclosure.

Fig. 6 is a technical framework diagram of an information pushing method according to an embodiment of the present disclosure.

Fig. 7 is a schematic diagram of an information pushing device according to an embodiment of the disclosure.

Fig. 8 is a block diagram of an electronic device, 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 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.

It should be noted that, all actions for acquiring signals, information or data in the present application are performed under the condition of conforming to the corresponding data protection rule policy of the country of the location and obtaining the authorization given by the owner of the corresponding device.

Fig. 1 is an exemplary schematic diagram of an information pushing method according to the present disclosure, as shown in fig. 1, where the information pushing method includes the following steps:

s101, acquiring keywords of information to be pushed.

The information pushing method provided by the disclosure can be applied to help operators recall interested target pushing objects for specific commodities quickly and accurately under a product operation scene, and do related operation activities for the target pushing objects, so that the operation efficiency is improved. In an embodiment of the disclosure, the execution subject may be a server, where the server may be an information push service platform.

The method comprises the steps of obtaining information to be pushed, and carrying out keyword recognition on the information to be pushed so as to obtain keywords of the information to be pushed. The information to be pushed is information to be pushed on the terminal device of the target pushing object, for example, the information to be pushed can be document information of a product to be pushed input by an operator, or can be document information of an APP to be pushed input by the operator. The keyword of the information to be pushed may include a model number of a certain commodity to be pushed, a name of a certain APP, and the like, which correspond to the information to be pushed.

For example, if the information to be pushed is "the commodity A" which is particularly good and suitable for all families, the people purchase the bar quickly, and after the keyword identification is performed on the information to be pushed, the keyword of the information to be pushed can be obtained as "the commodity A".

For example, if the information to be pushed is "the B type APP advertisement is few, the audio source is many, and the method is suitable for communication of musicians, and people download bars quickly", the keyword recognition is performed on the information to be pushed, and then the keyword of the information to be pushed can be obtained to be "the B type APP".

S102, matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library.

And matching the obtained keywords of the information to be pushed with a keyword text library so as to obtain recall reference information corresponding to the information to be pushed from the keyword text library. Alternatively, when matching the keywords with the keyword text library, the matching may be performed by adopting an inverted index manner. The recall reference information may include the number of times and the good score of the commodity or the APP corresponding to the keyword of the information to be pushed.

By way of example, taking a keyword of information to be pushed as an "A commodity", determining a keyword text library, wherein keywords of all commodities, sales data and good score data of all commodities are stored in the keyword text library, matching the obtained keywords of the information to be pushed with the keyword text library to obtain sales data and good score data corresponding to the "A commodity", and taking the sales data and good score data corresponding to the "A commodity" as recall reference information of the "A commodity".

By way of example, taking a keyword of information to be pushed as "B money APP", determining a keyword text library, wherein the keyword text library stores all keywords of the APP and downloading times and good score data of all the APP, and matching the obtained keyword of the information to be pushed with the keyword text library to obtain downloading times and good score data corresponding to the "B money APP", and taking the downloading times and good score data corresponding to the "B money APP" as recall reference information of the "B money APP".

S103, determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information.

And determining a target recall strategy corresponding to the information to be pushed from the candidate recall strategies according to the recall reference information corresponding to the information to be pushed.

By way of example, taking a keyword of information to be pushed as an "A commodity", determining a target recall strategy corresponding to the "A commodity" from the candidate recall strategies according to sales volume data and good score data corresponding to the "A commodity". For example, if the "commodity a" is a commodity with a monthly sale of 10 ten thousand and a good score of 99%, the probability that the "commodity a" is favored by most people is considered to be relatively high, and a strategy with a relatively high speed of acquiring the target pushing object is determined from the candidate recall strategies to be used as the target recall strategy; for example, if "commodity a" is a commodity of the month-old 100, then "commodity a" is considered as a commodity of the public, and a strategy with higher accuracy of acquiring the target push object is determined from the candidate recall strategies as the target recall strategy.

By way of example, taking the keyword of the information to be pushed as "B money APP", determining a target recall strategy corresponding to the "B money APP" from the candidate recall strategies according to the downloading times and the good score data corresponding to the "B money APP". For example, if the number of downloads of the "B-type APP" is 200 ten thousand and the qualification rate is 97%, the probability that the "B-type APP" is liked by most people is considered to be relatively high, and a strategy with relatively high speed of acquiring the target push object is determined from the candidate recall strategies to be used as the target recall strategy; for example, if the number of downloads of the "B-type APP" is 1000, the "B-type APP" is considered as the public APP, and a policy with higher accuracy of acquiring the target push object is determined from the candidate recall policies as the target recall policy.

S104, based on the target recall strategy, acquiring a target pushing object corresponding to the information to be pushed, and pushing the information to be pushed to the target pushing object.

And acquiring a target pushing object corresponding to the information to be pushed according to the determined target recall strategy, and pushing the information to be pushed to the target pushing object. The target recall policy may correspond to a target network model, input feature information of information to be pushed, for example, feature information of price, category, interaction times and the like corresponding to commodities of the information to be pushed, feature information of candidate push objects, for example, product information interacted by the candidate push objects in a set period of time, and feature information of age, gender, learning and the like of the candidate push objects, into the target network model, obtain a preference value of each candidate push object output by the target network model for the information to be pushed, sort the preference value according to the size, and select a set number of candidate push objects as target push objects and push the information to be pushed to the target push objects.

Fig. 2 is a schematic diagram of an information pushing method in a product operation scenario in the disclosed example, as shown in fig. 2, an operator inputs information to be pushed on an information pushing service platform, the information pushing service platform determines a target recall policy of the information to be pushed based on recall reference information matched with the information to be pushed in a keyword text box, and obtains a target pushing object corresponding to the information to be pushed based on the target recall policy.

As an implementation manner, as shown in fig. 2, after determining a target push object corresponding to the information to be pushed, the information push service platform returns the information of the target push object to an operator, and the operator determines to push the information to be pushed to the target push object.

As another implementation manner, after determining the target pushing object corresponding to the information to be pushed, the information pushing service platform automatically pushes the information to be pushed to the target pushing object.

According to the information pushing method provided by the embodiment of the disclosure, keywords of information to be pushed are obtained; matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library; determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information; based on the target recall strategy, a target pushing object corresponding to the information to be pushed is obtained, and the information to be pushed is pushed to the target pushing object. According to the recall reference information of the information to be pushed, the target recall strategy of the information to be pushed is automatically selected, recall speed and recall accuracy are considered, recall efficiency of the target pushed object is improved, and waste of resources is avoided.

Fig. 3 is an exemplary schematic diagram of an information pushing method according to the present disclosure, as shown in fig. 3, where the information pushing method includes the following steps:

s301, acquiring keywords of information to be pushed.

For the implementation of step S301, reference may be made to the description related to step S101 in the above embodiment, and the description is omitted here.

S302, matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library.

For the implementation of step S302, reference may be made to the description related to step S102 in the above embodiment, and the description is omitted here.

S303, determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information.

For the implementation of step S303, reference may be made to the description related to step S103 in the above embodiment, and the description is omitted here.

S304, a first candidate pushing object corresponding to the information to be pushed is obtained.

And acquiring a first candidate pushing object corresponding to the information to be pushed. Wherein,

for example, taking a keyword of information to be pushed as "a commodity", the first candidate pushing object may be registered for all users who may purchase "a commodity", for example, 1 million users share a certain software, and 1 million users purchase "a commodity", so that the remaining 9999 million users may be used as the first candidate pushing object.

S305, obtaining the push type of the information to be pushed, and screening the first candidate push object based on the push type to obtain a second candidate push object.

The obtaining of the push type of the information to be pushed may be, for example, obtaining a potential user interested in the information to be pushed, or obtaining a new user interested in the information to be pushed, or obtaining an old user having a purchase intention again for the information to be pushed, and screening the first candidate push object according to the push type to obtain the second candidate push object.

For example, if the push type is a potential user that obtains information to be pushed, all the first candidate push objects may be used as second candidate push objects.

For example, if the push type is a new user who obtains an interest in the information to be pushed, all the first candidate push objects that have previously purchased the commodity are excluded, and the remaining first candidate push objects that have not purchased the commodity are used as second candidate push objects.

For example, if the push type is an old user who has acquired the information to be pushed and has a intention to purchase again, the first candidate push object that has not purchased the commodity is excluded, and the remaining first candidate push objects that have purchased the commodity are used as the second candidate push objects.

Fig. 4 is a schematic diagram of interface creation in an operation scenario set forth in the present disclosure, and as shown in fig. 4, an operator may select an operation type, a push optimization parameter, a target push number, a blacklist, a whitelist, and the like, and input information to be pushed into a relevant box. As shown in fig. 4, the operator may select the push type of the information to be pushed as a potential user interested in obtaining the information to be pushed, or obtain a new user interested in the information to be pushed, or obtain an old user having a purchase intention for the information to be pushed.

S306, based on the target recall strategy and the second candidate pushing object, acquiring a target pushing object and pushing information to be pushed to the target pushing object.

Some embodiments of the present disclosure propose two push optimization parameters, which can be selected by operators, one to promote conversion and one to promote click-through rate. In this embodiment, the target recall policy may include a dynamic routing multi-point-of-interest network (MIND) model, a deep neural network (Deep Neural Networks, DNN) model, a dual-tower model, and the like, and taking the target recall policy to determine a target push object by using the DNN model as an example, in the training process of the DNN model, different training labels are adopted to respectively train a first DNN model corresponding to the conversion rate promotion and a second DNN model corresponding to the click rate promotion. And determining a target network model under a target recall strategy based on the push optimization parameters according to the push optimization parameters corresponding to the information to be pushed. If the target recall strategy is a DNN model, when the push optimization parameter is the click rate improvement, a second DNN model corresponding to the click rate improvement is adopted as a target network model; and when the pushing optimization parameter is the lifting conversion rate, adopting a first DNN model corresponding to the lifting conversion rate as a target network model.

After the target network model is determined, acquiring first characteristic information of information to be pushed and second characteristic information of a second candidate push object, inputting the first characteristic information and the second characteristic information into the target network model to acquire a first preference value of the second candidate push object to the information to be pushed, sorting the first preference values according to the order from large to small, and acquiring the target push object according to the sorted first preference value.

According to the method and the device, a target recall strategy of information to be pushed is automatically selected according to recall reference information of the information to be pushed, candidate pushing objects which do not meet the conditions are eliminated according to pushing types, so that the target pushing objects are more accurate.

Fig. 5 is an exemplary schematic diagram of an information pushing method according to the present disclosure, as shown in fig. 5, where the information pushing method includes the following steps:

s501, acquiring keywords of information to be pushed.

For the implementation of step S501, reference may be made to the description related to step S101 in the above embodiment, and the description is omitted here.

S502, matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library, wherein the recall reference information comprises the successful pushing times of the information to be pushed and the good score of the information to be pushed.

And matching the obtained keywords of the information to be pushed with a keyword text library so as to obtain recall reference information corresponding to the information to be pushed from the keyword text library. Alternatively, when matching the keywords with the keyword text library, the matching may be performed by adopting an inverted index manner. The recall reference information refers to the successful pushing times of the information to be pushed and the good score of the information to be pushed.

For example, taking a keyword of information to be pushed as "commodity A" as an example, recall reference information is sales data and good score data corresponding to the commodity A.

For example, taking the keyword of the information to be pushed as "B type APP" as an example, the recall reference information is the download times and the good score data corresponding to the "B type APP".

S503, determining a numerical interval corresponding to the successful pushing times and the good score.

By way of example, taking a keyword of information to be pushed as a commodity, different numerical intervals are set in advance for sales data and good score data, and numerical intervals to which sales data and good score of "commodity A" respectively belong are judged.

By taking keywords of information to be pushed as an APP as an example, different numerical intervals are set for downloading times and good score data in advance, and the numerical intervals to which the downloading times and the good score of the 'B-type APP' belong are judged.

S504, determining a pre-recall strategy from the candidate recall strategies as a target recall strategy in response to the successful pushing times being greater than or equal to the set times and the good score being greater than or equal to the set good score, and acquiring a target pushing object corresponding to the information to be pushed based on the pre-recall strategy.

If the successful pushing times of the keywords of the information to be pushed are greater than or equal to the set times and the good score is greater than or equal to the set good score, determining a pre-recall strategy from the candidate recall strategies as a target recall strategy, and acquiring a target pushing object corresponding to the information to be pushed based on the pre-recall strategy. The pre-recall strategy refers to a strategy of directly calling the push object information corresponding to the information to be pushed from the database, and after the push object information corresponding to the information to be pushed is called from the database, a target push object is determined from the push objects corresponding to the push object information so as to ensure recall speed.

Taking a keyword of information to be pushed as an example of "a commodity", if the monthly sales data of the "a commodity" is greater than or equal to 10 ten thousand and the qualification rate is greater than or equal to 99%, considering the "a commodity" as a high-quality commodity, for the high-quality commodity, acquiring push object information at fixed time according to a model corresponding to the promotion click rate optimization or the promotion conversion rate optimization, wherein the push object information comprises a preference value of each second candidate push object for the "a commodity", storing the push object information in a database, calling the push object information of the second candidate push objects from the database based on a pre-recall strategy, and sequencing the preference values included in the push object information to determine a set number of target push objects. The second candidate push object is obtained by screening the first candidate push object based on the push type.

S505, determining the model recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the qualification rate being less than the set qualification rate, and acquiring a target pushing object corresponding to the information to be pushed based on the model recall strategy.

If the successful pushing times of the keywords of the information to be pushed are greater than or equal to the set times and the good score is smaller than the set good score, determining the model recall strategy from the candidate recall strategies as a target recall strategy, and acquiring a target pushing object corresponding to the information to be pushed based on the model recall strategy.

The method comprises the steps of obtaining a target pushing object corresponding to information to be pushed based on a model recall strategy, wherein the steps of:

the method comprises the steps of obtaining first real-time characteristic information of information to be pushed, taking keywords of the information to be pushed as 'A commodity', and obtaining first real-time characteristic information such as real-time price, real-time category, real-time interacted times and the like of the 'A commodity'.

And acquiring second real-time characteristic information of the second candidate push object, wherein the interaction information of the second candidate push object with each commodity in a set time period and basic information such as age, gender, academic and city of the second candidate push object can be used as the second real-time characteristic information of the second candidate push object together. The second candidate pushing object is a pushing type based on the information to be pushed, and is obtained by screening from the first candidate pushing object corresponding to the information to be pushed.

The first real-time feature information and the second real-time feature information are input into a recall model determined based on push optimization parameters, wherein the push optimization parameters can comprise push click rate optimization or conversion rate optimization, and the recall model can comprise MIND model, DNN model, double-tower model and the like. And obtaining a second preference value of the second candidate push object to-be-pushed information output by the recall model, sorting the second preference value according to the order from big to small, and obtaining a target push object according to the sorted second preference value so as to ensure recall accuracy.

S506, determining the vector recall strategy from the candidate recall strategies as a target recall strategy in response to the successful pushing times smaller than the set times, and acquiring a target pushing object corresponding to the information to be pushed based on the vector recall strategy.

If the successful pushing times of the keywords of the information to be pushed are smaller than the set times, determining a vector recall strategy from the candidate recall strategies as a target recall strategy, and acquiring a target pushing object corresponding to the information to be pushed based on the vector recall strategy.

The method comprises the steps of obtaining a target pushing object corresponding to information to be pushed based on a vector recall strategy, wherein the steps of:

And calling a first target vector of information to be pushed (also called as first enhancement corresponding to the information to be pushed) from a database, wherein the first enhancement refers to a large sparse vector corresponding to the first characteristic information of the information to be pushed, converting the first sparse vector into a low-dimensional vector used for representing the information to be pushed) and a second target vector of a second candidate push object (also called as second enhancement corresponding to the second candidate push object), wherein the second enhancement refers to a large sparse vector corresponding to the second characteristic information of the second candidate push object, converting the large sparse vector corresponding to the second characteristic information of the second candidate push object into a low-dimensional vector used for representing the second candidate push object, taking a keyword of the information to be pushed as an example of 'commodity A', calling the first target vector of 'commodity A' from the database, calculating an inner product value of the first target vector and the second target vector, sorting the inner product value obtained after calculation according to the inner product value from large to small, and determining a target push object and recall speed according to the sorted inner product value, so as to ensure the recall speed between the target push object and recall. In this embodiment, the first target vector of the push information to be acquired and the second target vector of the second candidate push object may be determined in advance based on the target double-tower model determined by the push optimization parameter, and the first target vector of the push information to be acquired and the second target vector of the second candidate push object may be stored in the database for subsequent calling. The pushing optimization parameters may include lifting click rate optimization or lifting conversion rate optimization, and in the process of training the double-tower model, different training labels are adopted to respectively train a first double-tower model corresponding to the lifting conversion rate and a second double-tower model corresponding to the lifting click rate. When the pushing optimization parameter is the optimization of the click rate, taking the second double-tower model as a target double-tower model; and when the pushing optimization parameter is the lifting conversion rate optimization, taking the first double-tower model as a target double-tower model.

S507, pushing information to be pushed to the target pushing object.

After the target pushing object is determined, pushing information to be pushed to the target pushing object.

In order to clearly describe the implementation framework of the foregoing embodiment, fig. 6 is a technical framework diagram of an information pushing method according to the present disclosure, and as shown in fig. 6, the technical framework of the information pushing method mainly includes three parts, namely, feature generation, model training and recall, where the feature generation part is configured to obtain feature data and training tag data of a first candidate pushing object and information to be pushed from a data source, and store the feature data of the first candidate pushing object and the feature data of the information to be pushed into a database.

The model training mainly comprises a storage part and a calculation part, wherein the calculation part brings feature data and training label data into a depth model for training so as to obtain a recall model and Embedding corresponding to a first candidate push object generated after training and Embedding corresponding to information to be pushed, and the storage part stores the recall model obtained after training and Embedding corresponding to the first candidate push object generated after training and Embedding corresponding to the information to be pushed into a database, wherein Embedding corresponding to the information to be pushed refers to converting a large-scale vector corresponding to the first feature information of the information to be pushed into a low-dimensional vector used for representing the information to be pushed; the method comprises the steps of converting a large sparse vector corresponding to second characteristic information of a first candidate push object into a low-dimensional vector used for representing the first candidate push object. The database comprises a recall model after training, an Embeddding corresponding to the first candidate pushing object and an Embedding corresponding to the information to be pushed. And the recall model periodically acquires the target push object information corresponding to the high-quality product and stores the target push object information in the database, so that the target push object information corresponding to the high-quality product is also stored in the database, and the recall model can comprise a MIND model, a DNN model, a double-tower model and the like.

And the recall part is used for determining a target recall strategy from the candidate recall strategies based on the information to be pushed. The candidate recall strategy comprises three recall modes of pre-recall, model recall and vector recall. If the target recall strategy is pre-recall, target push object information corresponding to the high-quality product is called from a database, and preference values included in the push object information are ordered to determine a set number of target push objects so as to ensure recall speed; if the target recall strategy is model recall, inputting the first real-time characteristic information of the information to be pushed and the second real-time characteristic information of the second candidate push object into a recall model to obtain a second preference value of the second candidate push object to be pushed, which is output by the recall model, sorting the second preference values according to the order from big to small, and obtaining the target push object according to the sorted second preference value so as to ensure recall accuracy; if the target recall strategy is vector recall, invoking the Embedding corresponding to the first candidate push object and the Embedding corresponding to the information to be pushed from the database, calculating an inner product value of the Embedding corresponding to the first candidate push object and the Embedding corresponding to the information to be pushed, sorting the inner product values obtained after calculation according to the order from large to small, and determining the target push object according to the sorted inner product value so as to ensure balance between recall accuracy and speed.

Fig. 7 is a schematic diagram of an information pushing device according to the present disclosure, as shown in fig. 7, the information pushing device 700 includes an obtaining module 701, a matching module 702, a determining module 703, and a pushing module 704, where:

an obtaining module 701, configured to obtain a keyword of information to be pushed;

the matching module 702 is configured to match the keyword with the keyword text library, so as to obtain recall reference information corresponding to the information to be pushed from the keyword text library;

a determining module 703, configured to determine a target recall policy of the information to be pushed from the candidate recall policies based on the recall reference information;

and the pushing module 704 is configured to obtain a target pushing object corresponding to the information to be pushed based on the target recall policy, and push the information to be pushed to the target pushing object.

The information pushing device provided by the disclosure obtains keywords of information to be pushed; matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library; determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the recall reference information; based on the target recall strategy, a target pushing object corresponding to the information to be pushed is obtained, and the information to be pushed is pushed to the target pushing object. According to the recall reference information of the information to be pushed, the target recall strategy of the information to be pushed is automatically selected, recall speed and recall accuracy are considered, recall efficiency of the target pushed object is improved, and waste of resources is avoided.

In some embodiments, the pushing module 704 is further configured to: acquiring a first candidate pushing object corresponding to information to be pushed; the method comprises the steps of obtaining a pushing type of information to be pushed, and screening a first candidate pushing object based on the pushing type to obtain a second candidate pushing object; and acquiring the target push object based on the target recall strategy and the second candidate push object.

In some embodiments, the pushing module 704 is further configured to: obtaining pushing optimization parameters of information to be pushed, and determining a target network model under a target recall strategy based on the pushing optimization parameters; acquiring first characteristic information of information to be pushed and second characteristic information of a second candidate pushing object; inputting the first characteristic information and the second characteristic information into a target network model to acquire a first preference value of the second candidate push object to the information to be pushed; and sorting the first preference values from large to small, and acquiring a target pushing object according to the sorted first preference values.

In some embodiments, the recall reference information includes the number of successful pushes of the information to be pushed and the score of the information to be pushed, and the determining module 703 is further configured to: determining a numerical interval corresponding to successful pushing times and good score; and determining a target recall strategy of the information to be pushed from the candidate recall strategies based on the numerical intervals.

In some embodiments, the determining module 703 is further configured to: determining the pre-recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the good score being greater than or equal to the set good score; based on the pre-recall policy, obtaining a target push object corresponding to information to be pushed includes: calling push object information corresponding to the information to be pushed from a database; and determining the push object corresponding to the push object information as a target push object.

In some embodiments, the determining module 703 is further configured to: determining the model recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being greater than or equal to the set times and the good score being less than the set good score; based on the model recall strategy, obtaining a target push object corresponding to the information to be pushed comprises the following steps: acquiring first real-time characteristic information of information to be pushed; acquiring second real-time characteristic information of a second candidate push object; inputting the first real-time characteristic information and the second real-time characteristic information into a preset model to obtain a second preference value of a second candidate pushing object to be pushed, which is output by the preset model; and sorting the second preference values from large to small, and acquiring a target pushing object according to the sorted second preference values.

In some embodiments, the determining module 703 is further configured to: determining the vector recall strategy as a target recall strategy from the candidate recall strategies in response to the successful pushing times being smaller than the set times; the method for acquiring the target push object corresponding to the information to be pushed based on the vector recall strategy comprises the following steps: calling a first target vector of information to be pushed and a second target vector of a second candidate push object from a database; and performing inner product value calculation on the first target vector and the second target vector, sorting the inner product values obtained after calculation according to the sizes from large to small, and determining a target pushing object according to the sorted inner product values.

Fig. 8 is a block diagram of an electronic device 800, according to an example embodiment.

As shown in fig. 8, the electronic device 800 includes:

a memory 801 and a processor 802, a bus 803 connecting the different components (including the memory 801 and the processor 802), the memory 801 storing a computer program, the processor 802 implementing the information push method of the embodiments of the present disclosure when executing the program.

Bus 803 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, and a local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include Industry Standard Architecture (ISA) bus, micro channel architecture (MAC) bus, enhanced ISA bus, video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

Electronic device 800 typically includes a variety of electronic device readable media. Such media can be any available media that is accessible by electronic device 800 and includes both volatile and nonvolatile media, removable and non-removable media.

The memory 801 may also include computer system readable media in the form of volatile memory such as Random Access Memory (RAM) 804 and/or cache memory 805. Electronic device 800 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, storage system 806 may be used to read from and write to non-removable, nonvolatile magnetic media (not shown in FIG. 8, commonly referred to as a "hard disk drive"). Although not shown in fig. 8, a magnetic disk drive for reading from and writing to a removable non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive for reading from or writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In such cases, each drive may be coupled to bus 803 via one or more data medium interfaces. The memory 801 may include at least one program product having a set (e.g., at least one) of program modules configured to carry out the functions of the various embodiments of the disclosure.

A program/utility 808 having a set (at least one) of program modules 807 may be stored in, for example, the memory 801, such program modules 807 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment. Program modules 807 typically carry out the functions and/or methods in the embodiments described in this disclosure.

The electronic device 800 may also communicate with one or more external devices 809 (e.g., keyboard, pointing device, display 810, etc.), one or more devices that enable a user to interact with the electronic device 800, and/or any devices (e.g., network card, modem, etc.) that enable the electronic device 800 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 811. Also, the electronic device 800 may communicate with one or more networks such as a Local Area Network (LAN), a Wide Area Network (WAN) and/or a public network, such as the Internet, through a network adapter 812. As shown in fig. 8, the network adapter 812 communicates with other modules of the electronic device 800 over the bus 803. It should be appreciated that although not shown in fig. 8, other hardware and/or software modules may be used in connection with electronic device 800, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

The processor 802 executes various functional applications and data processing by executing programs stored in the memory 801.

It should be noted that, the implementation process and the technical principle of the electronic device in this embodiment refer to the foregoing explanation of the information pushing method in the embodiment of the disclosure, and are not repeated herein.

In order to implement the above-described embodiments, the embodiments of the present application also propose a non-transitory computer-readable storage medium storing computer instructions for causing a computer to implement the information push method as shown in the above-described embodiments. Alternatively, the non-transitory computer readable storage medium may be ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

In order to implement the above embodiments, the embodiments of the present application further provide a computer program product, including a computer program, which when executed by a processor implements the information pushing method as shown in the above embodiments.

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 disclosure is intended to cover any adaptations, uses, or adaptations of the disclosure following the general 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 information pushing method is characterized by comprising the following steps:

acquiring keywords of information to be pushed;

matching the keywords with a keyword text library to obtain recall reference information corresponding to the information to be pushed from the keyword text library;

determining a target recall strategy of the information to be pushed from candidate recall strategies based on the recall reference information;

and acquiring a target pushing object corresponding to the information to be pushed based on the target recall strategy, and pushing the information to be pushed to the target pushing object.

2. The method of claim 1, wherein the obtaining, based on the target recall policy, the target push object corresponding to the information to be pushed comprises:

acquiring a first candidate pushing object corresponding to the information to be pushed;

the pushing type of the information to be pushed is obtained, and the first candidate pushing object is screened based on the pushing type to obtain a second candidate pushing object;

And acquiring the target push object based on the target recall strategy and the second candidate push object.

3. The method of claim 2, wherein the obtaining the target push object based on the target recall policy and the second candidate push object comprises:

acquiring pushing optimization parameters of the information to be pushed, and determining a target network model under the target recall strategy based on the pushing optimization parameters;

acquiring first characteristic information of the information to be pushed and second characteristic information of the second candidate push object;

inputting the first characteristic information and the second characteristic information into the target network model to acquire a first preference value of the second candidate pushing object for the information to be pushed;

and sequencing the first preference values from big to small, and acquiring the target pushing object according to the sequenced first preference values.

4. The method of claim 2, wherein the recall reference information includes a number of successful pushes of the information to be pushed and a scoring rate of the information to be pushed, wherein determining a target recall policy for the information to be pushed from among candidate recall policies based on the recall reference information comprises:

Determining the number of successful pushing times and a numerical interval corresponding to the good score;

and determining the target recall strategy of the information to be pushed from the candidate recall strategies based on the numerical interval.

5. The method of claim 4, wherein the determining the target recall policy for the information to be pushed from among candidate recall policies based on the value interval comprises:

determining a pre-recall policy as the target recall policy from the candidate recall policies in response to the number of successful pushes being greater than or equal to a set number of times and the good score being greater than or equal to a set good score;

the obtaining, based on the pre-recall policy, the target push object corresponding to the information to be pushed includes:

invoking pushing object information corresponding to the information to be pushed from a database;

and determining the target push object from the push objects corresponding to the push object information.

6. The method of claim 4, wherein the determining the target recall policy for the information to be pushed from among candidate recall policies based on the value interval comprises:

determining a model recall policy as the target recall policy from the candidate recall policies in response to the number of successful pushes being greater than or equal to the set number of times and the good score being less than the set good score;

The obtaining, based on the model recall policy, the target push object corresponding to the information to be pushed includes:

acquiring first real-time characteristic information of the information to be pushed;

acquiring second real-time characteristic information of the second candidate pushing object;

inputting the first real-time characteristic information and the second real-time characteristic information into a preset model to obtain a second preference value of the second candidate pushing object output by the preset model for the information to be pushed;

and sorting the second preference values from large to small, and acquiring the target pushing object according to the sorted second preference values.

7. The method of claim 4, wherein the determining the target recall policy for the information to be pushed from among candidate recall policies based on the value interval comprises:

determining a vector recall policy as the target recall policy from the candidate recall policies in response to the successful push times being less than the set times;

the obtaining, based on the vector recall policy, the target push object corresponding to the information to be pushed includes:

calling a first target vector of the information to be pushed and a second target vector of the second candidate push object from a database;

And carrying out inner product value calculation on the first target vector and the second target vector, sorting the inner product values obtained after calculation according to the order from big to small, and determining the target pushing object according to the sorted inner product values.

8. An information pushing apparatus, characterized by comprising:

the acquisition module is used for acquiring keywords of the information to be pushed;

the matching module is used for matching the keywords with a keyword text library so as to acquire recall reference information corresponding to the information to be pushed from the keyword text library;

the determining module is used for determining a target recall strategy of the information to be pushed from candidate recall strategies based on the recall reference information;

and the pushing module is used for acquiring a target pushing object corresponding to the information to be pushed based on the target recall strategy and pushing the information to be pushed to the target pushing object.

9. The apparatus of claim 8, wherein the push module is further configured to:

10. The apparatus of claim 9, wherein the push module is further configured to:

11. The apparatus of claim 9, wherein the recall reference information comprises a number of successful pushes of the information to be pushed and a scoring rate of the information to be pushed, the determining module further configured to:

12. The apparatus of claim 11, wherein the determining module is further configured to:

13. The apparatus of claim 11, wherein the determining module is further configured to:

14. The apparatus of claim 11, wherein the determining module is further configured to:

15. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

16. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-7.

17. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any of claims 1-7.