CN113886732A

CN113886732A - Information recommendation method

Info

Publication number: CN113886732A
Application number: CN202010634747.8A
Authority: CN
Inventors: 王庆阳; 占飞
Original assignee: Tencent Cyber Tianjin Co Ltd
Current assignee: Tencent Cyber Tianjin Co Ltd
Priority date: 2020-07-02
Filing date: 2020-07-02
Publication date: 2022-01-04

Abstract

The invention provides an information recommendation method, an information recommendation device, electronic equipment and a computer-readable storage medium; the method comprises the following steps: presenting the information flow in a human-computer interaction interface; determining the degree of influence when the recommendation information is inserted at a plurality of insertion positions in the information stream; determining a target insertion position from the plurality of insertion positions according to the influence degrees corresponding to the plurality of insertion positions respectively; the recommendation information is displayed at a target insertion location in the information stream. By means of the invention, recommendation information can be presented in a suitable insertion position of an information stream.

Description

Information recommendation method

Technical Field

The present disclosure relates to internet technologies, and in particular, to an information recommendation method and apparatus, an electronic device, and a computer-readable storage medium.

Background

With the popularization of internet technology, electronic devices can provide richer information streams, such as text streams (news streams), picture streams, video streams (short video streams), and the like. More and more users acquire various information streams through electronic equipment, for example, browsing news through news clients, and the like, so that the convenience of life is greatly improved. Wherein various recommendation information, such as advertisements, applets, etc., may be inserted in the information stream.

In the related art, recommendation information is presented at random positions or fixed positions in information streams, and the recommendation mode inevitably causes interference or interruption to browsing the information streams and influences user experience; and results in an inefficient push of recommendation information, causing unnecessary consumption of resources of the server, including communication resources and computing resources.

Disclosure of Invention

The embodiment of the invention provides an information recommendation method, an information recommendation device, electronic equipment and a computer readable storage medium, which can present recommendation information in a proper insertion position of an information stream.

The technical scheme of the embodiment of the invention is realized as follows:

the embodiment of the invention provides an information recommendation method, which comprises the following steps:

presenting the information flow in a human-computer interaction interface;

determining a degree of influence when the recommendation information is inserted at a plurality of insertion locations in the information stream;

determining a target insertion position from the plurality of insertion positions according to the influence degrees corresponding to the plurality of insertion positions respectively;

displaying the recommendation information at the target insertion location in the information stream.

An embodiment of the present invention provides an information recommendation apparatus, including:

the presentation module is used for presenting the information flow in the human-computer interaction interface;

the processing module is used for determining the influence degree when the recommendation information is inserted into a plurality of insertion positions in the information stream;

the determining module is used for determining a target insertion position used when the recommendation information is presented from the plurality of insertion positions according to the influence degrees corresponding to the plurality of insertion positions respectively;

a display module to display the recommendation information at the target insertion location in the information stream.

In the above technical solution, the processing module is further configured to determine a group to which a target recommended object belongs according to historical recommendation information of the target recommended object;

acquiring a plurality of insertion position characteristics corresponding to the group to which the target recommendation object belongs;

and determining the influence degree when the recommendation information is inserted into a plurality of insertion positions in the information flow according to the characteristics of the plurality of insertion positions corresponding to the group to which the target recommendation object belongs.

In the above technical solution, the processing module is further configured to determine a tolerance degree and an acceptance degree of the target recommendation object to the historical recommendation information;

and determining the group to which the target recommendation object belongs according to the tolerance degree and the admission degree of the target recommendation object to the historical recommendation information.

In the above technical solution, the processing module is configured to determine a parameter corresponding to a tolerance of the target recommendation object to the historical recommendation information, where the parameter includes:

exposure of historical recommendation information of the target recommendation object, wherein the historical recommendation information is presented in the process that the target recommendation object browses a historical information stream;

the duration of browsing a history information stream by the target recommendation object, wherein the history recommendation information is inserted into the history information stream;

the processing module is used for determining parameters corresponding to the acceptance degree of the target recommendation object to the historical recommendation information, and the parameters comprise:

click through rate of the historical recommendation information of the target recommendation object;

click rate of historical recommendation information of the target recommendation object;

the processing module is further used for carrying out weighted summation on the exposure of the historical recommendation information of the target recommendation object and the duration of browsing the historical information stream by the target recommendation object, and taking the weighted summation result as the tolerance degree of the target recommendation object to the historical recommendation information;

and carrying out weighted summation on the click through rate of the historical recommendation information of the target recommendation object and the click quantity of the historical recommendation information of the target recommendation object, and taking the weighted summation result as the acceptance degree of the target recommendation object to the historical recommendation information.

In the above technical solution, the processing module is further configured to determine, from a plurality of tolerance intervals, a target tolerance interval in which the tolerance of the target recommendation object to the historical recommendation information is located, and determine a group corresponding to the target tolerance interval as a target group to which the target recommendation object belongs, where each tolerance interval corresponds to one group;

the tolerance degree intervals are obtained by dividing the value intervals of the tolerance degree; the value interval of the tolerance degree is composed of tolerance degree values of a plurality of historical recommendation objects to the historical recommendation information, and the value interval of the tolerance degree takes the maximum value and the minimum value of the tolerance degree as end values;

determining a target acceptance degree interval in which the acceptance degree of the target recommendation object to the historical recommendation information is located from a plurality of acceptance degree intervals, and determining a subgroup corresponding to the target acceptance degree interval as a subgroup to which the target recommendation object belongs in the target group, wherein each acceptance degree interval corresponds to one subgroup;

the plurality of acceptance degree intervals are obtained by dividing the value intervals of the acceptance degrees; the value interval of the acceptance degree is composed of the acceptance degree values of the plurality of historical recommendation objects to the historical recommendation information, and the value interval of the acceptance degree takes the maximum value and the minimum value of the acceptance degree as end values.

In the above technical solution, the apparatus further includes: the system comprises a preprocessing module, a storage module and a display module, wherein the preprocessing module is used for dividing a plurality of historical recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of historical recommendation objects to historical recommendation information;

and determining the characteristic of each inserting position in the historical information flow of each group, and taking the characteristic as the characteristic of the inserting position of the group.

In the above-described aspect, the characteristic of the insertion position includes an exposure rate of the insertion position; the preprocessing module is further configured to traverse the historical information streams sent to the historical recommendation objects in each group, and perform the following processing for each insertion position in the traversed historical information streams:

determining a first number of historical recommendation objects browsed to the insertion position in the group;

determining a second number of historical recommendation objects browsed to a first insertion position in the group;

determining a ratio of the first number to the second number as an exposure of the insertion sites.

In the above technical solution, the preprocessing module is further configured to divide the value intervals of the tolerance degrees of the plurality of history recommendation objects to the history recommendation information into a plurality of tolerance degree intervals;

each tolerance degree interval corresponds to one group, and the value interval of the tolerance degree takes the maximum value and the minimum value of the tolerance degree as end values;

determining a tolerance degree interval in which the tolerance degree of each history recommendation object to the history recommendation information is located, and determining a group corresponding to the tolerance degree interval as a target group to which the history recommendation object belongs;

dividing the value interval of the acceptance degree of the historical recommendation information of the historical recommendation object into a plurality of acceptance degree intervals;

each acceptance degree interval corresponds to one subgroup in the target group, and the value interval of the acceptance degree takes the maximum value and the minimum value of the acceptance degree as end values;

and determining an acceptance degree interval where the acceptance degree of the historical recommendation object to the historical recommendation information is located, and determining a subgroup corresponding to the determined acceptance degree interval as a subgroup to which the historical recommendation object belongs in the target group.

In the above technical solution, the processing module is further configured to perform the following processing for each of a plurality of insertion positions in the information stream:

determining information promotion benefits when the recommendation information is inserted in the insertion position according to the insertion position characteristics of the insertion position and the recommendation characteristics of the recommendation information;

wherein the insertion position is any one of a plurality of insertion positions at which the recommendation information is inserted in the information stream for the target recommendation object;

determining information interference loss when the recommended information is inserted at the insertion position according to the insertion position characteristics of the insertion position;

and using the residual part of the information promotion profit after the information interference loss offset processing as the influence degree when the recommendation information is inserted in the insertion position.

In the above technical solution, the processing module is further configured to use a difference between the information promotion income and the information interference loss when the recommendation information is inserted in the insertion location as an influence degree when the recommendation information is inserted in the insertion location; alternatively, the first and second electrodes may be,

multiplying the loss quantization coefficient by the information interference loss, and taking the result of the multiplication as the information interference loss after the loss quantization;

and determining a difference between the information promotion profit and the information interference loss after the loss quantization when the recommended information is inserted at the insertion position as an influence degree when the recommended information is inserted at the insertion position.

In the above technical solution, the insertion position feature of the insertion position includes an exposure rate of the insertion position and a click through rate of the history recommendation information inserted in the insertion position; the recommendation characteristics comprise exposure cost and estimated click through rate; the processing module is further configured to determine a ratio of a click through rate of the historical recommendation information inserted at the insertion position to an estimated click through rate of the recommendation information as an information promotion profit weight of the recommendation information;

and multiplying the information promotion profit weight of the recommendation information, the exposure cost of the recommendation information and the exposure rate of the insertion position, and taking the multiplication result as the information promotion profit corresponding to the recommendation information inserted in the insertion position.

In the above technical solution, the information interference loss includes interference loss and information loss; wherein the interference loss is used to quantify the following information: the number of new recommendation information that can be presented while the process of displaying the recommendation information at the insertion location and browsing the information stream continues;

wherein the information loss is used to quantify the information characterizing: the number of new recommendation information that cannot be presented continuously when the recommendation information is displayed at the insertion location and the process of browsing the information stream is stopped.

In the above technical solution, the insertion position characteristics of the insertion position include a probability that the information stream continues to be browsed after the recommended information is inserted in the insertion position, and an expected exposure amount of the insertion position; the processing module is further configured to multiply the probability of continuing to browse the information stream after inserting the recommendation information at the insertion position by the expected exposure amount at the insertion position, and take the multiplication result as the interference loss when inserting the recommendation information at the insertion position.

In the above technical solution, the insertion position characteristics include a probability of stopping browsing the information stream after the recommended information is inserted in the insertion position, and an expected exposure of a subsequent insertion position of the insertion position; wherein the priority of the information corresponding to the insertion position in the information flow is higher than the priority of the information corresponding to the subsequent insertion position in the information flow, and the priority is ordered in a manner consistent with the priority ordering manner of the information in the information flow; wherein, the priority ranking mode of the information in the information flow comprises at least one of the following modes: ascending or descending according to publication time; ascending or descending according to the time of the latest comment; according to the ascending order or descending order of the heat; according to the ascending order or the descending order of the click quantity; according to the ascending order or descending order of the forwarding amount; according to the arrangement sequence of the information in the information flow; the processing module is further used for multiplying the probability of stopping browsing the information stream after the recommended information is inserted in the insertion position by the expected exposure of the subsequent insertion position of the insertion position, and taking the multiplication result as the information loss when the recommended information is inserted in the insertion position.

In the above technical solution, the determining module is further configured to sort the influence degrees respectively corresponding to the plurality of insertion positions in a descending order, and screen out at least one influence degree sorted in the front;

and determining the inserting position corresponding to the screened influence degree as the target inserting position.

An embodiment of the present invention provides an electronic device for information recommendation, where the electronic device includes:

a memory for storing executable instructions;

and the processor is used for realizing the information recommendation method provided by the embodiment of the invention when the executable instructions stored in the memory are executed.

The embodiment of the invention provides a computer-readable storage medium, which stores executable instructions and is used for causing a processor to execute the information recommendation method provided by the embodiment of the invention.

The embodiment of the invention has the following beneficial effects:

the appropriate insertion position is selected to present the recommendation information through the influence degree, the fusion of the recommendation information and the information flow is promoted, the user experience of browsing the information flow is guaranteed, meanwhile, a good information recommendation effect is achieved, the user does not need to worry about interference or interruption of the browsing information flow when the recommendation information is presented, the purpose of effectively pushing the information is achieved, and meanwhile, the resources of the server are saved.

Drawings

Fig. 1 is a schematic view of an application scenario of an information recommendation system according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of an electronic device 500 for information recommendation according to an embodiment of the present invention;

3-7 are flow diagrams of information recommendation methods provided by embodiments of the invention;

FIG. 8 is a schematic diagram of a recommendation system provided by an embodiment of the present invention;

FIG. 9A is a schematic diagram of an interface for advertisement recommendation provided by an embodiment of the present invention;

FIG. 9B is a diagram illustrating an interface for advertisement recommendation provided by an embodiment of the present invention;

fig. 10 is a flowchart of an advertisement ranking algorithm provided by an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail with reference to the accompanying drawings, the described embodiments should not be construed as limiting the present invention, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

In the following description, references to the terms "first \ second" are only to distinguish similar recommended objects and do not denote a particular ordering with respect to the recommended objects, it being understood that "first \ second" may be interchanged under certain circumstances or sequences, to enable embodiments of the invention described herein to be practiced in other than the order shown or described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to be limiting of the invention.

Before further detailed description of the embodiments of the present invention, terms and expressions mentioned in the embodiments of the present invention are explained, and the terms and expressions mentioned in the embodiments of the present invention are applied to the following explanations.

1) Click Through Rate (CTR): the click arrival rate of the recommendation information (e.g. applet, picture advertisement, text advertisement, keyword advertisement, ranking advertisement, video advertisement, etc.), i.e. the ratio of the actual number of clicks (the number of pages to target) of the recommendation information to the presentation amount (Show Content) of the recommendation information.

2) Cost Per thousand (Cost Per Mile, CPM): the cost per thousand exposures is a cost calculation unit which delivers a media or a media schedule to 1000 people or a family, and is a method for measuring the actual utility of the investment cost of the recommendation information, namely, in the process of delivering the recommendation information (such as advertisements), the average cost of the advertisements needed by each thousand people to hear or see certain recommendation information respectively is measured. CPM depends on an "impression" scale, i.e. the number of times a person's eyes see a recommendation for a fixed period of time. The formula for CPM calculation is as follows: cost of thousand people (advertising cost/number of people arriving) x 1000, where advertising cost/number of people arriving is usually expressed in the form of a percentage, and whether its advertising investment is national or regional is usually considered when estimating this percentage.

3) Thousand show yields (Effective Cost Per Mile, eCPM): the unit of presentation may be a web page, an ad unit, or even a single ad. By default, eCPM refers to thousands of web presentation (Pageview) revenues, and eCPM is used to reflect a parameter of the profitability of a website and does not represent revenue. The eCPM calculation is formulated as follows: the web page display time is 1000, where the revenue is the advertisement unit price × the web page click-through rate × the web page display time, that is, the eCPM is the advertisement unit price × the web page click-through rate × 1000, and the eCPM is an index independent of the web page display time.

4) Conversion (Conversion Rate, CVR): and in a statistical period, the number of times of completing the conversion behavior accounts for the ratio of the total clicks of the recommendation information. The calculation formula is as follows: conversion rate (number of conversions/click rate) × 100%. For example: and 10 users see the result of information recommendation, wherein 5 users click certain recommendation information and jump to the target address, and 2 users have subsequent conversion behaviors, so that the conversion rate of the recommendation information is (2/5) × 100% ═ 40%.

5) Feed (feeds) flow: the information flow of content is continuously updated and presented to the user. feed is a content aggregator that combines several message sources that a user actively subscribes to, typically news websites and blogs, to help the user continuously obtain the latest feed content. There are many presentation forms of feed streams, including a time line (timeline) and a rank (rank), where the timeline is a presentation form of feed streams, and presents contents to users, such as microblogs and friend circles, according to the time sequence of content update of the feed streams; rank calculates the weight of the content according to some factors, so as to determine the sequence of content display, for example, the current microblog homepage information flow algorithm discards the original timeline, and adopts the latest intelligent sorting.

6) Insertion position: the insertion position may be a position between any two information in the information stream, for example, if a certain information stream is composed of 3 information (i.e. the first information, the second information and the third information), the position between the first information and the second information is an insertion position, and the position between the second information and the third information is an insertion position; the insertion position may be a position adjacent to the information in the information stream, for example, if a certain information stream is composed of 2 information (i.e., the first information and the second information), the position before the first information is an insertion position, the position after the first information and before the second information is an insertion position, and the position after the second information is an insertion position.

An exemplary application of the electronic device for information recommendation provided by the embodiment of the invention is described below.

The electronic device for information recommendation provided by the embodiment of the invention can be various types of terminals or servers, wherein the server can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server for providing cloud computing service; the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart sound box, a smart watch or any other smart device capable of browsing information. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the present invention is not limited thereto.

Taking the server as an example, the server determines a target insertion position used when presenting the recommendation information in the plurality of insertion positions according to the influence degrees corresponding to the recommendation information inserted in the plurality of insertion positions of the information stream, and sends the recommendation information and the target insertion position to the client, and the client presents the recommendation information in the target insertion position in the information stream. For example, for a news application, recommended advertisements are presented in a target insertion location in a news stream composed of multiple news; for short video applications, a recommended gamelet is presented in a target insertion location in a short video stream composed of multiple short videos.

Referring to fig. 1, fig. 1 is a schematic view of an application scenario of an information recommendation system 10 according to an embodiment of the present invention, a terminal 200 is connected to a server 100 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two.

A terminal where a client (e.g., a news client, a short video client, etc.) is located may send a request for information flow, for example, a user slides a news page through a human-computer interface of the terminal to browse news, and the terminal automatically obtains a news update request.

In some embodiments, the terminal 200 sends an information stream request to the server and invokes an information recommendation function provided by the server 100, the server 100 determines a target insertion position used when presenting the recommendation information in a plurality of insertion positions according to the influence degree corresponding to the insertion of the recommendation information in the plurality of insertion positions of the information stream by the information recommendation method provided by the embodiments of the present invention, and sends the recommendation information and the target insertion position to present the recommendation information in the target insertion position in the information stream, for example, a news client is installed on the terminal 200, after a user slides a news page, the terminal 200 automatically generates a news update request and sends the news update request to the server 100 through the network 300, the server 100 determines recall information corresponding to the portrait information of the user from a database according to the portrait information of the user and screens the recall information, the method comprises the steps of screening out recommendation information meeting the user interest, determining a target insertion position used when the recommendation information is presented in a plurality of insertion positions according to the influence degrees corresponding to the insertion of the recommendation information in the plurality of insertion positions of the news stream, sending the recommendation information and the target insertion position to a news client, and presenting the recommendation information in the target insertion position of the news stream through a display interface 210 of a terminal 200.

The following describes a structure of an electronic device for information recommendation according to an embodiment of the present invention, where the electronic device for information recommendation may be various terminals, such as a mobile phone and a computer.

Referring to fig. 2, fig. 2 is a schematic structural diagram of an electronic device 500 for information recommendation according to an embodiment of the present invention, and taking the electronic device 500 as a server as an example for explanation, the electronic device 500 includes: at least one processor 510, memory 550, at least one network interface 520, and a user interface 530. The various components in the electronic device 500 are coupled together by a bus system 540. It is understood that the bus system 540 is used to enable communications among the components. The bus system 540 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 540 in fig. 2.

The Processor 510 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The memory 550 may comprise volatile memory or nonvolatile memory, and may also comprise both volatile and nonvolatile memory. The non-volatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a Random Access Memory (RAM). The memory 550 described in connection with embodiments of the invention is intended to comprise any suitable type of memory. Memory 550 optionally includes one or more storage devices physically located remote from processor 510.

In some embodiments, memory 550 can store data to support various operations, examples of which include programs, modules, and data structures, or subsets or supersets thereof, as exemplified below.

An operating system 551 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

a network communication module 552 for communicating to other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), etc.;

in some embodiments, the information recommendation device provided in the embodiments of the present invention may be implemented in a software manner. Fig. 2 shows an information recommendation device 555 stored in the memory 550, which may be software in the form of programs and plug-ins, etc., and includes a series of modules including a presentation module 5551, a processing module 5552, a determination module 5553, a display module 5554, and a pre-processing module 5555; the presenting module 5551, the processing module 5552, the determining module 5553, the displaying module 5554, and the preprocessing module 5555 are configured to implement the information recommending function provided by the embodiment of the present invention.

As can be understood from the foregoing, the information recommendation method provided in the embodiments of the present invention may be implemented by various types of electronic devices for information recommendation, such as an intelligent terminal and a server.

The information recommendation method provided by the embodiment of the present invention is described below with reference to an exemplary application and implementation of the server provided by the embodiment of the present invention. Referring to fig. 3, fig. 3 is a schematic flowchart of an information recommendation method according to an embodiment of the present invention, and is described with reference to the steps shown in fig. 3.

In step 101, the client sends an information flow request to the server.

For example, a terminal (e.g., a social network client, a news client, a short video client, etc.) where the client is located may send an information flow request to the server, for example, a news page is slid through a human-computer interaction interface of the terminal, and after browsing the current news page, the terminal automatically obtains a news update request to request more news flows.

In step 102, the server sends the information stream to the client.

For example, after receiving the information flow request, the server sends the information flow to the client in response to the information flow request of the client, and the client can present the information flow in the human-computer interaction interface after receiving the information flow. In this embodiment, the sequence of sending the information stream and the subsequent sending target insertion location is not limited, that is, there is no obvious sequence in step 102 and step 105, for example, after the target insertion location is determined, the server may send the information stream, the target insertion location, and the recommendation information to the client together.

In step 103 the server determines the degree of influence when inserting recommendation information at a plurality of insertion locations in the information stream.

As an example of obtaining recommendation information, referring to fig. 8, fig. 8 is a schematic diagram of a recommendation system provided in an embodiment of the present invention, the recommendation system is configured to obtain recommendation information, such as an advertisement and a game applet, during a process that a target user uses a client running on a terminal, the client reports collected interaction behaviors (user logs) of the target user with respect to historical recommendation information to a server 200 as training sample data and user images and user features corresponding to the target user, the training sample data is behavior data of different users reported from respective terminals, training of a prediction model is performed based on the behavior data, the user images and the user features are fed back by a terminal corresponding to a certain user, the prediction model performs prediction of candidate recommendation information based on the obtained user images and user features corresponding to the target user, and performing shuffling processing and recommendation based on the prediction result, so that the content meeting the interest point of the user can be recommended, and diversity of recommended content can be realized through shuffling processing on the premise.

Referring to fig. 8, the terminal 200 is connected to the server 100 through the network 300, the server 100 includes an information generation module, a prediction module, and a mixing module, the information generation module uses various recommendation algorithms to quickly screen candidate recommendation information related to a user from the candidate recommendation information database 400, and submits the recalled candidate recommendation information to the prediction module, and the prediction module predicts and sorts the recalled candidate recommendation information according to a user image and user characteristics of a target user to obtain the sorted recommendation information. After the server obtains the recommendation information, in response to the recommendation information request, the server may determine the degree of influence corresponding to when the recommendation information is inserted at any one insertion position of the information stream, for example, 3 insertion positions in the information stream, and may determine the degree of influence corresponding to when the recommendation information is inserted at 3 insertion positions of the information stream, respectively.

Referring to fig. 4, fig. 4 is an optional flowchart of the information recommendation method according to the embodiment of the present invention, and in order to accurately obtain the degree of influence corresponding to when recommendation information is inserted into any insertion position of an information stream, fig. 4 shows that step 103 in fig. 3 can be implemented through steps 1031 to 1033 shown in fig. 4: in step 1031, determining a group to which the target recommendation object belongs according to the historical recommendation information of the target recommendation object; in step 1032, a plurality of insertion position features corresponding to the group to which the target recommendation object belongs are obtained; in step 1033, the influence degree when the recommendation information is inserted into the plurality of insertion positions in the information stream is determined according to the characteristics of the plurality of insertion positions corresponding to the group to which the target recommendation object belongs.

Wherein, the target recommendation object is an object which accords with the orientation condition of the recommendation information. The server may first obtain the historical recommendation information of the target user from the historical recommendation information of the target recommendation object, for example, a log file of the target user, determine a group to which the target recommendation object belongs, determine a plurality of insertion location features corresponding to the group to which the target recommendation object belongs, and finally refer to the plurality of insertion location features corresponding to the group to which the target recommendation object belongs to determine the degree of influence corresponding to when the recommendation information is inserted at the plurality of insertion locations of the information flow.

Referring to fig. 5, fig. 5 is an optional flowchart of the information recommendation method according to the embodiment of the present invention, and in order to accurately obtain the group to which the target recommendation object belongs, fig. 5 shows that step 1031 in fig. 4 can be implemented by steps 311 to 312 shown in fig. 5: in step 311, determining tolerance and acceptance of the target recommendation object to the historical recommendation information; in step 312, a group to which the target recommendation object belongs is determined according to the tolerance and the acceptance of the target recommendation object to the historical recommendation information.

In order to assign the target recommendation object to an accurate group, tolerance and acceptance of the target recommendation object to the historical recommendation information may be obtained according to the historical recommendation object of the target recommendation object.

In some embodiments, determining the tolerance and the acceptance of the target recommendation object to the historical recommendation information includes: carrying out weighted summation on exposure of historical recommendation information of the target recommendation object and duration of browsing historical information streams by the target recommendation object, and taking a weighted summation result as tolerance of the target recommendation object to the historical recommendation information; and carrying out weighted summation on the click through rate of the historical recommendation information of the target recommendation object and the click quantity of the historical recommendation information of the target recommendation object, and taking the weighted summation result as the acceptance degree of the target recommendation object to the historical recommendation information.

In connection with the above example, the parameters for determining the tolerance of the target recommendation object to the historical recommendation information include: exposure of historical recommendation information of the target recommendation object, wherein the historical recommendation information is presented in the process that the target recommendation object browses the historical information stream; and the target recommendation object browses the duration of the historical information stream, wherein the historical recommendation information is inserted into the historical information stream. The parameters for determining the acceptance degree of the target recommendation object to the historical recommendation information comprise: click through rate of historical recommendation information of the target recommendation object; the click rate of the historical recommendation information of the target recommendation object. The tolerance degree of the embodiment of the invention is not limited to the exposure and the duration, and can be any as long as the passive index of the target recommendation object to the recommendation information can be reflected; the acceptance degree of the embodiment of the invention is not limited to the click through rate and the click quantity, and only can the initiative index of the target recommendation object to the recommendation information be reflected.

The exposure of the historical recommendation information of the target recommendation object or the duration of the target recommendation object browsing the historical information stream can be directly used as the tolerance of the target recommendation object to the historical recommendation information; and taking the click through rate of the historical recommendation information of the target recommendation object or the click quantity of the historical recommendation information of the target recommendation object as the acceptance degree of the target recommendation object to the historical recommendation information.

In some embodiments, determining the group to which the target recommendation object belongs according to the tolerance and the acceptance of the target recommendation object to the historical recommendation information includes: determining a target tolerance degree interval in which the tolerance degree of the target recommendation object to the historical recommendation information is located from the tolerance degree intervals, and determining a group corresponding to the target tolerance degree interval as a target group to which the target recommendation object belongs; and determining a target acceptance degree interval in which the acceptance degree of the target recommendation object to the historical recommendation information is located from the plurality of acceptance degree intervals, and determining a subgroup corresponding to the target acceptance degree interval as a subgroup to which the target recommendation object belongs in the target group.

Each tolerance degree interval corresponds to one group, and the tolerance degree intervals are obtained by dividing the value intervals of the tolerance degrees; the value interval of the tolerance degree is composed of tolerance degree values of a plurality of historical recommendation objects to the historical recommendation information, and the maximum value and the minimum value of the tolerance degree are used as end values of the value interval of the tolerance degree. Each admission degree interval corresponds to one subgroup, and the admission degree intervals are obtained by dividing the value intervals of the admission degrees; the value interval of the acceptance degree is composed of the acceptance degree values of the plurality of historical recommendation objects to the historical recommendation information, and the value interval of the acceptance degree takes the maximum value and the minimum value of the acceptance degree as end values.

For example, indicators used to characterize tolerance include exposure; the index used for representing the admission degree comprises click through rate; the target group includes: a low-level exposure group, a medium-level exposure group, and a high-level exposure group; when the exposure of the target recommendation object is less than or equal to the low exposure threshold, dividing the target recommendation object into low exposure groups; when the exposure of the target recommendation object is larger than or equal to a high exposure threshold, dividing the recommendation object into high-level exposure groups; when the exposure of the target recommendation object is larger than a low exposure threshold and smaller than a high exposure threshold, dividing the target recommendation object into a middle exposure group; wherein the low exposure threshold is less than the high exposure threshold. When the target recommendation object is divided into the middle-level exposure group, the target recommendation object is divided into subgroups belonging to the target group according to the click passing rate of the target recommendation object on the recommendation information, for example, when the target recommendation object is divided into the middle-level exposure group, the target recommendation object is divided into the middle-level click group under the middle-level exposure group according to the click passing rate of the target recommendation object on the recommendation information.

For example, the tolerance degree intervals are a tolerance degree interval 1 (value interval is 0-10) and a tolerance degree interval 2 (value interval is 10-20), the tolerance degree of the current target recommendation object to the history recommendation information is 5, and the tolerance degree of the target recommendation object to the history recommendation information belongs to the tolerance degree interval 1; the plurality of acceptance degree intervals are an acceptance degree interval 1 (the value interval is 0-10) and an acceptance degree interval 2 (the value interval is 10-20), the acceptance degree of the current target recommendation object to the historical recommendation information is 15, and the acceptance degree of the target recommendation object to the historical recommendation information is the acceptance degree interval 2 belonging to the acceptance degree interval 1.

In some embodiments, before transmitting the information stream, the method further comprises: dividing the plurality of historical recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of historical recommendation objects to the historical recommendation information; and determining the characteristic of each inserting position in the historical information flow of each group, and taking the characteristic as the inserting position characteristic of the group.

In order to avoid the situation that a group to which the target recommendation object belongs is calculated in real time, a plurality of insertion position characteristics corresponding to the group to which the target recommendation object belongs are calculated. The method comprises the steps of calculating a plurality of insertion position characteristics corresponding to groups in an off-line mode, namely dividing a plurality of historical recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of historical recommendation objects to historical recommendation information, determining the characteristics of each insertion position in a historical information stream of each group, and storing the characteristics of each insertion position in a database as the insertion position characteristics of the group in advance. When the plurality of insertion position features corresponding to the group need to be acquired, for example, after the group to which the target recommendation object belongs is determined, the plurality of insertion position features corresponding to the group to which the target recommendation object belongs are acquired from the database, so that the time for calculating the plurality of insertion position features corresponding to the group to which the target recommendation object belongs is saved, and the plurality of insertion position features corresponding to the group to which the target recommendation object belongs are acquired in real time.

In some embodiments, determining characteristics of each insertion location in the historical information stream for each group comprises: traversing the history information stream sent to the history recommendation object in each group, and executing the following processing aiming at each insertion position in the traversed history information stream: determining a first number of historical recommendation objects browsed to insertion positions in a group; determining a second number of historical recommendation objects browsed to the first insertion position in the group; the ratio of the first number to the second number is determined as the exposure of the inserted position.

Wherein the characteristic of the inserted position includes an exposure of the inserted position. Exposure pb at insertion position t_tThe calculation method is as follows: the number of exposed users at the insertion position t divided by the first insertion positionNumber of exposed users.

Wherein the insertion location may further be characterized by: 1) the click through rate of the historical recommendation information inserted in the insertion position; 2) the characteristics of the insertion position comprise the probability of continuing browsing the information flow after the recommended information is inserted in the insertion position; 3) a desired exposure amount of the insertion site; 4) the probability of stopping browsing the information stream after inserting the recommendation information in the insertion position; 5) a desired exposure amount for a subsequent insertion position to the insertion position.

Wherein, 1) click through rate ctr of history recommendation information inserted at insertion position t_tThe calculation method is as follows: dividing the click rate of the historical recommendation information of the insertion position t by the exposure of the historical recommendation information of the insertion position t;

2) probability prob1 for continuing browsing information stream after inserting recommendation information at insertion position t_tNamely the probability that the user continues browsing after the recommended information is inserted into the position t in the information stream, the calculation method is as follows: the proportion of the number of the users browsing to the insertion position t +1 to the number of the users browsing to the insertion position t;

3) expected exposure amount expo1 of insertion position t_≥tNamely the expected exposure of the inserted position t and the subsequent inserted position in the information stream, the calculation method is as follows: the expected exposure of each position is equal to the average human exposure of the position multiplied by the exposure of the inserted position t, and the expected exposure is accumulated from the inserted position t to the back, namely, expo1_≥tTaking the value of (A);

4) probability prob2 for stopping browsing information stream after inserting recommendation information at insertion position t_tI.e. the probability that the user stops browsing after inserting the recommendation information at the insertion position t in the information stream, 1 minus prob1_tNamely the probability of jumping out of browsing;

5) expected exposure amount expo2 for insertion positions subsequent to insertion position t_>tI.e. the expected exposure for the subsequent insertion position of the insertion position t in the information stream, is calculated by: the expected exposure amount at each position is equal to the average human exposure amount at the position multiplied by the exposure rate at the current position, and the expected exposure amount is accumulated from the insertion position t +1, i.e., expo2_>tThe value of (a).

In some embodiments, dividing the plurality of historical recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of historical recommendation objects to the historical recommendation information includes: dividing the value intervals of the tolerance degrees of the plurality of historical recommendation objects to the historical recommendation information into a plurality of tolerance degree intervals; determining a tolerance degree interval of the tolerance degree of each history recommendation object to the history recommendation information, and determining a group corresponding to the tolerance degree interval as a target group to which the history recommendation object belongs; dividing the value interval of the acceptance degree of the historical recommendation information of the historical recommendation object into a plurality of acceptance degree intervals; and determining an acceptance degree interval where the acceptance degree of the historical recommendation object to the historical recommendation information is located, and determining a subgroup corresponding to the determined acceptance degree interval as a subgroup to which the historical recommendation object belongs in the target group.

Each tolerance degree interval corresponds to one group, and the value interval of the tolerance degree takes the maximum value and the minimum value of the tolerance degree as end values; each acceptance degree interval corresponds to one subgroup in the target group, and the value interval of the acceptance degree takes the maximum value and the minimum value of the acceptance degree as end values.

For example, indicators used to characterize tolerance include exposure; the index used for representing the admission degree comprises click through rate; the target group includes: a low-level exposure group, a medium-level exposure group, and a high-level exposure group; when the exposure of the target recommendation object is less than or equal to the low exposure threshold, dividing the target recommendation object into low exposure groups; when the exposure of the target recommendation object is larger than or equal to a high exposure threshold, dividing the recommendation object into high-level exposure groups; when the exposure of the target recommendation object is larger than a low exposure threshold and smaller than a high exposure threshold, dividing the target recommendation object into a middle exposure group; wherein the low exposure threshold is less than the high exposure threshold. And when the target recommendation object is divided into the middle-level exposure group, dividing the target recommendation object into the subgroups belonging to the target group according to the click through rate of the target recommendation object on the recommendation information.

Referring to fig. 6, fig. 6 is an optional flowchart of an information recommendation method according to an embodiment of the present invention, and fig. 6 shows that step 1033 in fig. 4 can be implemented by steps 331 to 333 shown in fig. 6: performing the following for each of a plurality of insertion locations in the information stream: in step 331, determining an information promotion benefit when the recommendation information is inserted at the insertion position according to the insertion position characteristics of the insertion position and the recommendation characteristics of the recommendation information; in step 332, determining an information interference loss when the recommended information is inserted at the insertion position according to the insertion position characteristics of the insertion position; in step 333, the remaining part of the information popularization profit after the information interference loss cancellation processing is set as the degree of influence when the recommendation information is inserted at the insertion position.

The insertion position is any one of a plurality of insertion positions where the recommendation information is inserted in the information stream for the target recommendation object, a group to which the target recommendation object belongs corresponds to the plurality of insertion positions, each insertion position corresponds to one insertion position feature, and the insertion position features of the same insertion position corresponding to different groups are different, for example, if the group to which the target recommendation object 1 belongs is group 1 and the group to which the target recommendation object 2 belongs is group 2, the insertion position feature of the insertion position 1 corresponding to the group 1 is different from the insertion position feature of the insertion position 1 corresponding to the group 2.

For example, the insertion position characteristics of the group to which the target recommendation object belongs and the recommendation characteristics of the recommendation information may characterize the degree of influence corresponding to the insertion position when the recommendation information is inserted. And determining information promotion revenue corresponding to the insertion position when the recommendation information is inserted in the insertion position based on the insertion position characteristics of the insertion position corresponding to the group to which the target recommendation object belongs and the recommendation characteristics of the recommendation information, determining information interference loss corresponding to the insertion position when the recommendation information is inserted in the insertion position based on the insertion position characteristics of the insertion position corresponding to the group to which the target recommendation object belongs, finally canceling the information interference loss in the information promotion revenue, and taking the rest part of the information promotion revenue after the information interference loss cancellation processing as the corresponding influence degree when the recommendation information is inserted in the insertion position. The larger the remaining portion is, the larger the influence degree corresponding to the insertion position of the recommendation information is.

In some embodiments, the step of using the remaining part of the information promotion profit after the information interference loss cancellation processing as the influence degree when the recommendation information is inserted at the insertion position includes: taking the difference value between the corresponding information promotion income and the information interference loss when the recommendation information is inserted in the insertion position as the influence degree when the recommendation information is inserted in the insertion position; or, multiplying the loss quantization coefficient by the information interference loss, and taking the product result as the information interference loss after loss quantization; the difference between the information promotion profit when the recommendation information is inserted at the insertion position and the corresponding information interference loss after the loss quantization is used as the influence degree when the recommendation information is inserted at the insertion position.

The loss quantization coefficient can enable the loss and the gain to be in the same dimension, a scaling coefficient is given to the loss, the loss and the gain are in the same dimension, and the loss quantization coefficient can be set based on empirical data.

Referring to fig. 7, fig. 7 is an optional flowchart of the information recommendation method according to the embodiment of the present invention, and fig. 7 shows that step 331 in fig. 6 can be implemented by steps 3311 to 3312 shown in fig. 7: in step 3311, determining the ratio of the click through rate of the historical recommendation information inserted at the insertion position to the estimated click through rate of the recommendation information as the information promotion income weight of the recommendation information; in step 3312, the information promotion profit weight of the recommendation information, the exposure cost of the recommendation information, and the exposure rate of the insertion position are multiplied, and the result of the multiplication is taken as the information promotion profit when the recommendation information is inserted at the insertion position.

The insertion position characteristics of the insertion position comprise the exposure rate of the insertion position and the click through rate of the historical recommendation information inserted in the insertion position; the recommended features include exposure cost and estimated click through rate.

Showing the information promotion income corresponding to the recommendation information inserted at the insertion position t, wherein ecpm_adExposure cost, pb, indicating recommendation information_tThe exposure rate representing the insertion position t is multiplied,

information promotion profit weight, ctr, representing recommendation information_tClick through rate, ctr, indicating history recommendation information inserted at the insertion position t_adAnd the estimated click through rate of the recommendation information is shown.

In some embodiments, wherein the information interference loss comprises an interference loss and an information loss, wherein the interference loss is used to quantify information characterizing: the number of new recommendation information that can be presented when the recommendation information is displayed at the insertion location and the process of browsing the information stream continues; wherein the information loss is used to quantify the following information: the number of new recommendation information that cannot be presented continuously when the recommendation information is displayed at the insertion location and the process of browsing the information stream is stopped. For example, the sum of the interference loss and the information loss is used as the information interference loss corresponding to the insertion position of the recommendation information.

In some embodiments, the insertion location characteristics of the insertion location include a probability of continuing to browse the information stream after the insertion of the recommendation information at the insertion location, and a desired exposure of the insertion location. prob1_t×expo1_tIndicates the interference loss corresponding to the insertion of the recommendation information at the insertion position t, prob1_tIndicating the probability of continuing to browse the information stream after inserting the recommendation information at the insertion location t, expo1_tIndicating the desired exposure at the insertion location t. And multiplying the probability of continuing browsing the information stream after inserting the recommendation information in the insertion position by the expected exposure amount in the insertion position, and taking the multiplication result as the corresponding interference loss when inserting the recommendation information in the insertion position.

In some embodiments, the insertion location characteristics include a probability of stopping browsing the information stream after the insertion of the recommendation information at the insertion location, and a period of subsequent insertion locations of the insertion locationExpecting exposure, wherein the priority of the information corresponding to the insertion position in the information flow is higher than the priority of the information corresponding to the subsequent insertion position in the information flow, and the priority is arranged in a mode consistent with the priority arrangement mode of the information in the information flow; the priority ranking mode of the information in the information flow comprises at least one of the following modes: ascending or descending according to publication time; ascending or descending according to the time of the latest comment; according to the ascending order or descending order of the heat; according to the ascending order or the descending order of the click quantity; according to the ascending order or descending order of the forwarding amount; according to the arrangement order of the information in the information flow. prob2_t×expo2_>tIndicates the corresponding information loss when the recommendation information is inserted at the insertion position t, prob2_tIndicating the probability of continuing to browse the information stream after inserting the recommendation, expo2_>tIndicating the desired exposure at the insertion location t. And multiplying the probability of stopping browsing the information stream after inserting the recommendation information in the insertion position by the expected exposure of the subsequent insertion position of the insertion position, and taking the multiplication result as the corresponding information loss when the recommendation information is inserted in the insertion position.

The priority ordering mode is consistent with the priority ordering mode of the information in the information stream, for example, the insertion position t corresponds to the first information, the subsequent insertion position is insertion position t +1, the insertion position t +1 corresponds to the second information, and the priority of the first information in the information stream is higher than that of the second information, that is, the first information is located before the second information, the insertion position t is located before the insertion position t +1, wherein the insertion modes of the insertion position t and the insertion position t +1 may be both inserted at the position before the information in the information stream; or inserted at a position after a certain information in the information stream; the insertion position t may be inserted at a position before the information in the information stream, and the insertion position t +1 may be inserted at a position after the information in the information stream.

For example, if the insertion position is insertion position 1, the insertion position subsequent to the insertion position is insertion position 2, and a certain information stream is composed of 3 pieces of information (i.e., information 1 and information 2), then the insertion position before information 1 is insertion position 1, and the insertion position before information 2 is insertion position 2; or the insertion position after the information 1 and before the information 2 is the insertion position 1, and the insertion position after the information 2 is the insertion position 2; alternatively, the insertion position before the information 1 is the insertion position 1, and the insertion position after the information 2 is the insertion position 2.

Wherein, the information in the information flow can be arranged according to the ascending order or descending order of the publication time; the information in the information stream may be arranged in ascending or descending order of the most recent review time; the information in the information stream may be arranged in ascending or descending order according to heat; the information in the information stream can be arranged according to the ascending order or the descending order of the click rate; the information in the information flow may be arranged in ascending or descending order according to the forwarding amount; the information in the information stream may be arranged according to the order of arrangement of the information in the information stream.

In step 104, the server determines a target insertion position among the plurality of insertion positions according to the influence degrees corresponding to the plurality of insertion positions.

For example, the influence degrees corresponding to the plurality of insertion positions are sorted in a descending order, at least one influence degree in the front of the order is screened out, and the insertion position corresponding to the screened influence degree is determined as a target insertion position used when the recommendation information is presented, so that the recommendation information is presented at the target insertion position.

In step 105, the server sends the recommendation information and the target insertion location to the client.

In step 106, the client presents the recommendation information in the target insertion location in the information stream.

For example, after receiving the information stream, the recommendation information, and the target insertion position, the client presents the recommendation information in the target insertion position in the information stream, that is, dynamically presents the recommendation information.

As an example, when the target user 1 sends an information flow request to the server through the client running in the terminal 1, and the target user 2 sends an information flow request to the server through the client running in the terminal 2, for example, the target user 1 and the target user 2 browse news simultaneously, as shown in fig. 9A, the server recommends an advertisement (recommendation information) to the target user 1 according to the user profile and the user characteristics of the target user 1, and determines a target insertion position used when presenting the car advertisement as a position 901 between the news 1 and the news 2 among a plurality of insertion positions according to the influence degrees corresponding to the plurality of insertion positions respectively; as shown in fig. 9B, the server also recommends the same car advertisement to the target user 2 based on the user profile and user characteristics of the target user 2, and determines a target insertion position used when presenting the car advertisement among the plurality of insertion positions as a position 902 between news 3 and news 4 based on the influence degrees corresponding to the plurality of insertion positions, respectively. As can be seen from fig. 9A and 9B, the server determines that the insertion positions of the recommendation information are different according to different users, that is, for each user, a suitable insertion position is selected to present the recommendation information, so that the fusion of the recommendation information and the information stream is promoted, and the user experience of browsing the information stream is ensured.

In the following, an exemplary application of the embodiments of the present invention in a practical application scenario will be described.

The embodiment of the invention can be applied to an application scene of information recommendation, such as advertisement recommendation, as shown in fig. 1, a terminal is connected with a server 100 deployed at a cloud end through a network 300, a news client application is installed on the terminal, after a user slides to the bottom of a news page, the terminal 200 automatically generates a news update request, and sends the news update request to the server 100 through the network 300, the server 100 determines a target insertion position used when the recommendation information is presented in a plurality of insertion positions according to the influence degree (total income) corresponding to the insertion position of the news stream when the recommendation information is inserted in the plurality of insertion positions, and sends the recommendation information and the target insertion position to a news client, and the recommendation information is presented in the target insertion position of the news stream.

In the recommendation field, a common ranking scenario is to rank contents of a single category such as advertisements, information, or commodities, and rank the contents based on a predicted value such as a CTR, CVR, or CPM that is a target of prediction. However, in an actual service scene, multiple types of content may be delivered in the same scene, for example, a certain number of advertisements may be inserted in a Feeds stream in a space, and in such a scene, the Feeds stream and the advertisements cannot be equalized, and a conventional CTR/CPM prediction mode is adopted for sorting. This ordering requirement is commonly referred to as shuffling, i.e., the co-ordering of multiple different categories of content. In the related art, the shuffling scheme of the spatial Feeds stream is in a fixed position form, that is, the advertisements are inserted into fixed positions in the Feeds stream, and the higher the eCPM value of the advertisements is, the more forward the inserted positions (insertion positions) are. The mixed arrangement mode is mainly from the aspects of product form and user experience, namely, on the premise of not disturbing the user experience, the inserted advertisements are ensured to maintain the order of eCPM from high to low.

However, the shuffling scheme in the related art has the following problems in practice: 1) the same mixed arrangement mode is adopted for all users, and the tolerance of different crowds to the advertisement is not considered; 2) with the fixed-position mixed arrangement scheme, the revenue and the loss caused by inserting the advertisement are not quantitatively measured, and the mixed arrangement scheme is not necessarily optimal.

In order to solve the above problems, the embodiment of the present invention provides a practical and comprehensive advertisement ranking algorithm (an information recommendation method) in combination with the actual service form of the product, which calculates the optimal insertion position of the advertisement by a specific algorithm formula, that is, calculates the influence caused by inserting the advertisement at different insertion positions, determines the insertion position corresponding to the largest influence as the optimal insertion position, for example, the influence caused by inserting the advertisement at insertion position 1 is 2, the influence caused by inserting the advertisement at insertion position 2 is 5, and the influence caused by inserting the advertisement at insertion position 3 is 3, and then the optimal insertion position is insertion position 2. In the algorithm scheme, a calculation mode of influences brought by inserting advertisements at different positions (including gains in information promotion (information promotion gains) brought by pushing the advertisements and losses brought by interference on information flow browsed by a user (information interference losses)) is designed, an optimal insertion position (target insertion position) is obtained by comparing the gains and the losses, and dynamic sequencing of the advertisements is realized.

The method and the device can be used for recommending scenes with mixed arrangement requirements, such as mixed arrangement of advertisements in Feeds streams in space, mixed arrangement of advertisements in information streams of news applications/microblog applications and mixed arrangement of advertisements in short video streams, and the advertisements are inserted into the optimal positions in the information streams (including image/text message streams and video streams (such as Feeds streams, information streams or short video streams)) browsed by the user under the condition that user experience is not disturbed as much as possible, so that the maximum benefit is obtained.

In the mixed arrangement of the spatial Feeds, the embodiment of the invention divides the crowd based on the tolerance degree of the advertisement, measures the interruption probability of the user browsing after the advertisement insertion according to the priori knowledge, quantitatively calculates the expected loss caused by the advertisement insertion, compares the loss and the profit by combining the potential profit caused by the advertisement, and solves the optimal position of the advertisement insertion on the premise of fixing the advertisement quantity, thereby taking the user experience and the goal of maximizing the advertisement profit into consideration.

As shown in fig. 10, the online operation of the advertisement ranking algorithm of the embodiment of the present invention mainly includes three parts: the first part is an off-line processing (preprocessing) part, which divides the crowd, calculates the characteristics of the required insertion positions off-line for different crowds and writes the characteristics into a memory, such as Kafka; the second part is an online processing part, and when a user browses the Feeds stream, the user can request advertisement information and related characteristics (such as e CPM and estimated CTR of the advertisement) in real time and pull the characteristics of the insertion position from Kafka; the third part is that the online processing part calculates the optimal position (target insertion position) for advertisement insertion based on the requested characteristics. The first part (population partitioning and feature calculation) and the third part are explained in detail below:

a first part: off-line processing

A) Crowd division

The purpose of dividing the groups is to perform mixed ranking more accurately. The exposure and CTR difference of different crowds is large, if the characteristics are calculated by uniformly using large disk data, the characteristic values are forcibly averaged, so that unnecessary deviation is brought to a user group with the activity deviating from the average value. Therefore, the crowd is divided in a statistical analysis mode, the indexes such as exposure, advertisement CTR and the like which are related to the strong user activity degree are adopted to layer the users, and the crowd in different layers is used as a classification result. The CTR of the exposure and the advertisement is selected as the basis of crowd division: the exposure represents the tolerance of the user to the browsing depth and the advertisement (the exposure is a passive index, namely the number of times of occurrence of recommendation information can be tolerated in the browsing information stream, namely the number of times of passive watching), and the higher the exposure is, the deeper the user browses and the higher the tolerance to the advertisement is; the CTR of the advertisement embodies the acceptance degree of the user to the advertisement (the CTR of the advertisement is an initiative index, namely the probability of actively watching the recommendation information in the process of browsing the information flow, namely the probability of actively clicking), and the higher the CTR is, the higher the acceptance degree of the user is represented, and otherwise, the lower the acceptance degree of the user is represented.

The browsing depth of the user can accord with a certain discrete probability distribution, and the user accords with a certain crowd distribution, so that a reasonable mapping relation can exist from the user to the browsing depth.

The exposure/CTR in the embodiments of the present invention is not limited to the exposure/CTR, and may be an index reflecting passive viewing/active clicking, for example, the exposure may be a staying time of a user in an information flow page where an advertisement is placed, and the CTR may be a click amount.

In addition, when the exposure is too low or too high in dividing the population, the advertisement CTR is not distinguished. The reason is that: on one hand, when the exposure is too low, the CTR of the advertisement is too sensitive, the fluctuation of the value range is too large, the randomness of the CTR is too large, and the remarkable statistical significance is not achieved; on the other hand, when the exposure is too high, the significance of the CTR of the advertisement for distinguishing the users is negligible, and the influence of the advertisement on the users with high exposure is small, so that the advertisement CTR does not need to be continuously used for subdividing the crowd.

Therefore, with regard to crowd division, the crowd is comprehensively divided from two dimensions, exposure and CTR of advertisements. Meanwhile, for the exposure which is too low and the exposure which is too high, the CTR of the advertisement can not be distinguished; the specific threshold value of the crowd layering is subject to the actual business situation.

B) Features for calculating insertion positions corresponding to different groups of people

After the population partition is completed, the following location features are calculated for each population:

1)ecpm_ad: advertisement score, estimated CPM value (exposure cost) of the advertisement;

2)ectr_ad: the estimated CTR of the advertisement;

3)ectr_t: the estimated value of the advertisement CTR inserted into the location t in the feeds stream (the click through rate of the historical recommendation information inserted into the location t) is calculated using the data of the past month, and the calculation method is as follows: dividing the advertisement click quantity of the insertion position t by the advertisement exposure quantity of the insertion position t;

4)pb_t: the exposure rate of the insertion position t in the feeds stream is calculated by using the data of the past month, and the calculation mode is as follows: dividing the number of exposure users at the insertion position t by the number of exposure users at the first insertion position;

5)expo1_≥t: the expected exposure for the insertion location t and subsequent insertion locations in the feeds stream is calculated using the past month data in the following manner: the expected exposure of each position is equal to the average human exposure of the position multiplied by the exposure of the inserted position t, and the expected exposure is accumulated from the inserted position t to the back, namely, expo1_≥tTaking the value of (A);

6)expo2_>t: the expected exposure for subsequent insertion locations to the insertion location t in the feeds stream is calculated using the past month data as: the expected exposure amount at each position is equal to the average human exposure amount at the position multiplied by the exposure rate at the current position, and the expected exposure amount is accumulated from the insertion position t +1, i.e., expo2_>tTaking the value of (A);

7)prob1_t: the probability that a user continues browsing after inserting the advertisement into the insertion position t in the feeds stream is calculated by using data of the past month, and the calculation mode is based on the meaning of the conditional probability and comprises the following steps: the proportion of the number of the users browsing to the insertion position t +1 to the number of the users browsing to the insertion position t;

8)prob2_t: the probability that a user jumps out of view (stops viewing) after inserting an advertisement at an insertion position t in a feeds stream is usedMonthly data were calculated, 1 minus prob1_tNamely the probability of jumping out of browsing;

9) w: loss quantization coefficients, scaling the loss so that the loss and the gain are in the same dimension; and assigning a scaling factor to the loss based on the total revenue and the total loss of each crowd for inserting the advertisement, so that the total loss and the total revenue are on the same dimension, and w is set based on empirical data.

And a third part: computing an optimal solution for an insertion location

Based on the above position characteristics, the mixed-row formula provided in the embodiment of the present invention is shown in formula (1):

wherein the content of the first and second substances,

indicating the weighted revenue (information promotion revenue) from inserting the advertisement position at the insertion position t,

represents a weight, which means: when the ecr_tGreater than ecr_adThe income is amplified when the user wants to use the system, and is reduced when the user wants to use the system; prob1_t×(expo1_≥t-expo2_>t) Indicating the loss (interference loss) caused when the user continues to browse after inserting the advertisement at the insertion position t; prob2_t×expo2_>tIndicates the loss (information loss) caused by the user jumping out of the viewing after inserting the advertisement at the insertion position t, and puts prob1_t×(expo1_≥t-expo2_>t) And prob2_t×expo2_>tAnd adding the two components, and scaling by a scaling factor w to obtain the total loss (information interference loss).

The formula solving range is the insertion position range of the feeds stream requested by the user at present, the position sequence number is counted from the first request of the user, and the value t of the maximum value of the formula can be obtained by using the formula (1) in the position range requested each time, namely the optimal insertion position of the advertisement insertion.

Therefore, from the product perspective, the mixed arrangement scheme takes the acceptance degree of different crowds to the advertisements and the tolerance degree of the advertisements into consideration, so that the real-time mixed arrangement of the advertisements in feeds is realized, and the user experience and the advertisement profit maximization are considered. From the technical point of view, a method for calculating the loss caused by advertisement insertion is provided: the probability calculation modes of continuous browsing and jumping-out browsing of the user after the advertisement is inserted are quantized, so that the expected loss caused by the advertisement insertion is comprehensively and reasonably calculated. The optimal advertisement insertion position can be obtained by comparing the profit and the loss of different insertion positions. The embodiment of the invention overcomes the problem of the advertisement mixed arrangement scheme with fixed rules, maximizes the advertisement income and considers the browsing experience of the user on the premise of not increasing the advertisement quantity; meanwhile, by designing a loss calculation mode, the inserted advertisement and the recommended content are compared in the same category to obtain the optimal advertisement insertion position.

The information recommendation method provided by the embodiment of the present invention has been described with reference to the exemplary application and implementation of the server provided by the embodiment of the present invention, and the following continues to describe a scheme for implementing information recommendation by cooperation of each module in the information recommendation device 555 provided by the embodiment of the present invention.

A presentation module 5551, configured to present the information stream in the human-computer interaction interface; a processing module 5552 determining a degree of influence when inserting recommendation information at a plurality of insertion locations in the information stream; a determining module 5553, configured to determine, according to the influence degrees corresponding to the multiple insertion positions, a target insertion position used when the recommendation information is presented from the multiple insertion positions; a display module 5554 for displaying the recommendation information at the target insertion location in the information stream.

In some embodiments, the processing module 5552 is further configured to determine a group to which a target recommended object belongs according to historical recommendation information of the target recommended object; acquiring a plurality of insertion position characteristics corresponding to the group to which the target recommendation object belongs; and determining the influence degree when the recommendation information is inserted into a plurality of insertion positions in the information flow according to the characteristics of the plurality of insertion positions corresponding to the group to which the target recommendation object belongs.

In some embodiments, the processing module 5552 is further configured to determine a tolerance level and an acceptance level of the target recommendation object for the historical recommendation information; and determining the group to which the target recommendation object belongs according to the tolerance degree and the admission degree of the target recommendation object to the historical recommendation information.

In some embodiments, the processing module 5552 is further configured to determine a parameter corresponding to a tolerance level of the target recommendation object to the historical recommendation information, where the parameter includes: exposure of historical recommendation information of the target recommendation object, wherein the historical recommendation information is presented in the process that the target recommendation object browses a historical information stream; the duration of browsing a history information stream by the target recommendation object, wherein the history recommendation information is inserted into the history information stream; the processing module 5552 is further configured to determine a parameter corresponding to the acceptance degree of the target recommendation object to the historical recommendation information, where the parameter includes: click through rate of the historical recommendation information of the target recommendation object; click rate of historical recommendation information of the target recommendation object; the processing module 5552 is further configured to perform weighted summation on the exposure of the historical recommendation information of the target recommendation object and the duration of browsing the historical information stream by the target recommendation object, and use the result of the weighted summation as the tolerance of the target recommendation object to the historical recommendation information; and carrying out weighted summation on the click through rate of the historical recommendation information of the target recommendation object and the click quantity of the historical recommendation information of the target recommendation object, and taking the weighted summation result as the acceptance degree of the target recommendation object to the historical recommendation information.

In some embodiments, the processing module 5552 is further configured to determine, from a plurality of tolerance intervals, a target tolerance interval in which the tolerance of the target recommendation object to the historical recommendation information is located, and determine a group corresponding to the target tolerance interval as a target group to which the target recommendation object belongs, where each tolerance interval corresponds to one group; the tolerance degree intervals are obtained by dividing the value intervals of the tolerance degree; the value interval of the tolerance degree is composed of tolerance degree values of a plurality of historical recommendation objects to the historical recommendation information, and the value interval of the tolerance degree takes the maximum value and the minimum value of the tolerance degree as end values; determining a target acceptance degree interval in which the acceptance degree of the target recommendation object to the historical recommendation information is located from a plurality of acceptance degree intervals, and determining a subgroup corresponding to the target acceptance degree interval as a subgroup to which the target recommendation object belongs in the target group, wherein each acceptance degree interval corresponds to one subgroup; the plurality of acceptance degree intervals are obtained by dividing the value intervals of the acceptance degrees; the value interval of the acceptance degree is composed of the acceptance degree values of the plurality of historical recommendation objects to the historical recommendation information, and the value interval of the acceptance degree takes the maximum value and the minimum value of the acceptance degree as end values.

In some embodiments, the apparatus further comprises: the preprocessing module 5555 is configured to divide the plurality of history recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of history recommendation objects to the history recommendation information; and determining the characteristic of each inserting position in the historical information flow of each group, and taking the characteristic as the characteristic of the inserting position of the group.

In some embodiments, the characteristic of the insertion locations comprises an exposure of the insertion locations; the preprocessing module 5555 is further configured to traverse the history information streams sent to the history recommendation objects in each group, and perform the following processing for each insertion position in the traversed history information stream: determining a first number of historical recommendation objects browsed to the insertion position in the group; determining a second number of historical recommendation objects browsed to a first insertion position in the group; determining a ratio of the first number to the second number as an exposure of the insertion sites.

In some embodiments, the preprocessing module 5555 is further configured to divide the value intervals of the tolerance degrees of the plurality of historical recommendation objects to the historical recommendation information into a plurality of tolerance degree intervals; each tolerance degree interval corresponds to one group, and the value interval of the tolerance degree takes the maximum value and the minimum value of the tolerance degree as end values; determining a tolerance degree interval in which the tolerance degree of each history recommendation object to the history recommendation information is located, and determining a group corresponding to the tolerance degree interval as a target group to which the history recommendation object belongs; dividing the value interval of the acceptance degree of the historical recommendation information of the historical recommendation object into a plurality of acceptance degree intervals; each acceptance degree interval corresponds to one subgroup in the target group, and the value interval of the acceptance degree takes the maximum value and the minimum value of the acceptance degree as end values; and determining an acceptance degree interval where the acceptance degree of the historical recommendation object to the historical recommendation information is located, and determining a subgroup corresponding to the determined acceptance degree interval as a subgroup to which the historical recommendation object belongs in the target group.

In some embodiments, the processing module 5552 is further configured to perform the following for each of a plurality of insertion locations in the information stream: determining information promotion benefits when the recommendation information is inserted in the insertion position according to the insertion position characteristics of the insertion position and the recommendation characteristics of the recommendation information; wherein the insertion position is any one of a plurality of insertion positions at which the recommendation information is inserted in the information stream for the target recommendation object; determining information interference loss when the recommended information is inserted at the insertion position according to the insertion position characteristics of the insertion position; and using the residual part of the information promotion profit after the information interference loss offset processing as the influence degree when the recommendation information is inserted in the insertion position.

In some embodiments, the processing module 5552 is further configured to use a difference between the information promotion revenue and the information interference loss when the recommendation information is inserted at the insertion position as the influence degree when the recommendation information is inserted at the insertion position; or, multiplying the loss quantization coefficient by the information interference loss, and taking the result of the multiplication as the information interference loss after the loss quantization; and determining a difference between the information promotion profit and the information interference loss after the loss quantization when the recommended information is inserted at the insertion position as an influence degree when the recommended information is inserted at the insertion position.

In some embodiments, the insertion position characteristics of the insertion position include an exposure rate of the insertion position and a click through rate of the history recommendation information inserted at the insertion position; the recommendation characteristics comprise exposure cost and estimated click through rate; the processing module 5552 is further configured to determine a ratio of a click through rate of the historical recommendation information inserted at the insertion position to an estimated click through rate of the recommendation information as an information promotion income weight of the recommendation information; and multiplying the information promotion profit weight of the recommendation information, the exposure cost of the recommendation information and the exposure rate of the insertion position, and taking the multiplication result as the information promotion profit corresponding to the recommendation information inserted in the insertion position.

In some embodiments, the information interference loss comprises an interference loss and an information loss; wherein the interference loss is used to quantify the following information: the number of new recommendation information that can be presented while the process of displaying the recommendation information at the insertion location and browsing the information stream continues; wherein the information loss is used to quantify the information characterizing: the number of new recommendation information that cannot be presented continuously when the recommendation information is displayed at the insertion location and the process of browsing the information stream is stopped.

In some embodiments, the insertion location characteristics of the insertion location include a probability of continuing to browse the information stream after the insertion location inserts the recommendation information, and a desired exposure of the insertion location; the processing module 5552 is further configured to multiply the probability of continuing to browse the information stream after inserting the recommendation information at the insertion position by the expected exposure at the insertion position, and to use the multiplication result as the interference loss when inserting the recommendation information at the insertion position.

In some embodiments, the insertion location characteristics include a probability of stopping browsing the information stream after the insertion location inserts the recommendation information, and a desired exposure for subsequent ones of the insertion locations; wherein, the priority of the information corresponding to the insertion position in the information flow is higher than the priority of the information corresponding to the subsequent insertion position in the information flow, and the priority is ordered in a manner consistent with the priority ordering manner of the information in the information flow; wherein, the priority ranking mode of the information in the information flow comprises at least one of the following modes: ascending or descending according to publication time; ascending or descending according to the time of the latest comment; according to the ascending order or descending order of the heat; according to the ascending order or the descending order of the click quantity; according to the ascending order or descending order of the forwarding amount; according to the arrangement sequence of the information in the information flow; the processing module 5552 is further configured to multiply the probability of stopping browsing the information stream after inserting the recommendation information at the insertion position by a desired exposure amount at a subsequent insertion position to the insertion position, and to use the multiplication result as the information loss when inserting the recommendation information at the insertion position.

In some embodiments, the determining module 5553 is further configured to sort the influence degrees corresponding to the plurality of insertion positions in a descending order, and filter out at least one influence degree sorted in the front; and determining the inserting position corresponding to the screened influence degree as the target inserting position.

Embodiments of the present invention provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the information recommendation method according to the embodiment of the invention.

Embodiments of the present invention provide a computer-readable storage medium storing executable instructions, which when executed by a processor, cause the processor to perform an information recommendation method provided by embodiments of the present invention, for example, the information recommendation method shown in fig. 3-7.

In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (H TML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).

By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.

The above description is only an example of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.

Claims

1. An information recommendation method, comprising:

presenting the information flow in a human-computer interaction interface;

2. The method of claim 1, wherein said determining a degree of influence when inserting recommendation information at a plurality of insertion locations in said information stream comprises:

determining a group to which a target recommendation object belongs according to historical recommendation information of the target recommendation object;

3. The method according to claim 2, wherein the determining the group to which the target recommendation object belongs according to the historical recommendation information of the target recommendation object comprises:

determining tolerance and acceptance of a target recommendation object to historical recommendation information;

4. A method according to claim 3, characterized in that the method comprises:

determining parameters corresponding to tolerance of the target recommendation object to historical recommendation information, wherein the parameters comprise:

determining parameters corresponding to the acceptance degree of the target recommendation object to the historical recommendation information, wherein the parameters comprise:

the determining tolerance and acceptance of the target recommendation object to the historical recommendation information includes:

carrying out weighted summation on the exposure of the historical recommendation information of the target recommendation object and the duration of browsing the historical information stream by the target recommendation object, and taking the weighted summation result as the tolerance of the target recommendation object to the historical recommendation information;

5. The method according to claim 3, wherein the determining the group to which the target recommendation object belongs according to the tolerance and the admission of the target recommendation object to the historical recommendation information comprises:

determining a target tolerance degree interval in which the tolerance degree of the target recommendation object to the historical recommendation information is located from a plurality of tolerance degree intervals, and determining a group corresponding to the target tolerance degree interval as a target group to which the target recommendation object belongs, wherein each tolerance degree interval corresponds to one group;

6. The method of claim 2,

before presenting the information flow in the human-computer interaction interface, the method comprises the following steps:

dividing a plurality of historical recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of historical recommendation objects to historical recommendation information;

7. The method of claim 6,

the feature of the inserted position comprises an exposure of the inserted position;

said determining characteristics of each insertion location in the historical information stream for each of said groups comprising:

traversing the history information stream sent to the history recommendation object in each group, and executing the following processing aiming at each insertion position in the traversed history information stream:

8. The method of claim 6, wherein the dividing the plurality of historical recommendation objects into a plurality of groups according to tolerance and acceptance of the plurality of historical recommendation objects to the historical recommendation information comprises:

dividing the value intervals of the tolerance degrees of the plurality of historical recommendation objects to the historical recommendation information into a plurality of tolerance degree intervals;

9. The method according to claim 2, wherein the determining the degree of influence when the recommendation information is inserted into a plurality of insertion positions in the information stream according to the characteristics of the plurality of insertion positions corresponding to the group to which the target recommendation object belongs comprises:

performing the following for each of a plurality of insertion locations in the information stream:

10. The method according to claim 9, wherein the using the remaining portion of the information promotion profit after the information interference loss cancellation processing as the degree of influence when the recommendation information is inserted at the insertion position comprises:

taking a difference between the information promotion profit and the information interference loss when the recommendation information is inserted at the insertion position as an influence degree when the recommendation information is inserted at the insertion position; alternatively, the first and second electrodes may be,

11. The method of claim 9,

the insertion position characteristics of the insertion position comprise the exposure rate of the insertion position and the click through rate of the historical recommendation information inserted in the insertion position; the recommendation characteristics comprise exposure cost and estimated click through rate;

the determining the information promotion benefit when the recommendation information is inserted in the insertion position according to the insertion position characteristics of the insertion position and the recommendation characteristics of the recommendation information includes:

determining the ratio of the click through rate of the historical recommendation information inserted in the insertion position to the estimated click through rate of the recommendation information as the information promotion income weight of the recommendation information;

multiplying the information promotion profit weight of the recommendation information, the exposure cost of the recommendation information, and the exposure rate of the insertion position, and taking the result of the multiplication as the information promotion profit when the recommendation information is inserted at the insertion position.

12. The method of claim 9,

the information interference loss comprises interference loss and information loss; wherein the interference loss is used to quantify the following information: the number of new recommendation information that can be presented while the process of displaying the recommendation information at the insertion location and browsing the information stream continues;

13. The method of claim 12,

the insertion position characteristics of the insertion position comprise the probability of continuing to browse the information stream after the recommended information is inserted in the insertion position and the expected exposure of the insertion position;

the determining information interference loss when the recommended information is inserted at the insertion position includes:

multiplying the probability of continuing to browse the information stream after the recommended information is inserted at the insertion position by the expected exposure at the insertion position, and taking the product result as the interference loss when the recommended information is inserted at the insertion position.

14. The method of claim 12,

the insertion position characteristics include a probability of stopping browsing the information stream after the insertion position inserts the recommendation information, and a desired exposure amount of a subsequent insertion position of the insertion position;

wherein the priority of the information corresponding to the insertion position in the information flow is higher than the priority of the information corresponding to the subsequent insertion position in the information flow, and the priority is ordered in a manner consistent with the priority ordering manner of the information in the information flow;

wherein, the priority ranking mode of the information in the information flow comprises at least one of the following modes:

ascending or descending according to publication time;

ascending or descending according to the time of the latest comment;

according to the ascending order or descending order of the heat;

according to the ascending order or the descending order of the click quantity;

according to the ascending order or descending order of the forwarding amount;

according to the arrangement sequence of the information in the information flow;

multiplying the probability of stopping browsing the information stream after the recommended information is inserted at the insertion position by the expected exposure of the subsequent insertion position of the insertion position, and taking the multiplication result as the information loss when the recommended information is inserted at the insertion position.

15. The method of claim 1, wherein said determining a target insertion location from said plurality of insertion locations based on respective degrees of influence associated with said plurality of insertion locations comprises:

sorting the influence degrees corresponding to the plurality of insertion positions in a descending order, and screening out at least one influence degree sorted in the front;