CN116467509A

CN116467509A - Information recommendation method, device, equipment, storage medium and program product

Info

Publication number: CN116467509A
Application number: CN202210033785.7A
Authority: CN
Inventors: 张月鹏; 杨永强; 贾子涵
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2022-01-12
Filing date: 2022-01-12
Publication date: 2023-07-21

Abstract

The application relates to an information recommendation method, an information recommendation device, information recommendation equipment, a storage medium and a program product, and relates to the technical field of information recommendation. The method comprises the following steps: based on the recall information of the recall in the recall process, a first recall information set is obtained; screening historical recall information with a historical scoring result higher than the first condition in the historical recall information recalled in the historical recall process to obtain a second recall information set; the historical scoring result is used for indicating the probability of the target object account executing target operation on the historical recall information at the historical recall time; based on the first recall information set and the second recall information set, providing the recommended recall information to the target object account; through the method, in the recall stage, the historical recall results meeting the conditions in the previous recommendation process are recycled, and the association between different information recommendation processes is realized, so that the utilization effect of recall information is improved, and the information recommendation effect is further improved.

Description

Information recommendation method, device, equipment, storage medium and program product

Technical Field

The embodiment of the application relates to the technical field of information recommendation, in particular to an information recommendation method, an information recommendation device, a storage medium and a program product.

Background

The development of multimedia technology and big data technology makes information recommendation more desirable. The information recommendation process may be implemented by a recommendation system. The primary recommendation process in the recommendation system comprises a recall stage, a sorting stage and a recommendation stage.

In the related art, the goal of the recall stage is to screen out a small amount of information from a large amount of information as recall information; the goal of the sort stage is typically to sort the recall information; the goal of the recommendation phase is typically to make information recommendations based on the ranking results.

The proportion of the information recommended in the recommendation stage to the information recalled in the recall stage is small, and due to factors such as information timeliness, difference exists in the recall information of different recommendation processes, so that the probability that the recall information which is not recommended in the recommendation stage is subsequently recommended is low, and the information recommendation effect is affected.

Disclosure of Invention

The embodiment of the application provides an information recommendation method, device, equipment, storage medium and program product, which can realize the association between different information recommendation processes, thereby improving the utilization effect of recall information and further improving the information recommendation effect. The technical scheme is as follows:

In one aspect, there is provided an information recommendation method, the method including:

based on the recall information of the recall in the recall process, a first recall information set is obtained;

screening the historical recall information with the historical scoring result higher than the first condition in the historical recall information recalled in the historical recall process to obtain a second recall information set; the historical scoring result is used for indicating the probability of the target object account executing target operation on the historical recall information at the historical recall moment;

and providing the recommended recall information to the target object account based on the first recall information set and the second recall information set.

In another aspect, there is provided an information recommendation apparatus, the apparatus including:

the first set acquisition module is used for acquiring a first recall information set based on the recall information of the present time recalled in the recall process;

the second set acquisition module is used for screening the historical recall information with the historical scoring result higher than the first condition in the historical recall information recalled in the historical recall process to obtain a second recall information set; the historical scoring result is used for indicating the probability of the target object account executing target operation on the historical recall information at the historical recall moment;

And the information recommendation module is used for providing the recommended recall information for the target object account based on the first recall information set and the second recall information set.

In a possible implementation manner, the second set obtaining module is configured to screen, in the history recall information recalled in the history recall process, the history recall information that is higher than the first condition and is not recommended to the target object account, so as to obtain the second recall information set.

In one possible implementation manner, the second set acquisition module includes:

the subset reading module is used for reading the historical information subsets corresponding to the n recording time periods in the historical information set respectively; the historical recall times of the historical recall information in the subset of historical information are within the historical time period corresponding to the subset of historical information; the historical scoring results of the historical recall information in the subset of historical information are higher than the first condition and are not recommended to the target object account; n is a positive integer;

and the collection acquisition sub-module is used for acquiring the second recall information collection from the n historical information subsets.

In one possible implementation manner, the set acquisition sub-module includes:

a subset obtaining unit, configured to obtain, when generating the history information subset of each of the n history information subsets for the recording period, a history recall information subset corresponding to each of the i recommendation processes executed in the recording period; the history recall information subset comprises target history recall information; the historical scoring result of the target historical recall information is in the top k names in the historical recall information subset; i, k is a positive integer;

the first set acquisition unit is used for acquiring an unreferenced information set, wherein the unreferenced information set comprises unreferenced information in the target history recall information corresponding to each i recommendation process;

and the subset generating unit is used for generating the historical information subset corresponding to the recording time period based on the non-recommended information set.

In a possible implementation manner, the subset generating unit is configured to, in response to the number of the non-recommended information in the non-recommended information set being greater than a first recall threshold, rank based on the historical scoring results of the respective non-recommended information, and obtain a first ranking result;

Based on the first sorting result, the un-recommended information with the quantity equal to the first recall threshold value is obtained to form the historical information subset corresponding to the recording time period.

In one possible implementation, the subset acquisition submodule includes:

a time acquisition unit for acquiring a recommended time; the recommending time is the time corresponding to the recommending process;

a subset obtaining unit, configured to obtain m target historical information subsets from n historical information subsets based on the recommended time; m is less than or equal to n, and m is a positive integer;

and the second set acquisition unit is used for acquiring a set formed by the history recall information in the m target history subsets as the second recall information set.

In a possible implementation manner, the subset obtaining unit is configured to:

determining a time interval based on the time interval of the recommended time in the recording time period in which the recommended time is located in the current recommendation process;

acquiring a recording time period which is in front of the recording time period in which the current recommending process is positioned and is different from the recording time period in which the current recommending process is positioned by the time interval as a target time period;

And acquiring the historical information subsets corresponding to the target time periods into m target historical information subsets.

In a possible implementation manner, the second set obtaining module is configured to screen, from the history recommendation information recommended by the history recall process, the history recommendation information of the target operation that is not received by the target object to obtain the second recall information set; the target object is an object corresponding to the target object account.

In one possible implementation manner, the information recommending module includes:

the scoring acquisition sub-module is used for acquiring the scoring result of the time of each recall information in the recall information set; the recall information set comprises the first recall information set and the second recall information set;

the recommendation information determining submodule is used for determining recommendation information based on the scoring result of the time of each recall information in the recall information set;

and the information recommending sub-module is used for providing the recommending information for the target object account.

In one possible implementation, the recommendation information determining submodule includes:

the sorting unit is used for sorting the recall information in the recall information set based on the current scoring result of each recall information in the recall information set to obtain a second sorting result;

And the recommendation information determining unit is used for determining the recommendation information based on the second sorting result.

In one possible implementation, the score acquisition sub-module includes:

the object feature acquisition unit is used for acquiring object features of the target object account;

and the scoring acquisition unit is used for respectively acquiring the scoring results of the time of each recall information based on the object characteristics of the target object account.

In a possible implementation manner, the scoring unit is configured to input the object feature of the target object account and target recall information into a scoring model, and obtain the current scoring result of the target recall information output by the scoring model; the target recall information is any one of the recall information;

the scoring model is obtained through training based on a training sample set, wherein the training sample set comprises sample object characteristics of a sample object account, sample information and scoring labels of the sample information relative to the sample object account.

In one possible implementation, the sample information in the training sample set is historical recommendation information; the sample object features are object features of the recommended object account corresponding to the historical recommendation information when the historical recommendation information is recommended; the scoring tag of the sample information relative to the sample object account is a recommendation result of the historical recommendation information relative to the recommendation object account;

The recommendation result comprises one of the recommended objects executing a target operation and the recommended objects not executing the target operation; the recommended object is an object corresponding to the recommended object account.

In one possible implementation manner, the first recall information set includes at least one of a first sub-information set, a second sub-information set and a third sub-information set; the first sub-information set is an information set composed of recall information obtained by screening from the candidate recall information, and the first sub-information set is a keyword determined based on historical behavior data of a target object account; the second sub-information set is an information set composed of recall information obtained by screening from the candidate recall information, and the semantic vector is determined based on historical behavior data of the target object account; the third sub-information set is an information set composed of recall information obtained by screening from the candidate recall information based on information attributes.

In another aspect, a computer device is provided, the computer device comprising a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the information recommendation method described above.

In another aspect, a computer readable storage medium having at least one computer program stored therein is provided, the computer program being loaded and executed by a processor to implement the above-described information recommendation method.

In another aspect, a computer program product or computer program is provided, the 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 to cause the computer device to perform the information recommendation methods provided in the various alternative implementations described above.

The technical scheme that this application provided can include following beneficial effect:

obtaining a first recall information set based on the recall information of the recall in the recall process; obtaining a second recall information set based on the relationship with the first condition; and determining the recommendation information in the current recommendation process from the information in the first recall information set and the information in the second recall information set, and recommending the recommendation information. Through the scheme, in the recall stage, the historical recall results meeting the conditions in the previous recommendation process are recycled, and the association between different information recommendation processes is realized, so that the utilization effect of recall information is improved, and the information recommendation effect is further improved.

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

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

FIG. 1 is a schematic diagram of an information recommendation process according to an exemplary embodiment of the present application;

FIG. 2 is a schematic diagram illustrating a configuration of an information recommendation system according to an exemplary embodiment;

FIG. 3 is a flowchart illustrating a method of information recommendation, according to an example embodiment;

FIG. 4 is a schematic diagram illustrating a generation process of a subset of history information according to an exemplary embodiment of the present application;

FIG. 5 illustrates a schematic diagram of a process of obtaining target history recall information, as shown in an exemplary embodiment of the present application;

FIG. 6 illustrates a schematic diagram of acquiring a set of un-recommended information according to an exemplary embodiment of the present application;

FIG. 7 is a schematic diagram of a historical information set generation process provided by an exemplary embodiment of the present application;

FIG. 8 illustrates a flowchart of an information recommendation method illustrated in an exemplary embodiment of the present application;

FIG. 9 is a diagram illustrating a correspondence between time intervals and sources of a subset of target history information according to an exemplary embodiment of the present application;

FIG. 10 illustrates an architecture diagram of an information recommendation system according to an exemplary embodiment of the present application;

FIG. 11 is a block diagram of an information recommendation device according to an exemplary embodiment of the present application;

FIG. 12 is a block diagram of a computer device shown in accordance with an exemplary embodiment;

fig. 13 is a block diagram of a computer device, according to an example embodiment.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present application as detailed in the accompanying claims.

The information recommendation method provided by the embodiment of the application can establish the relevance among independent recommendation processes, and multiplexing of history recall information in the history recommendation process is realized. Fig. 1 shows a schematic diagram of an information recommendation process according to an exemplary embodiment of the present application. As shown in FIG. 1, during the recall phase of the recommendation process, the computer device may perform information recall via two different sources of information. In the embodiment of the present application, the recall path from which the information source is the candidate recall information is illustrated as a first recall path 110, and the recall path from which the information source is the history recall information is illustrated as a second recall path 120. The first recall path 110 is configured to filter out candidate recall information conforming to the recall tag from a large number of candidate recall information according to a matching result of each candidate recall information and the recall tag, and form a first recall information set 130. The second recall path 120 is configured to recall and screen the history recall information obtained from the history recommending process before the current recommending process, obtain the history recall information that the history scoring result is higher but is not recommended to the target object account, or obtain the history recommending information that is recommended in the history recommending process but is not received the target operation executed by the target object, and form the second recall information set 140, where the target object is the object corresponding to the target object account. Then, the computer device uses the information in the first recall information set 130 and the information in the second recall information set 140 together as recall information in the current recommendation process, sorts the recall information in the sorting stage, and uses the sorted recall information of the first several names as recommendation information so as to provide the recommendation information to the target object account in the recommendation stage.

Fig. 2 is a schematic structural diagram of an information recommendation system 200 according to an exemplary embodiment, the information recommendation system 200 including: a server 220 and a number of terminals 240.

Server 220 includes at least one of a server, a plurality of servers, a cloud computing platform, and a virtualization center. The server 220 is configured to provide background services for the terminal 240; in this embodiment of the present application, the server 220 may be configured to execute an information recommendation process, and send the obtained recommendation information to the terminal 240, so that the terminal 240 performs information recommendation in a recommendation manner such as interface display or audio playing.

The terminal 240 may be a terminal device having an information recommendation function, for example, the terminal 240 may be a mobile phone, a tablet computer, an electronic book reader, smart glasses, a smart watch, an MP3 player (Moving Picture Experts Group Audio Layer III, moving picture experts compression standard audio layer 3), an MP4 (Moving Picture Experts Group Audio Layer IV, moving picture experts compression standard audio layer 4) player, a laptop portable computer, a desktop computer, and the like. The terminal 240 may include an application program with an information pushing function, so as to push information. Alternatively, the application may be an application that needs to be downloaded and installed, or may be a point-and-use application, which is not limited in the embodiment of the present application.

The terminal 240 is connected to the server 220 through a communication network. Optionally, the communication network is a wired network or a wireless network.

Alternatively, the wireless network or wired network described above uses standard communication techniques and/or protocols. The network is typically the Internet, but may be any network including, but not limited to, a local area network (Local Area Network, LAN), metropolitan area network (Metropolitan Area Network, MAN), wide area network (Wide Area Network, WAN), a mobile, wired or wireless network, a private network, or any combination of virtual private networks. In some embodiments, data exchanged over the network is represented using techniques and/or formats including HyperText Mark-up Language (HTML), extensible markup Language (Extensible Markup Language, XML), and the like. All or some of the links may also be encrypted using conventional encryption techniques such as secure socket layer (Secure Socket Layer, SSL), transport layer security (Transport Layer Security, TLS), virtual private network (Virtual Private Network, VPN), internet protocol security (Internet Protocol Security, IPsec), and the like. In other embodiments, custom and/or dedicated data communication techniques may also be used in place of or in addition to the data communication techniques described above.

Fig. 3 is a flowchart illustrating an information recommendation method that may be performed by a computer device, which may be implemented as a terminal or a server, schematically, which may be implemented as a terminal 240 or a server 220 as shown in fig. 1, according to an exemplary embodiment. As shown in fig. 3, the information recommendation method may include the steps of:

step 310, obtaining a first recall information set based on the recall information of the present recall in the present recall process.

In the embodiment of the application, when the recall information is recalled in the recall process, the computer equipment can screen the recall information from the candidate recall information based on the current matching result of the candidate recall information, so as to obtain a first recall information set; the matching result is used for indicating the matching degree of the candidate recall information and the recall tag; the recall tag is used for indicating rules for screening the recall information for the target object account from the candidate recall information.

In different application scenarios, the content types corresponding to the recall information in the recall process are different. Schematically, in a shopping recommendation scene, the recall information is commodity information; in the video recommendation scene, the recall information is video information; in the audio recommendation scene, the recall information is audio information; in the image pushing scene, the recall information is image information, and the like. Therefore, the content type of the recall information can be determined based on the recommended needs of different application scenarios, which is not limited in the present application.

In the embodiment of the present application, the number of candidate recall information may be far greater than the number of current recall information obtained based on the current matching result screening; that is, the process of obtaining the first recall information set may be regarded as primarily screening out part of the candidate recall information conforming to the recall tag from the massive candidate recall information, so as to reduce the number of information to be processed in the sorting stage and improve the information processing efficiency. The recall tag may be determined based on a target object or a screening condition, and illustratively, the recall tag is used for indicating a rule of screening the recall information for a target object account from each candidate recall information, where the target object account is a recommended object account to be subjected to information recommendation in the recommendation process; the computer device may employ different rules to screen the candidate recall information, for example, may screen the candidate recall information based on the target object account number, or may screen the candidate recall information based on information attributes, where the information attributes may include a vertical class (or referred to as a vertical domain) to which the information belongs, a hotspot ranking of the information, and so on; therefore, the recall tag may be a tag corresponding to the object feature, or may be a tag corresponding to the information attribute, or the like, which is not limited in this application.

When the first recall information set is acquired in the recall phase, the higher the matching degree between the candidate recall information and the recall tag is, the higher the probability of being screened to acquire the recall information of the present time is.

Step 320, screening the historical recall information with the historical scoring result higher than the first condition in the historical recall information recalled in the historical recall process to obtain a second recall information set; the historical scoring result is used for indicating the probability of the target object account executing target operation on the historical recall information at the historical recall time.

Wherein the first condition may be indicative of a historical scoring result threshold; alternatively, the first condition may also indicate a ranking threshold value after ranking based on the historical scoring results.

The history recall time is the time at which history of history recall information was recalled.

In the information recommendation process, in order to screen and obtain recommendation information from the recall information, the computer equipment needs to score the recall information in a sorting stage so as to determine the probability that a target object corresponding to a target object account performs target operation on the recall information after the recall information is recommended to the target object account, and further obtain the first names with higher scoring results from the recall information as recommendation information, and recommend the recommendation information to the target object account in the recommendation stage; thus, for historical recall information, each historical recall information has a historical scoring result corresponding thereto; the historical scoring result is used for indicating the probability that the historical recall information receives the target operation executed by the target object at the historical recall moment; because the number of recall information is larger than the number of recommended information in the recommended flow, the historical recall information has higher historical scoring result but is not pushed with the historical recall information of the target object account; or historical recommendation information of target operation of a target object corresponding to the target object account is not received in the historical recall information, wherein the historical recommendation information is recommended to the target object account; for example, 10 pieces of history recall information are sorted based on the history scoring results, and the history recall information which is pushed is determined to be the first two pieces of history scoring results with higher history scoring results, so that the history recall information which is ranked after the second name is not pushed in the history recommendation process; or determining that one of the two pushed historical recall information does not receive the target operation executed by the target object. Because the update frequency of the candidate recall information is higher, when information recall is carried out next time, the history recall information which is not pushed in the history recommendation process and has higher scoring result, or the history recommendation information which is pushed but does not receive the target operation executed by the target object, can not be recalled, and has lower probability of being recommended to the target object; although such historical recall information or historical recommendation information may be scored higher during the ranking stage; based on the above factors, if recall is performed using only the first recall path, the following problems may result: due to the update of the recall model or the influence of information timeliness, the history recall information or the history recommendation information with higher scores can be obtained in the sorting stage originally, and the sorting stage cannot be entered because the history recall information or the history recommendation information is not recalled in the recall stage of the recommendation process.

In this regard, in the embodiment of the present application, in order to effectively use the history recall information, the computer device may use an information set (a second recall information set) formed by the history recall information obtained by screening based on the history scoring result of the history recall information as an acquisition path of another recall information in the current recall flow, so as to establish a correlation between the history recommendation process and the current recommendation process, thereby implementing multiplexing of the history recall information, and improving the information recommendation efficiency and the recommendation effect.

In the embodiment of the present application, the target operation performed on the recall information by the target object corresponding to the target object account may refer to a positive feedback operation of the recall information by the target object, and the positive feedback operation may include at least one of operations of praise, collection, clicking, sharing, gift sending, and the like; the scoring result may be used to indicate a probability that the target object performs the target operation; illustratively, when the target operation performed by the target object on the recall information is a click operation, the scoring result may be used to indicate a probability that the recall information is clicked by the target object corresponding to the target object account.

In the embodiment of the application, the historical recall information may be recorded according to a recording time period. Because the probability that the same recommended object performs the target operation on the same history recall information may be different in different recording time periods, the history scoring result has a time attribute, and the history scoring results corresponding to the same history recall information may be different in different history recall times.

Because the interest point of the target object changes along with the change of time, in order to improve the relevance between the history recall information used by the recall and the recommendation process, in the embodiment of the application, the computer equipment can screen and obtain a second recall information set from the history recall information recalled in the history recall process based on the history recall time of the history recall information; illustratively, the computer device may screen for a second set of recall information from the historical recall information based on a recording time period in which a historical recall time of the historical recall information is located; therefore, the invalid recall condition caused by the large time difference between the history recall time of the obtained history recall information and the current recommendation process when the history recall information is randomly screened from the history recall process is avoided, and the recall effect on the history recall information is improved.

And step 330, providing the recommended recall information to the target object account based on the first recall information set and the second recall information set.

In one possible implementation, the computer device may obtain recommendation information from the recall information set and recommend the recommendation information; the recall information set comprises the current recall information in the first recall information set and the historical recall information in the second recall information set.

When the computer device is implemented as a server, the server may send the recall information set to the terminal so that the terminal determines recommendation information from the recall information set and recommends the recommendation information.

Or, the server may send the recommendation information to the terminal after determining the recommendation information from the recall information set, so that the terminal recommends the received recommendation information.

When the computer device is implemented as a terminal, the terminal determines recommendation information from the recall information set and recommends the recommendation information.

Alternatively, the computer device may recommend the recommendation information in a target recommendation manner. Illustratively, when the computer device is implemented as a server, the server may instruct the terminal to make information recommendation in a target recommendation manner. Optionally, the target recommendation mode may include at least one of an interface display mode or an audio playing mode; the terminal can recommend the recommendation information in a notification bar display mode; or the terminal can recommend the recommendation information in an interface display mode through a display interface in the installed application program; or the terminal can recommend the recommendation information in a popup window display mode; or the terminal can recommend the recommended information in an audio playing mode, and the like, and the recommending mode of the recommended information is not limited by the computer equipment.

In summary, according to the information recommendation method provided by the embodiment of the present application, the first recall information set is obtained based on the recall information of the present time recalled in the recall process of the present time; obtaining a second recall information set based on the relation between the historical scoring result of the historical recall information and the first condition; and determining the recommendation information in the current recommendation process from the information in the first recall information set and the information in the second recall information set, and recommending the recommendation information. Through the scheme, in the recall stage, the historical recall results meeting the conditions in the previous recommendation process are recycled, and the association between different information recommendation processes is realized, so that the utilization effect of recall information is improved, and the information recommendation effect is further improved.

Optionally, the second recall set includes history recommendation information of the target operation performed by the target object, which is not received in history recommendation information recommended by the history recommendation process; or, the second recall set includes at least one of the history recall information recalled in the history recall process, the history scoring result is higher than the first condition, and the history recall information is not recommended to the target object account. That is, in the current recommendation process, the computer device may screen the history recall information that is recalled in the history recall process, where the history scoring result is higher than the first condition, and the history recall information that is not recommended to the target object account is obtained, so as to obtain the second recall information set; and/or screening the historical recommendation information of the target operation which is not received by the target object in the historical recommendation information recommended by the historical recommendation process, so as to obtain a second recall information set.

When the second recall information set includes the history recommendation information of the target operation performed by the target object and is recommended in the history recommendation process, the computer apparatus may previously determine the history recommendation information of the target operation performed by the target object and store the history recommendation information and the history scoring result of the history recommendation information by counting the history recommendation information determined in each history recommendation process to generate a history recommendation information set including the recommended target operation performed by the target object but not received; in the recommending process, the computer equipment can obtain a second recall information set by screening information in the historical recommending information set.

When the second recall information set includes the history recall information recalled in the history recall process, the history scoring result is higher than the first condition, and the history recall information is not recommended to the target object account, because the number of the history recall information in the history recall process is more, in order to obtain the history recall information which is higher in the history scoring result and is not recommended, the computer equipment can screen the history recall information to generate a history information set including the history recall information which is higher in the history scoring result and is not recommended; in the recommending process, the computer equipment can obtain a second recall information set by screening information in the historical information set.

In the embodiment of the application, in the history recall information which is included in the second recall information set and recalled in the history recall process, the history scoring result is higher than the first condition, and the history recall information which is not recommended to the target object account is taken as an example, the information recommendation method provided by the application is described.

Optionally, when the computer device obtains the history recall information in the history information set, the computer device recalls and stores the information with the recording time period as a period. The history information set comprises n history information subsets corresponding to the recording time periods respectively; the history recall time of each history recall information contained in the history information subset is in a recording time period corresponding to the history information subset, and n is a positive integer. Over time, the computer device may generate a new subset of the historical information for a period of time and add the subset of the historical information to the set of historical information to update the set of historical information.

The following describes a process of generating a subset of the history information for each recording period in the history information set according to the embodiment shown in fig. 3; fig. 4 shows a schematic diagram of a process for generating a subset of history information according to an exemplary embodiment of the present application, as shown in fig. 4, the process includes the following steps:

Step 410, when generating a history information subset of each recording period in the n history information subsets, acquiring a history recall information subset corresponding to each of the i recommendation processes executed in the recording period; the history recall information subset comprises target history recall information; the historical scoring result of the target historical recall information is the top k names in the historical recall information subset; i, k are positive integers.

That is, during a recording period, the computer device may perform at least one recommendation process, each comprising a recall phase, a sort phase, and a recommend phase. In the embodiment of the present application, the number of times of the recommended procedure performed in each recording period may be set by the relevant person. Illustratively, the related personnel can control the frequency of information recommendation to the target object account by setting the execution times of the recommendation processes executed in the recording time period or setting the recommendation time of each recommendation process in the recording time period; or, the number of times of the recommended process that can be executed in each recording period may be determined based on the number of times of the recommended operation of the information triggered by the target object, and the number of times of the recommended process that can be executed in each recording period may be different based on the number of times of the recommended operation of the information triggered by the target object; illustratively, the information recommendation operation may include at least one of a start operation of the application by the target object and a refresh operation of the recommended content by the target object.

Taking a first recording time period in n recording time periods as an example, if other recording time periods exist before the first recording time period, in a recall stage of any recommendation process in the first recording time period, the obtained historical recall information set comprises a historical first recall information set and a historical second recall information set; if no other recording time period exists before the first recording time period, in the recall stage of any recommendation process in the first recording time period, the obtained historical recall information set comprises a historical first recall information set.

Taking the example that other recording time periods exist before the first recording time period and the first recall information set includes the history recall information acquired by the explicit recall branch, the implicit recall branch and the functional recall branch respectively, fig. 5 is a schematic diagram illustrating a process of acquiring the target history recall information in accordance with an exemplary embodiment of the present application, and as shown in fig. 5, recall item (information) acquired by the first recall path (including the explicit recall branch, the implicit recall branch and the functional recall branch) and the second recall path (the history recall path) all enter into a sorting stage to form a recall information set 510 (the recall information set 510 can be used as the history recall information set); the computer device ranks the individual recall information in the set of historical recall information 510 during a ranking stage; each recall path may be corresponding to a respective recall number quota, as shown in fig. 5, where the recall number quota corresponding to each recall branch in the first recall path is 4, the recall number quota corresponding to the second recall path (history recall path) is also 4, and the number of recall information in the recall information set is 16; after each recall information in the recall information set 510 is ranked according to the scoring result in the ranking stage, a ranking result can be obtained; based on the quantity information pushing setting of the pushing information, the computer equipment can push recall information of N top ranking (N=2 in fig. 5) of the ranking result indicating scoring result as recommendation information to the target object account, and acquire the information as history recommendation information; in addition, considering that when information pushing is performed, recall information with a higher scoring result needs to be pushed, in order to reduce occupation of storage space and occupation of processing resources, when obtaining target historical recall information, recall information with a ranking result indicating M (m=4 in fig. 5) before ranking of the scoring result may be obtained as target historical recall information 520, M > N obtained in a first recommendation process performed in the first recording period, where M, N are all positive integers.

Step 420, obtaining an un-recommended information set, where the un-recommended information set includes un-recommended information in target history recall information corresponding to each of the i recommendation processes.

And the computer equipment respectively corresponds to the obtained non-recommended information in the history recall information subsets of each recommendation process executed in the first recording time period to form a non-recommended information set. The non-recommended information refers to historical recall information which is not recommended to the target object account in the corresponding recommending process. In the same recommending process, when the history recall information sets are acquired, if the history recall information acquired by different recall branches are overlapped, only one of the overlapped history recall information is reserved; similarly, for the non-recommended information in the target history recall information corresponding to each recommending process, if the non-recommended information is overlapped, only one of the overlapped non-recommended information is reserved. Fig. 6 is a schematic diagram illustrating acquiring an un-recommended information set according to an exemplary embodiment of the present application, as shown in fig. 6, taking a case that 12 recommendation processes are performed in a first recording period, recording target history recall information corresponding to each of the 12 recommendation processes, and if there is coincidence between the target history recall information in the 12 recommendation processes, only one of the target history recall information is retained, that is, in the 12 recommendation processes, the target history recall information is item1 to item16; the historical recall information recommended to the target object account is as follows: item1, item2, item 5-8, item 11-12, item 15-16; the historical recall information not recommended to the target object account is: item3, item4, item9, item10, item13, item14; that is, the non-recommended information included in the non-recommended information set 610 corresponding to the first recording period is: item3, item4, item9, item10, item13, item14.

Step 430, generating a historical information subset corresponding to the recording time period based on the non-recommended information set.

In one possible implementation, the set of non-recommended information corresponding to the recording time period may be obtained as a subset of history information corresponding to the recording time period.

Alternatively, in another possible implementation, to avoid wasting storage resources and subsequent data processing resources of the computer by excessive amounts of history recall information recalled via the history recall path, the computer device is provided with a first recall threshold to limit a maximum amount of history recall information that can be included in each subset of history information when generating the subset of history information corresponding to each recording time period. In this case, in response to the number of non-recommended information in the set of non-recommended information being greater than the first recall threshold, the computer device ranks based on the historical scoring results for each of the non-recommended information, obtains a first ranking result, and obtains the number of non-recommended information equal to the first recall threshold based on the first ranking result, forming a subset of the historical information corresponding to the recording period.

That is, when the computer device stores the target historical recall information, the historical scoring result of the target historical recall information is also stored, so that when the historical information subset is obtained, the un-recommended information with higher historical scoring result can be selected based on the historical scoring result corresponding to each un-recommended information, and the historical information subset corresponding to the recording time period is formed. Taking the first recording period as an example, as shown in fig. 6, if the first recall threshold is 4, the first 4 non-recommended information with higher historical scoring results may be obtained: item3, item4, item10, item14, constitute a subset 620 of history information corresponding to the first recording period.

It should be noted that, the sorting manner of sorting based on the historical scoring results of each un-recommended information may be ascending sorting or descending sorting; when the sorting mode is ascending sorting, acquiring un-recommended information of the first recall threshold number of the first sorting result reciprocal to form a history information subset; and when the sorting mode is descending sorting, acquiring the first recall threshold number of un-recommended information of the first sorting result order number so as to form a history information subset.

If the number of recalls of the same non-recommended information is greater than one in at least one recommendation process in the first recording time period, the non-recommended information can obtain a plurality of scoring results after entering the sequencing stage for a plurality of times; in this case, the computer device may acquire a maximum value, or an average value, of a plurality of scoring results of the non-recommended information in the first recording period as a history scoring result of the non-recommended information in the first recording period.

Based on the description of the generation process of each subset of the history information in the history information set in the embodiment shown in fig. 4, fig. 7 shows a schematic diagram of the generation process of the history information set according to an exemplary embodiment of the present application, as shown in fig. 7, for n recording periods in the history information set, the recommendation process is performed i times in each recording period. Taking the generation process of the history subset of the recording period 1 among the n recording periods as an example: in the i recommendation processes of recording time period 1, each recommendation process corresponds to a historical recall information subset 710; the computer device obtains the history recall information for top M in each history recall information subset 710 as target history recall information 720; then, un-recommended information 730 is obtained from the target history recall information 720 corresponding to each recommending process, and un-recommended information in the i recommending processes is integrated to obtain an un-recommended information set 740; after the computer device obtains the non-recommended information, the computer device may filter the information in the non-recommended information set 740 based on the set first recall threshold used to limit the number of the history recall information in the history information subset, and obtain the set composed of the filtered non-recommended information as the history information subset of the recording period 1.

The generation process of the history information subset of the other recording period may refer to the generation process of the history information subset of the recording period 1, and will not be described herein.

After the computer device obtains the history information subsets corresponding to the n recording periods, a set formed by the n history information subsets is obtained as a history information set 750.

The subset of history information in the set of history information is updated over time. In the history information set, the probability that the history information subset with a longer time interval from the current time interval is recommended is lower, so in order to reduce the occupation of storage resources of the computer device, in one possible implementation manner, the computer device may be provided with a history information subset threshold value, so as to limit the maximum number of the history information subsets which can be stored in the history information set; optionally, the historical information subsets are stored in the historical information set in a sequence of the recording time periods from far to near, and in response to the number of the historical information subsets stored in the historical information set being equal to the historical information subset threshold and there being a new historical information subset of the recording time period to be added, the computer device removes the historical information subset of the recording time period furthest, and adds the new historical information subset of the recording time period to the historical information set.

It should be noted that the process of acquiring the history information set may be a preprocessing process, and may be performed between two information recommendation processes; or in the execution process of the primary information recommendation process before the current recommendation process, so that the acquisition process of the historical information set is not required to be executed in the current recommendation process, and the recall efficiency of recall information is improved. Alternatively, the above history information collection process may be executed in the current information recommendation process, which is not limited in this application.

After the historical information set is obtained, fig. 8 shows a flowchart of an information recommendation method according to an exemplary embodiment of the present application, where the information recommendation method may be performed by a computer device, which may be implemented as a terminal or a server, and schematically, may be implemented as the terminal 240 or the server 220 shown in fig. 1. As shown in fig. 8, the information recommendation method may include the steps of:

in the recall stage of the current recommendation process:

step 810, obtaining a first recall information set based on the recall information of the present recall in the recall process.

In one possible implementation, the first recall information set includes at least one of a first sub-information set, a second sub-information set, and a third sub-information set; the first sub-information set is an information set composed of the current recall information obtained by screening from the candidate recall information based on keywords determined by historical behavior data of the target object account; the second sub-information set is an information set composed of the recall information obtained by screening from the candidate recall information, and the semantic vector is determined based on the historical behavior data of the target object account; the third sub-information set is an information set composed of the current recall information obtained by screening from the candidate recall information based on the information attribute.

The manner in which information recall is performed for keywords determined based on historical behavioral data of the target object account may be referred to as explicit recall.

Optionally, the computer device may determine the object features based on historical behavioral data of the target object account; generating a keyword based on the object features, the keyword operable to explicitly describe the object features of the target object account; illustratively, the object characteristics may include various contents of a region, a sex, an age, an interest of the target object, a model of a terminal used, a purchasing power, and the like. In explicit recall, the recall tag is a keyword determined based on historical behavioral data of the target object account.

Alternatively, the object feature may be determined based on a selection operation of the object tag by the target object, and the object tag may include an object classification and an object interest. Illustratively, when the target object opens the application program for the first time, the terminal may provide a selectable object tag for the target object, and determine an object feature of a target object account corresponding to the target object based on a selection operation of the target object on the object tag by the target object; for example, when the target object first opens the video application, a plurality of object tags may be provided to the target object, and if the target object selects a tag such as "cartoon", "cartoon" from the plurality of object tags, the tag content of the object tag selected by the target object may be obtained as the object feature, and the keyword determined based on the object feature may be "second order element". Alternatively, the object features may be statistically determined based on historical operational records of the target object; illustratively, in the video recommendation application program, based on a series of historical behavior data such as a historical browsing record, a historical clicking record, a historical searching record and the like of the target object, the frequency of video of browsing, clicking and searching financial classes of the target object can be statistically determined to be higher, then the object characteristics can be determined to be preferential financial class content, and the keywords determined based on the object characteristics can be financial. The two types of determining object features may be used separately or in combination to determine the target object account keyword, which is not limited in this application.

Because the historical behavior data of the target object account can change along with time or along with the change of the interest points of the target object, along with the change of the historical behavior data of the target object account, the recall information acquired by an explicit recall mode can also change. In order to ensure the instantaneity of the recall information, optionally, the historical behavior data can be target object behavior data with a specified duration threshold from the current time point; illustratively, the historical behavior data may be behavior data of the target object within one week from the current point in time, or within three days; the specified duration threshold may be set by the relevant personnel, as this application is not limited.

In one possible implementation, the process of recalling candidate recall information by way of explicit recall may be implemented by a machine learning model. The embodiment of the application provides an explicit recall model which is used for determining the matching degree between candidate recall information and a target object account according to the keyword of the target object account and the candidate recall information so as to determine whether to recall the candidate recall information or not based on the matching degree; the process may be implemented as: inputting the keywords of the target object account and the candidate recall information into an explicit recall model to obtain the matching degree output by the explicit recall model; the explicit recall model may be obtained based on training of a first sample set, where the first sample set may include keywords of a sample object account, sample candidate recall information, and a matching degree tag corresponding to the sample candidate recall information; the sample candidate recall information comprises a positive sample and a negative sample, wherein the positive sample refers to sample candidate recall information matched with keywords of a sample object account, and the negative sample refers to sample candidate recall information not matched with keywords of the sample object account.

The manner in which information recall is performed on the basis of the semantic vector determined by the historical behavior data of the target object account may be referred to as implicit recall; during the implicit recall process, the computer device cannot explicitly describe the point of interest of the target object.

Implicit recall may be achieved by both U2I (User to Item) and I2I (Item to Item); the U2I is a method for recall by calculating the similarity between a user and item in a recommendation system; I2I is a class of methods in the recommendation system for recall by computing the similarity between items. In the implicit recall, the recall tag is a semantic vector determined based on historical behavior data of the target object account; for U2I, the recall tag is a semantic vector of the target object account; for I2I, the recall tag is a semantic vector of matching information. The matching information may include, among other things, historical recall information, historical information of the received target operation, and so forth.

Implicit recall typically utilizes a machine learning algorithm for information transformation and matching; taking U2I as an example, the computer equipment converts the target object and item into semantic vectors respectively through a machine learning algorithm, then calculates the vector similarity between the semantic vectors corresponding to the target object and item, and recalls information by using the vector similarity. Illustratively, assuming that the semantic vector of the target object 1 converted by the computer device through the machine learning algorithm is (1, 1), the semantic vector of item1 is (1,0.9,1), and the semantic vector of item2 is (1, 2), then the vector similarity between the semantic vector of the target object 1 and the semantic vector of item1 can be determined through vector similarity calculation, which is higher than the vector similarity between the semantic vector of the target object 1 and the semantic vector of item2, so that item1 is recalled among item1 and item2 is determined; the vector similarity may be calculated by one of cosine similarity (cos similarity) formula, dot product similarity, euclidean similarity, and the like, which is not limited in this application.

Optionally, the embodiment of the application provides an implicit recall model to implement vector transformation of the transformation object and determine vector similarity between different transformation objects; for U2I, the conversion object is a target object account number and candidate recall information respectively; for I2I, the conversion objects are matching information and candidate recall information, respectively. The implicit recall model may be obtained based on training of a second sample set, which may include a sample object account number, sample candidate recall information, and a vector similarity tag corresponding to the sample candidate recall information.

Because the update frequency of the candidate recall information in the recommendation system is higher, and the target object account number, the object characteristics of the same target object account number and the like are changed along with the change of time, in order to ensure the accuracy of information recall of the recall model (comprising an explicit recall model and an implicit recall model), the recall model can be updated according to the target training period.

Illustratively, the sample candidate recall information in each sample set (including the first sample set and the second sample set) may be historical candidate recall information, the sample object account in each sample set may be a historical recommendation object account corresponding to the historical candidate recall information, and the matching degree tag in the first sample set or the vector similarity tag in the second sample set may be used to indicate whether the historical candidate recall information is acquired as the historical recall information under the historical recommendation object account; for example, taking the target training frequency as 1 day as an example, the updating process of the explicit recall model may be to determine a first sample set based on the candidate recall information today, the keyword of the recommended object account, and the recall information, and train and update the explicit recall model through the first sample set today, so as to predict the matching degree of the candidate recall information acquired in the tomorrow and the keyword of the recommended object account through the updated explicit recall model, thereby determining the recall information acquired in the tomorrow through the explicit recall.

The manner in which information recall is performed based on information attributes may be referred to as functional recall; the function recall may include a hotspot recall, a plumb recall, and other recall modes; the hot spot recall means that information recall is carried out according to the accumulated result (information attribute) of the target operation received by each candidate recall information within a certain time period so as to recall the current hot spot information; illustratively, the hot spot information can be obtained from the candidate information according to the CTR (Click-Through-Rate) of the candidate information, and the hot spot information is recalled; the vertical recall refers to a manner of recalling information based on the information type (information attribute) to which each candidate recall information belongs, so as to recall information in a certain field. The target object can be enabled to receive the latest information content or the information content in a certain field at a probability through function recall; at this time, the recall tag is the information attribute.

In one possible implementation manner, in order to reduce the information processing pressure of the computer device in the sorting stage, the computer device may limit the number of pieces of current recall information that can be included in the first recall information set, that is, the first recall information set includes a first number of pieces of current recall information; responding to the fact that the first recall information set comprises a first sub information set, a second recall information set and a third sub information set, and the number of the recall information contained in each sub information set can be respectively set so as to limit the number of the recall information contained in the first recall information set; the number of the recall information in each sub-information set may be the same or different, but the sum of the number of the recall information in the sub-information set is smaller than or equal to the first number.

Step 820, reading the history information subsets corresponding to the n recording time periods in the history information set; the historical recall time of the historical recall information in the historical information subset is in a historical time period corresponding to the historical information subset; the historical scoring result of the historical recall information in the subset of historical information is higher than the first condition and is not recommended to the target object account.

The process of obtaining the history information set by the computer device may refer to the relevant content of the embodiment corresponding to fig. 4 or the embodiment corresponding to fig. 7, which will not be described herein.

Step 830, a second set of recall information is obtained from the n subsets of history information.

Optionally, the process of obtaining the second recall information set from the n historical information subsets may be implemented as steps S8301 to S8303:

s8301, obtaining the recommendation time corresponding to the current recommendation process.

The recommendation time may be the time when the target object triggers the current recommendation process, or may be the time when the related personnel preset the information recommendation to the target object actively.

S8302, acquiring m target historical information subsets from the n historical information subsets based on the recommended time; m is less than or equal to n, and m is a positive integer.

In one possible implementation manner, in a recording time period corresponding to a recommendation time of the current recommendation process, the target historical information subsets corresponding to the recommendation processes are the same or the same group of historical information subsets; for example, the m target historical information subsets corresponding to each recommendation process are all the historical information subsets (m=1) of the previous recording time period of the recording time; or, the m target historical information subsets corresponding to the recommendation processes are all a group of historical information subsets (m > 1) formed by the historical information subsets of m recording times before the recording time period. Illustratively, the recording time period is in days, and then the target historical information subset of the one-time recommendation process performed at any moment today is the historical information subset (m=1) acquired yesterday; or, a set of history information subsets consisting of daily history information subsets within 3 (m=3) days before today.

In another possible implementation manner, to fully apply each historical information subset in the historical information set, a correspondence between a time interval where the recommended time is located and a recording time period of the historical information subset may be set up, and the process of obtaining the target historical information subset may be implemented as follows:

Acquiring a time interval based on the time interval of the recommended moment in the recording time period in the current recommendation process;

acquiring a recording time period which is in front of the recording time period in which the current recommending process is located and is different from the recording time period in which the current recommending process is located by the time interval as a target time period;

Taking the example that the recommended time is set by a relevant person, the relevant person may set the recommended time to each time point, that is, the computer device performs information push at each time point of day. Each time-alignment point may be provided with a corresponding one (m=1) of the target history information subsets; for example, if the recording time period is set in units of days, the correspondence between the time interval in which the recommended time is located and the recording time period may be set as follows: when the recommended time is between 0 point and 2 points, the time interval between the target time period and the current recording time period is 8 days; when the recommended time is 3 to 5 points, the time interval between the target time period and the current recording time period is 7 days; and when the recommended time is between 6 and 8 points, the time interval between the target time period and the current recording time period is 6 days, and the like, until the recommended time is between 21 and 23 points, the time interval between the target time period and the current recording time period is 1 day.

Taking the recommending time as the time point of the information recommending operation triggered by the target object, the source of the target historical information subset can be determined based on the time interval in which the recommending time is located, fig. 9 is a schematic diagram showing the correspondence between the time interval and the source of the target historical information subset, which is shown in an exemplary embodiment of the present application, as shown in fig. 9, in the process of obtaining the target historical information subset, taking a recording time period corresponding to each time interval (m=1) as an example, when the recommending time is in the time interval of 0-3 points, the time interval between the target time period and the current recording time period is 8 days; when the recommended time is within the time interval of 3-6 points, the time interval between the target time period and the current recording time period is 7 days; when the recommended time is within the time interval of 6 to 9 points, the time interval between the target time period and the current recording time period is 6 days, and so on.

Alternatively, in one possible implementation, a correspondence between the time interval and a plurality of (m > 1) recording time periods may also be set; illustratively, when the recommended time is within a time interval of 0 to 3 points, the time interval between the recording time period in the target time period and the current recording time period is 16 days and 15 days, respectively; when the recommended time is within the time interval of 3 to 6 points, the time interval between the recording time period in the target time period and the current recording time period is 14 days and 13 days, respectively, and so on.

It should be noted that, the corresponding relationship between the time interval in which the recommended time is located and the target time period may be set by related personnel, that is, the recommended time is set, the sources of the target history subsets corresponding to different time intervals respectively, and the number of the target history subsets corresponding to different time intervals respectively may be changed based on the actual requirement.

S8303, acquiring a set formed by the history recall information in the m target history subsets as a second recall information set.

Optionally, when m=1, the target history subset is acquired as the second recall information set.

When m is more than 1, comprehensively acquiring a set formed by the history recall information of which the history scoring result is ranked in the top x names from m target history subsets, wherein the set is a second recall information set; wherein the value of x is equal to the maximum number of the historical recall information which can be contained in the second recall information set, and x is a positive integer.

Optionally, since the recall model is updated according to the target training period, the target history subset is obtained according to the recording time period; responding to the fact that the target training period of the recall model is the same as the period of the record time period, acquiring a set consisting of history recall information in m target history subsets as a second recall information set;

Or, in response to the difference between the target training period of the recall model and the period of the recording time period, after m target history subsets are obtained, the m target history subsets may be further screened by the target training period of the recall model, so as to screen and obtain the second recall information set from the target history subsets corresponding to the recording time period before the last update of the recall model. The process may be as follows:

acquiring a set formed by history recall information in the target history subsets before the last recall model updating time in the m target history subsets as a second recall information set; illustratively, assuming that the period of the recording period is 1 day and the target training period of the recall model is 2 days, after the target history subsets corresponding to the 3 recording periods are obtained, if the 3 rd recording period (yesterday) and the current recording period (today) belong to the same target training period, that is, the recall model used by yesterday and today is the same, the difference between the information recall results of yesterday and today may be smaller, so that the target history subset of the 3 rd recording period may be excluded, and a set composed of history recall information in the target history subsets of the remaining 2 recording periods may be obtained as the second recall information set.

In the scoring stage of the recommendation process:

step 840, obtaining the current scoring result of each recall information in the recall information set; the recall information set includes a first recall information set and a second recall information set.

In the embodiment of the application, the recall information of the present time in the first recall information set and the historical recall information in the second recall information set are jointly obtained to form a recall result of a recall stage of the present recommendation process, so that a recall information set is formed; and determining the recommendation information in the current round of recommendation process by acquiring the current scoring result of each recall information in the recall information set.

It should be noted that, the scoring result of each recall information is related to the recommended object account, and the corresponding scoring result may be different for different recommended object accounts for the same recall information; that is, the feedback of the same recall information by different object accounts is different. Therefore, when each recall information in the recall information set is scored, the recommendation object account can be combined for scoring to determine the recommendation information of the current recommendation object account (target object account), so that when the scoring result of the time of each recall information is obtained, the object characteristics of the target object account can be obtained first; and respectively acquiring the current scoring result of each recall information based on the object characteristics of the target object account.

In order to improve accuracy of the obtained scoring result of each recall information, in the embodiment of the application, the computer device may obtain the scoring result of each recall information through a machine learning method. Illustratively, the process of obtaining the scoring result of this time of each recall information may be implemented as follows: inputting the object characteristics of the target object account and the target recall information into a scoring model to obtain the scoring result of the target recall information output by the scoring model; the target recall information is any one of the recall information;

The scoring model may be at least one of an LR (Logistic Regression ) model, an FM (Factor Machine) model, and a DNN (Deep Neural Network ) model, among others.

In the model training process, sample object characteristics and sample information of a sample object account are input into a scoring model, so that a prediction scoring result of the sample information output by the scoring model relative to the sample object account is obtained; training a scoring model based on a scoring label corresponding to the prediction scoring result and the sample information; and repeating the process based on the combination of the sample object characteristics of different sample information and different sample object accounts, and performing iterative training on the scoring model until the training completion condition is reached, so as to obtain a trained scoring model, and predicting the scoring result of the target recall information relative to the target object account through the trained scoring model.

In the above process, the function value of the loss function may be calculated based on the prediction scoring result of the sample information relative to the sample object account and the scoring label of the sample information relative to the sample object account, so as to train the scoring model based on the function value of the loss function.

The training completion condition may include that the scoring accuracy of the trained scoring model reaches an accuracy threshold, the trained scoring model converges, or the number of iterations reaches a number of times threshold.

In one possible implementation manner, to improve the training effect on the scoring model, the historical information recommendation result may be obtained as a training sample set; that is, the sample information in the training sample set is history recommendation information; the sample object features are object features of a recommended object account corresponding to the historical recommendation information when the historical recommendation information is recommended; the scoring label of the sample information relative to the sample object account is a recommendation result of the historical recommendation information relative to the recommendation object account;

the recommendation result comprises one of the recommended object executing target operation and the recommended object not executing target operation; for example, if the recommendation result is that the recommendation object performs the target operation, the score label corresponding to the sample information may be 100, and if the recommendation result is that the recommendation object does not perform the target operation, the score label corresponding to the sample information is 0; the recommended object is an object corresponding to the recommended object account.

Step 850, determining recommendation information based on the current scoring result of each recall information in the recall information set.

In one possible implementation manner, when determining the recommended information based on the current scoring result of each recall information in the recall information set, the recall information in the recall information set may be ranked based on the current scoring result of each recall information in the recall information set, to obtain a second ranking result; based on the second ranking result, recommendation information is determined.

The method comprises the steps of carrying out ascending sort through recall information in a recall information set, or carrying out descending sort to obtain a second sort result; responding to the second sorting result, which is obtained by arranging all recall information in the recall information set in a descending order based on the scoring result, and acquiring N recall information with the top ranking indicated by the second sorting result as recommended information, wherein N is a positive integer; and meanwhile, acquiring the recall information of M top names of the second ranking result indication rank as target historical recall information in the current recommendation process, storing the target historical recall information in the current recommendation process and the scoring result of the target historical recall information, and generating a historical information subset of the current recording time period by combining the target historical recall information respectively acquired by other recommendation processes in the current recording time period so as to acquire a second recall information set by a historical recall path of the subsequent recommendation time.

In the recommendation stage of the current recommendation process:

step 860 provides the recommendation information to the target object account.

The computer equipment acquires recall information of N top ranking names indicated by the second sorting result as recommendation information, and recommends the recall information. Illustratively, when the computer device is a server, the computer device sends the determined recommendation information to the terminal, so that the terminal recommends the recommendation information through a self picture display function or a sound playing function.

When the history recall information is acquired, the second information set is acquired from the history information set based on the history recall time of the information, so that the history recall time of the acquired history recall information has a certain relevance in time with the current recommendation process, and the acquisition effect of the history recall information is improved.

When information recommendation is carried out, the corresponding relations of the time intervals between different time intervals and the target time periods for acquiring the historical information subsets are respectively set, so that each historical information subset can be utilized at the corresponding moment, and the information utilization rate of the historical recall information is improved.

In one possible implementation, when acquiring the recommendation information, the recommendation information is acquired from the first recall information set and the second recall information set respectively. In this case, when the recall information in the recall information set is ordered in the ordering stage, the current scoring result of the current recall information in the first recall information set can be obtained; sorting the recall information based on the scoring result of the recall information to obtain a third sorting result; determining the recall information of the first recommendation threshold value as recommendation information based on the third sequencing result; in the sorting stage, sorting the historical recall information based on the historical scoring result of the historical recall information in the second recall information set to obtain a fourth sorting result; and determining the historical recall information of the second recommendation threshold as recommendation information based on the fourth sorting result. The values of the first recommendation threshold and the second recommendation threshold may be the same or different, and the first recommendation threshold and the second recommendation threshold may be set by related personnel according to actual requirements. Illustratively, assuming that the value of the first recommendation threshold and the value of the second recommendation threshold are both 2, after the first recall information set and the second recall information set are obtained, in the sorting stage, the first recall information set, the first 2 pieces of current recall information with higher current scoring results, and the second recall information set, the first 2 pieces of history recall information with higher history scoring results, are jointly obtained to be the recommendation information of the current recommendation process, and the recommendation information is provided for the target account.

Fig. 10 is a schematic architecture diagram of an information recommendation system according to an exemplary embodiment of the present application, so as to implement the information recommendation method provided in the present application, and as shown in fig. 10, a recommendation process of the information recommendation system may be divided into a recall stage 1010, a sort stage 1020 and a recommendation stage 1030. In order to realize the information recommendation method provided by the application and multiplexing the history recall information, in the embodiment of the application, a plurality of components for counting and screening the history recall information are added in an information recommendation system, and a new recall path (history recall path) is used as a second recall path 1040, and together with an explicit recall branch, an implicit recall branch and a first recall path 1050 formed by a functional recall branch, information recall in a recall stage is realized; the plurality of components includes a history recall result recording component 1041, a history recall result statistics component 1042, and a history recall component 1043.

The history recall result recording component 1041 is configured to obtain and store a top recommendation result reported by the online recommendation system, and in combination with the embodiment shown in fig. 3, the history recall result recording component is configured to store recall information (recommendation information) of a top N and recall information of a top M; wherein M > N. The history recall result statistics component 1042 is used for carrying out statistics summary according to the period of the recording time period to obtain and store a history recall subset of each recording time period; that is, the history recall result statistics component 1042 is configured to count the un-recommended information in each recommendation process performed in each recording period, so as to generate the history recall subset corresponding to each recording period based on the set of un-recommended information of the recording period. The history recall component 1043 is configured to multiplex history recall information by re-pulling the stored subset of history recalls to a sorting stage in a new recommendation process to be used as a new recall.

The top recommendation result reported by the online recommendation system and obtained by the history recall result recording component 1041, and the history recall subset of each recording time period obtained by the history recall result statistics component 1042 may be stored in an offline storage system; alternatively, to ensure the security of the information, the information may be stored in the blockchain.

In one possible implementation manner, in order to ensure the safety and accuracy of the object information and the recommendation information of the recommendation object in the information recommendation process, historical behavior data, object feature data, historical recall information data and the like of the recommendation object involved in the information recommendation method disclosed by the application can be stored on a blockchain.

It will be appreciated that in the specific embodiments of the present application, related data such as object characteristics, historical behavioral data of recommended objects, etc. are relevant data, and when the above embodiments of the present application are applied to specific products or technologies, user permissions or consents need to be obtained, and the collection, use and processing of relevant data is required to comply with relevant laws and regulations and standards of relevant countries and regions.

According to the method, the historical recall path is added in the recommendation system, for example, 200G redis storage resources and a small amount of spark computing resources can be used on the premise that fewer storage resources are used, and therefore information recommendation effect and information recommendation efficiency are improved.

The information recommendation method provided by the embodiment of the application can be applied to any scene needing information recommendation, and the following two possible information recommendation scenes are provided by way of example:

1. a scene that an application program installed in a terminal actively pushes information to a target object;

in some possible cases, an application installed in the terminal needs to actively push information to a target object, for example, on the premise that the terminal does not open the application, the application pushes recommendation information to the target object at a predetermined time point through a popup window, a notification bar, and the like, so that the target object can acquire information content according to the recommendation information. In the process, at a preset time point, an application program initiates an information push instruction to a server of the application program through a terminal. After receiving the information pushing instruction, the server of the application program carries out information recall from two information sources of the candidate recall information and the history recall information through the information pushing method provided by the application program, sorts the recalled information to obtain pushing information, and returns the pushing information to the application program in the terminal. After the application program obtains the recommendation information, the application program pushes the push information to the target object through a notification bar or a popup window in the terminal.

2. A use scene of recommended functions such as "guess you like" provided in the application program;

in an application program, an information recommendation function is generally provided, for example, in a video application program, when an operation of opening the application program by a target object is received, or when a refresh operation of a recommendation page of the application program by the target object is received, the application program may recommend information to the target object based on historical behavior data of the target object, or real-time hot spot data, and the like. In the process, when the target object opens the application program or when the target object refreshes the page of the application program, the application program can send an information pushing instruction to a server of the application program, so as to obtain recommendation information obtained by the server of the application program based on the information pushing method provided by the application program, and recommend the recommendation information on a display interface of the application program.

Fig. 11 shows a block diagram of an information recommendation device according to an exemplary embodiment of the present application, and as shown in fig. 11, the information recommendation device includes:

a first set obtaining module 1110, configured to obtain a first set of recall information based on the current recall information recalled in the current recall process;

A second set obtaining module 1120, configured to screen, in the history recall information recalled in the history recall process, the history recall information having a history score result higher than that of the first condition, to obtain a second recall information set; the historical scoring result is used for indicating the probability of the target object account executing target operation on the historical recall information at the historical recall moment;

and the information recommending module 1130 is configured to provide the current recommended recall information to the target object account based on the first recall information set and the second recall information set.

In a possible implementation manner, the second set obtaining module 1120 is configured to filter, from the history recall information recalled in the history recall process, the history recall information that is higher than the first condition and is not recommended to the target object account, so as to obtain the second recall information set.

In one possible implementation, the second set acquisition module 1120 includes:

In one possible implementation manner, the set acquisition sub-module includes:

In one possible implementation, the subset acquisition submodule includes:

In a possible implementation manner, the second set obtaining module 1120 is configured to filter, from the history recommendation information recommended by the history recall process, the history recommendation information of the target operation that is not received by the target object to obtain the second recall information set; the target object is an object corresponding to the target object account.

In one possible implementation, the information recommendation module 1130 includes:

In one possible implementation, the score acquisition sub-module includes:

In one possible implementation manner, the first recall information set includes at least one of a first sub-information set, a second sub-information set and a third sub-information set; the first sub-information set is an information set composed of recall information obtained by screening from the candidate recall information, and the first sub-information set is a keyword determined based on historical behavior data of a recommended object account; the second sub-information set is an information set composed of recall information obtained by screening from the candidate recall information, and the semantic vector is determined based on historical behavior data of the recommended object account; the third sub-information set is an information set composed of recall information obtained by screening from the candidate recall information based on information attributes.

In summary, the information recommending apparatus provided in the embodiment of the present application obtains the first recall information set based on the recall information of the present time recalled in the recall process of the present time; obtaining a second recall information set based on the relation between the historical scoring result of the historical recall information and the first condition; and determining the recommendation information in the current recommendation process from the information in the first recall information set and the information in the second recall information set, and recommending the recommendation information. Through the scheme, in the recall stage, the historical recall results meeting the conditions in the previous recommendation process are recycled, and the association between different information recommendation processes is realized, so that the utilization effect of recall information is improved, and the information recommendation effect is further improved.

Fig. 12 shows a block diagram of a computer device 1200 shown in an exemplary embodiment of the present application. The computer device may be implemented as a server in the above-described aspects of the present application. The computer apparatus 1200 includes a central processing unit (Central Processing Unit, CPU) 1201, a system Memory 1204 including a random access Memory (Random Access Memory, RAM) 1202 and a Read-Only Memory (ROM) 1203, and a system bus 1205 connecting the system Memory 1204 and the central processing unit 1201. The computer device 1200 also includes a mass storage device 1206 for storing an operating system 1209, application programs 1210, and other program modules 1211.

The mass storage device 1206 is connected to the central processing unit 1201 through a mass storage controller (not shown) connected to the system bus 1205. The mass storage device 1206 and its associated computer-readable media provide non-volatile storage for the computer device 1200. That is, the mass storage device 1206 may include a computer readable medium (not shown) such as a hard disk or a compact disk-Only (CD-ROM) drive.

The computer readable medium may include computer storage media and communication media without loss of generality. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, erasable programmable read-Only register (Erasable Programmable Read Only Memory, EPROM), electrically erasable programmable read-Only Memory (EEPROM) flash Memory or other solid state Memory technology, CD-ROM, digital versatile disks (Digital Versatile Disc, DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage medium is not limited to the one described above. The system memory 1204 and mass storage device 1206 described above may be collectively referred to as memory.

According to various embodiments of the disclosure, the computer device 1200 may also operate through a network, such as the Internet, to a remote computer on the network. I.e., the computer device 1200 may be connected to the network 1208 via a network interface unit 1207 coupled to the system bus 1205, or alternatively, the network interface unit 1207 may be used to connect to other types of networks or remote computer systems (not shown).

The memory further includes at least one instruction, at least one program, a code set, or an instruction set, where the at least one instruction, the at least one program, the code set, or the instruction set is stored in the memory, and the central processor 1201 implements all or part of the steps in the information recommendation method shown in the foregoing embodiments by executing the at least one instruction, the at least one program, the code set, or the instruction set.

Fig. 13 illustrates a block diagram of a computer device 1300, according to an exemplary embodiment of the present application. The computer device 1300 may be implemented as the terminal described above, such as: smart phones, tablet computers, notebook computers or desktop computers. The computer device 1300 may also be referred to by other names of terminal devices, portable terminals, laptop terminals, desktop terminals, and the like.

In general, the computer device 1300 includes: a processor 1301, and a memory 1302.

In some embodiments, the computer device 1300 may further optionally include: a peripheral interface 1303 and at least one peripheral. The processor 1301, the memory 1302, and the peripheral interface 1303 may be connected by a bus or signal lines. The respective peripheral devices may be connected to the peripheral device interface 1303 through a bus, a signal line, or a circuit board. Specifically, the peripheral device includes: at least one of radio frequency circuitry 1304, a display screen 1305, a camera assembly 1306, audio circuitry 1307, and a power supply 1308.

In some embodiments, computer device 1300 also includes one or more sensors 1309. The one or more sensors 1309 include, but are not limited to: acceleration sensor 1310, gyroscope sensor 1311, pressure sensor 1312, optical sensor 1313, and proximity sensor 1314.

Those skilled in the art will appreciate that the architecture shown in fig. 13 is not limiting as to the computer device 1300, and may include more or fewer components than shown, or may combine certain components, or employ a different arrangement of components.

In an exemplary embodiment, a computer readable storage medium is also provided for storing at least one computer program loaded and executed by a processor to implement all or part of the steps in the above information push method. For example, the computer readable storage medium may be Read-Only Memory (ROM), random-access Memory (Random Access Memory, RAM), compact disc Read-Only Memory (CD-ROM), magnetic tape, floppy disk, optical data storage device, and the like.

In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product comprising at least one computer program loaded into a processor and executed all or part of the steps of the method shown in any of the embodiments of fig. 3, 4 or 8 described above.

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

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

Claims

1. An information recommendation method, the method comprising:

2. The method according to claim 1, wherein in the history recall information recalled in the history recall process, screening the history recall information with a history score higher than that of the first condition to obtain a second recall information set includes:

and screening the historical recall information which is recalled in the historical recall process and is not recommended to the target object account, wherein the historical scoring result is higher than the first condition, so as to obtain the second recall information set.

3. The method according to claim 2, wherein the screening the historical recall information that is not recommended to the target object account and has the historical scoring result higher than the first condition from the historical recall information recalled in the historical recall process to obtain the second recall information set includes:

reading the history information subsets corresponding to the n recording time periods in the history information set; the historical recall times of the historical recall information in the subset of historical information are within a historical time period corresponding to the subset of historical information; the historical scoring results of the historical recall information in the subset of historical information are higher than the first condition and are not recommended to the target object account; n is a positive integer;

And acquiring the second recall information set from the n historical information subsets.

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

when generating the history information subsets of each recording time period in n history information subsets, acquiring history recall information subsets corresponding to i recommendation processes executed in the recording time period; the history recall information subset comprises target history recall information; the historical scoring result of the target historical recall information is in the top k names in the historical recall information subset; i, k is a positive integer;

acquiring an unreferenced information set, wherein the unreferenced information set comprises unreferenced information in the target history recall information corresponding to each i recommendation process;

and generating the historical information subset corresponding to the recording time period based on the non-recommended information set.

5. The method of claim 4, wherein the generating the subset of history information corresponding to the recording time period based on the set of non-recommended information comprises:

responsive to the number of the non-recommended information in the non-recommended information set being greater than a first recall threshold, ranking based on the historical scoring results for each of the non-recommended information to obtain a first ranking result;

6. The method of claim 3, wherein said obtaining said second set of recall information from said n subsets of history information comprises:

acquiring a recommendation time corresponding to the recommendation process;

acquiring m target historical information subsets from n historical information subsets based on the recommended time; m is less than or equal to n, and m is a positive integer;

and acquiring a set consisting of the history recall information in the m target history subsets as the second recall information set.

7. The method of claim 6, wherein the obtaining m target historical information subsets from n historical information subsets based on the recommended time comprises:

8. The method according to claim 1, wherein in the history recall information recalled in the history recall process, screening the history recall information having a history score higher than the first condition to obtain a second recall information set includes:

screening historical recommendation information of the target operation which is not received by the target object from the historical recommendation information recommended by the historical recommendation process, and obtaining the second recall information set; the target object is an object corresponding to the target object account.

9. The method according to any one of claims 1 to 8, wherein the providing the current recommended recall information to the target object account based on the first set of recall information and the second set of recall information includes:

obtaining the scoring result of the time of each recall information in the recall information set; the recall information set comprises the first recall information set and the second recall information set;

determining recommendation information based on the current scoring result of each recall information in the recall information set;

And providing the recommendation information for the target object account.

10. The method of claim 9, wherein the determining recommendation information based on the current scoring results for each recall information in the set of recall information comprises:

based on the scoring result of the current time of each recall information in the recall information set, sequencing the recall information in the recall information set to obtain a second sequencing result;

and determining the recommendation information based on the second sorting result.

11. The method of claim 9, wherein the obtaining the current scoring result for each recall information in the collection of recall information comprises:

acquiring object characteristics of the target object account;

and respectively acquiring the scoring results of the time of each recall information based on the object characteristics of the target object account.

12. The method according to claim 11, wherein the obtaining the current scoring result of each recall information based on the object characteristics of the target object account includes:

inputting the object characteristics of the target object account and target recall information into a scoring model to obtain the scoring result of the target recall information output by the scoring model; the target recall information is any one of the recall information;

13. The method of claim 12, wherein the sample information in the training sample set is historical recommendation information; the sample object features are object features of the recommended object account corresponding to the historical recommendation information when the historical recommendation information is recommended; the scoring tag of the sample information relative to the sample object account is a recommendation result of the historical recommendation information relative to the recommendation object account;

14. The method of any one of claims 1 to 8, wherein the first set of recall information comprises at least one of a first set of sub-information, a second set of sub-information, and a third set of sub-information; the first sub-information set is an information set composed of recall information obtained by screening from candidate recall information based on keywords determined by historical behavior data of a target object account; the second sub-information set is an information set composed of recall information obtained by screening from the candidate recall information, and the semantic vector is determined based on historical behavior data of the target object account; the third sub-information set is an information set composed of recall information obtained by screening from the candidate recall information based on information attributes.

15. An information recommendation device, characterized in that the device comprises:

16. A computer device, characterized in that it comprises a processor and a memory, said memory storing at least one computer program, said at least one computer program being loaded and executed by said processor to implement the information recommendation method according to any of claims 1 to 14.

17. A computer readable storage medium having stored therein at least one computer program loaded and executed by a processor to implement the information recommendation method of any one of claims 1 to 14.

18. A computer program product, characterized in that the computer program product comprises at least one computer program, which is loaded and executed by a processor to implement the information recommendation method according to any of claims 1 to 14.