CN114329176A - Information recommendation method and device, computer equipment, storage medium and program product - Google Patents
Information recommendation method and device, computer equipment, storage medium and program product Download PDFInfo
- Publication number
- CN114329176A CN114329176A CN202111314406.3A CN202111314406A CN114329176A CN 114329176 A CN114329176 A CN 114329176A CN 202111314406 A CN202111314406 A CN 202111314406A CN 114329176 A CN114329176 A CN 114329176A
- Authority
- CN
- China
- Prior art keywords
- information
- recommended
- recommendation
- probability
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides an information recommendation method, an information recommendation device, computer equipment, a storage medium and a program product, and relates to the technical fields of computer networks, artificial intelligence, intelligent traffic, driving assistance and the like. Determining a first recall probability through historical interaction data based on the associated information and the similarity between the two information, so that the preference degree of the target object to-be-recommended information under the influence of the associated information is accurately quantified by combining the historical interaction condition and the information similarity condition; determining a second recall probability through the recommendation time deviation and the aging factor of the two kinds of information, measuring the preference change condition of the target object from the time effect, accurately predicting the positive feedback possibility of the information to be recommended along with the time, and accurately quantifying the preference possibility of the target object to the information to be recommended and the change trend of the preference degree of the target object along with the time; the information recommendation process is dynamically controlled based on the recommendation probability of different variation trends, invalid recommendations are reduced, and recommendation efficiency is improved.
Description
Technical Field
The application relates to the technical fields of computer networks, artificial intelligence, intelligent traffic, auxiliary driving and the like, in particular to an information recommendation method, an information recommendation device, computer equipment, a storage medium and a program product.
Background
With the rapid development of the internet technology, information in the network also expands exponentially, a user can be used as a producer to produce information and upload the information to an application platform, and the application platform can also screen the information and recommend the information to the user.
In the related art, the information recommendation process may include: the server recommends information of interest for the user based on the preference of the user; for example, if the user interest tag includes a game, information of game content is often recommended to the user. However, the information recommendation process is too rigid, for example, the recommendation may be repeated and the aesthetic fatigue of the user may be easily caused, thereby generating invalid recommendations, resulting in low efficiency of actual recommendation of the information recommendation process.
Disclosure of Invention
The application provides an information recommendation method, an information recommendation device, computer equipment, a storage medium and a program product, which can solve the problem of low actual recommendation efficiency in information recommendation in the related art. The technical scheme is as follows:
in one aspect, an information recommendation method is provided, and the method includes:
acquiring information to be recommended, and determining associated information which has an associated production relation with the information to be recommended, wherein the associated information is read information which is exposed to a target object;
determining a first recall probability of the information to be recommended based on the similarity between the information to be recommended and the associated information and historical interaction data of the associated information, wherein the first recall probability is used for indicating the preference degree of a target object on the information to be recommended under the influence of the associated information;
determining a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended, wherein the second recall probability is used for indicating the possibility that a target object positively feeds back the information to be recommended over time, and the aging factor is used for indicating the change trend of the second recall probability over time;
and determining the recommendation probability of the information to be recommended based on the first recall probability and the second recall probability, and recommending the information to be recommended to the target object based on the recommendation probability.
In a possible implementation manner, after recommending the information to be recommended to the target object based on the recommendation probability, the method further includes:
acquiring current interactive data of the target object to the information to be recommended;
determining an object preference value of the information to be recommended based on the current interaction data, the historical interaction data and the weight of each interaction behavior;
and updating the information recommendation set of the target object based on the object preference value of the information to be recommended.
In another aspect, an information recommendation apparatus is provided, the apparatus including:
the information determining module is used for acquiring information to be recommended and determining associated information which has an associated production relationship with the information to be recommended, wherein the associated information is read information which is exposed to a target object;
a first recall probability determination module, configured to determine a first recall probability of the information to be recommended based on a similarity between the information to be recommended and the associated information and historical interaction data of the associated information, where the first recall probability is used to indicate a preference degree of a target object on the information to be recommended under the influence of the associated information;
a second recall probability determining module, configured to determine a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended, where the second recall probability is used to indicate a possibility that a target object positively feeds back the information to be recommended over time, and the aging factor is used to indicate a variation trend of the second recall probability over time;
and the recommending module is used for determining the recommending probability of the information to be recommended based on the first recalling probability and the second recalling probability and recommending the information to be recommended to the target object based on the recommending probability.
In another aspect, a computer device is provided, the computer device comprising:
one or more processors;
a memory;
one or more computer programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs configured to: the above-described information recommendation method is performed.
In another aspect, a computer-readable storage medium is provided, which stores computer instructions that, when executed on a computer, enable the computer to perform the above-mentioned information recommendation method.
In another aspect, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the information recommendation method as described above.
The beneficial effect that technical scheme that this application provided brought is:
determining the first recall probability by determining the associated information of the information to be recommended and based on the historical interaction data of the associated information and the similarity between the two information, thereby accurately quantifying the preference degree of the target object on the information to be recommended under the influence of the associated information by combining the historical interaction condition and the information similarity condition; determining a second recall probability through the recommendation time deviation and the aging factor of the two information, and measuring the preference change condition of the target object from the time effect so as to accurately predict the probability of positive feedback of the information to be recommended along with the time; obtaining a recommendation probability based on the first recall probability and the second recall probability, and finally recommending based on the recommendation probability, so that the change trend of the preference possibility and the preference degree of the target object to-be-recommended information along with time is accurately quantified by combining the influence of the associated information and the influence of a time effect; the dynamic control of the information recommendation process based on the recommendation probabilities of different change trends is further realized, the preference information can be frequently recommended under the condition of higher recommendation probability, the recommendation probability can be timely monitored to be lowered so as to reduce the recommendation, and the aesthetic fatigue of the target object is perfectly avoided; and the method can also combine the time effect to influence the information which is not interested before the recommendation probability is increased and the information is recommended in due time; the effects of reducing the recommendation time interval of the preference information and increasing the recommendation time interval of the uninterested information are achieved; the probability that each recommendation is converted into the effective recommendation is increased, the invalid recommendations are reduced, and the recommendation efficiency is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings used in the description of the embodiments of the present application will be briefly described below.
Fig. 1 is a schematic diagram of an implementation environment of an information recommendation method provided in the present application;
fig. 2 is a schematic flowchart of an information recommendation method according to an embodiment of the present application;
fig. 3 is a schematic coordinate diagram of a recommendation probability provided in an embodiment of the present application;
fig. 4 is a schematic coordinate diagram of a recommendation probability provided in an embodiment of the present application;
fig. 5 is a schematic diagram of an information recommendation process provided in an embodiment of the present application;
fig. 6 is a schematic structural diagram of an information recommendation device according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of a computer device according to an embodiment of the present application.
Detailed Description
Reference will now be made in detail to embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are exemplary only for the purpose of explaining the present application and are not to be construed as limiting the present invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items.
To make the objects, technical solutions and advantages of the present application more clear, embodiments of the present application will be described in further detail below with reference to the accompanying drawings.
Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.
The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and the like.
Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.
A distributed cloud storage system (hereinafter, referred to as a storage system) refers to a storage system that integrates a large number of storage devices (storage devices are also referred to as storage nodes) of different types in a network through application software or application interfaces to cooperatively work by using functions such as cluster application, grid technology, and a distributed storage file system, and provides a data storage function and a service access function to the outside.
At present, a storage method of a storage system is as follows: logical volumes are created, and when created, each logical volume is allocated physical storage space, which may be the disk composition of a certain storage device or of several storage devices. The client stores data on a certain logical volume, that is, the data is stored on a file system, the file system divides the data into a plurality of parts, each part is an object, the object not only contains the data but also contains additional information such as data identification (ID, ID entry), the file system writes each object into a physical storage space of the logical volume, and the file system records storage location information of each object, so that when the client requests to access the data, the file system can allow the client to access the data according to the storage location information of each object.
The information recommendation method provided by the application relates to the technologies of artificial intelligence, machine learning, cloud storage and the like. For example, the embodiment of the application can utilize the artificial intelligence technology to realize an automatic recommendation process of high-value information for different users and different recommendation times. In one possible example, a prediction model may be trained in advance by using a machine learning technique, and a recommendation probability of information to be recommended may be predicted based on the prediction model. In a possible example, the information to be recommended, the information of the target object, and the like may be stored by using the cloud storage technology, for example, a file system of the cloud storage technology may be used to store data such as a video to be recommended, historical interaction data of the target object, and an interaction behavior log, and when any data needs to be accessed, the file system may implement an access process of the corresponding data according to the corresponding storage location information.
Fig. 1 is a schematic diagram of an implementation environment of an information recommendation method provided in the present application. As shown in fig. 1, the implementation environment includes: a server 101 and a terminal 102, wherein the server 101 can be a background server of an application program. The terminal 102 is installed with an application program, and the server 101 and the terminal 102 can perform data interaction based on the application program. For example, the server 101 may send information to be recommended to the terminal 102 based on the application program to implement a process of recommending information to the terminal.
Illustratively, the application program is provided with an information recommendation function, and the information recommendation function is used for screening information to recommend to the object of the application program. The object on the application program can produce information by the identity of a producer and also can recommend the information to the user by the identity browsing server of the user. The object may be a user of the application. For example, the producer can produce audio, video, text and other information and upload the information to the information interaction platform of the application program. The terminal 102 may be a terminal where the target object is located, the server 101 may be configured with a resource pool, where the resource pool includes information produced by a generator, and when receiving a recommendation request of the target object sent by the terminal 102, the server 101 may screen the resource pool for information and send the information to the terminal 102, so as to recommend information to the target object.
The application program may be any application program supporting an information recommendation function, and may be, for example, a live application program, a content interaction platform, an information application program, a game application program, a social application program, a video application program, a shopping application program, and the like. The application is provided with a display page on which the terminal 102 displays information recommended by the server 101, which may include but is not limited to: a video stream display page, a live broadcast room page, an information stream display control and the like; the user may perform interactive activities with the information based on the displayed page of the terminal 102; such as comments, customs clearance, praise, favorites, shopping, awards, and the like. The application program can be an independent application program, or an application plug-in installed in the independent application program, and the like; such as application plug-ins installed in browsers, applets installed in social applications, etc.
The server 101 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server or a server cluster providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a web service, cloud communication, a middleware service, a domain name service, a security service, a CDN (Content Delivery Network), a big data and artificial intelligence platform, and the like. Such networks may include, but are not limited to: a wired network, a wireless network, wherein the wired network comprises: a local area network, a metropolitan area network, and a wide area network, the wireless network comprising: bluetooth, Wi-Fi, and other networks that enable wireless communication. The terminals 102 may include, but are not limited to: the terminal 102 and the server 101 may be directly or indirectly connected through wired or wireless communication, but not limited thereto. The determination may also be based on the requirements of the actual application scenario, and is not limited herein.
The following description refers to terms related to embodiments of the present application:
and information to be recommended: any form of information produced by a producer and published to an internet platform; information in the form of, for example, text, pictures, audio or video; information content includes, but is not limited to: news, merchandise information, articles, advertisements, short videos, live broadcasts, and the like.
The producer: the author refers to an author that produces information on an internet platform, such as an arrival person of a short video application, a director of a live broadcast application, a publisher of a Content interaction application, a KOL (Key Opinion Leader), a producer for producing a PGC (Professional Generated Content), a UGC (User Generated Content, that is, User original Content), and the like.
The production relation is as follows: attribute information indicating a producer or information produced by a producer. For example, the production relationship may include, but is not limited to: producer, live broadcast room identification of information produced by the producer, information classification collection to which the information belongs, topic labels, similarity with other information and the like; the production relationship in the application is obtained after the consent of the producer and the related user.
Information recommendation set: each object is correspondingly associated with an information recommendation set, and the information recommendation set comprises at least one piece of read information and an association relation between the information and a historical preference value. It should be noted that, in the present application, the information recommendation set is obtained after the user agrees.
The information that has been read: refers to information that has been exposed to a target object.
And (4) recommending time: the time for recommending the associated information to the target object is referred to, and it should be noted that the recommended time of the associated information in the application is obtained after the user agrees.
Historical preference values: indicating the preference degree of the target object to the associated information; the higher the historical preference value, the higher the interest level of the target object in the associated information.
And (4) correlation information: the read information has an associated production relation with the information to be recommended; the read information of the target object in the application is obtained after the user agrees.
First recall probability: and the recommendation processing module is used for indicating the preference degree of the target object to the information to be recommended under the influence of the associated information.
Second recall probability: indicating the likelihood of the target object positively feeding back the information to be recommended over time.
Fig. 2 is a schematic flowchart of an information recommendation method according to an embodiment of the present application. The execution subject of the method may be a server. As shown in fig. 2, the method includes the following steps.
The information to be recommended is information to be recommended to a target object, and in the application, the server can filter the information to be recommended from a plurality of candidate information of the application program. For example, the server may perform the filtering based on the recall time of the candidate information, or may perform the filtering based on the current time period of the producer of the candidate information. Accordingly, the implementation manner of this step may include the following two manners.
In the first mode, the server acquires the information to be recommended, of at least one candidate information, of which the recall time is at the current time, based on the recall time of the at least one candidate information.
The recall time refers to the time of the candidate information to be recommended, and the server can acquire the candidate information of which the recall time is at the current time point or the current time period as the information to be recommended. For example, when the current time point is 18:00 pm on x months, the information to be recommended having the recall time of 18:00 pm on x months x days may be acquired, or the information to be recommended having the recall time of 18:00 pm to 18:05 pm on x months x days may be recalled.
In one possible implementation, the recall time of each candidate information may be a time to be recommended for the target object. For information to be recommended, the server may determine a recall time of the information to be recommended based on the read information that has been exposed to the target object. For example, the step of the server determining the recall time of the information to be recommended may include: the server determines associated information having an associated production relation with the information to be recommended based on a historical browsing record of a target object, and determines recall time of the information to be recommended based on recommendation time and a repeated recommendation interval threshold of the associated information; wherein the time period in which the recall time is later than the recommendation time is not lower than the repeated recommendation interval threshold. When there are a plurality of pieces of association information, the recall time may be determined based on the recommendation time of the last recommended association information. For example, the repeated recommendation interval threshold may be 0.5 hour, and if the server recommends the associated information at 17:00 pm on x days of x months, the information to be recommended is recommended again after 17:30, for example, the recall time of the information to be recommended may be 17: 30.
It should be noted that the recall time can be configured and modified based on the need. The above is exemplified by only determining the recall time by the recommendation time of the read information and the repeated recommendation interval threshold, but the specific value of the recall time is not particularly limited. For example, in a further possible embodiment, the server may further determine the recall time based on an associated recommendation acceptance and a recommendation time of the target object, where the associated recommendation acceptance may be a degree of acceptance of the information that is recommended to have an associated production relationship by the target object, and exemplarily, the server adjusts the repeated recommendation interval threshold according to the associated recommendation acceptance, and determines the recall time based on the recommendation time of the associated information and the adjusted repeated recommendation interval threshold. For example, the larger the associated recommendation acceptance, the smaller the value for adjusting the repeated recommendation interval threshold; the smaller the associated recommendation acceptance, the larger the value of adjusting the repeat recommendation interval threshold.
In the second mode, the server acquires information to be recommended, which is in a recalling period currently by a producer in at least one candidate message, based on a period in which the producer of the at least one candidate message is located.
The time period during which the producer is located is used to indicate the strength of the recommendation for information produced by the producer, and in one possible example, the time period during which the producer is located may include, but is not limited to, a cold lag period, a recall period, and the like. The recall period refers to a period of recommending information produced by a producer, the recommendation strength of the recall period is greater than the cold lag period, and the cold lag period can be a period of not recommending or slowing down the information produced by the recommended producer. In the application, the server can recommend the production information of the producer in the recall period. When there are a plurality of producers in the recall period among the producers of the candidate information, the server may further filter based on the start time of the producer entering the recall period, the information production time of the produced information, and the like; for example, the information to be recommended of the producer with the earliest start time, or the information to be recommended with the earliest production time among the plurality of information produced by the plurality of producers in the recall period is selected.
It should be noted that, the above description is only given for an example of determining information to be recommended based on the recall time of the candidate information or determining information to be recommended based on whether the producer is in the recall period. For example, the server may combine the two manners to determine the information to be recommended. For example, the server may filter out at least one candidate message, wherein the producer refers to a first candidate message currently in a recall period; and then according to the recall time of the first candidate information, acquiring the information to be recommended of which the recall time is at the current time in the first candidate information.
The associated information is read information that has been exposed to the target object. The server can screen out the associated information with the associated production relationship with the information to be recommended from the production relationship of the at least one piece of read information based on the production relationship of the at least one piece of read information of the target object and the production relationship of the information to be recommended. It should be noted that the production relationship of the at least one piece of read information of the related target object and the production relationship of the information to be recommended are obtained after the consent of the related user.
The production relationship is used to indicate attribute information of a producer or information produced by the producer. In one possible example, a producer can add a topic label to the produced information, configure the affiliated information classification collection, the live broadcast room identification and other information; the server can also calculate the feature similarity of the information and other information based on the configuration of a producer or the features of the information to be recommended; the associated production relationship may include, but is not limited to: the producers are the same, the live broadcast room identifications belong to the same information classification collection, the topic labels are overlapped, and the feature similarity exceeds at least one of a target threshold value. Based on several possible situations of information included in the associated production relationship, correspondingly, the server determines that the implementation manner of the associated information at least includes five manners shown in the following manner one to manner five.
In the first mode, the server determines the associated information of the producer in the read information of the target object, which is the same as the information to be recommended, based on the producer of the information to be recommended.
Correlating production relationships may include producer identity. Producers alike may include but are not limited to: the source terminal has the same terminal identification, the same object identification and the same actual creation team or creator. The source terminal may be a terminal that uploads information to an information interaction platform of the application program, for example, the associated information and the information to be recommended are both information uploaded by the same terminal ID; for example, the read information and the information to be recommended may be marked with a team identifier, and the associated information and the information to be recommended may be a series of videos created by the same team. For example, the object identifier refers to an object for uploading information to the information interaction platform, and the associated information and the information to be recommended may be information uploaded by the same object identifier.
Several possibilities encompassed by equation one are illustrated below in conjunction with examples one through three below:
for example, a series of medical science popularization videos issued by the XX medical science popularization account can be used as videos with associated production relationships. Alternatively, a video distributed by the same terminal ID may be a video having an associated production relationship.
Example two, videos, articles, advertisement links, and the like, which are all authored by the same authoring team (e.g., both XXX corporation, XX fellow team, good recommendation officer team named "XXX"), or the same author, may be published for the same object account or different object accounts, as information with associated production relationships.
Example three, a commentary video of a different movie, each commentary being commentary by a commentator R, may be treated as a video with an associated production relationship (which may be published by the same object account or a different object account). Alternatively, the lesson learning videos each given by the same teacher W are videos having an association relationship (may be issued by the same subject account or different subject accounts).
The embodiments of the present application are illustrated only by the above examples one to three, but do not limit the same specific manner of the manufacturer. For example, the producers may be the same, and the XX darts may be associated with the object account, and the videos issued by the XX darts on the team account and the personal account respectively may be videos of the same producer. It should be noted that the above-mentioned related information of any user, such as the account number, the terminal ID, the authoring team, and the like, is obtained after the user agrees.
And secondly, the server determines the associated information of the read information, which is the same as the information to be recommended, in the live broadcast room information based on the live broadcast room information of the information to be recommended.
The associated production relation comprises that the information of the live rooms is the same. The live room information may include, but is not limited to: the live broadcast rooms have the same identification, comprise the same main broadcast, have the same live broadcast field, have the same live broadcast theme or have the same brand of the related commodities in the live broadcast rooms.
Several possibilities encompassed by equation two are illustrated below in conjunction with examples one through six below:
for example one, for multiple live broadcasts in the same live broadcast room, for example, multiple live broadcasts with live broadcast rooms ID all being "xxxx" may be taken as live broadcasts with associated production relationships; when a new live broadcast is started by the live broadcast room ID, recommendation can be performed based on the live broadcast of the live broadcast room ID browsed by the target object.
Example two, for a live broadcast that is opened by the same object, for example, a live broadcast (for example, a first broadcast from 12:00 noon to 13:00 pm, and a second broadcast from 20: 00 evening to 21:00 evening) that are both opened by the live broadcast account of the object a may be used as a live broadcast with an associated production relationship; when the second field is to be recommended, the second field can be recommended in combination with the first field.
In the third example, for live broadcasts of the same object at different time points in the same session, for example, live broadcasts of the same object B are started, the live broadcast time is from 10:00 a.m. to 18:00 a.m., the target object browses the live broadcasts at 11 am and 12 pm, and when the live broadcasts of the same session are recommended to the target object again at 14 pm, the live broadcasts of the target object browsed at 11 am and 12 pm can be both used as live broadcasts with an associated production relationship, so that whether the live broadcasts of the same session are recommended again can be considered based on the unused time period of the live broadcasts of the same session.
Example four, for the live broadcasts including the same anchor, for example, anchor C and anchor D start game live broadcast M, anchor C also starts game live broadcast N in conjunction with game team E, game live broadcasts M and N may be in the same live broadcast room or different live broadcast rooms, and game live broadcast M and game live broadcast N including anchor C may also be taken as live broadcasts with associated production relationship.
For example five, for different live rooms with the same live theme, for example, the themes of the live room J and the live room K are both live games for the game Q, the live games for the game Q of the live room J and the live room K may be used as information with an associated production relationship.
Example six, for different live rooms associated with the same product, for example, if both live room F and live room G associated products are cosmetics under the same brand, the live broadcast of the live room F and live room G to the brand cosmetics may be used as information having an associated production relationship.
The live broadcast room information is the same and can be configured based on needs, and the embodiment of the present application is illustrated by only six examples in the second mode, but the specific mode of the same live broadcast room information is not limited. Of course, the same information in the live broadcast room may also include the same game name or the same movie name of the live broadcast of the game, and the same processing as the above example is omitted here for brevity. Any user-related information such as the live broadcast room identification, the anchor broadcast, the live broadcast theme and the like in the related live broadcast room information is obtained after agreement.
And thirdly, the server determines the associated information of the read information, which belongs to the same information classification set as the information to be recommended, based on the information classification set to which the information to be recommended belongs.
The associated production relationships include belonging to the same information taxonomy collection. The information classification collection is used for indicating classification results of information in categories, information contents, information topics or equal dimensions of the information. The server can create at least two information classification sets, and a producer can configure the information classification set to which the produced information belongs; or the server identifies the information produced by the producer and automatically configures the corresponding information classification collection based on the identification result. The sets belonging to the same information category may include, but are not limited to: the information classification sets are the same, the classification keywords of the information classification sets are overlapped, or the keyword overlapping degree of the information classification sets exceeds a specified threshold value.
The following example a illustrates equation three in conjunction with example a and example b.
Example one, a producer may configure a categorized collection of information to which the produced information belongs, e.g., a plurality of movie short videos belonging to a movie episode of "XXXX" may be taken as information having an associated production relationship; or the teaching course video belonging to the XX examination teaching collection is taken as the information with the associated production relationship; or the science popularization videos belonging to the "abc" medical science popularization set and the science popularization videos belonging to the "abd" medical science popularization set are taken as the information with the associated production relationship.
Example two, the server automatically identifies and configures the information classification collections to which the information belongs, for example, the server may configure the corresponding collections for news as collections of finance, sports, military, agriculture, etc.; the server may use news belonging to the finance and economics as information having an associated production relationship, or may use news belonging to the military affairs in the last week as information having an associated production relationship. The classified collection of information to which the related information belongs is obtained after the user agrees.
And fourthly, the server determines that the topic label in the read information and the topic label of the information to be recommended have overlapped associated information based on the topic label of the information to be recommended.
The associated production relationship comprises the fact that topic labels are overlapped. The presence of topic label overlap may include, but is not limited to: the topic labels comprise label keywords which are overlapped, the feature similarity of semantic features of the topic labels exceeds a specified similarity threshold value, and the like. Illustratively, the hashtag may be configured by the producer; or the server identifies the read information or the information to be recommended and automatically configures the corresponding topic label based on the identification result.
The following example a illustrates equation four. Example one, for example, video H is a video of a XXX sporting event, the hashtags of video H comprising: "# XXX race #", "# XXX race in which the champion xx is yyds (forever god) #", etc.; the topic tag of the video L includes: if the tag keywords of the topic tags of the # XXX event # "and the # XXX sports event champion xx #" are overlapped, the video H and the video L can be used as information with associated production relationship. The information on the topic labels and the feature similarity of the topic labels is obtained after the user agrees to the information.
And a fifth mode that the server determines the associated information of which the feature similarity with the information to be recommended exceeds a target threshold in the read information based on the similarity information of the information to be recommended.
Correlating the production relationship includes the feature similarity exceeding a target threshold. For example, when a producer produces and publishes information, the server can calculate the feature similarity between the information and other information, and if the feature similarity between the information and any information does not exceed a maximum similarity threshold, the server publishes the information to an information interaction platform of an application program. For example, the similarity information of the information to be recommended may include a plurality of feature similarities and a corresponding relationship between the read information corresponding to each feature similarity, where each feature similarity is a similarity between the information to be recommended and the corresponding read information. The server can screen the read information corresponding to the feature similarity exceeding the target threshold from the similarity information, and uses the screened read information as the associated information. For example, the server may obtain the similarity through a neural network model, for example, multi-modal features of the read information and the information to be recommended are respectively extracted through the neural network model, and feature similarity of the read information and the information to be recommended is calculated based on the multi-modal features of the read information and the information to be recommended. The multimodal features may be fusion features to various types of features in the information, such as text, video, audio, etc.
Mode five is illustrated below in conjunction with example one and example two below.
In an example, the video X and the video Y are both videos including images of each sight spot in the Z scenic spot, and the feature similarity reaches 90%, so that the video X and the video Y may be information having an associated production relationship.
In a second example, the hotspot video a and the hotspot article b are videos of a certain hotspot news in the hotspot ranking list, the titles, the included images, the subtitles in the videos, the images, the characters and other contents included in the hotspot article b are relatively close to each other, and the feature similarity obtained based on the texts, the images and other contents of the titles, the included images, the subtitles in the videos and the images and other contents of the hotspot article b reaches 85%, so that the hotspot video a and the hotspot article b can be used as information with an associated production relationship.
It should be noted that, in the embodiment of the present application, the above description is only given by way of example in one to five, but the determination method of the associated production relationship is not limited, for example, in another possible embodiment, two pieces of information having the same group characteristics of the group facing the two pieces of information may be used as the information having the associated production relationship based on the group characteristics of the group facing the two pieces of information, for example, the child arithmetic early education video a1 and the child art early education video a2 both facing the child, or the information having the associated production relationship. For example, the server may determine the associated information in any one of the above manners one to five, or may combine the above five manners, for example, combine the manner one and the manner four to obtain the first associated information that is the same as the producer of the information to be recommended and obtain the second associated information that overlaps with the new topic tag to be detected; of course, the server may combine multiple ways of determining the associated information based on multiple situations of permutation and combination of the above five ways, which is not described herein any more.
By determining the associated information of the information to be recommended based on the associated production relation, for example, the associated information meeting the requirements can be accurately obtained from multiple dimensions by the way of whether producers are the same, whether topic labels are overlapped, whether live broadcast information is the same, whether the information belongs to the same information classification set or whether the information has higher feature similarity, and the like, so that the accuracy of determining the associated information is improved, the information recommendation method can be suitable for recommendation processes of multiple scenes, and the applicability and the accuracy of the information recommendation method are improved.
The first recall probability is used for indicating the preference degree of the target object to the information to be recommended under the influence of the associated information. The server may determine the first recall probability based on interaction of the target object with the associated information. For example, the server may determine a historical preference value of a target object for the associated information based on historical interaction data of interaction between the target object and the associated information, and determine the first recall probability based on the historical preference value and the similarity.
Illustratively, the server may determine the first recall probability according to the similarity and the user preference value by the following formula (1);
formula (1):
wherein S is1Is shown asProbability of recall, P0Representing historical preference values, and m representing the similarity between the information to be recommended and the associated information, wherein the value range of m is [0,1 ]](ii) a For example, when the number of the associated information is multiple, m may be a similarity corresponding to the associated information whose recommended time is closest to the current time, or an average value of the similarities corresponding to the multiple associated information. The determination process of the user preference value may specifically include the process shown in the following step 2032.
In a possible implementation manner, the tag similarity for each interest tag may be further refined in combination with one or more interest tags of the target object, and the first recall probability is further obtained through a plurality of tag similarities. Accordingly, step 203 may be implemented by steps 2031-2033 below.
Step 2031, the server determines the tag similarity of the information to be recommended and the associated information for at least one interest tag based on the at least one interest tag of the target object.
Each interest tag may correspond to a tag similarity, which is used to indicate a similarity degree of the information to be recommended and the associated information for the interest tag. In this step, the server may obtain, through a feature extraction network, a first tag feature of the to-be-recommended information and a second tag feature of the associated information for the at least one interest tag, respectively; for each interest tag, a tag similarity for the interest tag is determined based on a first tag characteristic for the interest tag and a second tag characteristic for the interest tag. For example, an interest tag may represent an information category of information of interest to a target object; for example, the interest tag of the target object may be games, the eight trigrams, sports, cooking, traveling, and the like. The interest tag may be configured by the target object, or may be predicted by the server based on the historical browsing history and the interaction data of the target object.
In a possible implementation manner, the server may further measure the tag similarity of the information to be recommended and the associated information for each tag in combination with the possibility that the information to be recommended and the associated information are respectively matched with each interest tag. Illustratively, step 2031 may be implemented by the following steps S1 to S4.
Step S1, the server obtains, based on the feature extraction network, a first tag feature of the information to be recommended for the at least one interest tag and a second tag feature of the associated information for the at least one interest tag, respectively.
The feature extraction network is used for extracting tag features of the information for at least one interest tag. For example, the feature extraction network may include at least one sub-network, each sub-network corresponds to one interest tag, and each sub-network is configured to extract a tag feature of the information to be recommended or the associated information for the corresponding interest tag; the tag feature is used for representing the feature that the information to be recommended or the associated information is related to the interest tag. Illustratively, the first tag feature comprises a feature of the information to be recommended in at least one dimension. The second tag characteristic may include a characteristic of the associated information in at least one dimension. The feature extraction network is a pre-trained neural network. For example, if the information to be recommended is video 1 and the associated information includes video 2, the feature vector a1 for the game tag, the feature vector a2 for the eight trigram tag, and the feature vector A3 for the motion tag of video 1 may be extracted.
Step S2, the server obtains a first confidence of the information to be recommended for the at least one interest tag and a second confidence of the associated information for the at least one interest tag based on the multi-tag classification model.
The multi-label classification model is used for determining confidence degrees of the information to be recommended or the associated information respectively aiming at the interest labels based on the input information. The first confidence is used for indicating the possibility that the information to be recommended is respectively matched with the interest labels. The second confidence level is used to indicate the possibility that the associated information matches each interest tag respectively. The multi-label classification model can comprise a feature extraction module and a multi-label classifier; taking the first confidence coefficient of the information to be recommended as an example, the server may extract the features of the information to be recommended through the feature extraction module, input the extracted features into the feature classifier, and obtain the confidence coefficients of the information to be recommended and the interest tags through the multi-tag classifier, so as to obtain the first confidence coefficient. For example, a confidence matrix of the video 1 is obtained through a multi-classification model, and the confidence matrix includes probabilities that the video 1 is respectively matched with a plurality of interest tags such as game tags, bagua tags and motion tags. For example, the Multi-Label classification model is a model trained in advance, such as a Multi-Label classification model based on ML-KNN (Multi-Label K-Nearest Neighbor algorithm).
Step S3, the server determines a co-occurrence probability of the information to be recommended and the associated information for the at least one interest tag based on the first confidence level and the second confidence level.
The co-occurrence probability is used for indicating the probability that the information to be recommended and the associated information are matched with each interest tag together. For example, for each interest tag, the server may determine a co-occurrence probability of the information to be recommended and the associated information for the interest tag based on a confidence of the information to be recommended for the interest tag and a confidence of the associated information for the interest tag. For example, the product value of the confidence of the information to be recommended for the interest tag and the confidence of the associated information for the interest tag is used as the co-occurrence probability of the information to be recommended and the associated information for the interest tag. For example, video 1 has a confidence of 0.8 for game tags and 0.3 for the bagua tags; the confidence coefficient of the video 2 for the game label is 0.7, and the confidence coefficient for the eight trigrams label is 0.4; the co-occurrence probability for video 1 and video 2 may be 0.56 for the game tag and 0.12 for the eight trigram tag.
Step S4, the server determines at least one label similarity between the information to be recommended and the associated information based on the feature similarity between the first label feature and the second label feature and the co-occurrence probability.
The server may calculate the feature similarity based on the first tag feature corresponding feature vector and the second tag feature corresponding feature vector, for example, euclidean distance between vectors, cosine similarity, and the like. For each interest tag, the server may determine a product value between the co-occurrence probability and the feature similarity for the interest tag as the tag similarity for the interest tag.
Step 2032, the server determines the historical preference value of the associated information based on the historical interactive data of the associated information.
The historical preference value is used to indicate the degree of preference of the target object for the associated information. The larger the history preference value is, the larger the degree of preference of the target object to the associated information is indicated. The server can obtain historical interaction data of the target object on the associated information, and determine a historical preference value of the associated information based on at least one interaction behavior of the target object on the associated information and the weight of each interaction behavior. For example, a history preference value is obtained by performing a weighting calculation based on a behavior value and a corresponding weight corresponding to each interactive behavior.
In one possible implementation, the server may calculate the historical preference value based on the target object's weight on at least one interactive behavior of the associated information and each interactive behavior by the following formula (2):
formula (2): p0＝C1W1+C2W2+C3W3+……+CnWn；
Wherein, P0Indicates a historical preference value, W1、W2、W3……WnBehavior values respectively representing n interactive behaviors of the target object; the behavior value can represent the information preference degree corresponding to the interactive behavior, and the larger the behavior value of the interactive behavior is, the larger the information preference degree represented by the interactive behavior is represented; for example, if the target object browses a video, the video is interested after browsing, and the video is praised, the behavior value of the browsing behavior is smaller than that of the praise behavior. C1、C2、C3……CnRespectively and correspondingly representing weights corresponding to the n interactive behaviors; c1、C2、C3……CnWherein each weight takes on a value of [0, 1%]Real numbers in between.
In an exemplary manner, the first and second electrodes are,the server can normalize the historical preference value to ensure that the numerical value of the historical preference value is in a specified range; for example, for P obtained based on the above formula (2)0Carrying out normalization treatment to obtain P0Has a value range of [ -1, 1 [)]. Of course, W can also be paired1、W2、W3……WnNormalizing the behavior values of the interactive behaviors to ensure that the value range of each behavior value is [ -1, 1](ii) a The behavior value corresponding to the interactive behavior of the target object for positive feedback may be a positive value, and the behavior value corresponding to the interactive behavior of the negative feedback may be a negative value.
The server can obtain an interactive behavior log of the target object, and count at least one interactive behavior of the target object on the associated information based on the interactive behavior log. The interaction behavior may include, but is not limited to: exposure, clicking, collecting, adding cars, purchasing, staying time, praise, sharing, commenting, paying attention, enjoying, delivering gifts, closing, quickly drawing, uninteresting and the like; for example, clicking, collecting, adding cars, purchasing and the like on a commodity detail page of the e-commerce application; and clicking, agreeing, sharing, commenting, paying attention to, enjoying, delivering gifts, closing, quickly drawing, losing interest and the like on a live broadcast page of the live broadcast application, and certainly, the stay time of the target object on the page of the application program can be obtained. It should be noted that the above-mentioned interaction behavior log and any user-related data required for obtaining the interaction behavior log are obtained after the user agrees.
The server can calculate and store the historical preference value in advance, and the stored historical preference value can be directly read. For example, in this step, the server may obtain a historical preference value corresponding to the associated information from the information recommendation set of the target object.
The embodiment of the present application only takes the history preference value determination process shown in formula (2) as an example, but the actual determination method of the history preference value is not limited. Of course, the historical preference value output by the trained machine learning model based on the historical interactive data of the associated information can also be directly obtained; or directly using the accumulated value of the interaction times of the accumulated interaction behaviors as a historical preference value and the like, which are not listed one by one.
When the number of the associated information is plural, the historical preference value may be obtained based on historical interaction data of one or more associated information. For example one, the server may use an average value of object preference values of a plurality of pieces of associated information as the historical preference value, the object preference value of each piece of associated information indicating a degree of preference of the target object for the associated information; for example, the server may calculate an object preference value of each piece of associated information in advance by the above formula (2), and store the object preference value to the information recommendation set of the target object; in this step, the server may obtain the object preference value of each piece of associated information from the information recommendation set. For example two, the server may calculate a historical preference value according to the above formula (2) based on historical interaction data of a plurality of pieces of associated information; accordingly, for example two, W in equation (2) above1、W2、W3……WnRepresenting the behavior value of the target object on n interactive behaviors of the plurality of associated information; if the interactive behaviors of a plurality of pieces of associated information are repeated, the behavior values of the repeated interactive behaviors can be directly accumulated. For example, if the target object approves video 1 and approves and forwards video 2, the behavior value W corresponding to video 1 can be setjBehavior value W corresponding to video praisemAre added up to obtain C3(W31+W32)。
It should be noted that any data related to the user, such as the historical interaction data, the information recommendation set, the historical preference value, the object preference value, the interest tag of the target object, and the like of the target object on the read information, are obtained after the user agrees.
Step 2033, the server determines the first recall probability based on the at least one tag similarity and the historical preference value.
The server may accumulate the at least one tag similarity to obtain a similarity, and determine a first recall probability according to the accumulated similarity and the historical preference value by the above formula (1). Illustratively, the server may accumulate the similarity under at least one tag according to the similarity of at least one tag and the weight corresponding to each interest tag by using the following formula (3) to obtain the similarity between the information to be recommended and the associated information; determining the first recall probability according to the similarity obtained by accumulation and the user preference value through the formula (1);
formula (3): m ═ Σ ximi；
Wherein m isiRepresenting the tag similarity of the ith interest tag in the at least one tag similarity; x is the number ofiRepresenting the weight corresponding to the ith interest tag, wherein m represents the similarity between the information to be recommended and the associated information; for example, the accumulated value of the weights corresponding to the plurality of tag similarities may be 1.
The first recall probability is determined based on the historical preference value of the similarity, and the influence of the associated information on the information to be recommended is measured by combining the similarity of the information to be recommended and the associated information and the interaction of the associated information, so that the preference degree of the target object on the information to be recommended under the influence of the associated information is accurately quantified, and the accuracy of determining the first recall probability is improved. Moreover, the tag similarity of the two kinds of information aiming at the interest tags can be directionally determined by taking the interest tags of the target object as guidance, so that the similarity of the two kinds of information is individually measured around the interest attention points of the target object, and the accuracy of the similarity process is further improved; the recommendation requirements of users with different interests are met, the conversion rate of converting the recommended information into the effective information is improved, and the actual recommendation efficiency is further improved.
And 204, the server determines a second recall probability of the information to be recommended based on the recommendation time deviation between the information to be recommended and the associated information and the aging factor of the information to be recommended.
The second recall probability is used to indicate a likelihood that the target object is positively feeding back the information to be recommended over time. And the server carries out aging influence prediction on a second recall probability of the information to be recommended based on the recommendation time deviation and the aging factor to obtain the second recall probability. Wherein the recommended time offset may include a time period from the recommended time to the recall time.
In the application, the second recall probability can be determined by combining the change condition of the preference of the associated information by the target object and the prediction of the change of the information to be recommended along with the time. In one possible implementation mode, the change trend of the preference degree of the target object to the information to be recommended along with time is an ascending trend; the change trend of the preference degree of the target object to the associated information along with the time is a descending trend; step 204 may include: the server can predict a first time effective value of the information to be recommended based on a recommendation time deviation between the recall time of the information to be recommended and the recommendation time of the associated information and the aging factor, wherein the first time effective value is used for indicating an increase of the preference degree of the target object to the information to be recommended within the recommendation time deviation; the server determines a second time-efficient value of the information to be recommended based on the recommendation time deviation, the time-efficient factor and the first recall probability, wherein the second time-efficient value is a reduction amount of a preference degree of a target object to the associated information in the recommendation time deviation; the server determines the second recall probability based on the first time-averaged value and the second time-averaged value. Wherein the aging factor is used to indicate a trend of the second recall probability over time. The age factor may include a time bias factor and a rate of change factor.
For example, the mapping relationship between the first time-effective value and the recommended time offset, the aging factor, and the first recall probability may be in the form of the following formula (4), and the server may predict the first time-effective value based on the recommended time offset and the aging factor by the following formula (4);
formula (4):
wherein, U1May represent a first time-effective value, deltaT being the recommended time offset, deltaT ═ t2-t1；t2Recall time for information to be recommended, t1Is the recommended time of the associated information. Age factor includes changeA rate factor and a time offset factor. Alpha is a time deviation factor, beta is a change rate factor, and both alpha and beta are real numbers larger than 0.
For example, the mapping relationship between the second time-effective value and the recommended time offset, the aging factor, and the first recall probability may be in the form of the following formula (5), and the server may determine the second time-effective value based on the second time-effective value and the recommended time offset, the aging factor, and the first recall probability through the following formula (5);
formula (5):
wherein, U2May represent a second time-valid value; s1May represent a first recall probability, S1Can be obtained by the above formula (1); deltaT is the recommended time offset, deltaT ═ t2-t1；t2Recall time for information to be recommended, t1Is the recommended time of the associated information. Alpha is a time deviation factor, beta is a change rate factor, and both alpha and beta are real numbers larger than 0.
The server may determine a sum of the first time-valid value and the second time-valid value as the second recall probability. For example, the server may determine the second recall probability based on the recommended time offset, the first recall probability, and the aging factor by the following equation (6);
formula (6):
wherein S is2A second recall probability may be represented that increases over time. U shape1Can represent a first time-efficient value, U2Can represent a second time-efficient value, S1May represent a first recall probability; the function corresponding to equation (6) is a centrosymmetric function, and therefore α can also be referred to as a center point parameter.
In a possible implementation manner, the server may first obtain a value of the aging factor, and determine the second recall probability based on the aging factor;and the value of the aging factor can be dynamically adjusted according to different conditions. Step 204 may include: the server acquires a value of the aging factor based on the first characteristic of the information to be recommended or the second characteristic of the target object, wherein the aging factor is used for indicating the change trend of the second recall probability along with the time; the server determines the second recall probability based on the value of the aging factor, the recommended time deviation and the first recall probability. Wherein the aging factor comprises a change rate factor and a time deviation factor; the rate of change factor is used to indicate a rate of increase of the second recall probability over time, a value of the rate of change factor being inversely related to the rate of increase of the second recall probability; that is, a smaller value of the parameter of the rate of change factor indicates a higher rate of increase in the second recall probability over time. The time deviation factor is used for indicating a time period required when the second recall probability reaches an intermediate value, the intermediate value is an average value between a minimum value and a maximum value of the second recall probability, and the value of the time deviation factor is positively correlated with the size of the time period required when the time deviation factor reaches the intermediate value; that is, the smaller the parameter value of the time deviation factor, the smaller the recommended time deviation required for the second recall probability to reach the intermediate value, i.e., the earlier the time for the second recall probability to reach the intermediate value; the larger the parameter value of the time deviation factor, the larger the deviation of the recommended time required for the recommended probability to reach the intermediate value, i.e., the later the time for the recommended probability to reach the intermediate value. For example, as shown in equation (6), the second recall probability has a value range of (0, (1-S)1) That is, the maximum value of the second recall probability is close (1-S)1) The minimum value is close to 0, and (1-S) can be obtained1) The maximum value is 0 and the minimum value is 0.
The server may adjust a trend of the second recall probability over time by dynamically adjusting a value of the aging factor. Illustratively, three possibilities for adjustment are described below as examples one through three.
For example one, the information to be recommended may correspond to a recommendation cycle, and the recommendation cycle may include at least two recommendation periods, where the at least two recommendation periods correspond to at least two target object activity levels. The first characteristic may include a target period in which a recall time of the information to be recommended is located; the process of obtaining the value of the aging factor based on the first feature of the information to be recommended may include: the server determines a target time interval in which the recall time of the information to be recommended is located in the recommendation cycle based on at least two recommendation time intervals included in the recommendation cycle of the information to be recommended, and the server acquires a first value of a change rate factor and a second value of a time deviation factor which are matched with the target time interval. The first value is negatively correlated with the activity degree of the target object, and the second value is negatively correlated with the activity degree of the target object. Illustratively, the recommendation cycle includes an active period and a cold lag period, the active period target object being more active than the cold lag period. When the recall time of the information to be recommended is in the active period of the recommendation cycle, the server acquires a seventh value of the change rate factor and an eighth value of the time deviation factor; responding to the recall time in the cold lag time period of the recommended cycle, and acquiring a ninth value of the aging factor and a tenth value of the time deviation factor; wherein the seventh value is less than the ninth value; the eighth value is less than the tenth value.
Example two, the second characteristic may include an associated recommendation acceptance of the target object, the associated recommendation acceptance being used to indicate an acceptance degree of information recommending that the target object has an associated production relationship; the server may quantify the age factor value based on the associated recommended acceptance, and the process of the server obtaining the age factor value based on the second feature of the target object includes: the server determines the associated recommendation acceptance of the target object based on the historical interactive data of the read information of the target object; the server acquires a third value of a change rate factor matched with the correlation acceptance and a fourth value of a time deviation factor; and the third value is negatively correlated with the associated recommended acceptance, and the fourth value is negatively correlated with the associated recommended acceptance. For example, the relevance recommendation acceptance may be determined based on historical interaction data of the target object over a specified period of time in the past; for example, based on historical interaction data of the target object in a specified time period, determining associated recommendation information browsed by the target object in the specified time period; and determining the associated recommendation acceptance according to the number of the browsed total information and the number of the associated recommendation information. The associated recommendation information is information recommended in association with read associated information having an associated production relationship with the information when recommending the information. For example, the total amount of browsing information of the target object in the past week is 200, wherein the number of associated recommendation information is 160, and the associated recommendation acceptability is 80%. Certainly, the relevance recommendation acceptance may also be further measured based on the historical interaction data of the relevance recommendation information, for example, if the total browsing information amount of the target object in the past week is 200, the number of the relevance recommendation information is 160, and the number of the relevance recommendation information positively fed back by the target object in the relevance information is 120, the relevance recommendation acceptance is 60%. Of course, the association recommendation acceptance may also be determined based on other manners, which is not specifically limited in this embodiment of the application. For example, the specific value of the aging factor may be determined based on the target acceptability threshold. For example, in response to the association recommendation acceptance being higher than the target acceptance threshold, the server obtains an eleventh value of the rate-of-change factor and a twelfth value of the time deviation factor; responding to the repeated recommended acceptance not exceeding the target acceptance threshold, and acquiring a thirteenth value of the change rate factor and a fourteenth value of the time deviation factor by the server; wherein the eleventh value is less than the thirteenth value; the twelfth value is less than the fourteenth value.
Example three, the first characteristic may include at least one of the first recall probability or a current popularity; the server can position the recommendation level based on the first recall probability or the current heat and obtain the parameter value of the aging factor based on the recommendation level. The process may include: the server determines the recommendation level of the information to be recommended based on at least one of the first recall probability or the current popularity of the information to be recommended; the server obtains a fifth value of a change rate factor matched with the recommendation level and a sixth value of a time deviation factor based on the recommendation level of the information to be recommended, wherein the fifth value is in negative correlation with the recommendation level, and the sixth value is in negative correlation with the recommendation level. The larger the first recall probability or the higher the current popularity, the higher the recommendation level; correspondingly, the higher the recommendation level, the lower the value of the rate of change factor and the lower the value of the time deviation factor. The current popularity may be popularity of the information to be recommended in a user group of the application program, and a higher popularity indicates a higher attention degree or preference degree of the user group to the information to be recommended. When the popularity value is high, the server can configure the value of a small time deviation factor, so that the second recall probability reaches a high value in a relatively short time, and frequent recommendation of high popularity information is realized. By aiming at the information with higher historical preference value, the value of a smaller change rate factor can be configured, so that the second recall probability is slowly increased, the information preferred by the target object can not be recommended too frequently, and the problem of audit fatigue of the user is perfectly avoided.
In this step, the server may use a sum of the first recall probability and the second recall probability as the recommendation probability. The probability expression of the recommendation probability may be in the form of the following formula (7) based on the above formula (1) and the above formula (6), and the server may obtain the recommendation probability based on the first recall probability and the second recall probability by the following formula (7):
formula (7):
wherein S represents a recommendation probability, S1Representing the first recall probability, S1Is determined as in the above formula (1), S2Representing a second recall probability; deltaT is the recommended time deviation, alpha is a time deviation factor, beta is a change rate factor, the function corresponding to the formula (7) is a central symmetry function, and the value of the time deviation factor alpha is used forThe center position of the corresponding function is indicated, which is the center of symmetry of the function, and thus the time deviation factor α can be the center point parameter. Clearly, the larger β, the faster the rate of S growth; the larger alpha is, the larger deltaT is needed to reach the same S value.
As shown in FIG. 3, FIG. 3 is a graph of probability expression, where the horizontal axis is interval time, unit is hour, and the vertical axis is recommendation probability, and the graph includes S1A curve of the recommended probability as a function of time at 0.4, and S1The curve of the recommendation probability changing with time at-0.02 is obtained from fig. 3, and the recommendation probability in both curves increases with time and gradually approaches the value of the maximum recommendation probability, that is, 1; the central point parameter alpha of the two curves takes a value of 100, namely the horizontal axis coordinates of the central points of the two curves are both 100; the smaller the value of the central point parameter alpha is, the more left the position of the central point in the coordinate system is; the larger the value of the center point parameter alpha is, the more right the position of the center point is. FIG. 4 is a graph of probability expressions, S of two curves, as shown in FIG. 41The values of both are-0.05 and alpha are 150, but the beta value of the curve 1 is 20, the beta value of the curve 2 is 40, obviously, the S increasing rate in the curve 1 is obviously greater than the S increasing rate in the curve 2, namely, the S increasing rate is1When the values of (a) are the same and the value of (a) is the same, the smaller the value of (beta) is, the faster the value of (S) increases. While in FIG. 3, S is arranged1The curve with 0.4 has a rate of change factor beta of 10, S1Beta value of 15 for the-0.02 curve; from a combination of FIGS. 3 and 4, it can be seen that S1The smaller the value of (A), the faster S increases, e.g. S1The curve of 0.4 has an S increase rate smaller than S1The S increase rate in the curve of-0.02.
And when the recommendation probability meets the recommendation condition, the server recommends the information to be recommended to the target object, wherein the recommendation condition can be that the recommendation probability is greater than the target probability. For example, the server may perform a discarding operation, where the discarding operation refers to that for the target object, the target object is discarded to achieve the effect of not recommending the information to be recommended, but the resource pool still includes the information to be recommended, so that the server recommends the information to be recommended to other suitable users at appropriate time.
In one possible example, the server may perform random recommendation in combination with a random number, that is, the target probability may be a randomly generated numerical value; the process may include: the server acquires the random probability corresponding to the recommendation; the server recommends the information to be recommended to the target object when responding to the recommendation probability of the information to be recommended is higher than the random probability; and the server responds to the condition that the recommendation probability is not higher than the random probability, and does not recommend the information to be recommended. Certainly, the server may also directly obtain a specified default threshold value for recommendation, that is, the target probability may be a pre-stored default probability threshold value; when the recommendation probability is higher than the default probability threshold, the server recommends the information to be recommended to the target object; otherwise, the server does not recommend the information to be recommended.
In a possible implementation manner, the server may further update the information recommendation set of the target object based on the current interaction behavior of the target object on the information to be recommended. Illustratively, after the server recommends the information to be recommended to a target object, the server obtains current interactive data of the target object on the information to be recommended; the server determines an object preference value of the information to be recommended based on the current interaction data, the historical interaction data and the weight of each interaction behavior; and the server updates the information recommendation set of the target object based on the object preference value of the information to be recommended. The object preference value is used for indicating the preference degree of the target object for the exposed information to be recommended. The server may determine the object preference value of the information to be recommended based on the process similar to the process of determining the historical preference value of the associated information in step 2032, which is not described in detail herein. The terminal where the target object is located receives and displays information to be recommended sent by the server, and the target object and the information to be recommended displayed by the terminal where the target object is located can execute interactive behaviors such as praise, comment and forwarding; the server can obtain the current interaction behavior of the target object on the information to be recommended. For example, the information recommendation set is added with the association relationship between the information to be recommended and the object preference value corresponding to the information to be recommended. It should be noted that any user-related data, such as the historical interaction data, the current interaction data, and the weights of the interaction behaviors, are obtained after the user agrees.
In a possible example, the server may also manage the timeliness of the information recommendation set; for example, a failure time period of the history preference value corresponding to each read information in the information recommendation set may be configured. For example, the server may set a failure time period corresponding to the read information, and when the recommendation time of the read information reaches the failure time period from the current time, the server deletes the association relationship between the read information and the historical preference value from the information recommendation set. The failure period may be one week, one month, or five days, etc. The recommendation is carried out based on the recommendation probability, and the information recommendation set is updated in real time based on the recommended interactive behavior data, so that the accuracy and timeliness of the information recommendation set are guaranteed, the accuracy of predicting the recommendation probability each time is further improved, and the recommendation efficiency is improved.
It should be noted that the method and the device can be applied to information recommendation scenes with associated production relations in information recommendation, and for information with associated production relations, such as image-text videos, live broadcasts, audios and the like, when information to be recommended with associated production relations with read information of a target object needs to be recommended, the recommendation probability is predicted on the basis of the associated information, recommendation time deviation and the like in real time, so that the influence of time effect is accurately estimated, and the dynamic control recommendation process of the information is realized on the basis of the recommendation probability.
For example, for a producer, after issuing a content a produced by a certain author, when issuing another content B produced by the producer next time, the information recommendation method of the present application may be used to dynamically control the probability that information produced by the same producer is issued twice.
For example, after a typical recommendation scenario such as short video recommendation, a short video client produces multiple pieces of short video content, and a user consumes one or more pieces of short video of the client, the server needs to decide whether to continue to recommend pushing the short video of the client to the user. For people interested by the user, if the repeated recommendation frequency is too high, the aesthetic fatigue of the user is easily caused to become poor in interest, if the repeated recommendation frequency is too low, the interest preference of the user cannot be responded quickly, and the interest of the user is reduced and transferred along with the time; therefore, the information recommendation method can be adopted to predict the recommendation probability of the video recommended next time by combining the videos read by the users in the videos of the user and the historical recommendation time of the videos read by the users, so that the short video of the user can be frequently recommended under the condition of high recommendation probability, the recommendation probability can be timely monitored to be lowered to reduce the recommendation, and the problem of aesthetic fatigue of the users is perfectly avoided. For the tarnished people who are not interested by the user, the recommendation can be continued in a mode of reducing the recommendation frequency, so that the videos which are possibly newly created by the tarnished people can be also possibly loved by the user; and the influence of time effect on user interest can be combined, and when the recommendation probability is increased, videos which reach people and are not interested before are recommended timely. It should be noted that any data related to the user, such as the video read by the user, the historical recommendation time of the video read by the user, and the like, are obtained after the user agrees.
For the live broadcast recommendation scene, live broadcast contents of each anchor are different, and live broadcast contents of the same anchor at different moments are also different. The method is similar to the method for recommending the short video of the video tare, the method and the system are suitable for information recommendation scenes such as the above, and for producers such as tare, anchor and blogger, the process of repeatedly recommending the produced information of the same producer can be dynamically controlled, so that the experience and the information exposure efficiency of users are greatly improved.
Fig. 5 is a schematic diagram of an information recommendation process according to an embodiment of the present application. The information recommendation process of the embodiment of the present application is described below with reference to fig. 5; as shown in fig. 5, for the read information that has been exposed to the target object, object preference values of the read information may be calculated based on the interaction behavior of the target object with the exposure information, and the object preference values of the respective read information may be stored in the storage medium. When a target object requests recommendation information next time, whether the read information has associated information of information to be recommended or not can be judged based on the read information of the target object in the information recommendation set, and if the read information does not have the associated information, the information to be recommended is directly issued to a user; otherwise, namely the read associated information exists, the recommendation probability of the information to be recommended can be calculated based on historical interactive data, recommendation time, recall time of the information to be recommended and the like of the associated information, a random number is generated, and when the random number is smaller than the recommendation probability, the information to be recommended is issued to the target object, otherwise, discarding operation is executed; certainly, the second determination may also be performed based on some other considerations, for example, when the complete compressed packet data of the information to be recommended is successfully acquired, the information to be recommended may be exposed to the target object, and the interaction behavior of the exposed information may be further stored correspondingly based on whether the exposure is performed; if the complete compressed packet data is not successfully acquired, the target object is not exposed, and a discard operation is performed at this time. It should be noted that the data of the historical interaction data, the recommendation time, the information recommendation set, and the like of the related target object and any data related to the user required for obtaining the data are obtained after the user agrees.
It should be noted that any user data or data related to the user referred to in the embodiments of the present application is obtained after the user agrees.
According to the information recommendation method, the association information of the information to be recommended is determined, and the first recall probability is determined based on the historical interaction data of the association information and the similarity between the two information, so that the preference degree of the target object to be recommended under the influence of the association information is accurately quantized by combining the historical interaction condition and the information similarity condition; determining a second recall probability through the recommendation time deviation and the aging factor of the two information, and measuring the preference change condition of the target object from the time effect so as to accurately predict the probability of positive feedback of the information to be recommended along with the time; obtaining a recommendation probability based on the first recall probability and the second recall probability, and finally recommending based on the recommendation probability, so that the change trend of the preference possibility and the preference degree of the target object to-be-recommended information along with time is accurately quantified by combining the influence of the associated information and the influence of a time effect; the dynamic control of the information recommendation process based on the recommendation probabilities of different change trends is further realized, the preference information can be frequently recommended under the condition of higher recommendation probability, the recommendation probability can be timely monitored to be lowered so as to reduce the recommendation, and the aesthetic fatigue of the target object is perfectly avoided; and the method can also combine the time effect to influence the information which is not interested before the recommendation probability is increased and the information is recommended in due time; the effects of reducing the recommendation time interval of the preference information and increasing the recommendation time interval of the uninterested information are achieved; the probability that each recommendation is converted into the effective recommendation is increased, the invalid recommendations are reduced, and the recommendation efficiency is improved.
And the value of the aging factor is dynamically adjusted based on the characteristics of different information to be recommended or the characteristics of associated information, and the second recall probability is determined by adopting the aging factor which is accurately matched with different associated recommendation receptions at different time periods of a recommendation cycle, so that the second recall probability suitable for the current situation is accurately obtained, the accuracy of determining the second recall probability is improved, and the accuracy of the recommendation process is further improved.
Moreover, the tag similarity of the two kinds of information aiming at the interest tags is directionally determined by taking the interest tags of the target object as guidance, so that the similarity of the two kinds of information is individually measured around the interest attention points of the target object, and the accuracy of the similarity process is further improved; the recommendation requirements of users with different interests are met, the conversion rate of converting the recommended information into the effective information is improved, and the actual recommendation efficiency is further improved.
Fig. 6 is a schematic structural diagram of an information recommendation device according to an embodiment of the present application. As shown in fig. 6, the apparatus includes:
the information determining module 601 is configured to obtain information to be recommended, and determine associated information having an associated production relationship with the information to be recommended, where the associated information is read information that has been exposed to a target object;
a first recall probability determining module 602, configured to determine a first recall probability of the information to be recommended based on a similarity between the information to be recommended and the associated information and historical interaction data of the associated information, where the first recall probability is used to indicate a preference degree of a target object for the information to be recommended under the influence of the associated information;
a second recall probability determining module 603, configured to determine a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended, where the second recall probability is used to indicate a possibility that a target object positively feeds back the information to be recommended over time, and the aging factor is used to indicate a variation trend of the second recall probability over time;
the recommending module 604 is configured to determine a recommending probability of the information to be recommended based on the first recalling probability and the second recalling probability, and recommend the information to be recommended to the target object based on the recommending probability.
In a possible implementation manner, the second recall probability determining module 603 is configured to predict a first time effective value of the information to be recommended, based on a recommendation time deviation between the recall time of the information to be recommended and the recommendation time of the associated information, and the time effectiveness factor, where the first time effective value is used to indicate an increase amount of a preference degree of the target object to the information to be recommended within the recommendation time deviation; determining a second time-efficient value of the information to be recommended based on the recommendation time deviation, the time-efficient factor and the first recall probability, wherein the second time-efficient value is a reduction amount of a preference degree of a target object to associated information in the recommendation time deviation; determining the second recall probability based on the first time-averaged value and the second time-averaged value.
In a possible implementation manner, the second recall probability determining module 603 includes:
the acquisition unit is used for acquiring the value of the aging factor based on the first characteristic of the information to be recommended or the second characteristic of the target object;
the determining unit is used for determining the second recall probability based on the value of the aging factor, the recommended time deviation and the first recall probability;
wherein the aging factor includes a rate of change factor and a time offset factor;
the rate of change factor is used to indicate a rate of increase of the second recall probability over time, a value of the rate of change factor being inversely related to the rate of increase of the second recall probability;
the time deviation factor is used for indicating a time period required when the second recall probability reaches an intermediate value, the intermediate value is an average value between a minimum value and a maximum value of the second recall probability, and the value of the time deviation factor is positively correlated with the size of the time period required when the intermediate value is reached.
In one possible implementation, the first characteristic comprises a target period in which the recall time of the information to be recommended is located;
the acquisition unit is used for determining the target time interval of the recall time of the information to be recommended in the recommendation cycle based on at least two recommendation time intervals included in the recommendation cycle of the information to be recommended, wherein the at least two recommendation time intervals correspond to at least two target object activity degrees; acquiring a first value of a change rate factor matched with the target time period and a second value of a time deviation factor;
the first value is negatively correlated with the activity degree of the target object, and the second value is negatively correlated with the activity degree of the target object.
In one possible embodiment, the second characteristic includes an associated recommendation acceptance of the target object;
the acquisition unit is used for determining the associated recommendation acceptance of the target object based on the historical interactive data of the read information of the target object, wherein the associated recommendation acceptance is used for indicating the acceptance degree of the information which is recommended to have an associated production relationship; acquiring a third value of a change rate factor matched with the correlation acceptance and a fourth value of a time deviation factor;
and the third value is negatively correlated with the associated recommended acceptance, and the fourth value is negatively correlated with the associated recommended acceptance.
In one possible embodiment, the first characteristic comprises at least one of the first recall probability or a current popularity;
the acquisition unit is used for determining the recommendation level of the information to be recommended based on at least one of the first recall probability or the current popularity of the information to be recommended; acquiring a fifth value of a change rate factor and a sixth value of a time deviation factor matched with the recommendation grade based on the recommendation grade of the information to be recommended;
the fifth value is negatively correlated with the recommended level, and the sixth value is negatively correlated with the recommended level.
In one possible implementation, the first recall probability determining module 602 includes:
the tag similarity determining unit is used for determining at least one tag similarity of the information to be recommended and the associated information aiming at least one interest tag based on the at least one interest tag of the target object;
a historical preference value determining unit, configured to determine a historical preference value of the associated information based on historical interaction data of the associated information, where the historical preference value is used to indicate a preference degree of the target object for the associated information;
a first recall probability determination unit for determining the first recall probability based on the at least one tag similarity and the historical preference value.
In a possible implementation manner, the tag similarity determining unit is configured to obtain, based on a feature extraction network, a first tag feature of the information to be recommended for the at least one interest tag and a second tag feature of the associated information for the at least one interest tag, respectively; respectively acquiring a first confidence degree of the information to be recommended aiming at the at least one interest label and a second confidence degree of the associated information aiming at the at least one interest label based on a multi-label classification model, wherein the multi-label classification model is used for determining the confidence degrees of the information matched with the interest labels respectively based on input information; determining a co-occurrence probability of the information to be recommended and the associated information for the at least one interest tag based on the first confidence degree and the second confidence degree, wherein the co-occurrence probability is used for indicating the probability that the information to be recommended and the associated information are matched with each interest tag together; and determining at least one label similarity between the information to be recommended and the associated information based on the feature similarity between the first label feature and the second label feature and the co-occurrence probability.
In one possible embodiment, the information determining module 601 is configured to at least one of:
determining related information, which is the same as the information to be recommended, of the producer in the read information of the target object based on the producer of the information to be recommended, wherein the related production relation comprises the same producer;
determining the associated information of the read information, which is the same as the information to be recommended, of the live broadcast room information based on the live broadcast room information of the information to be recommended, wherein the associated production relation comprises the same live broadcast room information;
determining the associated information of the read information, which belongs to the same information classified collection as the information to be recommended, based on the information classified collection to which the information to be recommended belongs, wherein the associated production relation comprises that the live broadcast room identifiers are the same and belong to the same information classified collection;
determining related information of the read information, wherein the topic labels of the information to be recommended are overlapped, and the related production relationship comprises the fact that the topic labels are overlapped;
and determining associated information of which the feature similarity with the information to be recommended exceeds a target threshold in the read information based on the similarity information of the information to be recommended, wherein the associated production relationship comprises the feature similarity exceeding the target threshold.
In one possible embodiment, the information determining module 601 is configured to at least one of:
responding to a recommendation request of the target object, and acquiring information to be recommended, of at least one candidate information, of which the recall time is at the current time, based on the recall time of the at least one candidate information;
and in response to the recommendation request of the target object, acquiring information to be recommended, which is in the recalling period currently by the producer, in at least one candidate information based on the period in which the producer of the at least one candidate information is located.
In a possible implementation manner, the recommending module 604 is configured to obtain a random probability corresponding to the recommendation; recommending the information to be recommended to the target object when the recommendation probability of the information to be recommended is higher than the random probability; and in response to the recommendation probability not being higher than the random probability, not recommending the information to be recommended.
In one possible embodiment, the apparatus further comprises:
the acquisition module is used for acquiring the current interactive data of the target object to the information to be recommended;
the object preference value determining module is used for determining the object preference value of the information to be recommended based on the current interaction data, the historical interaction data and the weight of each interaction behavior;
and the updating module is used for updating the information recommendation set of the target object based on the object preference value of the information to be recommended.
The information recommendation device determines the first recall probability by determining the associated information of the information to be recommended and based on the historical interaction data of the associated information and the similarity between the two information, so that the preference degree of the target object to the information to be recommended under the influence of the associated information is accurately quantized by combining the historical interaction condition and the information similarity condition; determining a second recall probability through the recommendation time deviation and the aging factor of the two information, and measuring the preference change condition of the target object from the time effect so as to accurately predict the probability of positive feedback of the information to be recommended along with the time; obtaining a recommendation probability based on the first recall probability and the second recall probability, and finally recommending based on the recommendation probability, so that the change trend of the preference possibility and the preference degree of the target object to-be-recommended information along with time is accurately quantified by combining the influence of the associated information and the influence of a time effect; the dynamic control of the information recommendation process based on the recommendation probabilities of different change trends is further realized, the preference information can be frequently recommended under the condition of higher recommendation probability, the recommendation probability can be timely monitored to be lowered so as to reduce the recommendation, and the aesthetic fatigue of the target object is perfectly avoided; and the method can also combine the time effect to influence the information which is not interested before the recommendation probability is increased and the information is recommended in due time; the effects of reducing the recommendation time interval of the preference information and increasing the recommendation time interval of the uninterested information are achieved; the probability that each recommendation is converted into the effective recommendation is increased, the invalid recommendations are reduced, and the recommendation efficiency is improved.
And the value of the aging factor is dynamically adjusted based on the characteristics of different information to be recommended or the characteristics of associated information, and the second recall probability is determined by adopting the aging factor which is accurately matched with different associated recommendation receptions at different time periods of a recommendation cycle, so that the second recall probability suitable for the current situation is accurately obtained, the accuracy of determining the second recall probability is improved, and the accuracy of the recommendation process is further improved.
Moreover, the tag similarity of the two kinds of information aiming at the interest tags is directionally determined by taking the interest tags of the target object as guidance, so that the similarity of the two kinds of information is individually measured around the interest attention points of the target object, and the accuracy of the similarity process is further improved; the recommendation requirements of users with different interests are met, the conversion rate of converting the recommended information into the effective information is improved, and the actual recommendation efficiency is further improved.
The information recommendation apparatus of this embodiment can execute the information recommendation method shown in the above embodiments of this application, and the implementation principles thereof are similar and will not be described herein again.
Fig. 7 is a schematic structural diagram of a computer device provided in an embodiment of the present application. As shown in fig. 7, the computer apparatus includes: a memory and a processor; at least one program stored in the memory for execution by the processor, which when executed by the processor, implements:
determining the first recall probability by determining the associated information of the information to be recommended and based on the historical interaction data of the associated information and the similarity between the two information, thereby accurately quantifying the preference degree of the target object on the information to be recommended under the influence of the associated information by combining the historical interaction condition and the information similarity condition; determining a second recall probability through the recommendation time deviation and the aging factor of the two information, and measuring the preference change condition of the target object from the time effect so as to accurately predict the probability of positive feedback of the information to be recommended along with the time; obtaining a recommendation probability based on the first recall probability and the second recall probability, and finally recommending based on the recommendation probability, so that the change trend of the preference possibility and the preference degree of the target object to-be-recommended information along with time is accurately quantified by combining the influence of the associated information and the influence of a time effect; the dynamic control of the information recommendation process based on the recommendation probabilities of different change trends is further realized, the preference information can be frequently recommended under the condition of higher recommendation probability, the recommendation probability can be timely monitored to be lowered so as to reduce the recommendation, and the aesthetic fatigue of the target object is perfectly avoided; and the method can also combine the time effect to influence the information which is not interested before the recommendation probability is increased and the information is recommended in due time; the effects of reducing the recommendation time interval of the preference information and increasing the recommendation time interval of the uninterested information are achieved; the probability that each recommendation is converted into the effective recommendation is increased, the invalid recommendations are reduced, and the recommendation efficiency is improved.
In an alternative embodiment, a computer device is provided, as shown in FIG. 7, the computer device 700 shown in FIG. 7 comprising: a processor 701 and a memory 703. The processor 701 is coupled to a memory 703, such as via a bus 702. Optionally, the computer device 700 may further include a transceiver 704, and the transceiver 704 may be used for data interaction between the computer device and other computer devices, such as transmission of data and/or reception of data, and the like. It should be noted that the transceiver 704 is not limited to one in practical applications, and the structure of the computer device 700 is not limited to the embodiment of the present application.
The Processor 701 may be a CPU (Central Processing Unit), a general-purpose Processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other Programmable logic device, a transistor logic device, a hardware component, or any combination thereof. Which may implement or perform the various illustrative logical blocks, modules, and circuits described in connection with the disclosure. The processor 701 may also be a combination of computing functions, e.g., comprising one or more microprocessors, DSPs, and microprocessors, among others.
The Memory 703 may be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical Disc storage, optical Disc storage (including Compact Disc, laser Disc, optical Disc, digital versatile Disc, blu-ray Disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to these.
The memory 703 is used for storing application program codes (computer programs) for executing the present application, and is controlled by the processor 701. The processor 701 is configured to execute application program code stored in the memory 703 to implement the content shown in the foregoing method embodiments.
Among these, computer devices include, but are not limited to: a server, a service cluster or any electronic device with an information recommendation function.
The embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, which, when running on a computer, enables the computer to execute the corresponding content of the information recommendation method in the foregoing method embodiment.
Embodiments of the present application provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the information recommendation method.
It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and may be performed in other orders unless explicitly stated herein. Moreover, at least a portion of the steps in the flow chart of the figure may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
The foregoing is only a partial embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.
Claims (15)
1. An information recommendation method, characterized in that the method comprises:
acquiring information to be recommended, and determining associated information which has an associated production relation with the information to be recommended, wherein the associated information is read information which is exposed to a target object;
determining a first recall probability of the information to be recommended based on the similarity between the information to be recommended and the associated information and historical interaction data of the associated information, wherein the first recall probability is used for indicating the preference degree of a target object on the information to be recommended under the influence of the associated information;
determining a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended, wherein the second recall probability is used for indicating the possibility that a target object positively feeds back the information to be recommended over time, and the aging factor is used for indicating the change trend of the second recall probability over time;
and determining the recommendation probability of the information to be recommended based on the first recall probability and the second recall probability, and recommending the information to be recommended to the target object based on the recommendation probability.
2. The information recommendation method according to claim 1, wherein the determining a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended comprises:
predicting a first time effective value of the information to be recommended based on a recommendation time deviation between the recall time of the information to be recommended and the recommendation time of the associated information and the aging factor, wherein the first time effective value is used for indicating an increase of the preference degree of the target object to the information to be recommended within the recommendation time deviation;
determining a second time-efficient value of the information to be recommended based on the recommendation time deviation, the time-efficient factor and the first recall probability, wherein the second time-efficient value is the reduction of the preference degree of a target object to the associated information in the recommendation time deviation;
determining the second recall probability based on the first time-averaged value and the second time-averaged value.
3. The information recommendation method according to claim 1, wherein the determining a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended comprises:
acquiring a value of the aging factor based on the first characteristic of the information to be recommended or the second characteristic of the target object;
determining the second recall probability based on a value of the aging factor, the recommended time offset, and the first recall probability;
wherein the aging factor comprises a rate of change factor and a time offset factor;
the rate of change factor is used to indicate a rate of increase of the second recall probability over time, a value of the rate of change factor being inversely related to the rate of increase of the second recall probability;
the time deviation factor is used for indicating a time period required when the second recall probability reaches an intermediate value, the intermediate value is an average value between a minimum value and a maximum value of the second recall probability, and a value of the time deviation factor is positively correlated with a size of the time period required when the time deviation factor reaches the intermediate value.
4. The information recommendation method according to claim 3, wherein the first feature includes a target period in which a recall time of the information to be recommended is located; the obtaining of the value of the aging factor based on the first feature of the information to be recommended or the second feature of the target object includes:
determining a target time interval in which the recall time of the information to be recommended is positioned in a recommendation cycle based on at least two recommendation time intervals included in the recommendation cycle of the information to be recommended, wherein the at least two recommendation time intervals correspond to at least two target object activity degrees;
acquiring a first value of a change rate factor matched with the target time period and a second value of a time deviation factor;
the first value is negatively correlated with the activity degree of the target object, and the second value is negatively correlated with the activity degree of the target object.
5. The information recommendation method according to claim 3, wherein the second feature includes an associated recommendation acceptance of the target object; the obtaining of the value of the aging factor based on the first feature of the information to be recommended or the second feature of the target object includes:
determining the associated recommendation acceptance of the target object based on the historical interactive data of the read information of the target object, wherein the associated recommendation acceptance is used for indicating the acceptance degree of the information with associated production relation;
acquiring a third value of a change rate factor matched with the correlation acceptance and a fourth value of a time deviation factor;
and the third value is negatively correlated with the associated recommended acceptance, and the fourth value is negatively correlated with the associated recommended acceptance.
6. The information recommendation method according to claim 3, wherein the first feature comprises at least one of the first recall probability or a current popularity; the obtaining of the value of the aging factor based on the first feature of the information to be recommended or the second feature of the target object includes:
determining a recommendation level of the information to be recommended based on at least one of a first recall probability or a current popularity of the information to be recommended;
acquiring a fifth value of a change rate factor and a sixth value of a time deviation factor matched with the recommendation grade based on the recommendation grade of the information to be recommended;
and the fifth value is in negative correlation with the recommended grade, and the sixth value is in negative correlation with the recommended grade.
7. The information recommendation method according to claim 1, wherein the determining a first recall probability of the information to be recommended based on the similarity between the information to be recommended and the associated information and historical interaction data of the associated information comprises:
determining at least one tag similarity of the information to be recommended and the associated information for at least one interest tag based on the at least one interest tag of the target object;
determining a historical preference value of the associated information based on historical interaction data of the associated information, wherein the historical preference value is used for indicating the preference degree of the target object to the associated information;
determining the first recall probability based on the at least one tag similarity and the historical preference value.
8. The information recommendation method according to claim 7, wherein the determining at least one tag similarity of the information to be recommended and the associated information for the at least one interest tag based on the at least one interest tag of the target object comprises:
respectively acquiring a first tag feature of the information to be recommended aiming at the at least one interest tag and a second tag feature of the associated information aiming at the at least one interest tag based on a feature extraction network;
respectively acquiring a first confidence degree of the information to be recommended aiming at the at least one interest label and a second confidence degree of the associated information aiming at the at least one interest label based on a multi-label classification model, wherein the multi-label classification model is used for determining the confidence degrees of the information matched with the plurality of interest labels respectively based on input information;
determining a co-occurrence probability of the information to be recommended and the associated information for the at least one interest tag based on the first confidence degree and the second confidence degree, wherein the co-occurrence probability is used for indicating the probability that the information to be recommended and the associated information are matched with each interest tag together;
determining at least one label similarity between the information to be recommended and the associated information based on the feature similarity between the first label feature and the second label feature and the co-occurrence probability.
9. The information recommendation method according to claim 1, wherein the determining of the associated information having the associated production relationship with the information to be recommended includes at least one of:
determining associated information, which is the same as the information to be recommended, of a producer in the read information of the target object based on the producer of the information to be recommended, wherein the associated production relationship comprises the same producer;
determining associated information, which is the same as the information to be recommended, of the live broadcast room information in the read information based on the live broadcast room information of the information to be recommended, wherein the associated production relation comprises the same live broadcast room information;
determining associated information which belongs to the same information classification set as the information to be recommended in the read information based on the information classification set to which the information to be recommended belongs, wherein the associated production relation comprises that the live broadcast room identifiers are the same and belong to the same information classification set;
determining related information of the topic labels in the read information and the topic labels of the information to be recommended which are overlapped based on the topic labels of the information to be recommended, wherein the related production relationship comprises the fact that the topic labels are overlapped;
and determining associated information of which the feature similarity with the information to be recommended exceeds a target threshold in the read information based on the similarity information of the information to be recommended, wherein the associated production relationship comprises the feature similarity exceeding the target threshold.
10. The information recommendation method according to claim 1, wherein the obtaining of the information to be recommended includes at least one of:
responding to a recommendation request of the target object, and acquiring information to be recommended, of at least one candidate information, of which the recall time is at the current time, based on the recall time of the at least one candidate information;
and responding to the recommendation request of the target object, and acquiring information to be recommended, which is in a recalling period currently by a producer in at least one candidate information, based on the period in which the producer of the at least one candidate information is located.
11. The information recommendation method according to claim 1, wherein the recommending the information to be recommended to the target object based on the recommendation probability comprises:
obtaining a random probability corresponding to the recommendation;
recommending the information to be recommended to the target object when the recommendation probability of the information to be recommended is higher than the random probability;
and when the recommendation probability is not higher than the random probability, the information to be recommended is not recommended.
12. An information recommendation apparatus, characterized in that the apparatus comprises:
the information determining module is used for acquiring information to be recommended and determining associated information which has an associated production relationship with the information to be recommended, wherein the associated information is read information which is exposed to a target object;
a first recall probability determination module, configured to determine a first recall probability of the information to be recommended based on a similarity between the information to be recommended and the associated information and historical interaction data of the associated information, where the first recall probability is used to indicate a preference degree of a target object on the information to be recommended under the influence of the associated information;
a second recall probability determining module, configured to determine a second recall probability of the information to be recommended based on a recommendation time deviation between the information to be recommended and the associated information and an aging factor of the information to be recommended, where the second recall probability is used to indicate a possibility that a target object positively feeds back the information to be recommended over time, and the aging factor is used to indicate a variation trend of the second recall probability over time;
and the recommending module is used for determining the recommending probability of the information to be recommended based on the first recalling probability and the second recalling probability and recommending the information to be recommended to the target object based on the recommending probability.
13. A computer device, characterized in that the computer device comprises:
one or more processors;
a memory;
one or more computer programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs configured to: performing the information recommendation method according to any one of claims 1 to 11.
14. A computer-readable storage medium for storing computer instructions which, when executed on a computer, cause the computer to perform the information recommendation method of any one of claims 1 to 11.
15. A computer program product comprising a computer program, characterized in that the computer program realizes the information recommendation method of any one of claims 1 to 11 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111314406.3A CN114329176A (en) | 2021-11-08 | 2021-11-08 | Information recommendation method and device, computer equipment, storage medium and program product |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111314406.3A CN114329176A (en) | 2021-11-08 | 2021-11-08 | Information recommendation method and device, computer equipment, storage medium and program product |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114329176A true CN114329176A (en) | 2022-04-12 |
Family
ID=81044478
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111314406.3A Pending CN114329176A (en) | 2021-11-08 | 2021-11-08 | Information recommendation method and device, computer equipment, storage medium and program product |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114329176A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116028721A (en) * | 2023-03-30 | 2023-04-28 | 深圳市壹通道科技有限公司 | 5G message pushing system |
| CN117934126A (en) * | 2024-03-25 | 2024-04-26 | 珠海芯烨电子科技有限公司 | Personalized target information recommendation system based on user emotion analysis |
| CN118152668A (en) * | 2024-05-10 | 2024-06-07 | 腾讯科技（深圳）有限公司 | Media information processing method and device, equipment, storage medium and program product |
-
2021
- 2021-11-08 CN CN202111314406.3A patent/CN114329176A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116028721A (en) * | 2023-03-30 | 2023-04-28 | 深圳市壹通道科技有限公司 | 5G message pushing system |
| CN117934126A (en) * | 2024-03-25 | 2024-04-26 | 珠海芯烨电子科技有限公司 | Personalized target information recommendation system based on user emotion analysis |
| CN118152668A (en) * | 2024-05-10 | 2024-06-07 | 腾讯科技（深圳）有限公司 | Media information processing method and device, equipment, storage medium and program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020228514A1 (en) | Content recommendation method and apparatus, and device and storage medium | |
| CN111444428B (en) | Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium | |
| Shu et al. | Studying fake news via network analysis: detection and mitigation | |
| US20240119321A1 (en) | Generative interactive video method and system | |
| Bilal et al. | Social profiling: A review, taxonomy, and challenges | |
| CN111966914B (en) | Content recommendation method and device based on artificial intelligence and computer equipment | |
| US10977448B2 (en) | Determining personality profiles based on online social speech | |
| KR102497809B1 (en) | Advertiser and influencer mediation service provision device using artificial intelligence | |
| CN114329176A (en) | Information recommendation method and device, computer equipment, storage medium and program product | |
| KR20160057475A (en) | System and method for actively obtaining social data | |
| CN111339404A (en) | Content popularity prediction method and device based on artificial intelligence and computer equipment | |
| CN112131411A (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
| CN112131472B (en) | Information recommendation method, device, electronic equipment and storage medium | |
| KR102155342B1 (en) | System for providing multi-parameter analysis based commercial service using influencer matching to company | |
| CN108959323B (en) | Video classification method and device | |
| Bairavel et al. | Novel OGBEE-based feature selection and feature-level fusion with MLP neural network for social media multimodal sentiment analysis | |
| Salim et al. | Data analytics of social media 3.0: Privacy protection perspectives for integrating social media and Internet of Things (SM-IoT) systems | |
| Manoharan et al. | An Intelligent Fuzzy Rule‐Based Personalized News Recommendation Using Social Media Mining | |
| Jain et al. | Tweet recommender model using adaptive neuro-fuzzy inference system | |
| Balduini et al. | Recommending venues using continuous predictive social media analytics | |
| Farseev et al. | " 360° user profiling: past, future, and applications" by Aleksandr Farseev, Mohammad Akbari, Ivan Samborskii and Tat-Seng Chua with Martin Vesely as coordinator | |
| Ravi et al. | An intelligent location recommender system utilising multi-agent induced cognitive behavioural model | |
| Xiang et al. | Demographic attribute inference from social multimedia behaviors: a cross-OSN approach | |
| Kumar et al. | Session-based recommendations with sequential context using attention-driven LSTM | |
| Matsumoto et al. | Music video recommendation based on link prediction considering local and global structures of a network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |