CN117150141A - Method, device, equipment, medium and program product for determining recommended content - Google Patents
Method, device, equipment, medium and program product for determining recommended content Download PDFInfo
- Publication number
- CN117150141A CN117150141A CN202311196168.XA CN202311196168A CN117150141A CN 117150141 A CN117150141 A CN 117150141A CN 202311196168 A CN202311196168 A CN 202311196168A CN 117150141 A CN117150141 A CN 117150141A
- Authority
- CN
- China
- Prior art keywords
- recommended content
- candidate
- recommended
- content
- contents
- 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
- 238000000034 method Methods 0.000 title claims abstract description 85
- 230000004931 aggregating effect Effects 0.000 claims abstract description 26
- 238000004422 calculation algorithm Methods 0.000 claims description 31
- 238000004364 calculation method Methods 0.000 claims description 26
- 238000003860 storage Methods 0.000 claims description 18
- 230000002776 aggregation Effects 0.000 claims description 13
- 238000004220 aggregation Methods 0.000 claims description 13
- 230000015654 memory Effects 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 8
- 230000004044 response Effects 0.000 claims description 7
- 239000006185 dispersion Substances 0.000 claims description 5
- 238000009432 framing Methods 0.000 claims description 5
- 239000002699 waste material Substances 0.000 abstract description 8
- 230000000875 corresponding effect Effects 0.000 description 118
- 238000005516 engineering process Methods 0.000 description 27
- 238000013473 artificial intelligence Methods 0.000 description 20
- 238000012545 processing Methods 0.000 description 15
- 238000013528 artificial neural network Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000006243 chemical reaction Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 230000006399 behavior Effects 0.000 description 5
- 239000000463 material Substances 0.000 description 5
- 238000012216 screening Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 239000002537 cosmetic Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The application discloses a method, a device, equipment, a medium and a program product for determining recommended content, and relates to the field of content recommendation. The method comprises the following steps: acquiring a plurality of candidate recommended contents; aggregating the plurality of candidate recommended contents to obtain n first recommended content sets; based on the prediction recommendation results of the first recommendation model on the plurality of candidate recommendation contents, determining first coefficients respectively corresponding to the n first recommendation content sets; selectively discarding candidate recommended contents in the n first recommended content sets based on the n first coefficients to obtain n second recommended content sets; and determining the first recommended content displayed for the first account from the n second recommended content sets. The final recommendation content set participating in the decision is a set discarded by the candidate recommendation content, so that the computational waste of the candidate recommendation content is reduced when the recommendation content displayed to the first account is determined.
Description
Technical Field
The embodiment of the application relates to the field of content recommendation, in particular to a method, a device, equipment, a medium and a program product for determining recommended content.
Background
With the development of computer technology and networks, more and more advertisers select to put advertisements on line, and the advertisements to be put need to go through several main matching screening stages of recall, coarse arrangement and fine arrangement of an advertisement recommendation system, and finally the most suitable advertisements are selected for putting.
In the related art, for each advertisement display request of a user, a certain number of candidate advertisements are corresponding to recall, coarse ranking and fine ranking stages, and the more the candidate advertisements are, the larger the calculated amount of an advertisement recommendation system is, and the more accurate the recommendation effect is.
However, there are more or less repeat or redundant advertisements for recalled candidate advertisements, such as: repeated advertisements may exist in the plurality of candidate advertisements corresponding to the coarse-ranking stage, and repeated calculation of the advertisement recommendation system on the same advertisement is wasteful or unnecessary to some extent, wasting computational resources in the advertisement recommendation process.
Disclosure of Invention
The embodiment of the application provides a method, a device, equipment, a medium and a program product for determining recommended content, which can reduce the calculation power waste aiming at the recommended content in the process of determining the recommended content, and the technical scheme is as follows:
In one aspect, a method for determining recommended content is provided, the method comprising:
acquiring a plurality of candidate recommended contents, wherein the plurality of candidate recommended contents are recommended contents to be displayed to a first account;
aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the candidate recommended contents in the ith first recommended content set meet the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n;
determining first coefficients corresponding to the n first recommended content sets respectively based on a predicted recommendation result of a first recommendation model on the plurality of candidate recommended content, wherein the predicted recommendation result refers to recommendation probability of the candidate recommended content to the first account, and the ith first coefficient is used for representing prediction stability of the first recommendation model on the candidate recommended content in the ith first recommended content set;
selectively discarding candidate recommended contents in the n first recommended content sets based on n first coefficients to obtain n second recommended content sets;
and determining the first recommended content displayed for the first account from the n second recommended content sets.
In another aspect, there is provided a determining apparatus of recommended content, the apparatus including:
The data acquisition module is used for acquiring a plurality of candidate recommended contents, wherein the plurality of candidate recommended contents are recommended contents to be displayed to the first account;
the content aggregation module is used for aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the candidate recommended contents in the ith first recommended content set meet the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n;
the first determining module is used for determining first coefficients corresponding to the n first recommended content sets respectively based on the predicted recommendation results of the first recommended model on the plurality of candidate recommended content, wherein the predicted recommendation results refer to recommendation probabilities of recommending the candidate recommended content to the first account, and the ith first coefficient is used for representing the predicted stability of the first recommended model on the candidate recommended content in the ith first recommended content set;
the content discarding module is used for selectively discarding the candidate recommended content in the n first recommended content sets based on n first coefficients to obtain n second recommended content sets;
and the second determining module is used for determining the first recommended content displayed for the first account from the n second recommended content sets.
In another aspect, a computer device is provided, the computer device including a processor and a memory, where the memory stores at least one instruction, at least one program, a set of codes, or a set of instructions, the at least one instruction, the at least one program, the set of codes, or the set of instructions being loaded and executed by the processor to implement a method for determining recommended content according to any of the embodiments described above.
In another aspect, a computer readable storage medium is provided, in which at least one instruction, at least one program, a set of codes, or a set of instructions is stored, the at least one instruction, the at least one program, the set of codes, or the set of instructions being loaded and executed by a processor to implement a method for determining recommended content as described in any of the above embodiments.
In another aspect, a computer program product or computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the recommended content determination method according to any one of the above embodiments.
The technical scheme provided by the embodiment of the application has the beneficial effects that at least:
after a plurality of candidate recommended contents are obtained, aggregating the candidate recommended contents meeting the similarity requirement to obtain a recommended content set; secondly, determining a first coefficient of a recommended content set according to a predicted recommendation result (i.e. recommendation probability of recommending to a first account) of the plurality of candidate recommended contents by the recommendation model; and finally, selectively discarding the candidate recommended content in the recommended content set according to the first coefficient, and determining the recommended content finally displayed to the first account from the discarded recommended content set. On the one hand, the recommended content set finally participating in decision making is a set discarded by candidate recommended content, so that when the recommended content displayed to the first account is determined, the computational waste for the candidate recommended content is reduced; on the other hand, the first coefficient is used for representing the prediction stability of the first recommendation model on a plurality of candidate recommendation contents meeting the similarity requirement in the recommendation content set, the stability of the recommendation model is introduced as a reference standard, and the number of candidate recommendation contents to be discarded in the recommendation content set is measured, so that the discarding number can be dynamically adjusted according to the stability of the recommendation model, and the accuracy of the determined discarding number is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic illustration of an implementation environment provided by an exemplary embodiment of the present application;
FIG. 2 is a flow chart of a method for determining recommended content provided by an exemplary embodiment of the application;
FIG. 3 is a flow chart of a method for determining recommended content provided by another exemplary embodiment of the application;
FIG. 4 is a flow chart of a method for determining recommended content provided by yet another exemplary embodiment of the application;
FIG. 5 is a schematic diagram of a process for discarding a fine candidate recommended content queue according to one exemplary embodiment of the application;
FIG. 6 is a schematic diagram of a content presentation interface provided by an exemplary embodiment of the present application;
FIG. 7 is a block diagram of a recommendation content determining apparatus provided in an exemplary embodiment of the present application;
FIG. 8 is a block diagram of a determining apparatus of recommended content according to another exemplary embodiment of the present application;
Fig. 9 is a block diagram of a computer device according to an exemplary embodiment of the present application.
Detailed Description
For the purpose of promoting an understanding of the principles and advantages of the application, reference will now be made in detail to the embodiments of the application, some but not all of which are illustrated in the accompanying drawings. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The terms "first," "second," and the like in this disclosure are used for distinguishing between similar elements or items having substantially the same function and function, and it should be understood that there is no logical or chronological dependency between the terms "first," "second," and no limitation on the amount or order of execution.
First, a brief description will be given of terms involved in the embodiments of the present application.
Artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.
The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include, for example, sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, pre-training model technologies, operation/interaction systems, mechatronics, and the like. The pre-training model is also called a large model and a basic model, and can be widely applied to all large-direction downstream tasks of artificial intelligence after fine adjustment. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.
Machine Learning (ML) is a multi-domain interdisciplinary, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like. The pre-training model is the latest development result of deep learning, and integrates the technology.
With research and advancement of artificial intelligence technology, research and application of artificial intelligence technology is being developed in various fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned, autopilot, unmanned, digital twin, virtual, robotic, artificial intelligence generation content, conversational interactions, smart medicine, smart customer service, game AI, etc., and it is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.
Many common applications, such as: advertisement delivery is available in news applications, instant messaging applications, and the like. The advertisements in the application programs generally go through the main matching screening stages of recall, coarse ranking and fine ranking of the back advertisement recommendation system, and finally the advertisements displayed in the application programs are the most suitable advertisements in the recalled advertisements.
For a display request of a certain user, a certain number of candidate advertisements (namely candidate queues) are received in the stages of recall, coarse ranking, fine ranking and the like, and the more candidate advertisements in the candidate queues, the larger the calculated amount of an advertisement recommendation system is, and the more accurate the recommendation effect is. In practice, there may be more or less duplication or redundancy of candidate advertisements in the candidate queue recalled from the advertisement library per user request. The main reasons include that different advertisers may create the same advertising material; multiple dimensions of diversity are guaranteed during recall, including advertiser dimensions. If there cannot be duplicate candidate advertisements in the candidate queue, the diversity of the advertiser may have an impact, affecting the bid ecology. The presence of duplicate candidate advertisements within the candidate queue is an objectively existing fact in the current advertisement recommendation system.
From the user's perspective, the same advertising material is identical to the user's senses, and thus the user's behavior can be considered identical on the same advertising material. That is, the repeated computation of the advertisement recommendation system on the same advertisement is somewhat wasteful or unnecessary, wasting computational resources in the advertisement recommendation process.
The embodiment of the application provides a method for determining recommended content (such as advertisement), which dynamically adjusts the length of a candidate queue from the stability of a recommendation algorithm corresponding to a recommendation system so as to reduce the calculated amount of repeated materials and avoid the waste of calculation power of the recommendation system. The method for determining the recommended content comprises at least one of a plurality of scenes such as a music recommended scene, a news recommended scene, a video recommended scene and the like when the method is applied. It should be noted that the above application scenario is merely an illustrative example, and the method for determining recommended content provided in this embodiment may also be applied to other scenarios, which is not limited in this embodiment of the present application.
Secondly, an implementation environment related to the embodiment of the present application is described, and the method for determining recommended content provided in the embodiment of the present application may be implemented by the terminal alone, or may be implemented by the server, or implemented by the terminal and the server through data interaction, which is not limited in the embodiment of the present application. Alternatively, a method of determining recommended content by the terminal and the server being interactively performed will be described as an example.
Referring to fig. 1, the implementation environment relates to a terminal 110 and a server 120, and the terminal 110 and the server 120 are connected through a communication network 130. The communication network 130 may be implemented as a wired network or a wireless network, which is not limited in this embodiment of the present application.
In some embodiments, an application program having a function of determining recommended content is installed in the terminal 110. The application may be implemented as an instant messaging application, a video application, a news information application, a comprehensive search engine application, a social application, a gaming application, a shopping application, a map navigation application, etc., which is not limited in this regard.
Optionally, the terminal 110 is configured to send a plurality of candidate recommended contents to the server 120, where the plurality of candidate recommended contents are recommended contents to be displayed to a first account, and the first account is an account logged in an application program installed and running in the terminal 110.
After receiving the plurality of candidate recommended contents 121, the server 120 first aggregates the plurality of candidate recommended contents 121 to obtain n first recommended content sets 122, where candidate recommended contents included in each recommended content set meet a preset similarity requirement; then, n first coefficients 124 corresponding to the n first recommended content sets 122 are determined according to a plurality of predicted recommended results 123 of the first recommended model set in the server 120 for the plurality of candidate recommended contents 121; after obtaining the n first coefficients 124, the server 120 selectively discards candidate recommended content in the n first recommended content sets 122 based on the n first coefficients 124, to obtain n second recommended content sets 125; finally, the server 120 determines a first recommended content 126 to be presented to the first account from the n second recommended content sets 125.
In some embodiments, after the server 120 determines the first recommended content 126, the first recommended content 126 is sent to the terminal 110, and optionally, the terminal 110 will display the first recommended content 126 in the application program with the first account login.
The terminal 110 includes, but is not limited to, mobile terminals such as mobile phones, tablet computers, portable laptop computers, intelligent voice interaction devices, intelligent home appliances, vehicle terminals, and the like, and can also be implemented as desktop computers, and the like.
It should be noted that the server 120 can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and can also be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs (Content Delivery Network, content delivery networks), and basic cloud computing services such as big data and artificial intelligence platforms.
Cloud Technology (Cloud Technology) refers to a hosting Technology that unifies serial resources such as hardware, software, network and the like in a wide area network or a local area network to realize calculation, storage, processing and sharing of data. The cloud technology is based on the general names of network technology, information technology, integration technology, management platform technology, application technology and the like applied by the cloud computing business model, can form a resource pool, and is flexible and convenient as required. Cloud computing technology will become an important support. Background services of technical networking systems require a large amount of computing, storage resources, such as video websites, picture-like websites, and more portals. Along with the high development and application of the internet industry, each article possibly has an own identification mark in the future, the identification mark needs to be transmitted to a background system for logic processing, data with different levels can be processed separately, and various industry data needs strong system rear shield support and can be realized only through cloud computing. Optionally, the server 120 may also be implemented as a node in a blockchain system.
It should be noted that, in the present application, before collecting relevant data (such as candidate recommended content) of a user and during collecting relevant data of a user, a prompt interface, a popup window or output voice prompt information may be displayed, where the prompt interface, the popup window or the voice prompt information is used to prompt the user to collect relevant data currently, so that the present application only starts to execute the relevant step of obtaining relevant data of the user after obtaining the confirmation operation of the user to the prompt interface or the popup window, otherwise (i.e. when the confirmation operation of the user to the prompt interface or the popup window is not obtained), ends the relevant step of obtaining relevant data of the user, i.e. does not obtain relevant data of the user. In other words, all user data collected by the present application is collected with the user agreeing and authorized, and the collection, use and processing of relevant user data requires compliance with relevant laws and regulations and standards.
In connection with the description and the implementation environment, the method for determining recommended content provided by the present application will be described, and the method is applied to a server as an example, and as shown in fig. 2, the method includes the following steps 210 to 250.
Step 210, obtaining a plurality of candidate recommended content.
The plurality of candidate recommended contents are recommended contents to be displayed to the first account.
Illustratively, the plurality of candidate recommended content may be implemented as news content, musical compositions, video content, blog content, novel content, physical products, virtual products, etc., to be recommended, which the embodiments of the present application are not limited to.
Optionally, the data types corresponding to the plurality of candidate recommended content include at least one of text type, video type, audio type, picture type, and the like.
In some embodiments, the candidate recommended content is content delivered by a content delivery person; alternatively, the candidate recommended content is content automatically generated by artificial intelligence techniques, such as: and automatically generating a recommended content file and picture content related to the recommended content file according to the search content of the first account.
Optionally, the plurality of candidate recommended contents include recommended contents whose similarity meets the similarity requirement. Illustratively, the plurality of candidate recommended contents comprise candidate recommended content 1 and candidate recommended content 2, and the candidate recommended content 1 and the candidate recommended content 2 belong to the same recommended content put by different content putting personnel.
In some embodiments, a plurality of first candidate recommended content in a first queue is obtained; respectively predicting recommendation probability of a plurality of first candidate recommendation contents in a first queue through a first recommendation model to obtain prediction recommendation probability corresponding to the plurality of first candidate recommendation contents in the first queue; and screening the plurality of first candidate recommended contents in the first queue according to the prediction recommendation probabilities respectively corresponding to the plurality of first candidate recommended contents in the first queue to obtain a second queue, and taking the plurality of first candidate recommended contents contained in the second queue as the plurality of candidate recommended contents.
Optionally, assuming that the number of the plurality of candidate recommended contents is k, where k is a positive integer greater than 1, the first candidate recommended content with the largest k in the predicted recommendation probability is added to the second queue.
In some embodiments, stages such as recall, coarse ranking, fine ranking stages are required before determining recommended content to be presented to the first account. Illustratively, in the recall stage, a part of candidate recommended contents are screened out from a massive recommended content library according to the account information of the first account to form a coarse-ranking candidate recommended content queue, and the coarse-ranking stage is entered; in the coarse ranking stage, a part of better candidate recommended content is further screened out from the coarse ranking candidate recommended content queue according to the account information of the first account to form a fine ranking candidate recommended content queue, and the fine ranking stage is carried out; and in the fine-ranking stage, the recommended content finally displayed to the first account is further screened out from the fine-ranking candidate recommended content queue according to the account information of the first account.
Optionally, the plurality of candidate recommended contents are candidate recommended contents in a coarse-ranking candidate recommended content queue, wherein the coarse-ranking candidate recommended content queue refers to a candidate recommended content queue recalled from a recall stage and entering the coarse-ranking stage; alternatively, the plurality of candidate recommended contents are candidate recommended contents in a fine-ranking candidate recommended content queue, wherein the fine-ranking candidate recommended content queue refers to a candidate recommended content queue which enters a fine-ranking stage and is screened from a coarse-ranking stage.
Step 220, aggregating the plurality of candidate recommended contents to obtain n first recommended content sets.
The candidate recommended content in the ith first recommended content set meets the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n.
Illustratively, aggregating the plurality of candidate recommended contents refers to aggregating the plurality of candidate recommended contents according to the similarity, so that the similarity of the candidate recommended contents in the same recommended content set meets the similarity requirement.
Optionally, the preset similarity requirement is that the similarity between the candidate recommended contents is greater than or equal to the preset similarity, wherein the candidate recommended contents in the same recommended content set are the same when the preset similarity is 100%.
Optionally, the method of aggregating the plurality of candidate recommended content includes at least one of:
1. and carrying out aggregation according to the hash value corresponding to the candidate recommended content.
In some embodiments, performing hash computation on the plurality of candidate recommended contents according to a preset hash algorithm to obtain first hash values respectively corresponding to the plurality of candidate recommended contents; based on the first hash value, aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the similarity of the first hash value corresponding to the candidate recommended content in the ith first recommended content set is greater than or equal to a first preset similarity.
Alternatively, the preset hash algorithm may be implemented as at least one of a secure hash algorithm (Secure Hash Algorithm, SHA 1), a fifth version of a message digest algorithm (Message Digest algorithm, md5), etc. the hash value calculation algorithm is not limited in this embodiment of the present application.
Schematically, if the candidate recommended content is implemented as a text type, calculating a hash value of text data corresponding to the candidate recommended content as a first hash value; if the candidate recommended content is realized as the video type, calculating a hash value of video data corresponding to the candidate recommended content as a first hash value; if the candidate recommended content is implemented as an audio type, a hash value of audio data (e.g., a plurality of audio frames) corresponding to the candidate recommended content is calculated as a first hash value.
Optionally, the data types in the candidate recommended content include at least two of text data, audio data, picture data, video data and the like, and hash values respectively corresponding to the candidate recommended content of each data type are calculated; after hash values corresponding to candidate recommended contents of all data types are obtained, connecting the calculated hash values according to a preset data type sequence, and obtaining the connected hash values as first hash values. For example: the candidate recommended content comprises text data and audio data, text data hash values of the text data in the candidate recommended content are calculated, audio data hash values of the audio data in the candidate recommended content are calculated, and the text data hash values and the audio data hash values are connected according to a preset data type sequence (text type-audio type) to obtain a first hash value.
After hash values corresponding to the candidate recommended contents are calculated, the candidate recommended contents are aggregated, for example: the plurality of candidate recommended contents comprise candidate recommended content 1, candidate recommended content 2, candidate recommended content 3 and candidate recommended content 4, wherein the hash value of candidate recommended content 1 is hash value a, the hash value of candidate recommended content 2 is hash value b, the hash value of candidate recommended content 3 is hash value c, and the hash value of candidate recommended content 4 is hash value d. The similarity between the hash value a and the hash value b is greater than or equal to a first preset similarity, and the similarity between the hash value c and the hash value d is greater than or equal to the first preset similarity, so that the candidate recommended content 1 and the candidate recommended content 2 form a first recommended content set, and the candidate recommended content 3 and the candidate recommended content 4 form a first recommended content set.
Optionally, the plurality of candidate recommended contents respectively correspond to the recommended body content and the presentation page, and the method for calculating the first hash value further includes:
carrying out hash calculation on the recommended subject content corresponding to each of the plurality of candidate recommended contents according to a preset hash algorithm to obtain second hash values of the recommended subject content corresponding to each of the plurality of candidate recommended contents; carrying out hash calculation on the display pages corresponding to the candidate recommended contents respectively according to a preset hash algorithm to obtain third hash values of the display pages corresponding to the candidate recommended contents respectively; and fusing the second hash value and the third hash value to obtain a first hash value.
The recommended main content is, for example, a video to be recommended, a picture to be recommended, a text to be recommended, and the like. The display page is used for displaying the recommended main content, and illustratively, the display page can be implemented as a landing page for displaying the recommended main content.
Schematically, if the recommended main content is realized as a video and the display page is realized as a landing page for displaying the video, firstly calculating a hash value of the video according to a preset hash algorithm to serve as a second hash value, and then calculating a hash value of a landing picture corresponding to the landing page according to the preset hash algorithm to serve as a third hash value; and finally, connecting the second hash value and the third hash value according to a first preset sequence to obtain a first hash value.
In some embodiments, if the candidate recommended content is implemented as video data, then the video in the candidate recommended content may be framed to obtain multiple video frames and then hash calculated.
Optionally, the plurality of candidate recommended content includes the first video content. The method for obtaining the first hash value corresponding to the first video content further comprises the following steps:
framing the first video content to obtain a plurality of video frames; sub-hash values respectively corresponding to the plurality of video frames are obtained through a preset hash algorithm, and the sub-hash values respectively corresponding to the plurality of video frames are connected according to a second preset sequence to obtain a first hash value.
Or, framing the first video content to obtain a plurality of video frames; determining a key video frame from a plurality of video frames; and carrying out hash calculation on the key video frames according to a preset hash algorithm to obtain hash values of the key video frames as first hash values corresponding to the first video content.
The key video frames may be video frames corresponding to a certain preset time point in the first video content, for example: a first video frame in the first video content, an intermediate video frame in the first video content, a last video frame in the first video content, etc.
Alternatively, the key video frames are video frames containing a recommendation subject (e.g., a product to be recommended, etc.) corresponding to the first video content.
Alternatively, the number of key video frames may be one or more, for example: the key video frames comprise a first video frame and a last video frame in the first video content, hash values of the first video frame and the last video frame are calculated respectively, and the calculated hash values are connected to obtain a first hash value corresponding to the first video content.
In some embodiments, the recommended body content in the candidate recommended content is implemented as video data, and then the video in the recommended body content may be framed to obtain a plurality of video frames; sub-hash values respectively corresponding to the plurality of video frames are obtained through a preset hash algorithm, the sub-hash values respectively corresponding to the plurality of video frames are connected according to a second preset sequence to obtain a second hash value, and the second hash value and a third hash value are connected according to a first preset sequence to obtain a first hash value. Or framing the video in the recommended main content to obtain a plurality of video frames; determining a key video frame from a plurality of video frames; and carrying out hash calculation on the key video frames according to a preset hash algorithm to obtain hash values of the key video frames as second hash values, and connecting the second hash values with third hash values according to a first preset sequence to obtain first hash values.
The preset hash algorithm for calculating the first hash value, the second hash value, and the third hash value may be the same or different.
2. And aggregating according to the recommended content types corresponding to the candidate recommended content.
Optionally, the recommended content type refers to a type of a recommended content body (e.g., recommended video, recommended merchandise, etc.) in the candidate recommended content, and the recommended content type includes at least one of an industry type, a product type, a function type, etc. of the recommended content body, which is not limited in the embodiment of the present application.
Illustratively, the industry type of the recommended content body is as follows: finance industry, media industry, retail industry, travel industry, and the like; the product type of the recommended content body is as follows: digital electronic products, cosmetics, clothing, shoes, caps, etc.
In some embodiments, obtaining a recommended content type corresponding to each of the plurality of candidate recommended contents; based on the recommended content types, aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the similarity of the recommended content types corresponding to the candidate recommended contents in the ith first recommended content set is greater than or equal to a second preset similarity.
In other embodiments, aggregation is performed based on standardized product units (Standard Product Unit, SPU) to which candidate recommended content belongs.
Wherein SPU is the minimum unit of commodity information aggregation, and is a set of reusable and easily retrievable standardized information that describes the characteristics of a product. In popular terms, commodities with the same attribute value and characteristic can be called an SPU.
Optionally, obtaining standardized product units corresponding to the candidate recommended contents respectively; and based on the standardized product unit, aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the candidate recommended contents in the ith first recommended content set belong to the same standardized product unit.
It should be noted that the foregoing examples of the manner of aggregating candidate recommended content are merely illustrative, and the embodiments of the present application are not limited thereto.
Step 230, determining first coefficients corresponding to the n first recommended content sets respectively based on the predicted recommendation results of the first recommendation model on the plurality of candidate recommended content sets.
The prediction recommendation result refers to recommendation probability of recommending the candidate recommendation content to the first account.
Optionally, the first recommendation model includes at least one of a click rate prediction model and a conversion rate prediction model, which are schematic, the click rate prediction model is used for predicting click probability of the candidate recommendation content by the first account, and determining recommendation probability of recommending the candidate recommendation content to the first account by the click probability, wherein the click probability and the prediction probability are in positive correlation; the conversion rate prediction model is used for predicting conversion probability of preset behaviors (such as purchase, watching, subscribing, registering and the like) of the first account for the candidate recommended content, and determining recommendation probability of recommending the candidate recommended content to the first account according to the conversion probability, wherein the conversion probability and the prediction probability are in positive correlation.
Illustratively, the first recommendation model may be implemented as: multilayer perceptrons (Multilayer Perceptron, MLP), deep neural networks (Deep Neural Network, DNN), recurrent neural networks (Recurrent Neural Network, RNN), long Short-Term Memory (LSTM), convolutional neural networks (Convolutional Neural Network, CNN), etc., as embodiments of the application are not limited in this regard.
The ith first coefficient is used for representing the prediction stability of the first recommendation model on candidate recommendation contents in the ith first recommendation content set.
Illustratively, the prediction stability of the first recommendation model for the candidate recommended content in the first recommendation content set refers to the difference between the prediction results of the first recommendation model for the plurality of candidate recommended contents meeting the similarity requirement. If the first recommendation model has small change of the prediction results of the plurality of candidate recommendation contents meeting the similarity requirement, the prediction stability of the model is higher; if the first recommendation model has larger change of the prediction results of the plurality of candidate recommendation contents meeting the similarity requirement, the prediction stability of the model is lower.
In some embodiments, obtaining recommendation probability predictions of the first recommendation model for the plurality of candidate recommended contents, and taking prediction recommendation probabilities respectively corresponding to the plurality of candidate recommended contents as prediction recommendation results; according to the prediction recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, determining the prediction deviation corresponding to the ith first recommended content set as a first coefficient, wherein the prediction deviation refers to the degree of dispersion of the prediction recommendation probabilities of the candidate recommended contents in the ith first recommended content set.
Schematically, in the coarse ranking stage, a coarse ranking candidate recommended content queue is obtained, recommendation probability prediction is performed on p candidate recommended contents in the coarse ranking candidate recommended content queue through a first recommendation model, p is a positive integer greater than 1, a fine ranking candidate recommended content queue formed by q candidate recommended contents is obtained, q is a positive integer greater than 1, the q candidate recommended contents are the candidate recommended contents, and the prediction recommendation probability of the q candidate recommended contents in the coarse ranking stage, which is obtained through the first recommendation model, is the prediction recommendation probability corresponding to the candidate recommended contents respectively.
Alternatively, the above prediction bias may be implemented as at least one of the following:
1. the prediction bias includes a variance.
Optionally, according to the predicted recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, determining the variance corresponding to the ith first recommended content set as the first coefficient.
The variance measures the degree of difference between the data point and its average, and the variance is positively correlated with the degree of dispersion of the predicted recommended probability.
2. The prediction bias comprises a standard deviation.
Optionally, according to the predicted recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, determining the standard deviation corresponding to the ith first recommended content set as the first coefficient.
The standard deviation is the square root of the variance, and the larger the standard deviation is, the positive correlation is formed between the standard deviation and the discrete degree of the prediction recommendation probability.
3. The prediction bias includes a coefficient of variation.
Optionally, according to the predicted recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, determining a variation coefficient corresponding to the ith first recommended content set as the first coefficient.
Coefficient of variation (C) v ) The standard deviation is the ratio of the average value, usually expressed in percentage, and the variation coefficient is used for comparing the discrete degree of different data sets, so that the data of different units or different orders can be compared, and the variation coefficient and the discrete degree of the prediction recommendation probability are in positive correlation. The formula for calculating the coefficient of variation is shown in the following formula one:
Equation one:
wherein sigma is the standard deviation of the prediction recommendation probability of the candidate recommended content in the same first recommended content set, and mu is the average value of the prediction recommendation probability of the candidate recommended content in the same first recommended content set. The degree of dispersion is measured by using a variation coefficient, and natural class dimensions in a recommendation system are mainly considered, for example: the average click rate of the advertisement in the industry A is 10%, the estimated value of the advertisement in the industry A is about 10%, the average click rate of the advertisement in the industry B is 1%, the estimated value of the advertisement in the industry B is about 1%, and the influence of dimension can be avoided by adopting the variation coefficient.
And step 240, selectively discarding candidate recommended content in the n first recommended content sets based on the n first coefficients to obtain n second recommended content sets.
Schematically, candidate recommended contents in each first recommended content set are selectively discarded according to the first coefficient corresponding to each first recommended content set, namely, the stability index of the model.
Optionally, the first coefficient indicates that the stability of the first recommendation model is higher, and the first coefficient indicates that the number of candidate recommended content selectively discarded is greater; the first coefficient indicates that the stability of the first recommendation model is low, and the first coefficient indicates that the smaller the number of candidate recommended content that is selectively discarded.
In some embodiments, determining discard proportions corresponding to the n first recommended content sets respectively based on n first coefficients, where the first coefficients are in negative correlation with the discard proportions; and selectively discarding the candidate recommended content in the ith first recommended content set according to the discarding proportion corresponding to the ith first recommended content set.
Optionally, the different first coefficients correspond to different reject ratios. Illustratively, assuming that the first coefficient is 80%, the reject ratio is 20%; assuming that the first coefficient is 20%, the reject ratio is 80%.
Optionally, determining a rejection ratio corresponding to the ith first recommended content set based on the ith first coefficient; determining set weights corresponding to the ith first recommended content set according to the predicted recommendation probability of the first recommended model on the candidate recommended content in the ith first recommended content set; and selectively discarding the candidate recommended content in the ith first recommended content set according to the discarding proportion and the set weight corresponding to the ith first recommended content set.
Wherein the aggregate weight is inversely related to the number of selectively rejected candidate recommended content.
Optionally, determining a set weight corresponding to the ith first recommended content set according to an average value of the predicted recommendation probability of the candidate recommended content in the ith first recommended content set by the first recommendation model, wherein the average value and the set weight are in a positive correlation; or determining a set weight corresponding to the ith first recommended content set according to the maximum value of the predicted recommendation probability of the candidate recommended content in the ith first recommended content set by the first recommendation model, wherein the maximum value and the set weight are in positive correlation; and determining a set weight corresponding to the ith first recommended content set according to the minimum value of the predicted recommendation probability of the first recommended model on the candidate recommended content in the ith first recommended content set, wherein the minimum value and the set weight are in positive correlation.
Illustratively, assuming that the first coefficient is 80%, the reject ratio is 20%; if the set weight is 0.3, if the candidate recommended content in the set is 10, discarding 10×20% × (1-0.3) ≡2 (rounding up) candidate recommended contents; when the set weight is 0.7, if the candidate recommended content in the set is 10, 10×20% × (1-0.7) ≡1 (rounded up) candidate recommended content is discarded.
Step 250, determining a first recommended content displayed to the first account from the n second recommended content sets.
Optionally, after determining the n second recommended content sets, the candidate recommended content in the n second recommended content sets may be recombined into a candidate recommended content queue, a second recommendation model is input into the candidate recommended content queue, the recommendation probability of each candidate recommended content in the candidate recommended content queue for recommending to the first account is predicted, and the first recommended content displayed to the first account is determined according to the recommendation probability.
Alternatively, the number of the first recommended content may be one or a plurality.
Illustratively, assuming that the number of first recommended contents is one, the candidate recommended content with the highest recommendation probability is taken as the first recommended content from the candidate recommended content queue; assuming that the number of the first recommended contents is k, k is a positive integer greater than 1, k candidate recommended contents with the highest recommendation probability are taken as the first recommended contents from the candidate recommended content queues.
Wherein the second recommendation model includes at least one of a click rate prediction model, a conversion rate prediction model, and the like. Illustratively, the second recommendation model may be implemented as: multilayer perceptrons (Multilayer Perceptron, MLP), deep neural networks (Deep Neural Network, DNN), recurrent neural networks (Recurrent Neural Network, RNN), long Short-Term Memory (LSTM), convolutional neural networks (Convolutional Neural Network, CNN), etc., as embodiments of the application are not limited in this regard.
Illustratively, taking the example that the plurality of candidate recommended contents belong to the fine-ranking candidate recommended content queue as an illustration, the first recommendation model is used for predicting to obtain the fine-ranking candidate recommended content queue; the second recommendation model is used for predicting the recommendation probability of the selectively discarded fine candidate recommendation content queue to the first account.
In summary, according to the method for determining recommended content provided by the embodiment of the application, after a plurality of candidate recommended contents are obtained, the candidate recommended contents meeting the similarity requirement are aggregated, and a recommended content set is obtained; secondly, determining a first coefficient of a recommended content set according to a predicted recommendation result (i.e. recommendation probability of recommending to a first account) of the plurality of candidate recommended contents by the recommendation model; and finally, selectively discarding the candidate recommended content in the recommended content set according to the first coefficient, and determining the recommended content finally displayed to the first account from the discarded recommended content set. On the one hand, the recommended content set finally participating in decision making is a set discarded by candidate recommended content, so that when the recommended content displayed to the first account is determined, the computational waste for the candidate recommended content is reduced; on the other hand, the first coefficient is used for representing the prediction stability of the first recommendation model on a plurality of candidate recommendation contents meeting the similarity requirement in the recommendation content set, the stability of the recommendation model is introduced as a reference standard, and the number of candidate recommendation contents to be discarded in the recommendation content set is measured, so that the discarding number can be dynamically adjusted according to the stability of the recommendation model, and the accuracy of the determined discarding number is improved.
According to the method provided by the embodiment of the application, the stability of the first recommendation model is measured through the discrete degree of the predicted recommendation probability of the candidate recommended content in the first recommendation content set, so that the first coefficient for discarding the candidate recommended content in the first recommendation content set is determined, the discarding behavior is associated with the stability of the model, and the rationality of the determined discarding number of the candidate recommended content is improved.
According to the method provided by the embodiment of the application, the index for measuring the discrete degree of the prediction recommendation probability of the candidate recommended content in the first recommended content set is realized as the variation coefficient, so that the influence of dimension is avoided, and the accuracy of the determined first coefficient is improved.
According to the method provided by the embodiment of the application, the rejection proportion of the candidate recommended content in the first recommended content set and the first coefficient are in a negative correlation relationship, namely, the prediction stability of the model on the data in a certain set is higher, more candidate recommended content discarded in the set is obtained, if the prediction stability of the model on the data in a certain set is lower, fewer candidate recommended content discarded in the set is obtained, and the rejection quantity of the candidate recommended content corresponding to each set is more reasonable.
According to the method provided by the embodiment of the application, after the hash values of the plurality of candidate recommended contents are calculated, the plurality of candidate recommended contents are aggregated based on the hash values, and the plurality of candidate recommended contents with similar hash values are selectively discarded, so that the accuracy of the determined first candidate recommended content set is improved.
According to the method provided by the embodiment of the application, the hash values corresponding to the recommended main body content and the display page are fused to be used as the hash values corresponding to the candidate recommended content, and the recommended main body and the display page are bound and aggregated, so that the aggregated candidate recommended content has a certain similarity on the recommended main body and the display page, and the accuracy of selective discarding processing is improved.
According to the method provided by the embodiment of the application, when the candidate recommended content is realized as the video content, the key frame corresponding to the video content is extracted to calculate the hash value, and the hash value of the key frame is used as the first hash value for aggregation, so that the calculation resources in the process of calculating the first hash value corresponding to the video content are reduced.
According to the method provided by the embodiment of the application, the plurality of candidate recommended contents can be aggregated according to the recommended content types corresponding to the plurality of candidate recommended contents respectively, and the plurality of candidate recommended contents with similar recommended content types are selectively discarded, so that the accuracy of the determined first candidate recommended content set is improved.
In some embodiments, in determining the number of candidate recommended contents selectively discarded by the candidate recommended contents in each of the first recommended content sets, a lower discard number limit may be predetermined, and the number of candidate recommended contents discarded may be determined by the lower discard number limit. Illustratively, as shown in fig. 3, the embodiment shown in fig. 2 described above may also be implemented as steps 310 through 350, where step 240 may also be implemented as steps 341 through 346.
In step 310, a plurality of candidate recommended content is obtained.
The plurality of candidate recommended contents are recommended contents to be displayed to the first account.
Illustratively, the plurality of candidate recommended content may be implemented as news content, musical compositions, video content, blog content, novel content, etc., to be recommended, which the embodiments of the present application are not limited to.
Taking a plurality of candidate recommended contents as the contents in the fine candidate recommended content queue as an example for explanation:
illustratively, after receiving a content recommendation request for a first account, a terminal acquires account information corresponding to the first account and a coarse candidate recommendation content queue to be screened; the terminal sends account information corresponding to the first account and a rough candidate recommended content queue to be screened to a server, a first recommendation model is arranged in the server, the account information corresponding to the first account and the rough candidate recommended content queue to be screened are input into the first recommendation model, recommendation probability prediction of each candidate recommended content in the rough candidate recommended content queue can be obtained, and q candidate recommended content with highest recommendation probability can be screened from the rough candidate recommended content queue to form a fine candidate recommended content queue after the recommendation probability of each candidate recommended content is obtained. And the candidate recommended content in the fine-ranking candidate recommended content queue is the plurality of candidate recommended contents.
Step 320, aggregating the plurality of candidate recommended contents to obtain n first recommended content sets.
The candidate recommended content in the ith first recommended content set meets the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n.
Illustratively, aggregating the plurality of candidate recommended contents refers to aggregating the plurality of candidate recommended contents according to the similarity, so that the similarity of the candidate recommended contents in the same recommended content set meets the similarity requirement.
Step 330, determining first coefficients corresponding to the n first recommended content sets respectively based on the predicted recommendation results of the first recommendation model on the plurality of candidate recommended content sets.
The prediction recommendation result refers to recommendation probability of recommending the candidate recommended content to the first account, and the ith first coefficient is used for representing prediction stability of the first recommendation model on the candidate recommended content in the ith first recommended content set.
Illustratively, the prediction stability of the first recommendation model for the candidate recommended content in the first recommended content set refers to the difference between the prediction results of the first recommendation model for the plurality of candidate recommended contents meeting the similarity requirement.
Step 341, determining discard proportions corresponding to the n first recommended content sets respectively based on the n first coefficients.
Wherein the first coefficient is inversely related to the rejection ratio.
In some embodiments, the corresponding reject ratio for the first coefficient in different coefficient intervals is different. Optionally, the reject ratio corresponding to each of the plurality of first coefficient intervals and the plurality of first coefficient intervals is obtained, the plurality of first coefficient intervals are preset intervals, and the values of the first coefficients in different first coefficient intervals are different. And in response to the ith first coefficient belonging to the value range of the first coefficient section in the plurality of first coefficient sections, taking the rejection ratio corresponding to the first coefficient section as the rejection ratio corresponding to the ith first recommended content set.
Illustratively, assuming that the first coefficient is 80%, the first coefficient section is 70% to 100%, and the reject ratio corresponding to the first coefficient section is 20%, the reject ratio of the first recommended content set having the first coefficient of 80% is 20%.
In some embodiments, the plurality of discard proportions is a proportion determined from predicted recommendations of the first recommendation model over the historical time period. Optionally, the method of determining a plurality of reject ratios further comprises the steps of:
step one: the update time of the last first recommendation model is obtained.
Illustratively, the first recommendation model may be updated periodically or aperiodically, and the first recommendation model needs to be trained again (or understood as adjustment of model parameters) after updating. The update frequency of the first recommendation model can be determined according to requirements and resource conditions. The higher updating frequency can be more timely adapted to the interest change and the environment change of the user, and more accurate recommendation results are provided.
Step two: a time period of a preset duration is selected as a sampling time period between the update time and the current time.
Illustratively, assuming that the preset duration is 15 minutes before the last time the first recommendation model is updated is one hour, 15 minutes closest to the current time may be selected as the sampling period.
Step three: and acquiring m candidate recommended contents processed by the first recommendation model in the sampling time period and m prediction recommendation probabilities of the first recommendation model on the m candidate recommended contents.
m is a positive integer greater than 1.
Alternatively, the candidate recommended content processed by the first recommendation model may be content to be recommended to the first account; the candidate recommended content processed by the first recommendation model may also be content to be recommended to other accounts.
Optionally, the m candidate recommended contents are recommended contents input into the first recommendation model; or, the m candidate recommended contents are obtained by screening after the recommendation probability prediction is performed on the h candidate recommended contents by the first recommendation model, and h is a positive integer greater than 1.
Step four: and aggregating the m candidate recommended contents to obtain s first candidate recommended content sets.
s is a positive integer. Specific polymerization methods can refer to step 220, and are not described herein.
Step five: and determining the first coefficients corresponding to the s first candidate recommended content sets respectively, and sorting according to the first coefficients corresponding to the s first candidate recommended content sets respectively to obtain a first sorting.
Optionally, sorting according to the order of the first coefficients from small to large to obtain a first sorting; or, sorting is performed according to the order of the first coefficients from the big to the small to obtain a first sorting.
Step six: a plurality of first coefficient intervals is determined based on the first ordering.
Illustratively, the first coefficients are CV in small to large arrangements, respectively 1 ，CV 2 ，...，CV N . According to the service condition, dividing into 4 coefficient sections, wherein the threshold of the N/4,2 (N/4), 3 (N/4) section is that the first coefficient section is smaller than CV N/4 The second coefficient interval is equal to or greater than CV N/4 And is smaller than CV 2*(N/4) The third coefficient interval is equal to or greater than CV 2*(N/4) And is smaller than CV 3*(N/4) The fourth coefficient interval is equal to or greater than CV 3*(N/4) 。
Step seven: and determining reject ratios corresponding to the first coefficient intervals respectively.
Illustratively, each coefficient interval corresponds to a rejection ratio r1, r2, r3, r4, and the smaller the first coefficient, the closer the prediction recommendation probability is, and the higher the rejection ratio can be set.
In step 342, a first discard number is obtained.
The first reject number refers to a preset reject lower limit value.
Optionally, the first discard number is a discard lower limit value set by the developer.
Optionally, the first discard number is a discard lower limit value determined according to a predicted recommendation result of the first recommendation model over the historical period of time. Illustratively, based on step three of step 321:
assuming that the m candidate recommended contents are recommended contents input into the first recommendation model, the m candidate recommended contents may be implemented as candidate recommended contents in a coarse candidate recommended content queue, the m candidate recommended contents correspond to a plurality of coarse candidate recommended content queues, a first coarse candidate recommended content queue in the plurality of coarse candidate recommended content queues is obtained, the first coarse candidate recommended content queue refers to a coarse candidate recommended content queue including a similarity meeting a preset similarity, and a formula for calculating the first reject number d is shown as follows:
Formula II: d=c×v×r min
Where c is the average coarse ranking recommended content queue length for each candidate recommended content, v is the number of first coarse ranking candidate recommended content queues, r min Is the minimum reject ratio obtained in step seven in step 321.
The average thick-ranking recommended content queue length of each candidate recommended content refers to an average value of the queue length of the thick-ranking recommended content queue 1 and the queue length of the thick-ranking recommended content queue 2 assuming that the thick-ranking recommended content queue 1, the thick-ranking recommended content queue 2 and the thick-ranking recommended content queue 3 exist, wherein the thick-ranking recommended content queue 1 and the thick-ranking recommended content queue 2 contain the candidate recommended content a, and the average thick-ranking recommended content queue length corresponding to the candidate recommended content a is the average value of the queue length of the thick-ranking recommended content queue 1 and the queue length of the thick-ranking recommended content queue 2.
Assuming that m candidate recommended contents are contents obtained by screening h candidate recommended contents after recommendation probability prediction by a first recommendation model, the m candidate recommended contents can be implemented as candidate recommended contents in a fine candidate recommended content queue, the m candidate recommended contents correspond to a plurality of fine candidate recommended content queues, a first fine candidate recommended content queue in the plurality of fine candidate recommended content queues is obtained, the first fine candidate recommended content queue refers to a fine candidate recommended content queue containing similarity which accords with preset similarity, and the calculation of the first rejection number d can refer to the formula II.
Wherein c in the formula II is the average fine-ranking recommended content queue length of each candidate recommended content, v is the number of first fine-ranking candidate recommended content queues, r min Is the minimum reject ratio obtained in step seven in step 321.
And 343, determining a second reject number according to the reject ratio corresponding to the ith first recommended content set and the number of candidate recommended contents in the ith first recommended content set.
Illustratively, after determining the rejection ratio corresponding to the ith first recommended content set, calculating the product between the rejection ratio and the number of candidate recommended contents in the ith first recommended content set to obtain a second rejection number.
When the first reject number and the second reject number are calculated, if the result obtained by the calculation is not an integer, the rounding process may be performed according to at least one of a rounding method, a rounding-up algorithm, a rounding-down algorithm, and the like, to obtain the first reject number and the second reject number having the integer values.
In response to the second discard number being greater than or equal to the first discard number, the candidate recommended content in the ith first recommended content set is selectively discarded by the second discard number, step 344.
Illustratively, if the second discard number is greater than or equal to the first discard number, it is indicated that the second discard number has exceeded the discard lower limit, and the selective discard may be performed directly based on the second discard number.
In response to the second discard number being less than the first discard number, selectively discarding candidate recommended content in the ith first recommended content set by the first discard number, step 345.
Illustratively, if the second discard number is less than the first discard number, it is indicated that the second discard number does not reach the discard lower limit, and the selective discard may be performed based on the first discard number.
Optionally, if the number of candidate recommended contents in the ith first candidate recommended content set is smaller than the first discard number, the candidate recommended contents in the ith first candidate recommended content set are not selectively discarded.
Step 346, determining n second recommended content sets based on the selective discarding results of the n first recommended content sets.
Illustratively, after discarding candidate recommended content in each first recommended content set based on the respective corresponding discard numbers of each first recommended content set, n first recommended content sets after discarding the candidate recommended content are obtained as n second recommended content sets.
Step 350, determining the first recommended content displayed to the first account from the n second recommended content sets.
Optionally, after determining the n second recommended content sets, the candidate recommended content in the n second recommended content sets may be recombined into a candidate recommended content queue, a second recommendation model is input into the candidate recommended content queue, the recommendation probability of each candidate recommended content in the candidate recommended content queue for recommending to the first account is predicted, and the first recommended content displayed to the first account is determined according to the recommendation probability.
In summary, according to the method for determining recommended content provided by the embodiment of the application, after a plurality of candidate recommended contents are obtained, the candidate recommended contents meeting the similarity requirement are aggregated, and a recommended content set is obtained; secondly, determining a first coefficient of a recommended content set according to a predicted recommendation result (i.e. recommendation probability of recommending to a first account) of the plurality of candidate recommended contents by the recommendation model; and finally, selectively discarding the candidate recommended content in the recommended content set according to the first coefficient, and determining the recommended content finally displayed to the first account from the discarded recommended content set. On the one hand, the recommended content set finally participating in decision making is a set discarded by candidate recommended content, so that when the recommended content displayed to the first account is determined, the computational waste for the candidate recommended content is reduced; on the other hand, the first coefficient is used for representing the prediction stability of the first recommendation model on a plurality of candidate recommendation contents meeting the similarity requirement in the recommendation content set, the stability of the recommendation model is introduced as a reference standard, and the number of candidate recommendation contents to be discarded in the recommendation content set is measured, so that the discarding number can be dynamically adjusted according to the stability of the recommendation model, and the accuracy of the determined discarding number is improved.
The method provided by the embodiment of the application can determine a lower limit of the reject number, compare the reject number with the lower limit of the reject number after the reject number is calculated by the first coefficient, and reject candidate recommended contents in the set according to the calculated reject number if the reject number is greater than or equal to the lower limit of the reject number; if the reject number is smaller than the reject number lower limit, the candidate recommended content in the set is rejected according to the reject number lower limit, so that the reject obtained by calculation is prevented from being too small, the reject number is controlled within a reasonable range, and the accuracy of selective reject is improved.
In some embodiments, after selectively discarding the candidate recommended content in each first recommended content set, some candidate recommended content may be additionally recalled from the newly created or other recommended content library to be explored according to the selectively discarding amount, and the first recommended content may be determined from the additionally recalled candidate recommended content and each second candidate content set. Illustratively, as shown in FIG. 4, the embodiments described above and shown in FIG. 2 or FIG. 3 may also be implemented as steps 410 through 470.
In step 410, a plurality of candidate recommended content is obtained.
The plurality of candidate recommended contents are recommended contents to be displayed to the first account.
Illustratively, the plurality of candidate recommended content may be implemented as news content, musical compositions, video content, blog content, novel content, etc., to be recommended, which the embodiments of the present application are not limited to.
Step 420, aggregating the plurality of candidate recommended contents to obtain n first recommended content sets.
The candidate recommended content in the ith first recommended content set meets the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n.
Illustratively, aggregating the plurality of candidate recommended contents refers to aggregating the plurality of candidate recommended contents according to the similarity, so that the similarity of the candidate recommended contents in the same recommended content set meets the similarity requirement.
Step 430, determining first coefficients corresponding to the n first recommended content sets respectively based on the predicted recommendation results of the first recommendation model on the plurality of candidate recommended content sets.
The prediction recommendation result refers to recommendation probability of recommending the candidate recommended content to the first account, and the ith first coefficient is used for representing prediction stability of the first recommendation model on the candidate recommended content in the ith first recommended content set.
Illustratively, the prediction stability of the first recommendation model for the candidate recommended content in the first recommended content set refers to the difference between the prediction results of the first recommendation model for the plurality of candidate recommended contents meeting the similarity requirement.
Step 440, selectively discarding the candidate recommended content in the n first recommended content sets based on the n first coefficients, to obtain n second recommended content sets.
Optionally, the method of performing the selective discarding comprises at least one of:
1. random discard.
Optionally, after obtaining the reject number corresponding to the ith first candidate recommended content set based on the ith first coefficient, randomly rejecting the candidate recommended content corresponding to the reject number in the ith first recommended content set to obtain the ith second recommended content set.
2. And discarding the predicted recommendation probability of the candidate recommendation content according to the first recommendation model.
Discarding candidate recommended content with the number corresponding to the discarding number from the ith first recommended content set according to the predicted recommended probability of the first recommended model on the candidate recommended content in the ith first recommended content set (for example, discarding a plurality of candidate recommended content with the smallest predicted recommended probability, wherein the number of the plurality of candidate recommended content is the number corresponding to the discarding number), so as to obtain an ith second recommended content set.
3. Discarding according to the similarity of the candidate recommended content quality inspection.
Discarding the candidate recommended content of the number corresponding to the discarding number from the ith first recommended content set according to the similarity of the candidate recommended content in the ith first recommended content set, and obtaining an ith second recommended content set. Illustratively, the discard number is 2, and the first candidate recommended content set includes candidate recommended content 1, candidate recommended content 2, candidate recommended content 3 and candidate recommended content 4, where the similarity between the candidate recommended content 1 and the candidate recommended content 2 is greater than a third preset similarity (the third preset similarity is greater than the preset similarity), and one of the candidate recommended content 1 and the candidate recommended content 2 is randomly selected for discarding, or the candidate recommended content with a lower prediction recommendation probability is discarded. And if the similarity between the candidate recommended content 3 and the candidate recommended content 4 is greater than a third preset similarity (the third preset similarity is greater than the preset similarity), randomly selecting one to discard from the candidate recommended content 3 and the candidate recommended content 4, or discarding the candidate recommended content with lower prediction recommendation probability.
It should be noted that the number of the methods and the methods of use for selective discarding according to the embodiments of the present application are not limited. For example: after determining the discard number, half of the candidate recommended contents of the discard number may be randomly discarded, and then the other half of the candidate recommended contents of the discard number may be discarded according to the similarity.
Step 450, obtaining a preset recommended content library.
In some embodiments, the preset recommended content library is a different content library than the recommended content library that determines the plurality of candidate recommended content.
Optionally, the preset recommended content library is a newly created recommended content library.
Illustratively, the content developer can submit the latest recommended content through the recommended content delivery platform, and the latest recommended content can be formed into a preset recommended content library.
Optionally, the preset recommended content library is a recommended content library formed by recommended content recommended to the first account by the content in the historical period.
Step 460, recalling the recommended content with preset recall number from the preset recommended content library.
The preset recall number is a number determined according to the number of candidate recommended contents selectively discarded in the n first recommended content sets.
Optionally, the recall number is preset to be the number corresponding to the first recommended content set with the least number of candidate recommended content in the n first recommended content sets.
Or, the recall number is preset to be the number corresponding to the first recommended content set with the largest candidate recommended content in the n first recommended content sets.
Or, the preset recall number is an average value of the number of candidate recommended contents in the n first recommended content sets.
Alternatively, the preset recall is the first discard number in step 342, i.e., the discard lower limit.
In some embodiments, the preset recalled recommended content is recalled from the preset recommended content library by a preset rule. Optionally, the preset rules include filtering based on attributes of the recommended content (e.g., recommended content type, recommended content industry, etc.).
In other embodiments, the third recommendation model predicts a plurality of recommended contents in the preset recommended content library to obtain predicted recommendation probabilities corresponding to the plurality of recommended contents respectively, and takes a plurality of recommended contents with highest predicted recommendation probabilities as recommended contents with preset recall numbers, wherein the number of the plurality of recommended contents is the preset recall number.
Optionally, the third prediction recommendation model includes at least one of a click rate prediction model, a conversion rate prediction model, and the like.
Step 470, determining the first recommended content from the n second recommended content sets and the recommended content with the preset recall number.
Optionally, after determining n second recommended content sets, the candidate recommended content in the n second recommended content sets may be recombined into a candidate recommended content queue, and the candidate recommended content queue and the recommended content of the preset recall number are input into a second recommendation model, and the recommendation probability of each candidate recommended content in the candidate recommended content queue to the first account and the recommendation probability of the recommended content of the preset recall number to the first account are predicted and obtained, and the first recommended content displayed to the first account is determined according to the recommendation probability.
In summary, according to the method for determining recommended content provided by the embodiment of the application, after a plurality of candidate recommended contents are obtained, the candidate recommended contents meeting the similarity requirement are aggregated, and a recommended content set is obtained; secondly, determining a first coefficient of a recommended content set according to a predicted recommendation result (i.e. recommendation probability of recommending to a first account) of the plurality of candidate recommended contents by the recommendation model; and finally, selectively discarding the candidate recommended content in the recommended content set according to the first coefficient, and determining the recommended content finally displayed to the first account from the discarded recommended content set. On the one hand, the recommended content set finally participating in decision making is a set discarded by candidate recommended content, so that when the recommended content displayed to the first account is determined, the computational waste for the candidate recommended content is reduced; on the other hand, the first coefficient is used for representing the prediction stability of the first recommendation model on a plurality of candidate recommendation contents meeting the similarity requirement in the recommendation content set, the stability of the recommendation model is introduced as a reference standard, and the number of candidate recommendation contents to be discarded in the recommendation content set is measured, so that the discarding number can be dynamically adjusted according to the stability of the recommendation model, and the accuracy of the determined discarding number is improved.
According to the method provided by the embodiment of the application, the saved computer power resources of the candidate recommended content are discarded and can be distributed to other preset recommended content libraries, so that the high-efficiency utilization of the computer power resources is realized, and the recommending effect of the first account is improved.
Referring to fig. 5, a schematic process of discarding a fine candidate recommended content queue is shown, as shown in fig. 5, first, recommendation probability prediction is performed on a coarse candidate recommended content queue 510 through a first recommendation model to obtain a fine candidate recommended content queue 520, the fine candidate recommended content queue 520 is input into an aggregation system 530 to aggregate, and if the aggregation system 530 aggregates a plurality of candidate recommended contents in the fine candidate recommended content queue 520 according to md5 values, md5 values corresponding to each candidate recommended content are calculated according to md5 algorithm, candidate recommended contents with the same md5 value are aggregated to obtain a plurality of sets, and coefficients (e.g., mutation coefficients) corresponding to the sets are determined after the sets are obtained. The coefficient is used to measure the predictive stability of the first recommendation model for candidate recommended content in the collection.
After determining the coefficients corresponding to the sets, searching the discarding proportion corresponding to the coefficients corresponding to the sets according to the first interval, the second interval and the third interval, and assuming that the value of a certain coefficient is located in the first interval, indicating that the discarding proportion corresponding to the coefficient is 50%. Calculating the discarding proportion corresponding to the coefficient and the number of candidate recommended contents in the set corresponding to the coefficient as a discarding number; or setting a discard lower limit value, and taking the discard lower limit value as the discard number when the discard number is smaller than the discard lower limit value.
And discarding candidate recommended contents in the set corresponding to the coefficient according to the discarding number to obtain a plurality of sets subjected to discarding processing, and forming a final fine-ranking candidate recommended content queue 540 by the candidate recommended contents in the plurality of sets subjected to discarding processing.
Alternatively, the recommended content of the number corresponding to the discard number may be recalled from the recommended content library 550 according to the average value of the discard numbers corresponding to the respective sets; alternatively, the recommended content corresponding to the discard lower limit value is recalled from the recommended content library 550 by the average value of the discard lower limit value. Finally, the first recommended content 560, i.e., the content that is ultimately presented to the user, is determined from the fine candidate recommended content queue 540 and the additional recalled recommended content.
For illustration, taking an application program implemented as a news application program as an example, fig. 6 shows a schematic diagram of a content display interface determined by a method for determining recommended content according to an embodiment of the present application, as shown in fig. 6, after determining recommended content 601 by the method for determining recommended content provided by the embodiment of the present application, the recommended content 601 is displayed in the content display interface 600.
Under the condition that the calculation of the model has relatively high stability, the calculation scores of the recommended contents with the same or similar materials in the candidate queue are almost the same, namely the calculated recommended content ordering positions are close, and the behavior of the user on the same recommended content is the same, so that the calculation of the same or similar recommended content is superfluous. According to the method provided by the embodiment of the application, the model with high stability has higher probability of discarding the same or similar recommended content by introducing the stability of the model as a reference standard, so that the calculated amount of the same or similar recommended content in the next stage is reduced; moreover, the calculation power resource saved by discarding the recommended content can be used for calculating from the newly created recommended content or other recommended content needing to be explored, so that the exploration success rate is improved.
Fig. 7 is a block diagram showing a structure of an apparatus for determining recommended content according to an exemplary embodiment of the present application, and as shown in fig. 7, the apparatus includes:
the data acquisition module 710 is configured to acquire a plurality of candidate recommended contents, where the plurality of candidate recommended contents are recommended contents to be displayed to the first account;
the content aggregation module 720 is configured to aggregate the plurality of candidate recommended content to obtain n first recommended content sets, where n is a positive integer, i is greater than 0 and less than or equal to n, and candidate recommended content in the i-th first recommended content set meets a preset similarity requirement;
a first determining module 730, configured to determine first coefficients corresponding to the n first recommended content sets, respectively, based on a predicted recommendation result of a first recommendation model on the plurality of candidate recommended content, where the predicted recommendation result refers to a recommendation probability of recommending the candidate recommended content to the first account, and the i first coefficient is used to characterize a prediction stability of the first recommendation model on the candidate recommended content in the i first recommended content set;
the content discarding module 740 is configured to selectively discard candidate recommended content in the n first recommended content sets based on n first coefficients, to obtain n second recommended content sets;
A second determining module 750, configured to determine, from the n second recommended content sets, a first recommended content that is displayed to the first account.
Referring to fig. 8, in some embodiments, the first determining module 730 further includes:
a first obtaining unit 731, configured to obtain a prediction of recommendation probabilities of the plurality of candidate recommended contents by the first recommendation model, and take prediction recommendation probabilities corresponding to the plurality of candidate recommended contents respectively as the prediction recommendation results;
the first determining module 730 is configured to determine, according to the predicted recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, a predicted deviation corresponding to the ith first recommended content set as the first coefficient, where the predicted deviation refers to a degree of dispersion of the predicted recommendation probabilities of the candidate recommended contents in the ith first recommended content set.
In some embodiments, the first determining module 730 is configured to determine, according to the predicted recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, a coefficient of variation corresponding to the ith first recommended content set as the first coefficient.
In some embodiments, the content discard module 740 includes:
a determining unit 741, configured to determine rejection proportions corresponding to the n first recommended content sets respectively based on the n first coefficients, where the first coefficients and the rejection proportions have a negative correlation;
the content discarding module 740 is configured to selectively discard candidate recommended content in the ith first recommended content set according to a discarding proportion corresponding to the ith first recommended content set.
In some embodiments, the content discard module 740 includes:
a second acquisition unit 742 for acquiring the first reject number;
the determining unit 741 is configured to determine a second discard number according to the discard ratio corresponding to the ith first recommended content set and the number of candidate recommended contents in the ith first recommended content set;
the content discarding module 740 is configured to selectively discard candidate recommended content in the ith first recommended content set by the second discard number in response to the second discard number being greater than or equal to the first discard number;
the content discarding module 740 is configured to selectively discard candidate recommended content in the ith first recommended content set by the first discard number in response to the second discard number being smaller than the first discard number.
In some embodiments, the data obtaining module 710 is configured to obtain a preset recommended content library; the apparatus further comprises:
a content recall module 760, configured to recall, from the preset recommended content library, a preset number of recommended content, where the preset number of recalls is a number determined according to the number of candidate recommended content selectively discarded in the n first recommended content sets;
the second determining module 750 is configured to determine the first recommended content from the n second recommended content sets and the recommended content with the preset recall number.
In some embodiments, the content aggregation module 720 includes:
a calculating unit 721, configured to perform hash calculation on the plurality of candidate recommended contents according to a preset hash algorithm, so as to obtain first hash values corresponding to the plurality of candidate recommended contents respectively;
the content aggregation module 720 is configured to aggregate the plurality of candidate recommended content based on the first hash value to obtain the n first recommended content sets, where a similarity of a first hash value corresponding to the candidate recommended content in the i-th first recommended content set is greater than or equal to a first preset similarity.
In some embodiments, the plurality of candidate recommended content corresponds to a recommended subject content and a presentation page for presenting the recommended subject content, respectively; the calculating unit 721 is configured to:
performing hash calculation on the recommended subject contents respectively corresponding to the plurality of candidate recommended contents according to the preset hash algorithm to obtain second hash values of the recommended subject contents respectively corresponding to the plurality of candidate recommended contents;
carrying out hash calculation on the display pages corresponding to the candidate recommended contents respectively according to the preset hash algorithm to obtain third hash values of the display pages corresponding to the candidate recommended contents respectively;
and fusing the second hash value and the third hash value to obtain the first hash value.
In some embodiments, the plurality of candidate recommended content includes a first video content therein; the calculating unit 721 is configured to:
carrying out framing treatment on the first video content to obtain a plurality of video frames;
determining a key video frame from the plurality of video frames;
and carrying out hash calculation on the key video frames according to the preset hash algorithm to obtain hash values of the key video frames as first hash values corresponding to the first video content.
In some embodiments, the content aggregation module 720 includes:
a third obtaining unit 722, configured to obtain recommended content types corresponding to the plurality of candidate recommended contents respectively;
the content aggregation module 720 is configured to aggregate the plurality of candidate recommended content based on the recommended content types, to obtain the n first recommended content sets, where a similarity of a recommended content type corresponding to the candidate recommended content in the i-th first recommended content set is greater than or equal to a second preset similarity.
In summary, the determining device for recommended content provided in the embodiment of the present application obtains a plurality of candidate recommended contents, and then aggregates the candidate recommended contents meeting the similarity requirement to obtain a recommended content set; secondly, determining a first coefficient of a recommended content set according to a predicted recommendation result (i.e. recommendation probability of recommending to a first account) of the plurality of candidate recommended contents by the recommendation model; and finally, selectively discarding the candidate recommended content in the recommended content set according to the first coefficient, and determining the recommended content finally displayed to the first account from the discarded recommended content set. On the one hand, the recommended content set finally participating in decision making is a set discarded by candidate recommended content, so that when the recommended content displayed to the first account is determined, the computational waste for the candidate recommended content is reduced; on the other hand, the first coefficient is used for representing the prediction stability of the first recommendation model on a plurality of candidate recommendation contents meeting the similarity requirement in the recommendation content set, the stability of the recommendation model is introduced as a reference standard, and the number of candidate recommendation contents to be discarded in the recommendation content set is measured, so that the discarding number can be dynamically adjusted according to the stability of the recommendation model, and the accuracy of the determined discarding number is improved.
It should be noted that: the recommended content determining device provided in the above embodiment is only exemplified by the division of the above functional modules, and in practical application, the above functional allocation may be performed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the apparatus for determining recommended content provided in the above embodiment belongs to the same concept as the embodiment of the method for determining recommended content, and the specific implementation process is detailed in the method embodiment, which is not repeated here.
Fig. 9 shows a block diagram of an electronic device 900 provided by an exemplary embodiment of the application. The electronic device 900 may be a portable mobile terminal such as: smart phones, car terminals, tablet computers, MP3 players (Moving Picture Experts Group Audio Layer III, motion picture expert compression standard audio plane 3), MP4 (Moving Picture Experts Group Audio Layer IV, motion picture expert compression standard audio plane 4) players, notebook computers or desktop computers. Electronic device 900 may also be referred to by other names of user devices, portable terminals, laptop terminals, desktop terminals, and the like.
Generally, the electronic device 900 includes: a processor 901 and a memory 902.
Processor 901 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 901 may be implemented in at least one hardware form of DSP (Digital Signal Processing ), FPGA (Field-Programmable Gate Array, field programmable gate array), PLA (Programmable Logic Array ). The processor 901 may also include a main processor and a coprocessor, the main processor being a processor for processing data in an awake state, also referred to as a CPU (Central Processing Unit ); a coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 901 may integrate a GPU (Graphics Processing Unit, image processor) for taking care of rendering and drawing of content that the display screen needs to display. In some embodiments, the processor 901 may also include an AI (Artificial Intelligence ) processor for processing computing operations related to machine learning.
The memory 902 may include one or more computer-readable storage media, which may be non-transitory. The memory 902 may also include high-speed random access memory, as well as non-volatile memory, such as one or more magnetic disk storage devices, flash memory storage devices. In some embodiments, a non-transitory computer readable storage medium in memory 902 is used to store at least one instruction for execution by processor 901 to implement the method of determining recommended content provided by a method embodiment of the present application.
In some embodiments, the electronic device 900 also includes one or more sensors. The one or more sensors include, but are not limited to: proximity sensor, gyro sensor, pressure sensor.
A proximity sensor, also referred to as a distance sensor, is typically provided on the front panel of the electronic device 900. The proximity sensor is used to capture the distance between the user and the front of the electronic device 900.
The gyro sensor may detect a body direction and a rotation angle of the electronic device 900, and the gyro sensor may cooperate with the acceleration sensor to collect a 3D motion of the user on the electronic device 900. The processor 901 may implement the following functions according to the data collected by the gyro sensor: motion sensing (e.g., changing UI according to a tilting operation by a user), image stabilization at shooting, game control, and inertial navigation.
The pressure sensor may be disposed on a side frame of the electronic device 900 and/or on an underlying layer of the display screen. When the pressure sensor is disposed on the side frame of the electronic device 900, a holding signal of the electronic device 900 by a user may be detected, and the processor 901 performs left-right hand recognition or shortcut operation according to the holding signal collected by the pressure sensor. When the pressure sensor is arranged at the lower layer of the display screen, the processor 901 controls the operability control on the UI according to the pressure operation of the user on the display screen. The operability controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.
In some embodiments, electronic device 900 also includes other component parts, and those skilled in the art will appreciate that the structure shown in FIG. 9 is not limiting of electronic device 900 and may include more or fewer components than shown, or may combine certain components, or may employ a different arrangement of components.
The embodiment of the application also provides a computer device which can be implemented as a terminal or a server as shown in fig. 2. The computer device includes a processor and a memory, where at least one instruction, at least one program, a set of codes, or a set of instructions is stored, and the at least one instruction, at least one program, a set of codes, or a set of instructions is loaded and executed by the processor to implement the method for determining recommended content provided by the above method embodiments.
Embodiments of the present application also provide a computer readable storage medium having stored thereon at least one instruction, at least one program, a code set, or an instruction set, where the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the method for determining recommended content provided by the above method embodiments.
Embodiments of the present application also provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the recommended content determination method according to any one of the above embodiments.
Alternatively, the computer-readable storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), solid state disk (SSD, solid State Drives), or optical disk, etc. The random access memory may include resistive random access memory (ReRAM, resistance Random Access Memory) and dynamic random access memory (DRAM, dynamic Random Access Memory), among others. The foregoing embodiment numbers of the present application are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The foregoing description of the preferred embodiments of the present application is not intended to limit the application, but rather, the application is to be construed as limited to the appended claims.
Claims (14)
1. A method of determining recommended content, the method comprising:
acquiring a plurality of candidate recommended contents, wherein the plurality of candidate recommended contents are recommended contents to be displayed to a first account;
aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the candidate recommended contents in the ith first recommended content set meet the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n;
determining first coefficients corresponding to the n first recommended content sets respectively based on a predicted recommendation result of a first recommendation model on the plurality of candidate recommended content, wherein the predicted recommendation result refers to recommendation probability of the candidate recommended content to the first account, and the ith first coefficient is used for representing prediction stability of the first recommendation model on the candidate recommended content in the ith first recommended content set;
selectively discarding candidate recommended contents in the n first recommended content sets based on n first coefficients to obtain n second recommended content sets;
And determining the first recommended content displayed for the first account from the n second recommended content sets.
2. The method of claim 1, wherein determining the first coefficients for each of the n first recommended content sets based on the predicted recommended results of the first recommendation model for the plurality of candidate recommended content sets comprises:
acquiring recommendation probability predictions of the first recommendation model on the plurality of candidate recommendation contents, and taking the prediction recommendation probabilities respectively corresponding to the plurality of candidate recommendation contents as the prediction recommendation results;
according to the prediction recommendation probabilities respectively corresponding to the candidate recommended contents in the ith first recommended content set, determining a prediction deviation corresponding to the ith first recommended content set as the first coefficient, wherein the prediction deviation refers to the degree of dispersion of the prediction recommendation probabilities of the candidate recommended contents in the ith first recommended content set.
3. The method according to claim 2, wherein the determining, according to the predicted recommendation probabilities corresponding to the candidate recommended contents in the ith first recommended content set, the predicted deviation corresponding to the ith first recommended content set as the first coefficient includes:
And determining a variation coefficient corresponding to the ith first recommended content set as the first coefficient according to the prediction recommendation probability corresponding to the candidate recommended content in the ith first recommended content set.
4. A method according to any one of claims 1 to 3, wherein said selectively discarding candidate recommended content from said n first recommended content sets based on n first coefficients comprises:
determining rejection proportions corresponding to the n first recommended content sets respectively based on the n first coefficients, wherein the first coefficients and the rejection proportions are in negative correlation;
and selectively discarding the candidate recommended content in the ith first recommended content set according to the discarding proportion corresponding to the ith first recommended content set.
5. The method of claim 4, wherein the selectively discarding candidate recommended content in the ith first recommended content set according to the discard ratio corresponding to the ith first recommended content set comprises:
acquiring a first reject number;
determining a second reject number according to the reject ratio corresponding to the ith first recommended content set and the number of candidate recommended contents in the ith first recommended content set;
Selectively discarding candidate recommended content in the ith first recommended content set by the second discard number in response to the second discard number being greater than or equal to the first discard number;
selectively discarding candidate recommended content in the ith first recommended content set at the first discard number in response to the second discard number being less than the first discard number.
6. A method according to any one of claims 1 to 3, wherein the method further comprises:
acquiring a preset recommended content library;
recall, from the preset recommended content library, recommended content of a preset recall number, where the preset recall number is a number determined according to the number of candidate recommended content selectively discarded in the n first recommended content sets;
and determining the first recommended content from the n second recommended content sets and the recommended content with the preset recall number.
7. A method according to any one of claims 1 to 3, wherein aggregating the plurality of candidate recommended content to obtain n first recommended content sets includes:
carrying out hash calculation on the plurality of candidate recommended contents according to a preset hash algorithm to obtain first hash values respectively corresponding to the plurality of candidate recommended contents;
And based on the first hash values, aggregating the plurality of candidate recommended contents to obtain the n first recommended content sets, wherein the similarity of the first hash values corresponding to the candidate recommended contents in the ith first recommended content set is greater than or equal to a first preset similarity.
8. The method of claim 7, wherein the plurality of candidate recommended content corresponds to recommended subject content and a presentation page for presenting the recommended subject content, respectively;
the method carries out hash calculation on the plurality of candidate recommended contents according to a preset hash algorithm to obtain first hash values respectively corresponding to the plurality of candidate recommended contents, and comprises the following steps:
performing hash calculation on the recommended subject contents respectively corresponding to the plurality of candidate recommended contents according to the preset hash algorithm to obtain second hash values of the recommended subject contents respectively corresponding to the plurality of candidate recommended contents;
carrying out hash calculation on the display pages corresponding to the candidate recommended contents respectively according to the preset hash algorithm to obtain third hash values of the display pages corresponding to the candidate recommended contents respectively;
And fusing the second hash value and the third hash value to obtain the first hash value.
9. The method of claim 7, wherein the plurality of candidate recommended content comprises a first video content;
the hash calculation is performed on the plurality of candidate recommended contents according to a preset hash algorithm to obtain first hash values respectively corresponding to the plurality of candidate recommended contents, including:
carrying out framing treatment on the first video content to obtain a plurality of video frames;
determining a key video frame from the plurality of video frames;
and carrying out hash calculation on the key video frames according to the preset hash algorithm to obtain hash values of the key video frames as first hash values corresponding to the first video content.
10. A method according to any one of claims 1 to 3, wherein aggregating the plurality of candidate recommended content to obtain n first recommended content sets includes:
acquiring recommendation content types respectively corresponding to the plurality of candidate recommendation contents;
and based on the recommended content types, aggregating the plurality of candidate recommended content to obtain the n first recommended content sets, wherein the similarity of the recommended content types corresponding to the candidate recommended content in the ith first recommended content set is greater than or equal to a second preset similarity.
11. A recommended content determining apparatus, characterized in that the apparatus includes:
the data acquisition module is used for acquiring a plurality of candidate recommended contents, wherein the plurality of candidate recommended contents are recommended contents to be displayed to the first account;
the content aggregation module is used for aggregating the plurality of candidate recommended contents to obtain n first recommended content sets, wherein the candidate recommended contents in the ith first recommended content set meet the preset similarity requirement, n is a positive integer, and i is more than 0 and less than or equal to n;
the first determining module is used for determining first coefficients corresponding to the n first recommended content sets respectively based on the predicted recommendation results of the first recommended model on the plurality of candidate recommended content, wherein the predicted recommendation results refer to recommendation probabilities of recommending the candidate recommended content to the first account, and the ith first coefficient is used for representing the predicted stability of the first recommended model on the candidate recommended content in the ith first recommended content set;
the content discarding module is used for selectively discarding the candidate recommended content in the n first recommended content sets based on n first coefficients to obtain n second recommended content sets;
And the second determining module is used for determining the first recommended content displayed for the first account from the n second recommended content sets.
12. A computer device comprising a processor and a memory, wherein the memory has stored therein at least one program that is loaded and executed by the processor to implement the method of determining recommended content according to any of claims 1 to 10.
13. A computer-readable storage medium, wherein at least one program is stored in the storage medium, the at least one program being loaded and executed by a processor to implement the recommended content determination method of any one of claims 1 to 10.
14. A computer program product comprising a computer program which, when executed by a processor, implements a method of determining recommended content according to any of claims 1 to 10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311196168.XA CN117150141A (en) | 2023-09-15 | 2023-09-15 | Method, device, equipment, medium and program product for determining recommended content |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311196168.XA CN117150141A (en) | 2023-09-15 | 2023-09-15 | Method, device, equipment, medium and program product for determining recommended content |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117150141A true CN117150141A (en) | 2023-12-01 |
Family
ID=88886758
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311196168.XA Pending CN117150141A (en) | 2023-09-15 | 2023-09-15 | Method, device, equipment, medium and program product for determining recommended content |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117150141A (en) |
-
2023
- 2023-09-15 CN CN202311196168.XA patent/CN117150141A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230281448A1 (en) | Method and apparatus for information recommendation, electronic device, computer readable storage medium and computer program product | |
| US20230102337A1 (en) | Method and apparatus for training recommendation model, computer device, and storage medium | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| WO2021155691A1 (en) | User portrait generating method and apparatus, storage medium, and device | |
| CN112632403A (en) | Recommendation model training method, recommendation device, recommendation equipment and recommendation medium | |
| CN113569129A (en) | Click rate prediction model processing method, content recommendation method, device and equipment | |
| CN112052387A (en) | Content recommendation method and device and computer readable storage medium | |
| WO2024002167A1 (en) | Operation prediction method and related apparatus | |
| CN114417174B (en) | Content recommendation method, device, equipment and computer storage medium | |
| CN112269943B (en) | Information recommendation system and method | |
| WO2024041483A1 (en) | Recommendation method and related device | |
| CN116843022A (en) | Data processing method and related device | |
| CN116204709A (en) | Data processing method and related device | |
| CN116308640A (en) | Recommendation method and related device | |
| WO2023050143A1 (en) | Recommendation model training method and apparatus | |
| CN117150141A (en) | Method, device, equipment, medium and program product for determining recommended content | |
| CN115545738A (en) | Recommendation method and related device | |
| CN115482019A (en) | Activity attention prediction method and device, electronic equipment and storage medium | |
| CN113704608A (en) | Personalized item recommendation method and device, electronic equipment and storage medium | |
| CN116720003B (en) | Ordering processing method, ordering processing device, computer equipment and storage medium | |
| WO2023051678A1 (en) | Recommendation method and related device | |
| CN116523587A (en) | Data processing method and related device | |
| CN116401398A (en) | Data processing method and related device | |
| CN114504829A (en) | Property recommendation method, device, equipment and medium | |
| CN117216378A (en) | Data processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |