WO2020200199A1 - 个性化推荐的方法、终端设备和系统 - Google Patents

个性化推荐的方法、终端设备和系统 Download PDF

Info

Publication number
WO2020200199A1
WO2020200199A1 PCT/CN2020/082429 CN2020082429W WO2020200199A1 WO 2020200199 A1 WO2020200199 A1 WO 2020200199A1 CN 2020082429 W CN2020082429 W CN 2020082429W WO 2020200199 A1 WO2020200199 A1 WO 2020200199A1
Authority
WO
WIPO (PCT)
Prior art keywords
service content
service
user
terminal device
candidate set
Prior art date
Application number
PCT/CN2020/082429
Other languages
English (en)
French (fr)
Inventor
莫兰
胡迅
Original Assignee
华为技术有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to US17/601,009 priority Critical patent/US11843651B2/en
Priority to EP20784453.1A priority patent/EP3923163A4/en
Publication of WO2020200199A1 publication Critical patent/WO2020200199A1/zh

Links

Images

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
    • G06F16/907—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/909—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using geographical or spatial information, e.g. location
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L65/00—Network arrangements, protocols or services for supporting real-time applications in data packet communication
    • H04L65/60—Network streaming of media packets
    • H04L65/61—Network streaming of media packets for supporting one-way streaming services, e.g. Internet radio
    • H04L65/612—Network streaming of media packets for supporting one-way streaming services, e.g. Internet radio for unicast
    • 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
    • G06F16/903—Querying
    • G06F16/9035—Filtering based on additional data, e.g. user or group profiles
    • 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
    • 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
    • G06F16/955—Retrieval from the web using information identifiers, e.g. uniform resource locators [URL]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G06Q30/0271—Personalized advertisement
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0282—Rating or review of business operators or products
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • This application relates to the field of data processing, and more specifically, to a method, system, and terminal device for personalized recommendation.
  • the cloud side can only obtain limited user data, which may result in the personalized recommendation service content output by the cloud side The accuracy of is not high, which cannot meet the demands of users.
  • This application provides a method, system and terminal device for personalized recommendation, which helps to improve the accuracy of personalized recommendation.
  • a personalized recommendation method includes: a terminal device sends service content request information to a business system; the terminal device receives service content response information sent by the business system, and the service content response information includes service content Candidate set, the service content candidate set includes multiple service content, the service content candidate set is obtained by the business system according to the user behavior event reported by the terminal device; the terminal device obtains from the service content candidate set according to the data authorized by the user Determine one or more service contents; the terminal device displays the one or more service contents to the user.
  • the personalized recommendation method of the embodiment of this application can give full play to the advantages of end-side comprehensive and accurate user data without uploading additional user data. It only needs to securely filter and accurately match the service content that meets the user's demands in the downstream direction, and recommend based on the business system On the basis of the high recall rate of the service content, the advantage of user data saved by the terminal device is superimposed to solve the problem of high accuracy of personalized recommendation. Through the secondary personalized recommendation cooperation of end-cloud collaboration, a personalized recommendation service experience with high recall rate and high accuracy is provided.
  • the service content response information also includes the label information of each service content in the service content candidate set; wherein, the terminal device obtains information from the service according to the data authorized by the user.
  • the determination of one or more service contents in the content candidate set includes: the terminal device determines the one or more service contents according to the data authorized by the user and the label information of each service content in the service content candidate set.
  • the service content candidate set list transmitted by the business system to the terminal device also includes semantic tags corresponding to the service content candidate set. These information are sent to the terminal device together, and the terminal device can be further combined The data of the user knowledge base stored in the terminal device is matched and filtered with semantic tags, thereby improving the accuracy of personalized recommendations.
  • the terminal device determines the one or more service contents according to the data authorized by the user and the label information of each service content in the service content candidate set, including: The terminal device determines the correlation coefficient between each service content and the data authorized by the user according to the data authorized by the user and the label information of each service content; the terminal device determines the one or more service contents according to the correlation coefficient .
  • the terminal device can determine the correlation coefficient between the service content corresponding to the semantic label and the data of the user knowledge base according to the data of the user knowledge base and the semantic label stored in the terminal device, and obtain the correlation coefficient through calculation
  • the multiple correlation coefficients determine the service content ultimately displayed to users, which helps to improve the accuracy of personalized recommendations.
  • the terminal device can determine multiple correlation coefficients according to the label information of each service content in the multiple service content and the data authorized by the user; the terminal device can show the user that the multiple correlation coefficients are greater than Or the service content corresponding to the correlation coefficient equal to the preset value.
  • the service content is a massive real-time update service content of the business system.
  • the terminal device saves complete user knowledge base data, but the terminal device does not save or only saves part of the service content that is incomplete and not real-time. Therefore, the modeling of personalized recommendation algorithm based on terminal equipment should be based on the one-time recommendation of the business system to find massive real-time updated service content, and give full play to the advantages of the high-quality integrity of the local user data of the terminal equipment, thereby helping to improve personalized recommendation The accuracy rate.
  • the service content includes one or more of an advertising service, a video service, or a news service.
  • a personalized recommendation method includes: a service system receives service content request information sent by a terminal device; the service system determines a service content candidate set according to the service content request information, and the service content candidate set The service content includes multiple service contents; the service system sends service content response information to the terminal device, and the service content response information includes the service content candidate set; the terminal device determines one or more service content candidate sets according to the data authorized by the user. Service content; the terminal device displays the one or more service content to the user.
  • the personalized recommendation method of the embodiment of this application can give full play to the comprehensive and accurate user data advantages of the terminal device.
  • the terminal device does not need to upload additional user data to the business system, and only needs to filter safely and accurately match the services that meet the user's demands in the downstream direction.
  • the content is based on the high recall rate of the service content recommended by the business system once, and the comprehensive and accurate user data saved by the terminal device is superimposed to solve the problem of high accuracy of personalized recommendation.
  • a personalized recommendation service experience with high recall rate and high accuracy is provided.
  • the service content response information also includes the label information of each service content in the service content candidate set; wherein, the terminal device obtains information from the service according to the data authorized by the user.
  • the determination of one or more service contents in the content candidate set includes: the terminal device determines the one or more service contents according to the data authorized by the user and the label information of each service content in the service content candidate set.
  • the service content candidate set list transmitted by the business system to the terminal device also includes semantic tags corresponding to the service content candidate set.
  • the information is sent to the terminal device together, and the terminal device can be further combined
  • the data of the user knowledge base stored in the terminal device is matched and filtered with semantic tags, thereby improving the accuracy of personalized recommendations.
  • the terminal device determines the one or more service contents according to the data authorized by the user and the label information of each service content in the service content candidate set, including: The terminal device determines the correlation coefficient between each service content and the data authorized by the user according to the data authorized by the user and the label information of each service content; the terminal device determines the one or more service contents according to the correlation coefficient .
  • the terminal device can determine the correlation coefficient between the service content corresponding to the semantic label and the data of the user knowledge base according to the data of the user knowledge base and the semantic label stored in the terminal device, and obtain the correlation coefficient through calculation
  • the multiple correlation coefficients determine the service content ultimately displayed to users, which helps to improve the accuracy of personalized recommendations.
  • the method before the service system sends service content response information to the terminal device, the method further includes: the service system determines the value of each service content according to a preset algorithm Label Information.
  • the service content candidate set list sent by the business system to the terminal device for a recommendation output is also included. It may include semantic tags corresponding to the service content candidate set.
  • the business system brings this information to the terminal device, which helps the terminal device to further combine the data and content tags of the user knowledge base of the terminal device to match and filter, thereby helping to improve the accuracy of personalized recommendations.
  • the business system determines the service content candidate set according to the service content request information, including: the business system receives the user behavior event sent by the terminal device; the business system according to The user behavior event determines the user behavior attributes and the user's audience information; the business system determines the service content candidate set from the service contents stored in the business system according to the user behavior attributes and the user's audience information.
  • the business system mainly implements the primary selection of personalized recommendations, based on the limited user data of the business system, finds a large amount of service content that may match the needs of users, and realizes the functions of one-time screening, CTR prediction and ranking with high recall rate.
  • the service content is one or more of the massive real-time update service content of the business system.
  • the service content is an advertising service, a video service, or a news service.
  • the terminal device saves complete user knowledge base data, but the terminal device does not save or only saves part of the service content that is incomplete and not real-time. Therefore, the modeling of personalized recommendation algorithm based on terminal equipment should be based on the one-time recommendation of the business system to find massive real-time updated service content, and give full play to the advantages of the high-quality integrity of the local user data of the terminal equipment, thereby helping to improve personalized recommendation The accuracy rate.
  • a method for personalized recommendation is provided.
  • the method is applied to a personalized recommendation system.
  • the personalized recommendation system includes a terminal device and a business system.
  • the terminal device includes a processor and a memory, and the memory stores a Or multiple programs, the business system includes a business operation management system, an online business system, and an offline business system.
  • the method includes: the business operation management system sends the first service content candidate set to the offline business system; the offline business system is based on preset Algorithm to determine the label information of the service content of the first service content candidate set; the offline business system sends the label information of the first service content candidate set service content to the online business system; the online business system receives the service content request information sent by the terminal device; The online business system determines a second service content candidate set from the first service content candidate set according to the service content request information.
  • the second service content candidate set includes multiple service contents; the online business system sends a service content response to the terminal device
  • the service content response information includes the second service content candidate set and the label information of each service content in the second service content candidate set; the terminal device collects information from multiple One or more service contents are determined in the service content; the terminal device displays the one or more service contents to the user.
  • the terminal device determines one or more service contents from multiple service contents according to the data authorized by the user and the label information of each service content, including: The terminal device determines the correlation coefficient between each service content and the data authorized by the user according to the data authorized by the user and the label information of each service content; the terminal device determines the correlation coefficient between each service content and the data authorized by the user according to the correlation coefficient. Describe one or more service contents.
  • the method further includes: the offline business system receives the user behavior event sent by the terminal device; the offline business system determines the user behavior attribute and the user's audience based on the user behavior event Group information; the offline business system sends the user behavior attributes and the user's audience information to the online business system.
  • the online business system determines the second service content candidate set from the first service content candidate set according to the service content request information, including: According to the user behavior attribute, the audience information of the user, and the label information of the service content in the first service content candidate set, the second service content candidate set is determined from the first service content candidate set.
  • the service content is one or more of the massive real-time updated service content of the business system.
  • the service content is an advertising business, a video business, or a news business.
  • a personalized recommendation method includes: a service system determines label information of service content in a first service content candidate set; the service system receives service content request information sent by a terminal device; The content request information determines a second service content candidate set from the first service content candidate set, the second service content candidate set includes multiple service contents; the business system sends the multiple service contents and the multiple service contents to the terminal device Corresponding label information.
  • the method before the service system receives the service content request information sent by the terminal device, the method further includes: the service system receives the user behavior event sent by the terminal device; the service system According to the business behavior event, determine user behavior attributes and user audience information; wherein, the business system determines the second service content candidate set from the first service content candidate set, including: the business system determines the second service content candidate set according to the user behavior attributes and the user’s The audience group information and the label information of the service content in the first service content candidate set determine the second service content candidate set from the first service content candidate set.
  • the business system determining the label information of the service content in the first service content candidate set includes: the business system determines the first service content candidate according to a preset algorithm The label information of the centralized service content.
  • a terminal device in a fifth aspect, includes a touch screen, one or more processors, multiple application programs, and one or more programs; wherein, one or more programs are stored in the memory; When the one or more processors execute the one or more programs, the terminal device implements the above-mentioned first aspect and the personalized recommendation method in the possible implementation manners of the first aspect.
  • a business system in a sixth aspect, includes a business operation management system, an offline business system, and an online business system, wherein the business operation management system is used to receive the first service content candidate set; the business operation management system It is also used to send the first service content candidate set to the offline service system; the offline service system is used to determine the label information of the service content in the first service content candidate set; the offline service system is also used to receive the terminal device sending User behavior events, and determine user behavior attributes and audience information of the user according to the user behavior events; the offline business system is also used to concentrate the first service content candidates on the service content label information, the user’s behavior attributes and The user’s audience information is sent to the online business system; the online business system is used to receive the service content request information sent by the terminal device; the online business system is also used to base the user’s behavior attributes, the user’s audience information, and the first The label information of the service content in the service content candidate set determines the second service content candidate set from the first service content candidate set; the online business system is
  • a personalized recommendation system which includes the above-mentioned terminal device and the above-mentioned service system.
  • An eighth aspect provides a computer storage medium, including instructions, which when the instructions run on an electronic device, cause the electronic device to execute the personalized recommendation method in any of the foregoing possible implementations.
  • a ninth aspect provides a computer program product, which when the computer program product runs on an electronic device, causes the electronic device to execute the personalized recommendation method in any one of the foregoing possible implementations.
  • Fig. 1 is a schematic architecture diagram of a personalized recommendation system provided by an embodiment of the present application.
  • Fig. 2 is another schematic architecture diagram of a personalized recommendation system provided by an embodiment of the present application.
  • Fig. 3 is a schematic flowchart of a personalized recommendation method provided by an embodiment of the present application.
  • Fig. 4 is another schematic flowchart of a personalized recommendation method provided by an embodiment of the present application.
  • Fig. 5 is a schematic block diagram of a terminal device provided by an embodiment of the present application.
  • Fig. 6 is a schematic block diagram of a business system provided by an embodiment of the present application.
  • Fig. 7 is a schematic structural diagram of a personalized recommendation system provided by an embodiment of the present application.
  • references described in this specification to "one embodiment” or “some embodiments”, etc. mean that one or more embodiments of the present application include a specific feature, structure, or characteristic described in combination with the embodiment. Therefore, the phrases “in one embodiment”, “in some embodiments”, “in some other embodiments”, “in some other embodiments”, etc. appearing in different places in this specification are not necessarily All refer to the same embodiment, but mean “one or more but not all embodiments” unless it is specifically emphasized otherwise.
  • the terms “including”, “including”, “having” and their variations all mean “including but not limited to” unless otherwise specifically emphasized.
  • the terminal device in the embodiments of the present application may be a mobile phone, a tablet computer, a wearable device (for example, a smart watch), a vehicle-mounted device, an augmented reality (AR) device, a virtual reality (VR) device, and a notebook Computers, ultra-mobile personal computers (UMPC), netbooks, personal digital assistants (personal digital assistants, PDAs) and other equipment.
  • the terminal device of the embodiment of the present application may include a touch screen for displaying service content to the user.
  • the specific type of the terminal device is not limited in the embodiments of the present application.
  • the service content in the embodiment of the present application may be advertisement, news, video, music, etc.
  • the specific type of the service content is not limited in the embodiment of the present application.
  • ML Machine learning
  • NLP natural language processing
  • IR information retrieval
  • IR information retrieval
  • evaluation evaluation
  • accuracy accuracy
  • recall recall rate
  • Fig. 1 shows an architecture diagram of a personalized recommendation system provided by an embodiment of the present application.
  • the personalized recommendation system includes a terminal device and a business system.
  • the terminal device may include device-level artificial intelligence (artificial intelligence). , AI) engine and end-side business applications (application, APP); business systems can also be referred to as cloud-side devices, including cloud-side business operation management systems, cloud-side online business systems, and cloud-side offline business systems.
  • Business operation management system In addition to providing the operation management support capabilities required by business operators, the business operation management system in the embodiments of this application can review the compliance and access conditions of the service content that is finally presented to users (automatic review) , Manual review), configure business operation-related rules and strategies (recommendation algorithm models effective for different scenarios, audience groups or manual recommendation rules, weights, priorities, etc.).
  • Cloud-side online business system real-time processing and response to service content requests from terminal devices, filtering massive cloud-side service content, and calling the cloud-side device personalized recommendation engine for clickthrough rate (CTR) prediction and ranking decisions
  • CTR clickthrough rate
  • Cloud-side offline business system The focus is to support the recommendation decision-making of the online business system, and provide three key capabilities required: semantic label processing of service content, audience targeting based on crowd classification rules, and personalized offline recommendation model calculations. It mainly relies on two data sources: first, business activity event logs collected with user authorization; second, massive service content data.
  • End-side device AI engine A device system-level AI engine, which mainly includes user knowledge base (user basic attributes, device attributes, full-scenario business behavior attributes, habits or preferences), context awareness engine (user status, environment, context), The recommendation engine and the ranking decision rule engine that are called by the end-side business application (application, APP) based on user authorization.
  • application, APP application
  • the recommendation engine of the end-side AI engine achieves secondary more accurate audience matching security filtering, personalized CTR prediction and final ranking decision. Filter out the service content that is inappropriate or invalid to the user, and ensure that the service content finally presented to the user is matched and relevant. Based on the high recall rate on the cloud side, the problem of recommendation accuracy is solved.
  • End-side business APP It is the client carrier of the terminal equipment with business service capabilities. Request personalized service content of the business through the end-side business APP to present it to the user, and collect relevant user behavior events and report it to the end-side AI engine and the cloud-side offline business system authorized by the user to provide algorithms for optimizing business experience in the next step Model update.
  • FIG. 2 shows an architecture diagram of a personalized recommendation system provided by an embodiment of the present application.
  • the personalized recommendation system can be applied to the scenario of advertisement recommendation, and the personalized recommendation system can be specifically personalized The system of personalized advertisement recommendation.
  • the personalized advertisement recommendation system includes terminal devices and business systems.
  • the terminal devices can include device-level AI engines and end-side advertising software development kits (SDK).
  • SDK end-side advertising software development kits
  • the business systems can also be called cloud-side devices. Including advertising operation management system, online advertising platform and offline advertising platform.
  • Advertising operation management system In addition to providing the operational management support capabilities required by advertising business operators, the business operation management system in the embodiment of this application can review the compliance and access conditions of the advertising creative content that is finally presented to users ( Automatic review, manual review), configure rules and strategies related to advertising business operations (advertising recommendation algorithm models effective for different scenarios and audience groups or manual recommendation rules, weights, priorities, etc.).
  • Online advertising platform real-time processing and response to advertising requests from terminal devices, filtering and screening the content of massive advertising tasks, calling the personalized advertising recommendation engine of the business system for CTR prediction, ranking decisions and outputting to the end-side primary-selected advertising ideas Content list and related parameters.
  • Offline advertising platform The focus is to support the recommendation decision-making of the online advertising system and provide three key capabilities required: semantic label processing of advertising creative service content, audience targeting according to crowd classification rules, and personalized advertising offline recommendation model calculations. Mainly rely on two data sources: first, the business behavior event log and advertising user behavior event log collected with the authorization of the user; second, the service content data of massive advertising creatives that are not related to the user.
  • End-side device-level AI engine A device-system-level AI engine.
  • the capabilities related to the embodiments of this application mainly include: user knowledge base (user basic attributes, device attributes, full-scenario business behavior attributes and habits/preferences), context awareness engine (User status, environment, context), recommendation engine and ranking decision rule engine based on user authorization and approval of end-side business APP advertisement calls.
  • the recommendation engine of the end-side AI engine Based on the advertising creative service content candidate list and the structural information related to the advertising creative service content returned by the first cloud-side advertising personalized recommendation, the recommendation engine of the end-side AI engine achieves secondary more accurate audience matching security filtering and personalized advertising CTR prediction and final ranking decision, filter out unsuitable or invalid advertising creative service content to users, and ensure that the final advertising creative service content presented to users is matched and relevant. Based on the high recall rate on the cloud side, the problem of high accuracy of personalized advertisement recommendation is solved.
  • End-side advertising SDK It can be integrated by business apps and is a client-side carrier with advertising business service capabilities in terminal devices. Through the end-side advertising SDK, request personalized advertising service content to be presented to users, and collect related user behavior events and advertising event logs, and report to the end-side AI engine and offline advertising system for the next step to provide optimized personalized advertising business algorithm models Update.
  • FIG. 3 is a schematic flowchart of a personalized recommendation method 100 provided by an embodiment of the present application. As shown in FIG. 3, the method 100 includes:
  • the cloud-side business operation management system will import and upload service content data in real time or regularly.
  • the service content data includes text, pictures, uniform resource locator (URL) addresses, and the like.
  • URL uniform resource locator
  • advertisers can import and upload some creative materials for advertisements in real time or regularly through the business operation management system.
  • the cloud-side online business system and the cloud-side offline business system are not visible to advertisers, and the cloud-side business operation management system is visible to advertisers.
  • JD.com promotes a type of computer, through the business operation management system, it can upload some text information, picture information, URL address, etc. about this type of computer to the cloud-side offline business system in real time.
  • Vipshop promotes a type of beauty products, it can upload some text information, picture information and URL addresses of such beauty products to the cloud-side offline business system in real time through the business operation management system.
  • BMW promotes a car, it can upload some text information, picture information, URL address, etc. about this car to the cloud-side offline business system in real time through the business operation management system.
  • the business operation management system may specifically be a cloud-side advertising operation system
  • the cloud-side business system may specifically be a cloud-side offline advertising platform
  • the cloud-side online business system may specifically be a cloud-side real-time advertising platform.
  • the cloud-side offline business system uses related learning algorithms (such as machine learning algorithms, deep learning algorithms, etc.) to calculate and mine service content, and generate structured and semantic content tags; these content tags are used in the cloud-side online business system and
  • the terminal equipment matches and filters the service content and calculates the content relevance.
  • the cloud-side offline business system can perform tag mining calculations on the computer-related information. For example, through analysis, it is found that such computers are "xxx brand laptops", “ xxx brand game console” and so on.
  • the cloud-side offline business system can perform tag mining calculations on the beauty-related information. For example, through analysis, it is possible to obtain such beauty makeup as “lipstick” and “perfume”. "and many more.
  • the cloud-side offline business system can perform tag mining calculations on the car-related information. For example, it can be analyzed to find that this type of car is "SUV”, "BMW X7", etc. Wait.
  • the cloud-side offline business system realizes tag mining of service content, which provides a basis for the following cloud-side content matching filtering or content correlation calculation.
  • the cloud-side business operation management system delivers personalized recommendation scenarios of service content, relevant policy rules, and audience information supported by the service content to the cloud-side offline business system and the cloud-side online business system.
  • the cloud-side business operation management system is visible to the service content provider.
  • the service content provider can configure the relevant policy rules of the personalized recommendation scenario of the related service content and the audience information supported by the service content, and pass these contents through the cloud-side business operation management system Distributed to the cloud-side offline business system and the cloud-side online business system.
  • major e-commerce platforms can place advertisements through the cloud-side business operation management system, that is, deliver corresponding advertising tasks or ideas.
  • the number of advertising tasks or ideas is not limited.
  • thousands of advertising tasks may be issued to a cloud-side offline business system (for example, a cloud-side offline advertising platform) or a cloud-side online business system (or, a cloud-side online advertising platform).
  • Table 1 shows a set of advertising tasks based on different audience information.
  • S104 The end-side business App collects user behavior events authorized by the user, and reports to the offline business system (including the big data platform).
  • the big data platform in the cloud-side offline business system can store and process user behavior events reported by the end-side business App.
  • user behavior events include but are not limited to display, click, slide, or download.
  • the terminal device may report some business behavior events of the user in the browser to the cloud-side offline business system.
  • the terminal device in S104 in the embodiment of the present application may also report some user business behavior events in a certain APP to the business system, provided that the user authorizes and agrees to upload these business behavior events.
  • user behavior events may be user behavior events of the user the previous day, or user behavior events of the user a few days ago, which is not limited in the embodiment of the present application.
  • S105 The cloud-side offline business system calculates user behavior attributes (only part of the data authorized by the user) and the user's audience orientation (based on audience orientation Options available on the cloud side), and trains a cloud-side personalized recommendation model.
  • the cloud-side offline business system can calculate the user's behavior attributes, the user's audience orientation, and other information based on the reported user behavior events.
  • 70% of the business behavior events reported by the end-side are that the user searches for "Huawei”, “Apple”, “vivo”, etc. in the terminal device, and the offline business system can determine that the user behavior attribute is interested in technology , And can also determine that the user’s audience is targeted at 20-30 year olds.
  • the offline business system can determine that the user behavior attribute is interested in cars, and can also determine the user's audience orientation For men aged 20-50.
  • the offline business system can determine that the user’s behavior attribute is interested in beauty, and can also determine the user’s
  • the target audience is women aged 20-40.
  • the offline business system can also update the cloud-side personalized recommendation model based on the reported user behavior events.
  • S106 The cloud-side offline business system synchronizes the following information to the cloud-side online business system: user behavior attributes, service content and related semantic structured tags, audience targeting results, and a personalized recommendation model after training.
  • the offline business system calculates the user behavior attributes, the user's audience-oriented information and updates the personalized recommendation model through related user behavior events, the user behavior attributes, the user's audience-oriented information, and the personalized recommendation model can be sent to the cloud side Online business system.
  • S107 The cloud-side online business system receives service content request information from the terminal device.
  • the terminal device when the user opens the browser, the terminal device is triggered to send service content request information to the cloud-side online business system for requesting the business system to recommend personalized service content.
  • S108 The cloud-side online business system performs matching and filtering on the service content according to the service content request information.
  • the cloud-side online business system may match and filter some of the advertisements according to the user behavior attributes calculated in S105, the result of the user's audience targeting, and the rules of different audience scenes when the advertiser places advertisements in S103.
  • the user determines in S105 that the user is interested in technology and cars, and the audience is targeted to men aged 20-50.
  • the advertising tasks shown in Table 1 you can set "shampoo”, “lipstick”, " Ads such as “makeup” and “potato chips” are filtered out.
  • S109 The cloud-side online business system calls the cloud-side personalized recommendation algorithm model to perform CTR prediction and result sorting.
  • the cloud-side online business system can call the personalized recommendation algorithm model synchronized to the cloud-side online business system in S106 to perform CTR prediction and result sorting on the filtered service content.
  • S110 The cloud-side online business system obtains a personalized recommendation preliminary selection result based on rules.
  • the cloud-side online business system can also obtain personalized recommendations based on some preset rules, such as geographic and model rules.
  • the cloud-side online system can further filter the filtered service content based on the two dimensions of the cloud-side personalized recommendation algorithm model and preset rules, so as to obtain the initial personalized recommendation. Election results.
  • the cloud-side online system can also filter the filtered service content based on one of the two dimensions of the side's personalized recommendation algorithm model and preset rules, so as to obtain a preliminary selection result of personalized recommendation.
  • the embodiment of the application does not make any limitation on this.
  • the cloud-side online business system may integrate multiple algorithm or rule recommendation results, and output the cloud-side preliminary recommendation results.
  • the cloud-side business system mainly implements the initial selection of personalized recommendations. Based on the limited user data on the cloud side, it finds a large amount of service content that may match the needs of users, and realizes the functions of one-time screening, CTR prediction and ranking with high recall rate. .
  • the cloud side has full, real-time service content and authorized collection of user business data, based on limited user data to perform personalized recommendations on the cloud side, find the most complete and real-time service content candidate set that may match the user’s needs, and recommend it on the cloud side If the accuracy is not too high, the focus is on the high recall rate.
  • the cloud-side online business system sends service content response information to the end-side APP.
  • the service content response information includes a personalized recommendation primary selection result: multiple service content candidate sets and semantic tags corresponding to the service content candidate sets.
  • Table 2 shows a preliminary selection result of a personalized recommendation output from the cloud side.
  • the results of the preliminary selection of personalized recommendation shown in Table 2 are illustrated by taking the service content as the creative content of the advertisement as an example.
  • the multiple service content candidate sets may be the ID of each service content and the URL address corresponding to each service content.
  • the cloud-side online business system After the cloud-side online business system performs the first recommendation result according to the personalized recommendation algorithm or rule, it can also determine the semantic label corresponding to the first recommendation result. For example, the multiple advertising tasks shown in Table 2 contain many The cloud-side online business system can tag these advertising tasks with semantic tags based on the tag mining of the service content in S101 and S102.
  • the business system carries the semantic label corresponding to the recommendation result in the first recommendation result issued to the terminal device, which can facilitate the terminal device to perform secondary screening.
  • the business system downloads the terminal device
  • the recommended result sent is the final recommended result, and the recommended result only contains the service content candidate set.
  • the terminal device can request the corresponding service content from the business system according to the URL address therein, thereby displaying the corresponding service content to the user.
  • the business system delivers multiple service content candidate sets (including corresponding IDs and URL addresses) to the terminal device, and at the same time also delivers semantic tags corresponding to the service content, so that the terminal device can access the cloud side easily.
  • the output service content is subjected to secondary screening, which helps to improve the accuracy of personalized recommendations.
  • the personalized recommendation method of the embodiment of this application can give full play to the comprehensive and accurate user data advantages of the terminal device, without uploading additional user data and tags, and only need to securely filter and accurately match the service content that meets the user’s demands in the downstream direction, based on business Based on the high recall rate of the system's one-time advertising recommendation service content, the data advantage of the terminal equipment is superimposed, thereby solving the problem of high accuracy rate of personalized recommendation.
  • a personalized recommendation service experience with high recall rate and high accuracy is provided.
  • the personalized recommendation primary selection result output by the cloud-side business system may also include audience information and/or launch price of the service content.
  • Table 3 shows another personalized recommendation primary selection result output by the cloud side.
  • the results of the preliminary selection of personalized recommendation shown in Table 3 are illustrated by taking the service content as the creative content of the advertisement as an example.
  • the end-side APP determines whether the user authorizes the end-side APP to call the end-side AI engine.
  • the end-side APP can call the end-side AI engine to implement the security filtering of service content and the configuration of the final decision of secondary personalized recommendation.
  • the end-side APP sends personalized recommendation request information to the end-side AI engine.
  • the personalized recommendation request information includes multiple service content candidate sets and semantic tags corresponding to the service content candidate sets.
  • the end-side APP can send multiple service content candidate sets of the service content output by the cloud-side online business system and the tags of the service content candidate set to the end-side AI engine, requesting the AI engine to perform secondary personalization on the results of the personalized recommendation primary selection recommend.
  • the end-side APP may send the ID corresponding to the service content and the corresponding semantic label to the end-side AI engine.
  • the end-side APP can send the ID corresponding to the service content, the corresponding semantic label, and the audience to the end-side AI engine Information and launch price.
  • S114 The end-side AI engine performs secondary matching, screening, and filtering on the service content in the result of the personalized recommendation primary selection based on the input multiple service content candidate sets, combined with user related information.
  • the end-side AI engine mainly includes: a user knowledge base with the most comprehensive and accurate user data and attributes, a situational awareness engine that perceives the context and environment in real time, and an end-side personalized recommendation engine.
  • the context awareness engine is mainly used for calculation and input.
  • the real-time status parameters of the current user on the end-side are finally sent to the end-side personalized recommendation engine as input parameters.
  • the end-side AI engine can filter the recommendation results sent by the end-side APP.
  • the end-side AI engine can calculate the correlation coefficient based on the recommendation result and the user knowledge base.
  • the user’s knowledge base includes user attribute characteristics: gender, age, interest, blog post characteristics, attention user characteristics, location-based service (location-based service, LBS) location information (ie real-time location), education, occupation, family, consumption Level etc.
  • the end-side AI engine can now calculate the correlation coefficient with the service content Connect 1-Connect 7 based on these user attribute characteristics.
  • the user’s knowledge base includes information: the user’s age is 20-30 years old, the gender is male, the annual consumption amount is 50,000-80,000, and the interest is technology products, and certain preset conditions can be passed
  • the correlation coefficient shown in Table 4 is calculated.
  • the first recommendation result includes the launch price
  • commercialization factors may also be considered at the same time, and the calculation of the correlation coefficient may also consider the launch price.
  • Table 5 shows a calculation result of a correlation coefficient.
  • the end-side AI engine can output correlation coefficients in descending order: "smart appliances”, “domestic cars”, “tablets”, “lightweight notebooks” "Computer”, “Japanese company car”, “beauty phone” and “German car”.
  • the end-side AI engine invokes the algorithm model of the end-side personalized recommendation engine, performs secondary CTR prediction and sorting, realizes the final selection of personalized recommendations, and finally decides to output safe, accurate, and personalized recommendations that match end-side user needs Service Content.
  • CTR prediction is one of the most core algorithms in calculating advertisements.
  • CTR prediction is to predict the click situation of each advertisement and predict whether the user will click or not.
  • CTR prediction is related to many factors, such as historical click-through rate, advertising location, time, and users.
  • the algorithm model of the personalized recommendation engine is a model obtained by comprehensively considering various factors and characteristics and training on a large amount of historical data.
  • the training samples for CTR prediction are generally obtained from historical logs (log) and offline feature libraries. The sample label is relatively easy. User clicks are marked as 1, and no clicks are marked as 0. Many characteristics will be considered, such as the demographic characteristics of the user, the characteristics of the advertisement itself, and the characteristics of the advertisement display.
  • One-Hot codes are sampled for category features. For example, there are three types of occupations: students, white-collar workers, and workers. Then a vector of length 3 will be used to represent them respectively: [1, 0, 0], [0, 1, 0 ], [0, 0, 1]. This can make the feature dimension expand greatly, and the feature will be very sparse. At present, the advertising feature libraries of many companies are hundreds of millions.
  • the ranking may be effective cost per mile (eCPM) ranking.
  • eCPM refers to the advertising revenue that can be obtained for every thousand impressions.
  • the unit of display can be a webpage, an ad unit, etc.
  • advertisers can analyze the effects of advertisements, optimize and adjust advertisements in a targeted manner, thereby increasing revenue.
  • the end-side AI engine determines that the Connect 1-Connect 7 advertising tasks do not meet the user's preferences, or they are all disgusting to the user Can filter out these advertising tasks and notify the end-side APP not to display any advertising content.
  • the personalized recommendation method of the embodiment of this application can give full play to the comprehensive and accurate user data advantages of the terminal device, without uploading additional user data and tags, and only need to securely filter and accurately match the service content that meets the user’s demands in the downstream direction, based on business Based on the high recall rate of the one-time recommended service content, the system superimposes the data advantages of terminal equipment to solve the problem of high accuracy rate of personalized recommendation.
  • a personalized recommendation service experience with high recall rate and high accuracy is provided.
  • the end-side AI engine sends personalized recommendation response information to the end-side APP, where the personalized recommendation response information includes a list of service content that ultimately meets the demands of the user.
  • the end-side AI engine may send the second recommendation result obtained according to the algorithm model of the end-side personalized recommendation engine in S115 to the end-side APP.
  • the end-side AI engine can select the ID of the service content with the top two prediction results and send it to the end-side APP.
  • the end-side AI engine determines that the IDs of the service content output by the secondary prediction and sorting are Connect 4 and Connect 6.
  • the end-side APP downloads service content data from the cloud-side online business system to the end-side APP according to the service content list displayed by the final decision.
  • the end-side APP After the end-side APP receives the ID of the service content output by the final recommendation decision output by the end-side AI engine, it downloads the service content output by the final recommendation decision from the cloud-side online business system to the end-side APP according to the URL address corresponding to the ID.
  • the end-side APP after receiving the first two digits (Connect 4 and Connect 6) of the secondary prediction and ranking output, the end-side APP requests the cloud-side online business system to download information about "smart home appliances” and “smart appliances” according to URL addresses 4 and URL6.
  • the advertising task of Tablet PC is the first two digits (Connect 4 and Connect 6) of the secondary prediction and ranking output.
  • S118 The end-side APP outputs and displays the final service content to the user.
  • the end-side APP After receiving the final personalized service content, the end-side APP displays the service content to the user.
  • the two advertising tasks of "smart appliances” and “tablets” are displayed on the user's browser interface.
  • S119 can refer to the process of S104 above, and for brevity, details are not described herein again.
  • the cloud-side offline business system updates the user behavior attributes and the cloud-side personalized recommendation model in real time.
  • the end-side APP can upload the user's business behavior events for the two advertising tasks of "smart home appliances" and “tablets" to the cloud-side offline business system.
  • the end-side AI engine can update the user knowledge base, and the cloud-side offline business system can calculate Some user behavior attributes, user audience orientation and update cloud-side personalized recommendation model.
  • two personalized recommendation algorithm modeling with each focus, giving full play to the advantages of data differentiation of the terminal and cloud.
  • the selection of the algorithm and feature set of the two personalized recommendation algorithm modeling is independent and independent. It is only the input of the candidate set of the personalized recommendation of the terminal device, and it depends on the output of the personalized recommendation result of the business system.
  • the terminal device has complete user knowledge base data, but the end-side APP does not save or only saves incomplete and non-real-time part of the service content. Therefore, the model based on the end-side personalized recommendation algorithm should be based on the one-time recommendation on the cloud side with a high recall rate. On the basis of finding more complete service content and multiple candidate sets, we should give full play to the advantages of the high-quality integrity of the local user data of the terminal device, and focus on solving High accuracy rate is recommended.
  • the semantic and structured content tag mining calculation (real-time increment, full volume) of the service content is unified on the cloud side. Therefore, in addition to the cloud-side personalized recommendation process, the service content tags are used to achieve rule matching and filtering between the service content and the user’s business attributes, and it is passed to the end-side APP in the cloud-side one-time recommendation output service content candidate set list, and at the same time It also includes semantic tags corresponding to the service content candidate set, and this information is brought to the end-side APP for further matching and filtering with the end-side user knowledge base data and content tags on the end-side.
  • mobile advertising also requires higher requirements for the promotion of mobile advertising.
  • the ultimate goal of mobile advertising should be "advertising as a service", and the content of advertising services provided to users is exactly what users need, not harassing information. Therefore, an accurate and in-depth understanding of user needs, providing high-quality native content that matches user needs, and promoting service content are the long-term goals of a sustainable and sound business development.
  • FIG. 4 is a schematic flowchart of a personalized recommendation method 200 according to an embodiment of the present application.
  • the method 200 may be used in an advertisement recommendation scenario. As shown in FIG. 4, the method 200 includes:
  • the cloud-side advertising operation system imports and uploads massive amounts of advertising creative content in real time or regularly.
  • the cloud-side offline advertising platform uses related learning algorithms (for example, machine learning algorithms, deep learning algorithms, etc.) to calculate and mine the content of advertising creative services to generate structured and semantic content tags; these content tags are used for cloud-side online businesses
  • the system and terminal equipment match and filter the creative content of advertisements and calculate the content relevance.
  • the cloud-side advertising operation system delivers the personalized recommendation scenarios of advertising creative content, relevant policy rules, and audience information supported by service content to the cloud-side offline advertising platform and the cloud-side online advertising platform.
  • the cloud-side advertising operation system is visible to advertisers. Advertisers can configure relevant policy rules for personalized recommendation scenarios of related advertisements and audience information of creative content of advertisements, and distribute these contents to cloud-side offline advertisements through the cloud-side advertising operation system Platform and cloud-side online advertising platform
  • advertising business operators configure relevant policy rules for personalized recommendation scenarios to support different scenarios and audience information; use a combination of different personalized advertising recommendation algorithms, and support related content relevance and exclusive matching Filtering rules.
  • the advertiser will also define the audience information (or related audience group rules) specified by the corresponding delivery task in the cloud-side advertising operation system (on the advertiser-oriented delivery platform). Both types of strategies will be issued to the cloud-side offline advertising platform and the cloud-side online advertising platform.
  • S204 The end-side business APP collects user behavior events authorized by the user and reports them to the cloud-side offline advertising platform (including the big data platform).
  • the big data platform in the cloud-side offline advertising platform can store and process user behavior events reported by the end-side business APP.
  • user behavior events include but are not limited to display, click, slide, or download, etc.
  • the cloud-side offline advertising platform calculates user behavior attributes (only part of the data authorized by the user) and the user's audience group (based on audience-oriented Options available on the cloud side), and trains a cloud-side personalized recommendation model.
  • the big data platform in the cloud-side offline advertising platform calculates user behavior attributes, the user's audience orientation, and updates the cloud-side personalized advertisement recommendation model based on part of the data reported by the terminal device that is authorized by the user.
  • the audience targeting information calculated by the cloud side offline advertising platform will not be particularly accurate, and the accuracy of the cloud side personalized recommendation model will not be too high.
  • S205 can refer to the description of S105 above, and for the sake of brevity, details are not repeated here.
  • the cloud-side offline advertising platform synchronizes the following information to the cloud-side online advertising platform: user behavior attributes, service content and related semantic structured tags, audience targeting results, and a personalized advertising recommendation model after training.
  • S206 may refer to the description of S106 above, and for the sake of brevity, details are not repeated here.
  • the cloud-side online advertisement platform receives advertisement request information from the advertisement SDK of the end-side App.
  • the terminal device when the user opens the browser, the terminal device is triggered to send an advertisement request to the cloud-side online advertisement platform.
  • S208 The cloud-side online advertising platform performs matching filtering on the creative content of the advertisement.
  • the cloud-side online advertising platform matches and filters the creative content of advertising according to the corresponding scene, advertising request parameters, user behavior attributes, and audience targeting results.
  • the cloud-side online advertising platform may refer to the example in S108 for the process of screening the creative content of the advertisement. For the sake of brevity, it will not be repeated here.
  • S209 The cloud-side online advertising platform calls the cloud-side personalized recommendation algorithm model to perform CTR prediction and result sorting.
  • the cloud-side online advertising platform can call the personalized recommendation model synchronized to the cloud-side online advertising platform in S206 to perform CTR prediction and result sorting on the filtered advertising creative content.
  • the cloud-side online advertising platform obtains a personalized recommendation preliminary selection result based on rules.
  • the cloud-side online advertising platform can also further filter the filtered advertising creative content according to some preset rules, such as geographical and model rules, so as to obtain personalized recommendation preliminary results.
  • the advertising business system mainly implements the primary selection of personalized advertising recommendation. Based on the limited user data on the cloud side (including user advertising behavior data), it finds a large amount of creative advertising content that may match the needs of users, and achieves a high recall rate.
  • One-time screening and recommendation CTR prediction and ranking functions are used.
  • the cloud-side online advertising platform sends advertisement request response information to the advertisement SDK, where the advertisement request response information includes the personalized recommendation primary selection result.
  • the personalized recommendation preliminary selection results include, but are not limited to: multiple candidate sets of advertising creative content, semantic structured tags of advertising creative content, or real-time advertising bidding information, and so on. It should be understood that the results of the preliminary selection of personalized recommendation can refer to Table 2 or Table 3 above. For brevity, details are not repeated here.
  • the advertising business system carries the semantic tags corresponding to the creative content of the advertisement in the first recommendation result issued to the terminal device, which can facilitate the secondary screening of the terminal device.
  • the advertising business system in the prior art provides The recommendation result delivered by the terminal device is the final recommendation result, and the recommendation result only contains a candidate set of advertising creative content. After receiving these advertisement creative content candidate sets, the terminal device can request the corresponding advertisement creative content from the business system according to the URL address therein, thereby displaying the corresponding advertisement creative content to the user.
  • the business system delivers multiple advertisement creative content candidate sets (including corresponding IDs and URL addresses) to the terminal device, and at the same time also delivers the semantic tags corresponding to the service content, so that the terminal device can easily access the cloud.
  • the service content output from the side is screened twice, which helps to improve the accuracy of personalized recommendations.
  • the personalized recommendation method of the embodiment of this application can give full play to the comprehensive and accurate user data advantages of the terminal device, without uploading additional user data and tags, and only need to securely filter and accurately match the creative content of advertisements that meet the user's demands in the downstream direction.
  • the data advantage of terminal equipment is superimposed to solve the problem of high accuracy rate of personalized recommendation.
  • the advertising SDK determines whether the user is authorized to call the end-side AI engine.
  • the advertising SDK When the advertising SDK confirms that the user is authorized to call the end-side AI engine, the advertising SDK calls the end-side AI engine to implement the security filtering of creative content and the second personalized recommendation decision.
  • the advertising SDK sends personalized advertisement recommendation request information to the end-side AI engine, and the personalized recommendation request information includes the personalized recommendation primary selection result output by the cloud-side online advertising platform.
  • the advertising SDK may also send the placement price (or bidding information) to the end-side AI engine.
  • the end-side AI engine performs secondary matching and filtering on the advertising creative content in the result of the personalized recommendation primary selection based on the inputted candidate set of advertising creative content and combined with user related information.
  • the end-side AI engine invokes the algorithm model of the end-side personalized recommendation engine, performs secondary CTR prediction and sorting, realizes the final selection of personalized recommendations, and finally decides to output safe, accurate and personalized advertisements that match end-side user needs Creative content.
  • the personalized recommendation method of the embodiment of this application can take full advantage of the comprehensive and accurate user data saved by the terminal device, without uploading additional user data and tags, and only need to securely filter and accurately match the advertising creative service content in the downstream direction that meets the needs of users , Based on the high recall rate of the one-time advertisement recommendation service content of the advertising business system, superimpose the user data advantage of the terminal device to solve the problem of high accuracy rate of personalized advertisement recommendation.
  • the secondary personalized advertisement recommendation cooperation of terminal cloud collaboration it provides a personalized advertisement recommendation service experience with high recall rate and high accuracy.
  • the end-side AI engine sends personalized advertisement recommendation request response information to the advertising SDK, where the personalized advertisement recommendation request response information includes a list of advertisement creative content that ultimately meets the demands of the user.
  • the advertising SDK downloads the advertising creative content data from the cloud side to the end-side APP according to the finally decided advertising creative content list.
  • the advertisement SDK outputs and displays the final personalized advertisement creative content to the user.
  • the advertising SDK Based on the user's authorization and consent to collect, the advertising SDK reports user behavior events to the cloud-side online advertising platform.
  • S220 The cloud-side offline advertising platform updates the user advertisement attributes and the cloud-side personalized advertisement recommendation model in real time.
  • the advertising SDK can also report user behavior time to the end-side AI engine, and the end-side AI engine can update user data and related attributes in the user knowledge base in real time.
  • the method 200 further includes:
  • S221 The cloud-side advertisement online platform calculates the consumption details of advertisement placement.
  • the personalized recommendation method of the embodiment of the present application is based on the high recall rate of the cloud-side one-time advertisement recommendation algorithm model, and superimposes the advantages of end-side data to solve the problems of brand safety and high accuracy of personalized advertisement recommendation.
  • End-cloud collaboration two personalized advertising recommendation algorithm modeling with each focus, give full play to the advantages of the data differentiation of end-cloud.
  • the selection of the algorithm and feature set of the two personalized advertising recommendation algorithm modeling is independent and independent. It is only the input of the candidate set recommended by the end-side personalized advertisement, and it depends on the output of the candidate set list of the cloud-side personalized advertisement recommendation.
  • the personalized recommendation method of the embodiment of the present application helps to solve the problem of low audience targeting accuracy caused by cloud-side data quality and integrity issues. Leverage the respective data advantages of terminal-cloud collaboration to form a complete audience-oriented collaboration.
  • the cloud side has full/real-time advertising tasks and creative service content and authorization to collect part of user business data and user advertising behavior events. Based on limited user business data, the cloud side personalized advertising recommendation algorithm is modeled to find possible matching user requirements, The most complete and real-time advertising service content candidate set, focusing on solving the high recall rate of advertising recommendations when the recommendation accuracy is not high.
  • the end-side has complete user knowledge base data, but the end-side business does not save or only saves incomplete and non-real-time service content. Therefore, modeling based on the end-side personalized advertisement recommendation algorithm needs to be based on the cloud-side one-time ad recommendation high recall rate to return to the multiple candidate sets of advertising creative service content, give full play to the advantages of the high-quality integrity of end-side local user data, and focus on solving The final advertisement recommends a high accuracy rate.
  • the personalized recommendation method of the embodiment of the present application helps solve the problem of relying solely on the end-side recommendation algorithm model and data, and can only solve the limited scenarios of end-side local content recommendation.
  • the problem is that it is impossible to obtain massive and rich and accurate cloud-side service content; it also helps to solve the problem of low accuracy of recommendation results caused by cloud-side data quality and integrity problems of the cloud-side recommendation algorithm model, especially the content is sensitive to security Sexual issues (users are not suitable for content), so as to meet user privacy and security, and more accurate personalized recommendation user experience.
  • the semantic and structured content tag mining calculation (real-time increment, full volume) of the advertising creative service content is unified on the cloud side. Therefore, in addition to the cloud-side personalized advertisement recommendation process, the creative content of the creative content and the user’s business attributes are matched and filtered using the tags of the creative service content on the cloud side, and the creative service content that is passed to the cloud side of the end-side advertising SDK for one-time advertisement recommendation output.
  • the candidate set list also includes tags corresponding to the candidate set of advertising creative service content. This information is brought to the end-side advertising SDK for further matching and filtering by combining the data of the end-side user knowledge base with the advertising creative content tags on the end-side.
  • the service content may be video services, advertising services, news services, etc.
  • the foregoing method 200 only uses an advertising service as an example for illustration. In the embodiments of the present application, the specific form of the service content is described. It does not make any restrictions.
  • FIG. 5 shows a schematic structural diagram of a terminal device 300 according to an embodiment of the present application.
  • the terminal device 300 includes a touch screen 310, a memory 320, and a processor 330, wherein one or more computer programs are Stored in the memory 320, one or more computer programs include instructions.
  • the instructions are executed by the processor 330, the terminal device 300 is caused to perform the following operations:
  • Receive service content response information sent by the business system includes a service content candidate set, the service content candidate set includes multiple service contents, and the service content candidate set is the user behavior reported by the business system according to the terminal device Event
  • the one or more service contents are displayed to the user through the touch screen 310.
  • the service content response information also includes label information of each service content in the service content candidate set, and when the instruction is executed by the processor 330, the terminal device 300 is caused to perform the following operations:
  • the one or more service contents are determined according to the data authorized by the user and the label information of each service content in the service content candidate set.
  • the terminal device 300 when the instruction is executed by the processor 330, the terminal device 300 is caused to perform the following operations:
  • the one or more service contents are determined.
  • the terminal device 300 when the instruction is executed by the processor 330, the terminal device 300 is caused to perform the following operations:
  • the service content corresponding to the correlation coefficient that is greater than or equal to the preset value among the plurality of correlation coefficients is displayed to the user through the touch screen.
  • the service content is a massive real-time update service content of the business system.
  • the service content includes one or more of advertising services, video services or news services.
  • the terminal device 300 may correspond to the terminal device in the above-mentioned personalized recommendation method 100 or method 200, and the processor 330 may be used to execute the operation of the terminal device in the above-mentioned method 100 or method 200.
  • Fig. 6 shows a schematic block diagram of a business system 400 provided by an embodiment of the present application.
  • the business system 400 includes a business operation management system 410, an offline business system 420, and an online business system 430, in which,
  • the business operation management system 410 is configured to receive the first service content candidate set released by the service content distributor;
  • the business operation management system 410 is further configured to send the first service content candidate set to the offline business system 420;
  • the offline service system 420 is configured to determine the label information of the service content in the first service content candidate set, and send it to the online service system 430;
  • the online business system 430 is configured to receive service content request information sent by a terminal device
  • the online business system 430 is further configured to determine a second service content candidate set from the first service content candidate set according to the service content request information;
  • the online service system 430 is further configured to send service content request response information to the terminal device, where the service content request response information includes the second service content candidate set and label information of the service content in the second service content candidate set.
  • the offline business system 420 is also used for:
  • the user behavior event determine the user behavior attribute, the user's audience orientation information, and train a personalized recommendation model.
  • the online business system 430 is specifically used for:
  • the user's audience orientation information and the label information of the service content matching and filtering the service content in the first service content candidate set;
  • the business system 400 may correspond to the business system in the method 100 or the advertising business system in the method 200;
  • the business operation management system 410 may correspond to the business operation management system in the method 100, or may correspond to For the advertising operation management system in method 200;
  • offline business system 420 may correspond to the cloud-side offline business system in method 100, or may correspond to the offline advertising platform in method 200;
  • online business system 430 may correspond to the method 100 in The cloud-side online business system may also correspond to the online advertising platform in method 200.
  • FIG. 7 shows a schematic block diagram of a personalized recommendation system 500 provided by an embodiment of the present application.
  • the personalized recommendation system includes a terminal device 510 and a service system 520, where the terminal device may be the aforementioned The terminal device 300, the service system may be the aforementioned service system 400.
  • the disclosed system, device, and method may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components can be combined or It can be integrated into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • each unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the technical solution of this application essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disk or optical disk and other media that can store program code .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Multimedia (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Library & Information Science (AREA)
  • Computational Linguistics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Transfer Between Computers (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种个性化推荐的方法、系统和终端设备，该方法包括：终端设备向业务系统发送服务内容请求信息；终端设备接收业务系统发送的服务内容响应信息，该服务内容响应信息包括服务内容候选集，该服务内容候选集中包括多个服务内容，该服务内容候选集为业务系统根据终端设备上报的用户行为事件得到的；终端设备根据用户授权的数据，从该服务内容候选集中确定一个或者多个服务内容；终端设备向用户展示该一个或者多个服务内容。该方法有助于提高个性化推荐的准确性。

Description

个性化推荐的方法、终端设备和系统
本申请要求在2019年4月3日提交中国国家知识产权局、申请号为201910266830.1、发明名称为“个性化推荐的方法、终端设备和系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及数据处理领域，并且更具体地，涉及一种个性化推荐的方法、系统和终端设备。
背景技术
随着移动互联网的极速发展，为用户随时随地获取所需的丰富信息及服务提供了极大便利。然而如何在现有多方激烈竞争的业务生态快速胜出，快速抓住用户的注意力并且获取更大的商业化利益，就需要深度思考如何做到真正的以用户为中心，提供更实时、精准、个性化的精品优质的服务内容。
现有的个性化推荐场景中，虽然端侧保存有丰富的用户个人隐私数据，但是出于隐私保护的目的，云侧只能获得有限的用户数据，可能导致云侧输出的个性化推荐服务内容的准确性不高，从而无法满足用户的诉求。
发明内容
本申请提供一种个性化推荐的方法、系统和终端设备，有助于提高个性化推荐的准确性。
第一方面，提供了一种个性化推荐方法，该方法包括：终端设备向业务系统发送服务内容请求信息；该终端设备接收该业务系统发送的服务内容响应信息，该服务内容响应信息包括服务内容候选集，该服务内容候选集中包括多个服务内容，该服务内容候选集为该业务系统根据该终端设备上报的用户行为事件得到的；该终端设备根据用户授权的数据，从该服务内容候选集中确定一个或者多个服务内容；该终端设备向用户展示该一个或者多个服务内容。
本申请实施例的个性化推荐方法，可以充分发挥端侧全面准确的用户数据优势，无需上传额外的用户数据，只需在下行方向安全过滤和精准匹配符合用户诉求的服务内容，基于业务系统推荐的服务内容的高召回率的基础上，叠加终端设备保存的用户数据的优势，解决个性化推荐高准确率的问题。通过端云协同的二次个性化推荐配合，提供高召回率、高准确率的个性化推荐服务体验。
结合第一方面，在第一方面可能的实现方式中，该服务内容响应信息中还包括该服务内容候选集中每个服务内容的标签信息；其中，该终端设备根据用户授权的数据，从该服务内容候选集中确定一个或者多个服务内容，包括：该终端设备根据该用户授权的数据和该服务内容候选集中每个服务内容的标签信息，确定该一个或者多个服务内容。
本申请实施例的个性化推荐方法，业务系统传给终端设备的服务内容候选集列表中， 同时还包括对应服务内容候选集的语义化标签，这些信息一同发送给终端设备，终端设备可以进一步结合终端设备保存的用户知识库的数据与语义化标签进行匹配过滤，从而提升个性化推荐的准确率。
结合第一方面，在第一方面可能的实现方式中，该终端设备根据该用户授权的数据和该服务内容候选集中每个服务内容的标签信息，确定该一个或者多个服务内容，包括：该终端设备根据该用户授权的数据和该每个服务内容的标签信息，确定该每个服务内容与该用户授权的数据的相关系数；该终端设备根据该相关系数，确定该一个或者多个服务内容。
本申请实施例的个性化推荐方法，终端设备可以根据终端设备保存的用户知识库的数据与语义化标签，确定该语义化标签对应的服务内容与用户知识库的数据的相关系数，通过计算得到的多个相关系数确定最终向用户展示的服务内容，有助于提升个性化推荐的准确率。
在一些可能的实现方式中，终端设备可以根据多个服务内容中每个服务内容的标签信息和用户授权的数据，确定多个相关系数；终端设备可以向用户展示所述多个相关系数中大于或者等于预设值的相关系数对应的服务内容。
结合第一方面，在第一方面可能的实现方式中，该服务内容为该业务系统海量实时更新的服务内容。
本申请实施例的个性化推荐方法，终端设备保存有完整的用户知识库的数据，但终端设备未保存或只保存不完整、不实时的部分服务内容。因此，基于终端设备个性化推荐算法建模要基于业务系统一次推荐找到海量实时更新的服务内容的基础上，充分发挥终端设备本地用户数据高质量完整性的优势，从而有助于提升个性化推荐的准确率。
结合第一方面，在第一方面可能的实现方式中，该服务内容包括广告业务、视频业务或者新闻业务中的一种或者多种。
第二方面，提供了一种个性化推荐方法，该方法包括：业务系统接收终端设备发送的服务内容请求信息；该业务系统根据该服务内容请求信息，确定服务内容候选集，该服务内容候选集包括多个服务内容；该业务系统向该终端设备发送服务内容响应信息，该服务内容响应信息包括该服务内容候选集；该终端设备根据用户授权的数据，从该服务内容候选集中确定一个或者多个服务内容；该终端设备向用户展示该一个或者多个服务内容。
本申请实施例的个性化推荐方法，可以充分发挥终端设备具有全面准确的用户数据优势，终端设备无需向业务系统上传额外的用户数据，只需在下行方向安全过滤和精准匹配符合用户诉求的服务内容，基于业务系统一次推荐服务内容的高召回率的基础上，叠加终端设备保存的全面准确的用户数据优势，解决个性化推荐高准确率的问题。通过端云协同的二次个性化推荐配合，提供高召回率高准确率的个性化推荐服务体验。
结合第二方面，在第二方面可能的实现方式中，该服务内容响应信息中还包括该服务内容候选集中每个服务内容的标签信息；其中，该终端设备根据用户授权的数据，从该服务内容候选集中确定一个或者多个服务内容，包括：该终端设备根据该用户授权的数据和该服务内容候选集中每个服务内容的标签信息，确定该一个或者多个服务内容。
本申请实施例的个性化推荐方法，业务系统传给终端设备的服务内容候选集列表中，同时还包括对应服务内容候选集的语义化标签，这些信息一同发送给终端设备，终端设备可以进一步结合终端设备保存的用户知识库的数据与语义化标签进行匹配过滤，从而提升个性化推荐的准确率。
结合第二方面，在第二方面可能的实现方式中，该终端设备根据该用户授权的数据和该服务内容候选集中每个服务内容的标签信息，确定该一个或者多个服务内容，包括：该终端设备根据该用户授权的数据和该每个服务内容的标签信息，确定该每个服务内容与该用户授权的数据的相关系数；该终端设备根据该相关系数，确定该一个或者多个服务内容。
本申请实施例的个性化推荐方法，终端设备可以根据终端设备保存的用户知识库的数据与语义化标签，确定该语义化标签对应的服务内容与用户知识库的数据的相关系数，通过计算得到的多个相关系数确定最终向用户展示的服务内容，有助于提升个性化推荐的准确率。
结合第二方面，在第二方面可能的实现方式中，该业务系统向该终端设备发送服务内容响应信息之前，该方法还包括：该业务系统根据预设的算法，确定该每个服务内容的标签信息。
本申请实施例的个性化推荐方法，为了提升服务内容跟用户相关性的匹配筛选过滤过程，统一在业务系统完成对服务内容的语义化、结构化的内容标签挖掘计算。因此，除了在业务系统个性化推荐过程中，使用服务内容的标签实现服务内容与用户业务属性的规则匹配过滤外，业务系统传给终端设备的一次推荐输出的服务内容候选集列表中，同时还可以包括对应服务内容候选集的语义化标签。业务系统将这些信息一同带给终端设备，有助于终端设备进一步结合终端设备的用户知识库的数据与内容标签进行匹配过滤，从而有助于提升个性化推荐的准确率。
结合第二方面，在第二方面可能的实现方式中，该业务系统根据该服务内容请求信息，确定服务内容候选集，包括：该业务系统接收该终端设备发送的用户行为事件；该业务系统根据该用户行为事件，确定用户行为属性以及用户的受众群体信息；该业务系统根据该用户行为属性以及该用户的受众群体信息，从该业务系统保存的服务内容中确定该服务内容候选集。
本申请实施例中，业务系统主要实现个性化推荐的初选，基于业务系统有限的用户数据，找到可能匹配用户需求的海量服务内容，实现高召回率的一次筛选、CTR预测和排序功能。
结合第二方面，在第二方面可能的实现方式中，该服务内容为该业务系统海量实时更新的服务内容中的一种或者多种。
结合第二方面，在第二方面可能的实现方式中，该服务内容为广告业务、视频业务或者新闻业务。
本申请实施例的个性化推荐方法，终端设备保存有完整的用户知识库的数据，但终端设备未保存或只保存不完整、不实时的部分服务内容。因此，基于终端设备个性化推荐算法建模要基于业务系统一次推荐找到海量实时更新的服务内容的基础上，充分发挥终端设备本地用户数据高质量完整性的优势，从而有助于提升个性化推荐的准确率。
第三方面，提供了一种个性化推荐的方法，该方法应用于个性化推荐系统，该个性化推荐系统包括终端设备和业务系统，其中，终端设备包括处理器和存储器，该存储器存储有一个或者多个程序，业务系统包括业务运营管理系统、在线业务系统和离线业务系统，该方法包括:业务运营管理系统向所述离线业务系统发送第一服务内容候选集；离线业务系统根据预设的算法，确定第一服务内容候选集中服务内容的标签信息；离线业务系统向在线业务系统发送第一服务内容候选集中服务内容的标签信息；在线业务系统接收所述终 端设备发送的服务内容请求信息；在线业务系统根据所述服务内容请求信息，从第一服务内容候选集中确定第二服务内容候选集，第二服务内容候选集包括多个服务内容；在线业务系统向所述终端设备发送服务内容响应信息，服务内容响应信息包括第二服务内容候选集和第二服务内容候选集中每个服务内容的标签信息；该终端设备根据用户授权的数据和所述每个服务内容的标签信息，从多个服务内容中确定一个或者多个服务内容；该终端设备向用户展示所述一个或者多个服务内容。
结合第三方面，在第三方面可能的实现方式中，该终端设备根据用户授权的数据和所述每个服务内容的标签信息，从多个服务内容中确定一个或者多个服务内容，包括：该终端设备根据所述用户授权的数据和所述每个服务内容的标签信息，确定所述每个服务内容与所述用户授权的数据的相关系数；该终端设备根据所述相关系数，确定所述一个或者多个服务内容。
结合第三方面，在第三方面可能的实现方式中，该方法还包括：离线业务系统接收该终端设备发送的用户行为事件；离线业务系统根据该用户行为事件，确定用户行为属性以及用户的受众群体信息；离线业务系统向在线业务系统发送所述用户行为属性以及所述用户的受众群体信息。
结合第三方面，在第三方面可能的实现方式中，该在线业务系统根据所述服务内容请求信息，从第一服务内容候选集中确定第二服务内容候选集，包括：该在线业务系统根据所述用户行为属性、所述用户的受众群体信息和所述第一服务内容候选集中服务内容的标签信息，从所述第一服务内容候选集中确定所述第二服务内容候选集。
结合第三方面，在第三方面可能的实现方式中，该服务内容为该业务系统海量实时更新的服务内容中的一种或者多种。
结合第三方面，在第三方面可能的实现方式中，该服务内容为广告业务、视频业务或者新闻业务。
第四方面，提供了一种个性化推荐的方法，该方法包括：业务系统确定第一服务内容候选集中的服务内容的标签信息；该业务系统接收终端设备发送的服务内容请求信息；根据该服务内容请求信息，从第一服务内容候选集中确定第二服务内容候选集，该第二服务内容候选集包括多个服务内容；该业务系统向终端设备发送该多个服务内容以及该多个服务内容对应的标签信息。
结合第四方面，在第四方面可能的实现方式中，该业务系统接收终端设备发送的服务内容请求信息之前，该方法还包括：该业务系统接收该终端设备发送的用户行为事件；该业务系统根据该业务行为事件，确定用户行为属性以及用户受众群体信息；其中，该业务系统从第一服务内容候选集中确定第二服务内容候选集，包括：该业务系统根据该用户行为属性、该用户的受众群体信息和该第一服务内容候选集中服务内容的标签信息，从所述第一服务内容候选集中确定所述第二服务内容候选集。
结合第四方面，在第四方面可能的实现方式中，该业务系统确定第一服务内容候选集中的服务内容的标签信息，包括：该业务系统根据预设的算法，确定该第一服务内容候选集中的服务内容的标签信息。
第五方面，提供了一种终端设备，该终端设备包括触摸屏，一个或者多个处理器，多个应用程序以及一个或者多个程序；其中，一个或者多个程序被存储在该存储器中；当一个或者多个处理器在执行该一个或者多个程序时，使得终端设备实现上述第一方面及第一 方面可能的实现方式中的个性化推荐的方法。
第六方面，提供了一种业务系统，该业务系统包括业务运营管理系统、离线业务系统和在线业务系统，其中，该业务运营管理系统用于接收第一服务内容候选集；该业务运营管理系统还用于将该第一服务内容候选集发送给该离线业务系统；该离线业务系统用于确定该第一服务内容候选集中服务内容的标签信息；该离线业务系统还用于接收该终端设备发送的用户行为事件，并根据该用户行为事件确定用户行为属性、该用户的受众群体信息；该离线业务系统还用于将该第一服务内容候选集中服务内容的标签信息、该用户的行为属性和该用户的受众群体信息发送给在线业务系统；该在线业务系统用于接收终端设备发送的服务内容请求信息；该在线业务系统还用于根据用户行为属性、该用户的受众群体信息以及该第一服务内容候选集中服务内容的标签信息，从该第一服务内容候选集中确定第二服务内容候选集；该在线业务系统还用于向该终端设备发送该第二服务内容候选集以及该第二服务内容候选集中每个服务内容的标签信息。
第七方面，提供了一种个性化推荐的系统，包括上述终端设备和上述业务系统。
第八方面提供了一种计算机存储介质，包括指令，当指令在电子设备上运行时，使得电子设备执行上述任一项可能的实现中的个性化推荐的方法。
第九方面提供了一种计算机程序产品，当计算机程序产品在电子设备上运行时，使得电子设备执行上述任一项可能的实现中的个性化推荐的方法。
附图说明
图1是本申请实施例提供的个性化推荐的系统的示意性架构图。
图2是本申请实施例提供的个性化推荐的系统的另一示意性架构图。
图3是本申请实施例提供的个性化推荐的方法的示意性流程图。
图4是本申请实施例提供的个性化推荐的方法的另一示意性流程图。
图5是本申请实施例提供的终端设备的示意性框图。
图6是本申请实施例提供的业务系统的示意性框图。
图7是本申请实施例提供的个性化推荐的系统的示意性结构图。
具体实施方式
以下实施例中所使用的术语只是为了描述特定实施例的目的，而并非旨在作为对本申请的限制。如在本申请的说明书和所附权利要求书中所使用的那样，单数表达形式“一个”、“一种”、“所述”、“上述”、“该”和“这一”旨在也包括例如“一个或多个”这种表达形式，除非其上下文中明确地有相反指示。还应当理解，在本申请以下各实施例中，“至少一个”、“一个或多个”是指一个、两个或两个以上。术语“和/或”，用于描述关联对象的关联关系，表示可以存在三种关系；例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B的情况，其中A、B可以是单数或者复数。字符“/”一般表示前后关联对象是一种“或”的关系。
在本说明书中描述的参考“一个实施例”或“一些实施例”等意味着在本申请的一个或多个实施例中包括结合该实施例描述的特定特征、结构或特点。由此，在本说明书中的不同之处出现的语句“在一个实施例中”、“在一些实施例中”、“在其他一些实施例中”、“在另外一些实施例中”等不是必然都参考相同的实施例，而是意味着“一个或多个但不 是所有的实施例”，除非是以其他方式另外特别强调。术语“包括”、“包含”、“具有”及它们的变形都意味着“包括但不限于”，除非是以其他方式另外特别强调。
本申请实施例中的终端设备，可以是手机、平板电脑、可穿戴设备(例如，智能手表)、车载设备、增强现实(augmented reality，AR)设备、虚拟现实(virtual reality，VR)设备、笔记本电脑、超级移动个人计算机(ultra-mobile personal computer，UMPC)、上网本、个人数字助理(personal digital assistant，PDA)等设备。本申请实施例的终端设备可以包括触摸屏，用于向用户展示服务内容。本申请实施例中对终端设备的具体类型并不作任何限定。
本申请实施例中的服务内容，可以是广告、新闻、视频、音乐等等，本申请实施例中对服务内容的具体类型并不作任何限定。
在介绍本申请实施例的技术方案之前，首先介绍几个和本申请相关的概念。
机器学习(machine learning，ML)，自然语言处理(natural language processing，NLP)，信息检索(information retrieval，IR)等领域，评估(evaluation)是一个必要的工作，而其评价指标包括准确率(accuracy)和召回率(recall)，其中，准确率又称为“精度”、“正确率”，召回率又称为“查全率”。
以检索为例，可以把检索情况用表1表示。
表1检索情况
  相关 不相关
检索到的 A B
未检索到的 C D
如果希望被检索到的内容越多越好，这是追求“查全率”，即A/(A+C)，越大越好。
如果希望检索到的内容中，真正想要的、也就是相关的越多越好，不相关的越少越好，这是追求“准确率”，即A/(A+B)，越大越好。
例如，邮箱中有100封邮件，实际有10条为垃圾邮件。但垃圾邮件识别软件一共识别出40封垃圾邮件，其中包括实际为垃圾邮件的10封，另外30封被误识别为垃圾邮件，则根据表1所示，A为10，B为30，C为0，召回率为100％，正确率为25％。召回率与准确率虽然没有必然的关系(从上面公式中可以看到)，在实际应用中，是相互制约的。要根据实际需求，找到一个平衡点。
图1示出了本申请实施例提供的个性化推荐的系统的架构图，如图1所示，个性化推荐系统包括终端设备和业务系统，其中，终端设备可以包括设备级人工智能(artificial intelligence，AI)引擎和端侧业务应用(application，APP)；业务系统也可以称之为云侧设备，包括云侧业务运营管理系统、云侧在线业务系统和云侧离线业务系统。
业务运营管理系统：除了提供业务运营人员所需的运营管理支撑能力外，本申请实施例中业务运营管理系统可以对最终呈现给用户的服务内容的合规性和准入条件进行审核(自动审核，人工审核)、配置业务运营相关的规则和策略(不同场景、受众群体生效的推荐算法模型或人工推荐规则、权重、优先级等)。
云侧在线业务系统：实时处理和响应终端设备的服务内容请求，对海量云侧服务内容进行过滤筛选，调用云侧设备个性化推荐引擎进行点击通过率(click through rate，CTR)预测、排序决策输出给终端设备经过初选的服务内容列表及相关参数。完成业务系统对服 务内容的第一次初选，提供更全面更实时的可能匹配用户个性化诉求的服务内容候选集，主要是解决个性化推荐的服务内容高召回率问题。
云侧离线业务系统：重点在于支撑在线业务系统的推荐决策，提供所需的三个重点能力：服务内容的语义化标签处理、根据人群分类规则进行受众定向、个性化离线推荐模型计算。主要依赖两种数据来源：第一，经用户授权同意采集的业务行为事件日志；第二，海量服务内容数据。
端侧设备AI引擎：属于设备系统级的AI引擎，主要包括用户知识库(用户基本属性、设备属性、全场景业务行为属性、习惯或者偏好)、情景感知引擎(用户状态、环境、上下文)、基于用户授权同意端侧业务应用(application，APP)调用的推荐引擎和排序决策规则引擎。基于第一次云侧个性化推荐返回的服务内容候选列表及服务内容相关标签信息，在端侧AI引擎的推荐引擎实现二次更精准受众匹配安全过滤以及个性化CTR预测和最终的排序决策，过滤掉对用户不适宜或无效的服务内容，确保最终呈现给用户服务内容都是匹配相关的。基于云侧高召回率基础上，解决推荐准确率的问题。
端侧业务APP：是业务服务能力在终端设备的客户端载体。通过端侧业务APP请求业务个性化的服务内容呈现给用户，并收集相关用户行为事件上报给端侧AI引擎及经用户授权同意的云侧离线业务系统，用于下一步提供优化业务体验的算法模型更新。
图2示出了本申请实施例提供的个性化推荐的系统的架构图，如图2所示，该个性化推荐系统可以适用于广告推荐的场景下，该个性化推荐的系统可以具体为个性化广告推荐的系统。个性化广告推荐的系统包括终端设备和业务系统，其中，终端设备可以包括设备级AI引擎和端侧广告软件开发工具包(software development kit，SDK)，业务系统也可以称之为云侧设备，包括广告运营管理系统、在线广告平台和离线广告平台。
广告运营管理系统：除了提供广告业务运营人员所需的运营管理支撑能力外，本申请实施例中业务运营管理系统可以对最终呈现给用户的广告创意内容的合规性和准入条件的审核(自动审核，人工审核)、配置广告业务运营相关的规则和策略(不同场景、受众群体生效的广告推荐算法模型或人工推荐规则、权重、优先级等)。
在线广告平台：实时处理和响应终端设备的广告请求，对海量广告投放任务内容进行过滤筛选，调用业务系统的个性化广告推荐引擎进行CTR预测、排序决策输出给端侧的经过初选的广告创意内容列表及相关参数。完成业务系统对广告投放任务内容的第一次初选，提供更全面更实时的可能匹配用户个性化诉求的广告创意服务内容候选集，主要是解决个性化广告推荐服务内容高召回率问题。
离线广告平台：重点在于支撑在线广告系统的推荐决策，提供所需的三个重点能力：广告创意服务内容的语义化标签处理、根据人群分类规则进行受众定向、个性化广告离线推荐模型计算。主要依赖两种数据来源：第一，经用户授权同意采集的业务行为事件日志和广告用户行为事件日志；第二，与用户无关的海量广告创意的服务内容数据。
端侧设备级AI引擎：属于设备系统级的AI引擎，与本申请实施例相关能力主要有：用户知识库(用户基本属性、设备属性、全场景业务行为属性和习惯/偏好)、情景感知引擎(用户状态、环境、上下文)、基于用户授权同意端侧业务APP广告调用的推荐引擎和排序决策规则引擎。基于第一次云侧广告个性化推荐返回的广告创意服务内容候选列表及广告创意服务内容相关标签结构化信息，在端侧AI引擎的推荐引擎实现二次更精准受众匹配安全过滤以及个性化广告CTR预测和最终的排序决策，过滤掉对用户不适宜或 无效的广告创意服务内容，确保最终呈现给用户广告创意服务内容都是匹配相关的。基于云侧高召回率基础上，解决个性化广告推荐高准确率的问题。
端侧广告SDK：可以被业务APP集成，是广告业务服务能力在终端设备的客户端载体。通过端侧广告SDK请求个性化的广告服务内容呈现给用户，并收集相关用户行为事件和广告事件日志，上报给端侧AI引擎及离线广告系统，用于下一步提供优化个性化广告业务算法模型更新。
图3是本申请实施例提供的个性化推荐方法100的示意性流程图，如图3所示，该方法100包括：
S101，云侧的业务运营管理系统会实时或定期导入和上传服务内容数据。
可选地，服务内容数据包括文字、图片、同一资源定位符(uniform resource locator，URL)地址等。
示例性的，广告主可以通过业务运营管理系统实时或者定期导入和上传一些广告创意素材。
本申请实施例中，云侧在线业务系统和云侧离线业务系统对广告主并不可见，对广告主可见的是云侧业务运营管理系统。
示例性的，京东推广了一类电脑，则可以通过业务运营管理系统实时向云侧离线业务系统上传一些关于此类电脑的文字信息、图片信息以及URL地址等。
示例性的，唯品会推广了一类美妆产品，则可以通过业务运营管理系统实时向云侧离线业务系统上传一些关于此类美妆产品的文字信息、图片信息以及URL地址等。
示例性的，宝马推广了一款汽车，则可以通过业务运营管理系统实时向云侧离线业务系统上传一些关于此款汽车的文字信息、图片信息以及URL地址等。
应理解，本申请实施例中，该业务运营管理系统具体可以为云侧广告运营系统，云侧业务系统具体可以为云侧离线广告平台，云侧在线业务系统具体可以为云侧实时广告平台。
S102，云侧离线业务系统使用相关学习算法(如：机器学习算法、深度学习算法等)，计算挖掘服务内容，生成结构化、语义化的内容标签；这些内容标签用于云侧在线业务系统和终端设备对服务内容的匹配过滤、内容相关性计算。
示例性的，云侧离线业务系统在收到上传的关于电脑的信息后，可以对该电脑相关的信息进行标签挖掘计算，例如，通过分析得到此类电脑为“xxx品牌的笔记本电脑”、“xxx品牌的游戏机”等等。
示例性的，云侧离线业务系统在收到上传的关于美妆的信息后，可以对该美妆相关的信息进行标签挖掘计算，例如，通过分析得到此类美妆为“口红”、“香水”等等。
示例性的，云侧离线业务系统在收到上传的关于汽车的信息后，可以对该汽车相关的信息进行标签挖掘计算，例如，通过分析得到此类汽车为“SUV”、“宝马X7”等等。
通过S101和S102，云侧离线业务系统实现了对服务内容的标签挖掘，为下面云侧内容匹配过滤或者内容的相关性计算提供了依据。
S103，云侧业务运营管理系统将服务内容个性化推荐场景、相关策略规则和服务内容支持的受众信息下发给云侧离线业务系统和云侧在线业务系统。云侧业务运营管理系统对服务内容投放商可见，服务内容投放商可以配置相关服务内容的个性化推荐场景的相关策略规则和服务内容支持的受众信息，并将这些内容通过云侧业务运营管理系统下发给云侧 离线业务系统和云侧在线业务系统。
示例性的，各大电商平台可以通过云侧业务运营管理系统投放广告，即下发对应的广告任务或者创意，本申请实施例中对广告任务或者创意的数量并不作任何限定，在同一时间段，可能有成千上万的广告任务下发给云侧离线业务系统(例如，云侧离线广告平台)或者云侧在线业务系统(或者，云侧在线广告平台)。
例如，表1示出了一组根据不同受众信息的广告投放任务。
表1不同受众信息的广告投放任务
Figure PCTCN2020082429-appb-000001
S104，端侧业务App采集经用户授权同意的用户行为事件，上报给离线业务系统(含大数据平台)。
云侧离线业务系统中的大数据平台可以对端侧业务App上报的用户行为事件进行存储和处理。
本申请实施例中，用户行为事件包括但不限于展示、点击、滑动或者下载等。
示例性的，终端设备可以将用户在浏览器中的一些业务行为事件上报给云侧离线业务系统。
应理解，本申请实施例中S104中终端设备也可以上报一些用户在某个APP中的业务行为事件给业务系统，前提是用户授权同意这些业务行为事件的上传。
还应理解的是，这些用户行为事件可以是用户前一天的用户行为事件，或者用户前几天的用户行为事件件，本申请实施例对此并不作任何限定。
S105，云侧离线业务系统计算用户行为属性(只有经用户授权的部分数据)和用户的受众定向(基于云侧具备的受众定向Options)，训练云侧个性化推荐模型。
云侧离线业务系统可以根据上报的用户行为事件计算用户的行为属性、用户的受众定向等信息。
例如，端侧上报的业务行为事件中70％的事件为用户在终端设备中搜索“华为”、“苹 果”、“vivo”等等，则离线业务系统可以确定该用户行为属性为对科技感兴趣，并且还可以确定用户的受众定向为20-30岁的年轻人。
又例如，端侧上报的业务行为事件中60％的事件为用户在浏览器点击关于汽车的新闻，则离线业务系统可以确定该用户行为属性为对汽车感兴趣，并且还可以确定用户的受众定向为20-50岁的男性。
又例如，端侧上报的业务行为事件中80％的事件为用户点击、购买口红、护肤品等等，则离线业务系统可以确定该用户行为属性为对美妆感兴趣，并且还可以确定用户的受众定向为20-40岁的女性。
同时离线业务系统还可以根据上报的用户行为事件，对云侧个性化推荐模型进行更新。
S106，云侧离线业务系统同步以下信息到云侧在线业务系统：用户行为属性、服务内容及其相关语义化结构化标签、受众定向的结果、训练后的个性化推荐模型。
在离线业务系统通过相关用户行为事件计算出用户行为属性、用户的受众定向的信息以及更新个性化推荐模型后，可以将用户行为属性、用户的受众定向的信息以及个性化推荐模型发送给云侧在线业务系统。
S107，云侧在线业务系统收到来自于终端设备的服务内容请求信息。
示例性的，当用户打开浏览器时，触发终端设备向云侧在线业务系统发送服务内容请求信息，用于请求业务系统进行个性化服务内容的推荐。
S108，云侧在线业务系统根据服务内容请求信息，对服务内容进行匹配过滤。
示例性的，云侧在线业务系统可以根据S105中计算得到的用户行为属性、用户的受众定向的结果以及S103中广告商投放广告时的不同受众群体场景规则，对其中的一些广告进行匹配过滤。
示例性的，用户在S105中确定用户对科技和汽车感兴趣，并且受众定向为20-50岁的男性，则对于表1所示的广告任务，可以将“洗发水”、“口红”、“彩妆”、“薯片”等广告内容过滤掉。
S109，云侧在线业务系统调用云侧的个性化推荐算法模型，进行CTR预测及结果排序。
云侧在线业务系统可以调用S106中同步到云侧在线业务系统的个性化推荐算法模型，对过滤后的服务内容进行CTR预测及结果排序。
应理解，云侧在线业务系统对服务内容进行CTR预测及结果排序的过程可以参考现有技术中的做法，本申请实施例中对此不做任何限定。
S110，云侧在线业务系统基于规则，获得个性化推荐初选结果。
云侧在线业务系统还可以根据一些预设的规则，例如地域、机型规则，获得个性化推荐的初选结果。
应理解，本申请实施例中，并不对S109和S110的先后顺序进行限定。
还应理解，本申请实施例中，云侧在线系统可以基于云侧的个性化推荐算法模型和预设的规则这两个维度对过滤后的服务内容进行进一步筛选，从而获得个性化推荐的初选结果。云侧在线系统也可以基于侧的个性化推荐算法模型和预设的规则这两个维度中的一个维度对过滤后的服务内容进行筛选，从而获得个性化推荐的初选结果。本申请实施例对此并不作任何限定。
本申请实施例中，云侧在线业务系统可以综合多种算法或者规则推荐结果，输出云侧初步推荐结果。
本申请实施例中，云侧业务系统主要实现个性化推荐的初选，基于云侧有限的用户数据，找到可能匹配用户需求的海量服务内容，实现高召回率的一次筛选、CTR预测和排序功能。
云侧有全量、实时的服务内容及授权采集用户业务部分数据，基于有限的用户数据进行云侧个性化推荐，找到可能匹配用户诉求、最完整的、实时的服务内容候选集，在云侧推荐精度不太高情况下，重点解决推荐高召回率。
S111，云侧在线业务系统向端侧APP发送服务内容响应信息，服务内容响应信息中包括个性化推荐初选结果：多个服务内容候选集及服务内容候选集对应的语义化标签。
例如，表2示出了一种云侧输出的个性化推荐初选结果。表2所示的个性化推荐初选结果是以服务内容为广告创意内容为例进行说明。
表2个性化推荐初选结果
Figure PCTCN2020082429-appb-000002
该多个服务内容候选集可以为每个服务内容的ID以及每个服务内容对应的URL地址。
云侧在线业务系统在根据个性化推荐算法或者规则进行第一次推荐结果后，也可以确定第一次推荐的结果对应的语义化标签，例如，表2所示的多个广告任务中含有很多的图片，云侧在线业务系统可以根据S101和S102中服务内容的标签挖掘给这些广告任务打上对应的语义化标签。
与现有技术不同的是，业务系统在给终端设备下发第一次推荐结果中携带推荐结果对应的语义化标签，可以方便终端设备进行二次筛选，现有技术中业务系统给终端设备下发的推荐结果即为最终的推荐结果，推荐结果中仅包含服务内容候选集。终端设备接收到这些服务内容候选集后，可以根据其中的URL地址后就向业务系统请求对应的服务内容，从而将对应的服务内容展示给用户。
本申请实施例中，业务系统给终端设备下发的多个服务内容候选集(包括对应的ID及URL地址)的同时，还会下发服务内容对应的语义化标签，方便终端设备对云侧输出的服务内容进行二次筛选，从而有助于提高个性化推荐的准确性。
本申请实施例的个性化推荐方法，可以充分发挥终端设备全面准确的用户数据优势，无需上传额外的用户数据和标签，只需在下行方向安全过滤和精准匹配符合用户诉求的服务内容，基于业务系统一次广告推荐服务内容的高召回率基础上，叠加终端设备的数据优 势，从而解决个性化推荐高准确率的问题。通过端云协同的二次个性化推荐配合，提供高召回率、高准确率的个性化推荐服务体验。
可选地，云侧业务系统输出的个性化推荐初选结果还可以包括服务内容的受众信息和/或投放价格。
例如，表3示出了另一种云侧输出的个性化推荐初选结果。表3所示的个性化推荐初选结果是以服务内容为广告创意内容为例进行说明。
表3个性化推荐初选结果
Figure PCTCN2020082429-appb-000003
S112，端侧APP确定用户是否授权同意端侧APP调用端侧AI引擎。
在用户授权同意的情况下，端侧APP可以调用端侧AI引擎实现服务内容的安全过滤以及二次个性化推荐最终决策的配置。
S113，在用户授权同意的情况下，端侧APP向端侧AI引擎发送个性化推荐请求信息，该个性化推荐请求信息中包括多个服务内容候选集及服务内容候选集对应的语义化标签。
端侧APP可以向端侧AI引擎发送云侧在线业务系统输出的服务内容的多个服务内容候选集、及服务内容候选集的标签，请求AI引擎对个性化推荐初选结果做二次个性化推荐。
示例性的，端侧APP可以向端侧AI引擎发送服务内容对应的ID及对应的语义化标签。可选地，当云侧在线业务系统输出的个性化推荐初选结果中包含受众信息和投放价格时，端侧APP可以向端侧AI引擎发送服务内容对应的ID、对应的语义化标签、受众信息和投放价格。
S114，端侧AI引擎根据输入的多个服务内容候选集，结合用户相关信息，对个性化 推荐初选结果中的服务内容进行二次匹配筛选过滤。
端侧AI引擎主要包含：具备最全面最准确的用户数据及属性的用户知识库，实时感知情景上下文和环境的情景感知引擎，以及端侧个性化推荐引擎。
情景感知引擎主要用于计算和输入，当前用户在端侧的实时状态参数，最终也作为输入参数发送给端侧个性化推荐引擎。
由于端侧AI引擎具有最全面最准确的用户知识库的数据以及实时感知情景上下文和环节的情景感知引擎，端侧AI引擎可以对端侧APP发送的推荐结果进行过滤筛选。
示例性的，端侧AI引擎可以根据推荐结果和用户知识库进行相关系数的计算。
例如，用户的知识库中包括用户属性特征：性别、年龄、兴趣、博文特征、关注用户特征、基于位置服务(location based service，LBS)位置信息(即实时位置)、学历、职业、家庭、消费水平等。端侧AI引擎可以现根据这些用户属性特征计算与服务内容Connect 1-Connect 7的相关系数。
示例性的，用户的知识库中包括的信息：用户的年龄为20-30岁，性别为男性，每年的消费金额为5-8万，兴趣为科技类产品，则可以通过一定的预设条件计算得到表4所示的相关系数。
表4
Figure PCTCN2020082429-appb-000004
由表4可以看出，端侧AI引擎可以输出相关系数由高到低的排序为：“平板电脑”、“智能家电”、“美颜手机”、“轻薄笔记本电脑”、“国产汽车”、“日企汽车”、“德国汽车”。
一种的可能的实现方式中，在第一次推荐结果中包含投放价格时，还可以同时考虑到商业化因素，相关系数的计算还可以考虑投放价格。
例如，表5示出了一种相关系数的计算结果。
表5
Figure PCTCN2020082429-appb-000005
Figure PCTCN2020082429-appb-000006
由表5可以看出，在考虑商业化因素的情况下，端侧AI引擎可以输出相关系数由高到低的排序为：“智能家电”、“国产汽车”、“平板电脑”、“轻薄笔记本电脑”、“日企汽车”、“美颜手机”和“德国汽车”。
S115，端侧AI引擎调用端侧个性化推荐引擎的算法模型，进行二次CTR预测和排序，实现个性化推荐的终选，最终决策输出匹配端侧用户需求的安全、精准、个性化的推荐服务内容。
CTR预测是计算广告中最核心的算法之一，CTR预测是对每次广告的点击情况做出预测，预测用户是点击还是不点击。其中，CTR预测和很多因素相关，比如历史点击率、广告位置、时间、用户等。个性化推荐引擎的算法模型就是综合考虑各种因素、特征，在大量历史数据上训练得到的模型。CTR预测的训练样本一般从历史日志(log)、离线特征库获得。样本标签相对容易，用户点击标记为1，没有点击标记为0。特征则会考虑很多，例如用户的人口学特征、广告自身特征、广告展示特征等。这些特征中会用到很多类别特征，例如用户所属职业、广告展示的IP地址等。一般对于类别特征会采样One-Hot编码，例如职业有三种：学生、白领、工人，那么会会用一个长度为3的向量分别表示他们：[1，0，0]、[0，1，0]、[0，0，1]。可以这样会使得特征维度扩展很大，同时特征会非常稀疏。目前很多公司的广告特征库都是上亿级别的。
本申请实施例中，排序可以为千次展示收益(effective cost per mile，eCPM)排序。eCPM指的是每一千次展示可以获得的广告收入，展示的单位可以是网页，广告单元等。根据ePCM，广告主可以分析出投放广告的效果，并针对性地进行优化、调整广告，进而来提高收入。
应理解，本申请实施例中，如果S114和S115中经过端侧AI引擎的二次筛选，端侧AI引擎确定Connect 1-Connect 7的广告任务均不符合用户的喜好，或者，均为用户反感的广告内容，可以将这些广告任务都过滤掉，并通知端侧APP不展示任何广告内容。
本申请实施例的个性化推荐方法，可以充分发挥终端设备全面准确的用户数据优势，无需上传额外的用户数据和标签，只需在下行方向安全过滤和精准匹配符合用户诉求的服务内容，基于业务系统一次推荐服务内容的高召回率的基础上，叠加终端设备的数据优势，解决个性化推荐高准确率的问题。通过端云协同的二次个性化推荐配合，提供高召回率高准确率的个性化推荐服务体验。
S116，端侧AI引擎向端侧APP发送个性化推荐响应信息，该个性化推荐响应信息包括最终满足用户诉求的服务内容列表。
端侧AI引擎可以将S115中根据端侧个性化推荐引擎的算法模型得到的第二次推荐结果发送给端侧APP。
示例性的，端侧AI引擎可以选择预测结果排在前两位的服务内容的ID，发送给端侧APP。
示例性的，端侧AI引擎确定二次预测和排序输出的服务内容的ID为Connect 4和Connect 6。
S117，端侧APP根据最终决策展示的服务内容列表，从云侧在线业务系统下载服务内容数据到端侧APP。
端侧APP在收到端侧AI引擎输出的最终推荐决策输出的服务内容的ID后，根据ID对应的URL地址从云侧在线业务系统下载最终推荐决策输出的服务内容到端侧APP。
示例性的，端侧APP在收到二次预测和排序输出的前两位(Connect 4和Connect 6)后，根据URL地址4和URL6向云侧在线业务系统请求下载关于“智能家电”和“平板电脑”的广告任务。
S118，端侧APP输出展示最终的服务内容给用户。
端侧APP在接收到最终的个性化服务内容后，向用户展示服务内容。
例如，在用户的浏览器界面展示“智能家电”和“平板电脑”这2个广告任务。
S119，基于用户授权同意采集，端侧APP将用户业务操作行为事件上报云侧离线业务系统，同时端侧AI引擎本地保存。
应理解，S119可以参考上述S104的过程，为了简洁，在此不再赘述。
S120，云侧离线业务系统实时更新用户行为属性及云侧个性化推荐模型。
端侧APP可以将用户对“智能家电”和“平板电脑”这2个广告任务的业务行为事件上传给云侧离线业务系统，端侧AI引擎可以更新用户知识库，云侧离线业务系统可以计算用户部分行为属性，用户的受众定向及更新云侧个性化推荐模型。
本申请实施例中，通过端云协同配合，各有侧重的两次个性化推荐算法建模，充分发挥端云各自数据差异化优势。其中，两次个性化推荐算法建模的算法及特征集的选择是独立的，不依赖的。只是终端设备的个性化推荐的候选集的输入，依赖业务系统的个性化推荐结果的输出。
终端设备有完整的用户知识库的数据，但端侧APP未保存或只保存不完整、不实时的部分服务内容。因此，基于端侧个性化推荐算法建模要基于云侧一次推荐高召回率返回找到更全的服务内容多个候选集基础上，充分发挥终端设备本地用户数据高质量完整性的优势，重点解决推荐高准确率。
为了提升服务内容跟用户相关性的匹配筛选过滤过程，统一在云侧完成对服务内容的语义化、结构化的内容标签挖掘计算(实时增量、全量)。因此，除了在云侧个性化推荐过程中，使用服务内容的标签实现服务内容与用户业务属性的规则匹配过滤外，传给端侧APP的云侧一次推荐输出的服务内容候选集列表中，同时还包括对应服务内容候选集的语义化标签，这些信息一同带给端侧APP，用于在端侧进一步结合端侧用户知识库的数据与内容标签进行匹配过滤。
移动广告作为用户消费信息服务内容的关键组成之一，也需要对移动广告的推广提出更高的要求。对于移动广告最终目标应该是“广告即服务”，提供给用户的广告服务内容正是用户所需的服务内容，而不是骚扰信息。因此，精准深入理解用户需求、提供匹配用户需求的优质原生内容、推广服务内容是一个可持续良性发展的业务的长期持续目标。同时，一定要严格保障和执行用户隐私保护相应法规，在合规合法情况下，合理使用用户个人数据，提供更好的个性化精准推荐的服务体验。
图4是本申请实施例提供的个性化推荐方法200的示意性流程图，该方法200可以用于广告推荐的场景，如图4所示，该方法200包括：
S201，云侧广告运营系统实时或定期导入和上传海量的广告创意内容。
S202，云侧离线广告平台使用相关学习算法(例如，机器学习算法、深度学习算法等)，计算挖掘广告创意服务内容，生成结构化、语义化的内容标签；这些内容标签用于云侧在线业务系统和终端设备对广告创意内容的匹配过滤、内容相关性计算。
S203，云侧广告运营系统将广告创意内容个性化推荐场景、相关策略规则和服务内容支持的受众信息下发给云侧离线广告平台和云侧在线广告平台。
云侧广告运营系统对广告主可见，广告主可以配置相关广告的个性化推荐场景的相关策略规则和广告创意内容的受众信息，并将这些内容通过云侧广告运营系统下发给云侧离线广告平台和云侧在线广告平台
基于云侧广告运营系统，广告业务运营人员配置相关的个性化推荐场景的相关策略规则，支持不同场景、受众信息；使用不同个性化广告推荐算法的组合，并支持相关内容相关性、排他性的匹配过滤规则。另外，广告主在云侧广告运营系统(面向广告主投放平台上)也会定义相应的投放任务指定的受众信息(或者，相关受众群体规则)。两类策略都会下发给云侧离线广告平台和云侧在线广告平台。
S204，端侧业务APP采集经过用户授权同意的用户行为事件，上报给云侧离线广告平台(含大数据平台)。
云侧离线广告平台中的大数据平台可以对端侧业务APP上报的用户行为事件进行存储和处理。
本申请实施例中，用户行为事件包括但不限于展示、点击、滑动或者下载等等。
S205，云侧离线广告平台计算用户行为属性(只有经过用户授权的部分数据)和用户的受众群体(基于云侧具备的受众定向Options)以及训练云侧个性化推荐模型。
示例性的，云侧离线广告平台中的大数据平台根据终端设备上报的经过用户授权的部分数据，计算用户行为属性、用户的受众定向以及更新云侧个性化广告推荐模型。
由于云侧有限的用户的业务行为事件作为输入，云侧离线广告平台计算的受众定向信息不会特别准确，云侧个性化推荐模型的准确性不会太高。
应理解，S205可以参考上述S105的描述，为了简洁，在此不再赘述。
S206，云侧离线广告平台同步以下信息到云侧在线广告平台：用户行为属性、服务内容及其相关语义化结构化标签、受众定向的结果、训练后的个性化广告推荐模型。
应理解，S206可以参考上述S106的描述，为了简洁，在此不再赘述。
S207，云侧在线广告平台接收来自于端侧App的广告SDK的广告请求信息。
示例性的，用户在打开浏览器时，触发终端设备向云侧在线广告平台发送广告请求。
S208，云侧在线广告平台对广告创意内容进行匹配过滤。
云侧在线广告平台根据相应场景、广告请求参数、用户行为属性、受众定向的结果、对广告投放创意内容进行匹配过滤。
云侧在线广告平台对广告创意内容的筛选过程可以参考上述S108中的举例，为了简洁，在此不再赘述。
S209，云侧在线广告平台调用云侧的个性化推荐算法模型，进行CTR预测及结果排序。
云侧在线广告平台可以调用S206中同步到云侧在线广告平台的个性化推荐模型，对过滤后的广告创意内容进行CTR预测及结果排序。
S210，云侧在线广告平台基于规则，获得个性化推荐初选结果。
应理解，本申请实施例中，并不对S209和S210的先后顺序进行限定。
云侧在线广告平台还可以根据一些预设的规则，例如地域、机型规则，对过滤后的广告创意内容进行进一步筛选，从而获得个性化推荐初选结果。
本申请实施例中，广告业务系统主要实现个性化广告推荐的初选，基于云侧有限的用户数据(含用户广告行为数据)，找到可能匹配用户需求的海量广告创意内容，实现高召回率的一次筛选、推荐CTR预测和排序功能。
S211，云侧在线广告平台向广告SDK发送广告请求响应信息，该广告请求响应信息包括该个性化推荐初选结果。
该个性化推荐初选结果包括但不限于：多个广告创意内容候选集、广告创意内容的语义化结构化标签或者广告实时竞价信息等等。应理解，个性化推荐初选结果可参考上述表2或者表3，为了简洁，在此不再赘述。
与现有技术不同的是，广告业务系统在给终端设备下发第一次推荐结果中携带广告创意内容对应的语义化标签，可以方便终端设备进行二次筛选，现有技术中广告业务系统给终端设备下发的推荐结果即为最终的推荐结果，推荐结果中仅包含广告创意内容候选集。终端设备接收到这些广告创意内容候选集后，可以根据其中的URL地址后就向业务系统请求对应的广告创意内容，从而将对应的广告创意内容展示给用户。
本申请实施例中，业务系统给终端设备下发的多个广告创意内容候选集(包括对应的ID及URL地址)的同时，还会下发服务内容对应的语义化标签，方便终端设备对云侧输出的服务内容进行二次筛选，从而有助于提高个性化推荐的准确性。
本申请实施例的个性化推荐方法，可以充分发挥终端设备全面准确的用户数据优势，无需上传额外的用户数据和标签，只需在下行方向安全过滤和精准匹配符合用户诉求的广告创意内容，基于业务系统一次广告推荐广告创意内容的高召回率基础上，叠加终端设备的数据优势，从而解决个性化推荐高准确率的问题。通过端云协同的二次个性化推荐配合，提供高召回率、高准确率的个性化推荐服务体验。
S212，广告SDK确定用户是否授权同意调用端侧AI引擎。
广告SDK在确定用户授权同意调用端侧AI引擎的情况下，广告SDK调用端侧AI引擎实现广告创意内容的安全过滤及二次个性化推荐决策。
S213，在用户授权同意的情况下，广告SDK向端侧AI引擎发送个性化广告推荐请求信息，该个性化推荐请求信息中包括云侧在线广告平台输出的个性化推荐初选结果。
一个实施例中，广告SDK还可以将投放价格(或者，竞价信息)发送给端侧AI引擎。
S214，端侧AI引擎根据输入的广告创意内容候选集，结合用户相关信息对个性化推荐初选结果中的广告创意内容进行二次匹配筛选过滤。
应理解，端侧AI引擎结合用户相关数据对广告创意内容进行匹配筛选过滤的过程可以参考上述S114中的描述，为了简洁，在此不再赘述。
S215，端侧AI引擎调用端侧个性化推荐引擎的算法模型，进行二次CTR预测和排序，实现个性化推荐的终选，最终决策输出匹配端侧用户需求的安全、精确、个性化的广告创意内容。
本申请实施例的个性化推荐方法，可以充分发挥终端设备保存全面准确的用户数据优势，无需上传额外的用户数据和标签，只需在下行方向安全过滤和精准匹配符合用户诉求的广告创意服务内容，基于广告业务系统一次广告推荐服务内容的高召回率基础上，叠加 终端设备的用户数据优势，从而解决个性化广告推荐高准确率的问题。通过端云协同的二次个性化广告推荐配合，提供高召回率高准确率的个性化广告推荐服务体验。
S216，端侧AI引擎向广告SDK发送个性化广告推荐请求响应信息，该个性化广告推荐请求响应信息包括最终满足用户诉求的广告创意内容列表。
S217，广告SDK根据最终决示的广告创意内容列表，从云侧下载广告创意内容数据到端侧APP。
S218，广告SDK输出展示最终的个性化广告创意内容给用户。
S219，基于用户授权同意采集，广告SDK将用户行为事件上报给云侧在线广告平台。
S220，云侧离线广告平台实时更新用户广告属性及云侧个性化广告推荐模型。
应理解，S216-S220中描述可以参考上述S116-S120中的描述。
广告SDK还可以将用户行为时间上报给端侧AI引擎，端侧AI引擎可以实时更新用户知识库中的用户数据及相关属性。
示例性的，该方法200还包括：
S221，云侧广告在线平台计算广告投放的消耗明细。
本申请实施例的个性化推荐方法，基于云侧一次广告推荐算法模型的高召回率基础上，叠加端侧数据优势，解决个性化广告推荐的品牌安全、高准确率的问题。通过端云协同的二次个性化推荐配合，提供高同时具备召回率、高准确率的个性化广告推荐体验。
端云协同配合，各有侧重的两次个性化广告推荐算法建模，充分发挥端云各自数据差异化优势。其中两次个性化广告推荐算法建模的算法及特征集的选择是独立的，不依赖的。只是端侧个性化广告推荐的候选集的输入，要依赖云侧个性化广告推荐候选集列表的输出。相比于现有技术中的端云协同配合推荐的方法，本申请实施例的个性化推荐方法，有助于解决由于云侧数据质量和完整问题而造成的受众定向准确性不高的问题，发挥端云协同各自的数据优势，形成完整受众定向协同配合。
云侧有全量/实时的广告投放任务和创意服务内容及授权采集用户业务部分数据和用户广告行为事件，基于有限的用户业务数据进行云侧个性化广告推荐算法建模，找到可能匹配用户诉求、最完整的、实时的广告服务内容候选集，在推荐精度不太高情况下，重点解决广告推荐高召回率。
端侧有完整的用户知识库的数据，但端侧业务没有保存或只保存的不完整、不实时的服务内容。因此，基于端侧个性化广告推荐算法建模需要基于云侧一次广告推荐高召回率返回广告创意服务内容多个候选集基础上，充分发挥端侧本地用户数据高质量完整性的优势，重点解决最终广告推荐高准确率。相比于现有技术中的仅基于端侧推荐的方法，本申请实施例的个性化推荐方法，有助于解决单独依赖端侧推荐算法模型和数据，只能解决端侧本地内容推荐有限场景问题，无法做到获取海量丰富的精准的云侧服务内容；还有助于解决云侧推荐算法模型由于云侧数据质量和完整问题而造成的推荐结果准确率不高问题，特别是内容安全敏感性问题(用户不适宜内容)，做到既满足用户隐私安全，更精准的个性化推荐用户体验。
为了提升广告创意服务内容跟用户相关性的匹配筛选过滤过程，统一在云侧完成对广告创意服务内容的语义化、结构化的内容标签挖掘计算(实时增量、全量)。因此，除了在云侧个性化广告推荐过程中，使用广告创意服务内容的标签实现创意内容与用户业务属性的规则匹配过滤外，传给端侧广告SDK的云侧一次广告推荐输出的创意服务内容候选 集列表，同时还包括对应广告创意服务内容候选集的标签，这些信息一同带给端侧广告SDK，用于在端侧进一步结合端侧用户知识库的数据与广告创意内容标签进行匹配过滤。
应理解，无论是业务系统个性化精准推荐应用场景，还是广告系统的个性化精准推荐场景，都适用以上相同的端云协同个性化精准推荐的设计原则和流程。
还应理解，本申请实施例中，服务内容可以为视频业务、广告业务、新闻业务等等，上述方法200中仅仅是以广告业务为例进行说明，本申请实施例中对服务内容的具体形式并不作任何限定。
以上介绍了本申请实施例提供的个性化推荐的方法，下面介绍本申请实施例提供的终端设备、业务系统以及个性化推荐系统。
图5示出了根据本申请实施例的终端设备300的示意性结构图，如图5所示，该终端设备300包括触摸屏310、存储器320和处理器330，其中，一个或多个计算机程序被存储在存储器320中，一个或多个计算机程序包括指令。当指令被处理器330执行时，使得终端设备300执行以下操作：
向业务系统发送服务内容请求信息；
接收该业务系统发送的服务内容响应信息，该服务内容响应信息包括服务内容候选集，该服务内容候选集中包括多个服务内容，该服务内容候选集为该业务系统根据该终端设备上报的用户行为事件得到的；
根据用户授权的数据，从该服务内容候选集中确定一个或者多个服务内容；
通过触摸屏310向用户展示该一个或者多个服务内容。
可选地，该服务内容响应信息中还包括该服务内容候选集中每个服务内容的标签信息，当指令被处理器330执行时，使得终端设备300执行以下操作：
根据该用户授权的数据和该服务内容候选集中每个服务内容的标签信息，确定该一个或者多个服务内容。
可选地，当指令被处理器330执行时，使得终端设备300执行以下操作：
根据该用户授权的数据和该每个服务内容的标签信息，确定该每个服务内容与该用户授权的数据的相关系数；
根据该相关系数，确定该一个或者多个服务内容。
可选地，当指令被处理器330执行时，使得终端设备300执行以下操作：
根据多个服务内容中每个服务内容的标签信息和用户授权的数据，确定多个相关系数；
通过触摸屏向用户展示所述多个相关系数中大于或者等于预设值的相关系数对应的服务内容。
可选地，该服务内容为该业务系统海量实时更新的服务内容。
可选地，该服务内容包括广告业务、视频业务或者新闻业务中的一种或者多种。
应理解，终端设备300可以对应上述个性化推荐的方法100或者方法200中的终端设备，处理器330可以用于执行上述方法100或者方法200中终端设备的操作。
图6示出了本申请实施例提供的业务系统400的示意性框图，如图6所示，该业务系统400包括业务运营管理系统410、离线业务系统420和在线业务系统430，其中，
业务运营管理系统410，用于接收服务内容投放商投放的第一服务内容候选集合；
业务运营管理系统410，还用于向离线业务系统420发送该第一服务内容候选集合；
离线业务系统420，用于确定第一服务内容候选集中的服务内容的标签信息，并发送给在线业务系统430；
在线业务系统430，用于接收终端设备发送的服务内容请求信息；
在线业务系统430，还用于根据该服务内容请求信息，从第一服务内容候选集合中确定第二服务内容候选集合；
在线业务系统430，还用于向终端设备发送服务内容请求响应信息，该服务内容请求响应信息包括该第二服务内容候选集合以及该第二服务内容候选集中服务内容的标签信息。
可选地，该离线业务系统420，还用于：
接收终端设备发送的用户行为事件；
根据该用户行为事件，确定用户行为属性、用户的受众定向信息以及训练个性化推荐模型。
可选地，该在线业务系统430，具体用于：
根据该用户行为属性、用户的受众定向信息以及服务内容的标签信息，对第一服务内容候选集中的服务内容进行匹配过滤；
对匹配过滤后的服务内容进行CTR预测以及排序，获得该第二服务内容候选集。
应理解，该业务系统400可以对应于方法100中的业务系统，也可以对应于方法200中的广告业务系统；该业务运营管理系统410可以对应于方法100中的业务运营管理系统，也可以对应于方法200中的广告运营管理系统；离线业务系统420可以对应于方法100中的云侧离线业务系统，也可以对应于方法200中的离线广告平台；在线业务系统430可以对应于方法100中的云侧在线业务系统，也可以对应于方法200中的在线广告平台。
图7示出了本申请实施例提供的个性化推荐的系统500的示意性框图，如图7所示，该个性化推荐的系统包括终端设备510和业务系统520，其中，终端设备可以为上述终端设备300，该业务系统可以为上述业务系统400。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(read-only memory，ROM)、随机存取存储器(random access memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (17)

  1. 一种个性化推荐方法，其特征在于，包括：
    终端设备向业务系统发送服务内容请求信息；
    所述终端设备接收所述业务系统发送的服务内容响应信息，所述服务内容响应信息包括服务内容候选集，所述服务内容候选集中包括多个服务内容，所述服务内容候选集为所述业务系统根据所述终端设备上报的用户行为事件得到的；
    所述终端设备根据用户授权的数据，从所述服务内容候选集中确定一个或者多个服务内容；
    所述终端设备向用户展示所述一个或者多个服务内容。
  2. 根据权利要求1所述的方法，其特征在于，所述服务内容响应信息中还包括所述服务内容候选集中每个服务内容的标签信息；
    其中，所述终端设备根据用户授权的数据，从所述服务内容候选集中确定一个或者多个服务内容，包括：
    所述终端设备根据所述用户授权的数据和所述服务内容候选集中每个服务内容的标签信息，确定所述一个或者多个服务内容。
  3. 根据权利要求2所述的方法，其特征在于，所述终端设备根据所述用户授权的数据和所述服务内容候选集中每个服务内容的标签信息，确定所述一个或者多个服务内容，包括：
    所述终端设备根据所述用户授权的数据和所述每个服务内容的标签信息，确定所述每个服务内容与所述用户授权的数据的相关系数；
    所述终端设备根据所述相关系数，确定所述一个或者多个服务内容。
  4. 根据权利要求1至3中任一项所述的方法，其特征在于，所述多个服务内容为所述业务系统海量实时更新的服务内容。
  5. 根据权利要求4所述的方法，其特征在于，所述服务内容包括广告业务、视频业务或者新闻业务中的一种或者多种。
  6. 一种个性化推荐方法，其特征在于，包括：
    业务系统接收终端设备发送的服务内容请求信息；
    所述业务系统根据所述服务内容请求信息，确定服务内容候选集，所述服务内容候选集包括多个服务内容；
    所述业务系统向所述终端设备发送服务内容响应信息，所述服务内容响应信息包括所述服务内容候选集；
    所述终端设备根据用户授权的数据，从所述服务内容候选集中确定一个或者多个服务内容；
    所述终端设备向用户展示所述一个或者多个服务内容。
  7. 根据权利要求6所述的方法，其特征在于，所述服务内容响应信息中还包括所述服务内容候选集中每个服务内容的标签信息；
    其中，所述终端设备根据用户授权的数据，从所述服务内容候选集中确定一个或者多个服务内容，包括：
    所述终端设备根据所述用户授权的数据和所述服务内容候选集中每个服务内容的标签信息，确定所述一个或者多个服务内容。
  8. 根据权利要求7所述的方法，其特征在于，所述终端设备根据所述用户授权的数据和所述服务内容候选集中每个服务内容的标签信息，确定所述一个或者多个服务内容，包括：
    所述终端设备根据所述用户授权的数据和所述每个服务内容的标签信息，确定所述每个服务内容与所述用户授权的数据的相关系数；
    所述终端设备根据所述相关系数，确定所述一个或者多个服务内容。
  9. 根据权利要求7或8所述的方法，其特征在于，所述业务系统向所述终端设备发送服务内容响应信息之前，所述方法还包括：
    所述业务系统根据预设的算法，确定所述每个服务内容的标签信息。
  10. 根据权利要求6至9中任一项所述的方法，其特征在于，所述业务系统根据所述服务内容请求信息，确定服务内容候选集，包括：
    所述业务系统接收所述终端设备发送的用户行为事件；
    所述业务系统根据所述用户行为事件，确定用户行为属性以及用户的受众群体信息；
    所述业务系统根据所述用户行为属性以及所述用户的受众群体信息，从所述业务系统保存的服务内容中确定所述服务内容候选集。
  11. 根据权利要求6至10中任一项所述的方法，其特征在于，所述多个服务内容为所述业务系统海量实时更新的服务内容中的一种或者多种。
  12. 根据权利要求11所述的方法，其特征在于，所述服务内容为广告业务、视频业务或者新闻业务。
  13. 一种个性化推荐的系统，其特征在于，所述个性化推荐的系统包括终端设备和业务系统，所述终端设备包括处理器和存储器，所述存储器中存储有一个或者多个程序，所述业务系统包括业务运营管理系统、离线业务系统和在线业务系统，其中，
    所述业务运营管理系统，用于向所述离线业务系统发送第一服务内容候选集；
    所述离线业务系统，用于根据预设的算法，确定所述第一服务内容候选集中服务内容的标签信息；
    所述离线业务系统，还用于向所述在线业务系统发送所述第一服务内容候选集中服务内容的标签信息；
    所述在线业务系统，用于接收所述业务应用发送的服务内容请求信息；
    所述在线业务系统，用于根据所述服务内容请求信息，从所述第一服务内容候选集中确定第二服务内容候选集，所述第二服务内容候选集包括多个服务内容；
    所述在线业务系统，还用于向所述终端设备发送服务内容响应信息，所述服务内容响应信息包括所述第二服务内容候选集和所述第二服务内容候选集中每个服务内容的标签信息；
    当所述一个或者多个程序被所述处理器执行时，使得所述终端设备执行以下操作：
    根据用户授权的数据和所述每个服务内容的标签信息，从所述多个服务内容中确定一个或者多个服务内容；
    向用户展示所述一个或者多个服务内容。
  14. 根据权利要求13所述的系统，其特征在于，当所述一个或者多个程序被所述处 理器执行时，使得所述终端设备执行以下操作：
    根据所述用户授权的数据和所述每个服务内容的标签信息，确定所述每个服务内容与所述用户授权的数据的相关系数；
    根据所述相关系数，确定所述一个或者多个服务内容。
  15. 根据权利要求13或14所述的系统，其特征在于，所述离线业务系统还用于：
    接收所述终端设备发送的用户行为事件；
    根据所述用户行为事件，确定用户行为属性以及用户的受众群体信息；
    向所述在线业务系统发送所述用户行为属性以及所述用户的受众群体信息。
    其中，所述在线业务系统具体用于：
    根据所述用户行为属性、所述用户的受众群体信息和所述第一服务内容候选集中服务内容的标签信息，从所述第一服务内容候选集中确定所述第二服务内容候选集。
  16. 根据权利要求13至15中任一项所述的系统，其特征在于，所述第一服务内容候选集中的服务内容为海量实时更新的服务内容。
  17. 根据权利要求16所述的系统，其特征在于，所述多个服务内容为广告业务、视频业务或者新闻业务中的一种或者多种。
PCT/CN2020/082429 2019-04-03 2020-03-31 个性化推荐的方法、终端设备和系统 WO2020200199A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US17/601,009 US11843651B2 (en) 2019-04-03 2020-03-31 Personalized recommendation method and system, and terminal device
EP20784453.1A EP3923163A4 (en) 2019-04-03 2020-03-31 PERSONALIZED RECOMMENDATION PROCESS, TERMINAL EQUIPMENT AND SYSTEM

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910266830.1A CN110059249B (zh) 2019-04-03 2019-04-03 个性化推荐的方法、终端设备和系统
CN201910266830.1 2019-04-03

Publications (1)

Publication Number Publication Date
WO2020200199A1 true WO2020200199A1 (zh) 2020-10-08

Family

ID=67318272

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/082429 WO2020200199A1 (zh) 2019-04-03 2020-03-31 个性化推荐的方法、终端设备和系统

Country Status (4)

Country Link
US (1) US11843651B2 (zh)
EP (1) EP3923163A4 (zh)
CN (1) CN110059249B (zh)
WO (1) WO2020200199A1 (zh)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200314614A1 (en) * 2019-03-29 2020-10-01 Apple Inc. Systems and methods for autonomous vehicle communication
CN110059249B (zh) * 2019-04-03 2022-11-25 华为技术有限公司 个性化推荐的方法、终端设备和系统
CN110688997B (zh) * 2019-09-24 2023-04-18 北京猎户星空科技有限公司 一种图像处理方法及装置
CN111178970B (zh) * 2019-12-30 2023-06-30 微梦创科网络科技（中国）有限公司 广告投放的方法及装置、电子设备和计算机可读存储介质
CN111325036B (zh) * 2020-02-19 2023-10-13 中国人民解放军军事科学院军事科学信息研究中心 一种面向新兴技术预测的佐证事实的抽取方法及系统
CN114255056A (zh) * 2020-09-19 2022-03-29 华为技术有限公司 广告显示方法及电子设备
CN114500731A (zh) * 2020-10-27 2022-05-13 华为技术有限公司 一种广告显示方法及电子设备
CN112463841B (zh) * 2020-11-03 2023-04-21 贵州江南航天信息网络通信有限公司 一种基于工业大数据的智能化决策与精准推送的方法及引擎
CN112506975A (zh) * 2020-11-30 2021-03-16 橙脑教育科技(上海)有限公司 一种信息推荐方法、装置和系统
CN113032676B (zh) * 2021-03-31 2022-11-08 上海天旦网络科技发展有限公司 基于微反馈的推荐方法和系统
CN117151828B (zh) * 2023-10-30 2024-01-30 建信金融科技有限责任公司 推荐物品池的处理方法、装置、设备和介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070214207A1 (en) * 2006-03-09 2007-09-13 Fengzhou Xie Method and system for accurate issuance of data information
CN101334792A (zh) * 2008-07-10 2008-12-31 中国科学院计算技术研究所 一种个性化服务推荐系统和方法
CN103368898A (zh) * 2012-03-26 2013-10-23 中兴通讯股份有限公司 一种实现信息推送的方法和系统
CN103457978A (zh) * 2012-06-01 2013-12-18 日立民用电子株式会社 便携式终端以及具备其的网购内容筛选系统
CN109376309A (zh) * 2018-12-28 2019-02-22 北京百度网讯科技有限公司 基于语义标签的文档推荐方法和装置
CN110059249A (zh) * 2019-04-03 2019-07-26 华为技术有限公司 个性化推荐的方法、终端设备和系统

Family Cites Families (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090300016A1 (en) * 2008-06-02 2009-12-03 Microsoft Corporation Techniques for filtering data
CN101615961B (zh) * 2008-06-24 2011-05-18 华为技术有限公司 媒体内容推荐方法和装置
US20110035280A1 (en) * 2009-08-04 2011-02-10 Visa U.S.A. Inc. Systems and Methods for Targeted Advertisement Delivery
US20110041157A1 (en) * 2009-08-13 2011-02-17 Tandberg Television Inc. Systems and Methods for Selecting Content For a Subscriber of a Content Service Provider
CN102411596A (zh) 2010-09-21 2012-04-11 阿里巴巴集团控股有限公司 一种信息推荐方法及系统
CN102956009B (zh) 2011-08-16 2017-03-01 阿里巴巴集团控股有限公司 一种基于用户行为的电子商务信息推荐方法与装置
US20130055056A1 (en) * 2011-08-30 2013-02-28 Google Inc. System and Method for Searching Belongings Using Social Graph Information
US8521594B1 (en) * 2011-10-18 2013-08-27 Google Inc. Directed content to anonymized users
EP2650800A1 (en) * 2012-04-13 2013-10-16 Alcatel Lucent Method and device for controlling access of users to physical objects belonging to other users through social networks
US8849843B1 (en) * 2012-06-18 2014-09-30 Ez-XBRL Solutions, Inc. System and method for facilitating associating semantic labels with content
US10691743B2 (en) * 2014-08-05 2020-06-23 Sri International Multi-dimensional realization of visual content of an image collection
CN103810162B (zh) 2012-11-05 2017-12-12 腾讯科技（深圳）有限公司 推荐网络信息的方法和系统
CN103593434A (zh) * 2013-11-12 2014-02-19 北京奇虎科技有限公司 应用推荐方法及装置、服务器设备
US20150348106A1 (en) * 2014-05-29 2015-12-03 Google Inc. Dynamic content item creation
CN104298719B (zh) 2014-09-23 2018-02-27 新浪网技术(中国)有限公司 基于社交行为进行用户的类别划分、广告投放方法和系统
US10671679B2 (en) * 2014-12-30 2020-06-02 Oath Inc. Method and system for enhanced content recommendation
CN104965890B (zh) * 2015-06-17 2017-05-31 深圳市腾讯计算机系统有限公司 广告推荐的方法和装置
CN106326277B (zh) * 2015-06-30 2019-07-16 上海证大喜马拉雅网络科技有限公司 一种基于用户行为的音频个性化推荐方法和系统
EP3440556A4 (en) * 2016-04-08 2019-12-11 Pearson Education, Inc. SYSTEM AND METHOD FOR AUTOMATIC GENERATION OF CONTENT AGGREGATION
EP4033431A1 (en) * 2016-06-02 2022-07-27 Kodak Alaris Inc. Method for producing and distributing one or more customized media centric products
CN107123057A (zh) 2017-03-22 2017-09-01 阿里巴巴集团控股有限公司 用户推荐方法及装置
CN108322523B (zh) * 2018-01-19 2021-11-23 维沃移动通信有限公司 一种应用推荐方法、服务器及移动终端
CN108763314B (zh) * 2018-04-26 2021-01-19 深圳市腾讯计算机系统有限公司 一种兴趣推荐方法、装置、服务器及存储介质
CN109034907A (zh) * 2018-08-06 2018-12-18 北京三快在线科技有限公司 广告数据投放方法及装置、电子设备、存储介质
CN109359242B (zh) * 2018-09-25 2019-12-03 潍坊工程职业学院 一种数据推送方法和装置
CN109375309B (zh) 2018-11-30 2020-05-01 东莞市谷麦光学科技有限公司 一种复合led导光板及其制备方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070214207A1 (en) * 2006-03-09 2007-09-13 Fengzhou Xie Method and system for accurate issuance of data information
CN101334792A (zh) * 2008-07-10 2008-12-31 中国科学院计算技术研究所 一种个性化服务推荐系统和方法
CN103368898A (zh) * 2012-03-26 2013-10-23 中兴通讯股份有限公司 一种实现信息推送的方法和系统
CN103457978A (zh) * 2012-06-01 2013-12-18 日立民用电子株式会社 便携式终端以及具备其的网购内容筛选系统
CN109376309A (zh) * 2018-12-28 2019-02-22 北京百度网讯科技有限公司 基于语义标签的文档推荐方法和装置
CN110059249A (zh) * 2019-04-03 2019-07-26 华为技术有限公司 个性化推荐的方法、终端设备和系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3923163A4

Also Published As

Publication number Publication date
EP3923163A1 (en) 2021-12-15
US11843651B2 (en) 2023-12-12
US20220150292A1 (en) 2022-05-12
EP3923163A4 (en) 2022-03-16
CN110059249A (zh) 2019-07-26
CN110059249B (zh) 2022-11-25

Similar Documents

Publication Publication Date Title
WO2020200199A1 (zh) 个性化推荐的方法、终端设备和系统
US11798034B1 (en) Directed content to anonymized users
RU2629449C2 (ru) Устройство, а также способ выбора и размещения целевых сообщений на странице результатов поиска
US20160364736A1 (en) Method and system for providing business intelligence based on user behavior
CN107871244B (zh) 一种广告效果的检测方法和装置
KR100908982B1 (ko) 지능형 정보 제공 시스템 및 방법
US20170286539A1 (en) User profile stitching
CN100545843C (zh) 用于搜索网站的方法和意见跟踪服务器
US20230245151A1 (en) Systems and methods for determining segments of online users from correlated datasets
US9940634B1 (en) Content consumption monitor
US20170364931A1 (en) Distributed model optimizer for content consumption
US20160275545A1 (en) Methods and systems for using device connectivity information in on-line advertising
CN111242661A (zh) 优惠券发放方法和装置以及计算机系统和介质
US8954567B1 (en) Predicting mediaplan traffic
US10331713B1 (en) User activity analysis using word clouds
US20110196871A1 (en) Targeting Online Ads by Grouping and Mapping User Properties
US20150127468A1 (en) User engagement based nonguaranteed delivery pricing
RU118772U1 (ru) Модуль сбора информации для профилирования интернет-пользователей
KR102246387B1 (ko) 마케팅을 위한 인플루언서 분석 솔루션 제공 방법
KR101673372B1 (ko) 템플릿을 기반으로 하는 멀티미디어 네트워크 서비스 시스템 및 방법
US9870577B1 (en) Conditional bids in an auction
US20150242886A1 (en) Ad impression availability and associated adjustment values
US20180033054A1 (en) Content unit creation
US20170085672A1 (en) Commercial-Interest-Weighted User Profiles
JP5847867B2 (ja) 広告主向け参考情報提供装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20784453

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 20784453.1

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2020784453

Country of ref document: EP

Effective date: 20210906

NENP Non-entry into the national phase

Ref country code: DE