WO2017096877A1 - 一种推荐方法和装置 - Google Patents

一种推荐方法和装置 Download PDF

Info

Publication number
WO2017096877A1
WO2017096877A1 PCT/CN2016/089244 CN2016089244W WO2017096877A1 WO 2017096877 A1 WO2017096877 A1 WO 2017096877A1 CN 2016089244 W CN2016089244 W CN 2016089244W WO 2017096877 A1 WO2017096877 A1 WO 2017096877A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
recommended content
behavior data
content
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2016/089244
Other languages
English (en)
French (fr)
Inventor
祁立
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Le Holdings Beijing Co Ltd
LeTV Information Technology Beijing Co Ltd
Original Assignee
Le Holdings Beijing Co Ltd
LeTV Information Technology Beijing Co Ltd
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 Le Holdings Beijing Co Ltd, LeTV Information Technology Beijing Co Ltd filed Critical Le Holdings Beijing Co Ltd
Priority to US15/248,497 priority Critical patent/US20170169349A1/en
Publication of WO2017096877A1 publication Critical patent/WO2017096877A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/048—Fuzzy inferencing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
    • G06F16/63—Querying
    • G06F16/635—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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—Recommending goods or services

Definitions

  • Embodiments of the present invention relate to the field of communications technologies, and in particular, to a recommended method and apparatus.
  • the existing music recommendation scheme can analyze the historical behavior data of the user such as playing, collecting, and paying attention to the music, and know the user's preference, and then recommend the music that meets the user's preference for the user.
  • the existing music recommendation scheme is based on the accumulation of certain historical behavior data after the user uses the music platform for a period of time, for the new user, because there is no historical behavior data, or the historical behavior data is less.
  • the existing music recommendation scheme cannot accurately know the user's preference based on the historical behavior data, so the accuracy of the music recommended for the user is low, and the recommendation effect is not satisfactory.
  • the embodiment of the present invention provides a recommendation method and device, which are used to solve the defect that the accuracy of the music recommended by the user in the existing music recommendation scheme is low, and the accuracy of the recommended content can be improved.
  • An embodiment of the present invention provides a recommendation method, including:
  • the ecological historical behavior data of the user includes at least one of the following historical behavior data: historical behavior data of the user in at least two applications on the at least one terminal, And the user at least on at least two terminals Historical behavior data in an application;
  • the recommended content is recommended to the user.
  • An embodiment of the present invention provides a recommendation apparatus, including:
  • a generating unit configured to generate at least one recommended content according to the ecological historical behavior data of the user;
  • the ecological historical behavior data of the user includes at least one of the following historical behavior data: the user is in at least two applications on the at least one terminal Historical behavior data, and historical behavior data of at least one application of the user on at least two terminals;
  • a recommendation unit configured to recommend the recommended content to the user.
  • An embodiment of the present invention provides a computer program, comprising computer readable code, when the computer readable code is run on a smart terminal, causing the smart terminal to perform the above recommended method.
  • Embodiments of the present invention provide a computer readable medium in which the above computer program is stored.
  • the recommendation method and the device provided by the embodiment of the present invention may generate the recommended content according to the ecological historical behavior data of the user, where the ecological historical behavior data may specifically include: historical behavior of the user in at least two applications on the at least one terminal Data, and historical behavior data of the user in at least one application on at least two terminals; comparing the historical behavior data of the user to the music playing, collecting, and paying according to the existing recommendation scheme, knowing the user's preference, and further The user recommends music that meets the user's preference; since the ecological historical behavior data in the embodiment of the present invention can be derived from multiple terminals or multiple applications, the ecological historical behavior data is more abundant, and the user's preference is based on the rich ecological historical behavior data.
  • the analysis result is more accurate, so the accuracy of the recommended content can be improved; when the user is a new user, the user can be recommended according to the historical behavior data of the user in other applications or other terminals, so that the new user can be solved.
  • Historical behavior in the app It is time to empty, or less historical behavior data, the recommended low accuracy problem.
  • FIG. 1 is a flow chart showing the steps of a first embodiment of a preferred method of the present invention
  • FIG. 2 is a flow chart of steps of a second embodiment of a preferred method of the present invention.
  • FIG. 3 is a flow chart of steps of a third embodiment of a preferred method of the present invention.
  • FIG. 4 is a schematic structural view of a first embodiment of a recommending device according to the present invention.
  • FIG. 5 is a schematic structural diagram of Embodiment 2 of a recommended device according to the present invention.
  • FIG. 6 is a schematic structural view of a third embodiment of a recommending device according to the present invention.
  • Figure 7 is a schematic structural view of a fourth embodiment of a recommending device of the present invention.
  • Figure 8 shows schematically a block diagram of a smart terminal for performing the method according to the invention
  • Fig. 9 schematically shows a storage unit for holding or carrying program code implementing the method according to the invention.
  • FIG. 1 a flow chart of steps in a first embodiment of a preferred method of the present invention is shown.
  • Step 101 Generate at least one recommended content according to the user's ecological historical behavior data.
  • the ecological historical behavior data of the user may specifically include at least one of the following historical behavior data: the user is in at least two applications on the at least one terminal. Historical behavior data, and historical behavior data of at least one application of the user on at least two terminals;
  • the embodiment of the present invention can be applied to any application program such as a music software application and a video software application of the smart terminal, so as to accurately and recommend the user to the user through the application program.
  • Application program such as a music software application and a video software application of the smart terminal, so as to accurately and recommend the user to the user through the application program.
  • User favorite music, video and other content User favorite music, video and other content.
  • the foregoing ecological historical behavior data may be used to represent an operation record generated by a user in an application, and may specifically include the following three situations:
  • Case 1 Historical behavior data of at least two applications of the user on one terminal; for example: multiple applications of the user on the mobile terminal of the mobile phone (music application, video application, wallpaper application, browser application, and game application) Historical behavior data in applications such as programs;
  • Case 2 historical behavior data of the user in an application on at least two terminals; for example: a music application on a mobile terminal, a tablet, a smart TV, or the like, or a video application, or a wallpaper application Historical behavior data in an application, such as a program, or a browser application, or a game application;
  • Case 3 historical behavior data of at least two applications of the user on at least two terminals; for example, a music application, a video application, a wallpaper application of a user on a plurality of terminals such as a mobile phone mobile terminal, a tablet computer, a smart TV, and the like Historical behavior data in applications such as programs, browser applications, and game applications.
  • the obtaining manner of the foregoing ecological historical behavior data may include: obtaining a user's online record through the gateway to obtain the user's ecological historical behavior data; and/or acquiring the user's behavior log flow from the third-party application platform.
  • the manner of obtaining the ecological historical behavior data is not specifically limited in the embodiment of the present invention.
  • the historical operation performed by the user in each terminal and each application can be learned according to the ecological historical behavior data of the user, and the historical operation is further analyzed to obtain the user's preference, and then the user can be obtained according to the user's preference. Recommended content.
  • the method may further include the following steps:
  • the candidate recommended content when the number of the recommended content generated according to the ecological historical behavior data is less than the first threshold, the candidate recommended content may be obtained to supplement the recommended content.
  • the candidate recommendation content may be a public recommendation content that is created by the application background staff and/or automatically generated by the application according to the click volume, and may include: the label is a popular recommendation content, and the label is a new product.
  • the recommended content and label are recommended contents such as recommended content of a certain region.
  • the embodiment of the present invention may further include the following steps:
  • the recommended reason for the recommended content is recommended to the user; for example, the recommended content "Bacchus”, the corresponding recommendation reason is "watching the movie "Red Sorghum””.
  • Step 102 Recommend the recommended content to the user.
  • the recommendation method provided by the embodiment of the present invention may generate the recommended content according to the ecological historical behavior data of the user, where the ecological historical behavior data may specifically include: a history of the user in at least two applications on the at least one terminal. Behavior data, and historical behavior data of at least one application of the user on at least two terminals; comparing the historical behavior data of the user to the music playing, collecting, and paying according to the existing recommendation scheme, and knowing the user's preference, and further The user is recommended to the music that meets the user's preference; since the ecological historical behavior data in the embodiment of the present invention is more abundant, the analysis result of the user's preference is more accurate according to the rich ecological historical behavior data, thereby improving the accuracy of the recommended content.
  • the ecological historical behavior data may specifically include: a history of the user in at least two applications on the at least one terminal. Behavior data, and historical behavior data of at least one application of the user on at least two terminals; comparing the historical behavior data of the user to the music playing, collecting, and paying according to the existing recommendation scheme, and knowing the
  • the user When the user is a new user, the user can be recommended according to the historical behavior data of the user in the third-party application or other terminal, so that the historical behavior data of the new user in the application can be solved, or historical behavior When the data is small, the recommended accuracy is low. question.
  • FIG. 2 a flow chart of the steps of the second embodiment of the preferred method of the present invention is shown.
  • Step 201 Calculate a feature feature of the user according to the ecological historical behavior data of the user, and generate a first recommended content according to the image feature; and/or
  • Step 202 Calculate a similar user of the user according to the ecological historical behavior data of the user, and generate a second recommended content according to the recommended content of the similar user; and/or
  • Step 203 Acquire and according to the behavior object in the ecological historical behavior data of the user.
  • Step 204 Generate at least one recommended content according to at least one of the first recommended content, the second recommended content, and the third recommended content.
  • Step 205 Recommend the recommended content to the user.
  • the embodiment of the present invention refines the step of generating recommended content according to the user's ecological historical behavior data through steps 201 to 204, and analyzes the ecological historical behavior data of the user to obtain the first And recommending the content, and calculating the user's portrait feature and the similar user of the user, respectively generating the first recommended content and the second recommended content, to generate the recommended content according to the first recommended content, the second recommended content, and the third recommended content. .
  • the user's eco-history behavior data may be analyzed to calculate a user's portrait feature, and then the first recommended content is generated according to the image feature;
  • the image feature of the user may be a set of tags for characterizing the user.
  • the image may include basic attributes such as age, gender, and region, and may also include interest features of the user, for example, a language tag of the played music. Properties such as the type of music to be played.
  • the obtained portrait features of the user specifically include: female, 24 years old, and the language label of the played music is Europe and America, and the type of music played.
  • the tag is an anime episode or the like
  • the recommended content obtained according to the portrait feature of the user may specifically include: a popular song in the current female young people group, a music label in the US and Europe, a music tag labeled as anime, and the like.
  • step 202 the user's eco-history behavior data may be analyzed, and the similar users of the user are calculated, thereby obtaining the recommended content of the similar user to generate the second recommended content.
  • the similar user of the user may be a user who has the same interests and interests as the current user, and may specifically calculate a similar user of the current user by using a user based algorithm.
  • the specific process may be:
  • the historical behavior data is obtained by the current user's interest feature, wherein the interest feature may specifically include the user's operational characteristics of the historical behavior object, such as: user viewing, and/or search, and/or click, and/or attention, and/or collection.
  • Passing a certain historical content establishing the user's interest feature vector with the above interest feature as a dimension, and using the above interest feature vector to calculate the similarity between other users and the current user, and determining that the similarity is greater than the first threshold is the current user's similarity
  • the user generates the second recommended content according to the recommended content of the similar user of the current user;
  • the interest feature vector 1 is established according to the ecological historical behavior data of the user A, and the interest feature vector i of other users different from the current user is obtained, where i may be different from the current user.
  • the recommended content 1 of the user B and the recommended content 2 of the user D are obtained, and the recommended content 1 and the recommended content 2 are combined as the recommended content.
  • the recommended content associated with the behavior object may be obtained according to the behavior object in the ecological historical behavior data of the user, and the third recommended content is generated according to the associated recommended content.
  • the behavior object in the above historical behavior data may be "red sorghum”, which can be obtained and "Red Sorghum” related music, such as the episode “Bacchus”, “Sister, you boldly go forward”, etc., you can continue to find related singers, creators or other music albums based on these episodes, etc., to get more A plurality of associated music to generate a third recommended content based on the above music.
  • the user reads the novel "Soldier Assault" through the e-book software, and then the TV drama adapted from the novel of the same name can be obtained according to the novel name "Soldier Assault” recorded in the ecological historical behavior data. Then find the titles, endings and episodes in the TV series, and even other TV dramas in which the same actors participate, and get some related music to generate the third recommended content based on the above music.
  • the user browses some websites through the browser software, leaving a plurality of URL (Uniform Resoure Locator) history records, and then can record according to the ecological historical behavior data.
  • the URLs of these websites are obtained to obtain the background music on the corresponding webpage as the associated music to generate a third push based on the above music. Recommended content.
  • the user also operates a game software, so that the related soundtrack in the game can be obtained according to the name of the game recorded in the ecological historical behavior data, and even the adapted music is obtained.
  • a soundtrack of the same name cartoon or the like to generate a third recommended content based on the above music.
  • the above exemplifies the manner in which the associated recommended content is obtained from the ecological historical behavior data.
  • the content recorded by the ecological historical behavior data and the manner of obtaining the associated music may be determined according to a specific scenario, and the embodiment of the present disclosure does not limit the ecological historical behavior data and the manner of acquiring the associated music. .
  • the user's ecological historical behavior data may also be filtered to filter out ecological historical behavior data that does not meet the user's preference. For example, when the duration of the user watching a movie recorded in the eco-historical behavior data is too short (3 min), the user can be considered that the user does not like the movie, so it can be filtered out. It can be understood that the specific filtering is performed by the embodiment of the present invention. The rules are not restricted.
  • the process of generating at least one recommended content according to at least one of the first recommended content, the second recommended content, and the third recommended content may include: determining the first recommended content. Or the second recommended content or the third recommended content is the recommended content; or the above three recommended contents are combined in any combination to generate the recommended content.
  • the step of generating at least one recommended content according to at least one of the foregoing first recommended content, the second recommended content, and the third recommended content may specifically include:
  • the first recommended content, and/or the second recommended content, and/or the third recommended content are selected according to a preset ratio to obtain at least one recommended content.
  • the first preset ratio is 20% in the embodiment of the present invention
  • the second preset ratio is 20%
  • the third preset ratio is 20%.
  • the embodiment of the present invention may specifically: obtain the recommended content of the first recommended content by a percentage of 20%, and obtain the recommended content of the second recommended content by a percentage of 20%, according to the percentage
  • the ratio of sixty is obtained for the recommended content of the third recommended content, and the three parts of the recommended content obtained above are combined to obtain the recommended content;
  • the first preset ratio is 20% and the second preset ratio is 80% in the embodiment of the present invention.
  • the recommended content of the first recommended content is obtained in a proportion of 20%, and the recommended content of the second recommended content is obtained according to 80%. If the recommended content of the two parts is insufficient, the third recommended content is used for supplementing. .
  • the preset ratio may be determined by a person skilled in the art according to actual application requirements, for example, if a person skilled in the art believes that the accuracy of the first recommended content is higher, the first The ratio of the recommended content is set relatively higher.
  • the preset ratio may be determined according to the behavior data of the recommended content by the user, for example, the first recommended content, the second recommended content, and the third recommended content may be separately targeted to the user.
  • the browsing behavior or the listening behavior of the recommended content is counted, and the ratio of the recommended content in the first recommended content, the second recommended content, and the third recommended content to the total recommended content browsed or listened to by the user is determined according to the foregoing statistical result. Use this ratio as the current preset ratio.
  • a method for determining a preset ratio according to actual application requirements and a method for determining the preset ratio according to behavior data of the recommended content by the user may be used in combination.
  • the method may be used to determine the initial manner.
  • the ratio is set, and as the user's behavior data of the recommended content is accumulated, the current preset ratio can be adjusted by the method 2, etc., and it can be understood that the specific determination manner of the preset ratio is not limited in the embodiment of the present invention. .
  • FIG. 3 a flow chart of the steps of the third embodiment of the preferred method of the present invention is shown.
  • Step 301 Generate at least one recommended content according to the user's ecological historical behavior data.
  • the ecological historical behavior data of the user may specifically include at least one of the following historical behavior data: the user is in at least two applications on the at least one terminal. Historical behavior data, and historical behavior data of at least one application of the user on at least two terminals;
  • Step 302 Extract recommended content features of the recommended content.
  • Step 303 Enter the recommended content feature of the recommended content, and/or the user feature, and/or the interaction feature of the user and the historical content into a factoring machine (FM), by the FM
  • FM factoring machine
  • the model outputs the user's preference for the recommended content; wherein, by analyzing the ecological historical behavior data, the user feature and the interactive feature of the user and the historical content are obtained;
  • Step 304 Sort the recommended content according to the user's preference for the recommended content output by the FM model.
  • Step 305 Recommend, to the user, recommended content according to the user's preference for the recommended content.
  • the embodiment of the present invention adds steps 302 to 304 to sort the recommended content according to the user's favorite degree, and recommend the sorted recommended content to the user through step 305, so as to make the most suitable user.
  • the recommended content is in the forefront to provide a better experience for the user.
  • the recommended content feature may specifically include various attributes such as a label of the recommended content (for example, post-90, rock, Europe, and the like), for example, a song whose name is “nunchaku” is recommended.
  • the content features may include: post-90, rap, Chinese style, mainland, and the like; the user feature may specifically be a user's portrait feature, etc.; the user's interaction with the historical content may specifically include: the user clicks on the historical content, And/or collections, and/or red hearts and other operations.
  • the embodiment of the present invention may input the recommended content feature of the recommended content, and/or the user feature, and/or the interactive feature of the user and the historical content into the factor molecular machine FM model, and the FM model performs the foregoing multiple vector dimensions according to the foregoing
  • the user's preference for the recommended content is calculated and outputted, and the user's preference for the recommended content is compared, and the recommended content may be sorted according to the user's preference for the recommended content.
  • the foregoing FM model may specifically be:
  • the embodiment of the present invention may further include the following steps to train the foregoing FM model:
  • Step S1 extracting a comprehensive feature from the user's ecological historical behavior data; wherein the comprehensive feature may specifically include at least one of the following features: a user feature, a historical content feature, and an interaction feature between the user and the historical content;
  • Step S2 integrating the integrated feature into the FM model to train the FM model.
  • the historical content feature may be a feature of the historical content obtained from the ecological historical behavior data, where the historical content is used to represent the content that the user has operated, and may specifically include: user viewing, and/or searching, and / or click, and / or attention, and / or favorite content.
  • the FM model is trained by using the comprehensive feature extracted from the user's ecological historical behavior data to obtain a model formula capable of predicting the favorite content according to the user's preference.
  • a recommendation method provided by an embodiment of the present invention may be configured to train an FM model according to the extracted comprehensive feature, to sort the recommended content generated according to the ecological historical behavior data according to the trained FM model according to the trained FM model, to obtain The optimally ranked recommended content is recommended to the user.
  • the FM model according to the embodiment of the present invention can be based on the user characteristics obtained from the analysis of the ecological historical behavior data and the interaction characteristics of the user and the historical content, and the recommended analysis from the recommended content.
  • a plurality of feature vectors such as content features, that is, predicting the user's preference for the recommended content based on the plurality of vector dimensions, so that the predicted user's preference for the recommended content is more accurate, and then sorting according to the favorite degree, and obtaining Optimal sorting results.
  • FIG. 4 a schematic structural diagram of a first embodiment of a recommendation apparatus according to the present invention is shown, which may include: a generating unit 401 and a recommending unit 402;
  • the generating unit 401 may be configured to generate at least one recommended content according to the ecological historical behavior data of the user;
  • the ecological historical behavior data of the user includes at least one of the following historical behavior data: at least two of the users on the at least one terminal Historical behavior data in an application, and historical behavior data in at least one application of the user on at least two terminals;
  • a recommendation unit 402 configured to recommend the recommended content to the user
  • the recommended device may further include:
  • a candidate unit configured to determine whether the number of recommended content is less than a first threshold, and if the number of recommended content is less than the first threshold, acquiring candidate recommended content to supplement the recommended content;
  • the candidate content is used to indicate the public recommendation content recommended to all users.
  • the embodiment of the present invention may further include:
  • the recommendation reason unit may be configured to recommend the recommendation reason of the recommended content to the user.
  • FIG. 5 a schematic structural diagram of a second embodiment of a recommendation apparatus according to the present invention is shown, which may include: a generating unit 501 and a recommending unit 502;
  • the generating unit 501 may be configured to generate at least one recommended content according to the ecological historical behavior data of the user;
  • the ecological historical behavior data of the user includes at least one of the following historical behavior data: at least two of the users on the at least one terminal Historical behavior data in an application, and Historical behavior data of the user in at least one application on at least two terminals;
  • a recommendation unit 502 configured to recommend the recommended content to the user
  • the generating unit 501 may specifically include:
  • the first generating subunit 5011 may be configured to calculate a portrait feature of the user according to the ecological historical behavior data of the user, and generate a first recommended content according to the portrait feature; and/or
  • the second generation sub-unit 5012 may be configured to calculate a similar user of the user according to the ecological historical behavior data of the user, and generate a second recommended content according to the recommended content of the similar user; and/or
  • the third generation sub-unit 5013 may be configured to acquire, according to the behavior object in the ecological historical behavior data of the user, the recommended content associated with the behavior object, and generate a third recommended content according to the associated recommended content. ;
  • the generating subunit 5014 is configured to generate at least one recommended content according to at least one of the first recommended content, the second recommended content, and the third recommended content.
  • the generating the recommended content sub-unit 5014 may specifically include:
  • the obtaining module may be configured to select the first recommended content, and/or the second recommended content, and/or the third recommended content according to a preset ratio to obtain at least one recommended content.
  • FIG. 6 a schematic structural diagram of a third embodiment of a recommendation apparatus of the present invention is shown, which may include: a generating unit 601, a first extracting unit 602, a calculating unit 603, a sorting unit 604, and a recommending unit 605;
  • the generating unit 601 may be configured to generate at least one recommended content according to the ecological historical behavior data of the user;
  • the ecological historical behavior data of the user includes at least one of the following historical behavior data: at least two of the users on the at least one terminal Historical behavior data in an application, and historical behavior data in at least one application of the user on at least two terminals;
  • the first extracting unit 602 may be configured to extract recommended content features of the recommended content.
  • the calculating unit 603 may be configured to input the recommended content feature of the recommended content, and/or the user feature, and/or the interaction feature of the user and the historical content into the factor molecular machine FM model, and output the a user's preference for the recommended content; wherein, by analyzing the ecological historical behavior data, the user feature and the interactive feature of the user and the historical content are obtained;
  • the sorting unit 604 is configured to sort the recommended content according to the user's preference for the recommended content output by the FM model
  • a recommendation unit 605 configured to recommend the recommended content to the user
  • the above recommendation unit 605 may specifically include:
  • the recommendation sub-unit 6051 may be configured to recommend the recommended content according to the user's preference for the recommended content to the user.
  • Embodiment 4 of a recommendation apparatus of the present invention may specifically include: a generating unit 701, a first extracting unit 702, a calculating unit 703, a sorting unit 704, a second extracting unit 705, and a training unit. 706 and recommendation unit 707; wherein
  • the generating unit 701 may be configured to generate at least one recommended content according to the ecological historical behavior data of the user;
  • the ecological historical behavior data of the user includes at least one of the following historical behavior data: at least two of the users on the at least one terminal Historical behavior data in an application, and historical behavior data in at least one application of the user on at least two terminals;
  • the first extracting unit 702 may be configured to extract recommended content features of the recommended content.
  • the calculating unit 703 may be configured to input the recommended content feature of the recommended content, and/or the user feature, and/or the interaction feature of the user and the historical content into the factor molecular machine FM model, and output the a user's preference for the recommended content; wherein, by analyzing the ecological historical behavior data, the user feature and the interactive feature of the user and the historical content are obtained;
  • the sorting unit 704 may be configured to sort the recommended content according to the user's preference for the recommended content output by the FM model;
  • the second extracting unit 705 may be configured to extract the integrated feature from the ecological historical behavior data of the user; wherein the comprehensive feature includes at least one of the following features: a user feature, a historical content feature, and a user interaction with the historical content. feature;
  • the training unit 706 can be configured to fuse the comprehensive feature into the FM model to train the FM model;
  • a recommendation unit 707 configured to recommend the recommended content to the user
  • the recommendation unit 707 may specifically include:
  • the recommendation subunit 7071 may be configured to recommend the recommended content according to the user's preference for the recommended content to the user.
  • the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.
  • the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, ie may be located A place, or it can be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment. Those of ordinary skill in the art can understand and implement without deliberate labor.
  • FIG. 8 illustrates that a smart terminal in accordance with the present invention can be implemented.
  • the smart terminal conventionally includes a processor 810 and a computer program product or computer readable medium in the form of a memory 820.
  • the memory 820 may be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), an EPROM, a hard disk, or a ROM.
  • the memory 820 has A storage space 830 of program code 831 that performs any of the method steps above.
  • storage space 830 for program code may include various program code 831 for implementing various steps in the above methods, respectively.
  • the program code can be read from or written to one or more computer program products.
  • Such computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards or floppy disks.
  • Such a computer program product is typically a portable or fixed storage unit as described with reference to FIG.
  • the storage unit may have a storage section, a storage space, and the like arranged similarly to the storage 820 in the intelligent terminal of FIG.
  • the program code can be compressed, for example, in an appropriate form.
  • the storage unit includes computer readable code 831', ie, code readable by a processor, such as 810, that when executed by the smart terminal causes the smart terminal to perform each of the methods described above step.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Automation & Control Theory (AREA)
  • Fuzzy Systems (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种推荐方法和装置，其中的方法具体包括：根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据（101）；将所述推荐内容推荐给所述用户（102）。由此，能够提高推荐内容的准确度。

Description

一种推荐方法和装置
本申请要求在2015年12月9日提交中国专利局、申请号为201510908328.8、发明名称为“一种推荐方法和装置”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
本发明实施例涉及通信技术领域，特别是涉及一种推荐方法和装置。
背景技术
随着智能终端和网络技术的发展，用户可通过各式各样的网站、应用程序等途径播放音乐。然而，各类音乐平台均为用户提供了数以万计的音乐资源，用户若想从中找到自己所喜欢的音乐恰似大海捞针。因此，这就需要能够根据用户的音乐喜好对用户进行音乐推荐。
现有的一种音乐推荐方案可以根据用户对音乐的播放、收藏、关注等历史行为数据分析，获知用户的喜好，进而为用户推荐符合用户喜好的音乐。
但是，现有的音乐推荐方案是在用户使用音乐平台后一段时间后，积累了一定历史行为数据的基础上进行的，对于新用户而言，由于其没有历史行为数据，或者历史行为数据较少，在这样的场景下，现有的音乐推荐方案不能根据历史行为数据准确获知用户的喜好，因此为用户推荐的音乐的准确度较低，推荐效果不理想。
发明内容
本发明实施例提供一种推荐方法和装置，用以解决现有的音乐推荐方案中为用户推荐的音乐的准确度较低的缺陷，能够提高推荐内容的准确度。
本发明实施例提供一种推荐方法，包括：
根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少 一个应用中的历史行为数据；
将所述推荐内容推荐给所述用户。
本发明实施例提供一种推荐装置，包括：
生成单元，用于根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
推荐单元，用于将所述推荐内容推荐给所述用户。
本发明实施例提供一种计算机程序，其包括计算机可读代码，当所述计算机可读代码在智能终端上运行时，导致所述智能终端执行上述的推荐方法。
本发明实施例提供一种计算机可读介质，其中存储了上述的计算机程序。
本发明实施例提供的一种推荐方法和装置，可以根据用户的生态历史行为数据生成推荐内容，其中，生态历史行为数据具体可以包括：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；相对于现有的推荐方案根据用户对音乐的播放、收藏、关注等历史行为数据分析，获知用户的喜好，进而为用户推荐符合用户喜好的音乐；由于本发明实施例中的生态历史行为数据可以来源于多个终端或者多个应用，因此生态历史行为数据更为丰富，根据丰富的生态历史行为数据对用户的喜好的分析结果更为精准，因此能够提高推荐内容的准确度；在用户为新用户的时候，可以根据用户在其他应用程序或者其他终端上的历史行为数据为用户进行推荐，因此，可以解决新用户在应用程序中的历史行为数据为空、或者历史行为数据较少时，推荐的准确度低的问题。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍，显而易见地，下 面描述中的附图是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明的一种推荐方法实施例一的步骤流程图；
图2为本发明的一种推荐方法实施例二的步骤流程图；
图3为本发明的一种推荐方法实施例三的步骤流程图；
图4为本发明的一种推荐装置实施例一的结构示意图；
图5为本发明的一种推荐装置实施例二的结构示意图；
图6为本发明的一种推荐装置实施例三的结构示意图；
图7为本发明的一种推荐装置实施例四的结构示意图；
图8示意性地示出了用于执行根据本发明的方法的智能终端的框图；以及
图9示意性地示出了用于保持或者携带实现根据本发明的方法的程序代码的存储单元。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
方法实施例一
参照图1，示出了本发明的一种推荐方法实施例一的步骤流程图，具体可以包括：
步骤101、根据用户的生态历史行为数据生成至少一项推荐内容；上述用户的生态历史行为数据具体可以包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
本发明实施例可以应用于智能终端的音乐软件应用程序、视频软件应用程序等任意应用程序，以通过上述应用程序向用户准确地、推荐符合用 户喜好的音乐、视频等内容。
本发明实施例中，上述生态历史行为数据可用于表示用户在应用程序中产生的操作记录，其具体可以包括如下三种情形：
情形1、用户在一个终端上的至少两个应用中的历史行为数据；例如：用户在手机移动终端的多个应用(音乐应用程序、视频应用程序、壁纸应用程序、浏览器应用程序及游戏应用程序等应用)中的历史行为数据；
情形2、用户在至少两个终端上的一个应用中的历史行为数据；例如：用户在手机移动终端、平板电脑、智能电视等多个终端上的音乐应用程序、或者视频应用程序、或者壁纸应用程序、或者浏览器应用程序、或者游戏应用程序等应用程序中的历史行为数据；
情形3、用户在至少两个终端上的至少两个应用中的历史行为数据；例如：用户在手机移动终端、平板电脑、智能电视等多个终端上的音乐应用程序、视频应用程序、壁纸应用程序、浏览器应用程序、游戏应用程序等应用程序中的历史行为数据。
本发明实施例中，上述生态历史行为数据的获取方式具体可以包括：可以通过网关获取用户上网记录，以获取用户的生态历史行为数据；和/或，从第三方应用平台获取用户的行为日志流，以获取用户的生态历史行为数据；和/或，根据储存在用户本地终端上的数据cookie获取用户的生态历史行为数据，本发明实施例对获取生态历史行为数据的方式不做具体限定。
本发明实施例中，可以根据用户的生态历史行为数据获知用户在各终端以及各应用中进行的历史操作，并进一步对上述历史操作进行分析获知用户的喜好，进而可以根据用户的喜好获取符合用户喜好的推荐内容。
在本发明的一种可选实施例中，所述方法具体还可以包括步骤：
判断上述的推荐内容的数目是否小于第一阈值，若上述推荐内容的数目小于所述第一阈值，则获取侯补推荐内容对上述推荐内容进行补充；其中，上述侯补推荐内容具体可以用于表示推荐给所有用户的公共推荐内容。
本发明实施例中，当根据生态历史行为数据生成的上述推荐内容的数目小于第一阈值时，可以获取侯补推荐内容对上述推荐内容进行补充，上 述候补推荐内容可以为应用程序后台工作人员创建的、和/或，应用程序根据点击量而自动生成的推荐给所有用户的公共推荐内容，具体可以包括：标签为热门的推荐内容、标签为新品的推荐内容、标签为某一地域的推荐内容等推荐内容。
在本发明的一种可选实施例中，本发明实施例中具体还可以包括步骤：
将所述推荐内容的推荐理由推荐给所述用户；例如：推荐内容“酒神曲”，其对应的推荐理由为“观看了电影“红高粱””。
步骤102、将上述推荐内容推荐给上述用户。
综上，本发明实施例提供的一种推荐方法，可以根据用户的生态历史行为数据生成推荐内容，其中，生态历史行为数据具体可以包括：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；相对于现有的推荐方案根据用户对音乐的播放、收藏、关注等历史行为数据分析，获知用户的喜好，进而为用户推荐符合用户喜好的音乐；由于本发明实施例中的生态历史行为数据更为丰富，根据丰富的生态历史行为数据对用户的喜好的分析结果更为精准，因此能够提高推荐内容的准确度；在用户为新用户的时候，可以根据用户在第三方应用程序或者其他终端上的历史行为数据为用户进行推荐，因此，可以解决新用户在应用程序中的历史行为数据为空、或者历史行为数据较少时，推荐的准确度低的问题。
方法实施例二
参照图2，示出了本发明一种推荐方法实施例二的步骤流程图，具体可以包括：
步骤201、根据用户的生态历史行为数据计算所述用户的画像特征，并根据所述画像特征生成第一推荐内容；和/或
步骤202、根据所述用户的生态历史行为数据计算所述用户的相似用户，并根据所述相似用户的推荐内容生成第二推荐内容；和/或
步骤203、根据所述用户的生态历史行为数据中的行为对象，获取与 所述行为对象相关联的推荐内容，并根据所述相关联的推荐内容生成第三推荐内容；
步骤204、根据所述第一推荐内容、所述第二推荐内容和所述第三推荐内容中的至少一种，生成至少一项推荐内容；
步骤205、将上述推荐内容推荐给上述用户。
相比于方法实施例一，本发明实施例通过步骤201至步骤204对根据用户的生态历史行为数据生成推荐内容的步骤进行了细化，通过对用户的生态历史行为数据进行分析，以获得第三推荐内容，并通过计算出用户的画像特征和用户的相似用户，分别生成第一推荐内容和第二推荐内容，以根据上述第一推荐内容、第二推荐内容及第三推荐内容生成推荐内容。
步骤201中，可以通过对用户的生态历史行为数据进行分析，计算得用户的画像特征，进而根据上述画像特征生成第一推荐内容；
上述用户的画像特征具体可以为刻画用户特征的标签(tag)集合，例如，其具体可以包括年龄、性别、地域等基础属性，也可以包括用户的兴趣特征，例如，播放的音乐的语种标签、播放的音乐的种类标签等属性。
在本发明的一种应用示例中，假设通过对用户的生态历史行为数据加以分析，获得的用户的画像特征具体包含：女，24岁，播放的音乐的语种标签为欧美，播放的音乐的种类标签为动漫插曲等，则根据上述用户的画像特征获得的推荐内容具体可以包括：当前女性年轻人群体中的热门歌曲、标签为欧美的音乐、标签为动漫的音乐等等推荐内容。
步骤202中，可以通过对用户的生态历史行为数据进行分析，计算的到上述用户的相似用户，进而获得上述相似用户的推荐内容，以生成第二推荐内容；
本发明实施例中，上述用户的相似用户可以为与当前用户具有相同兴趣爱好的用户，具体可以通过基于用户(user based)算法计算出当前用户的相似用户，具体过程可以为：通过用户的生态历史行为数据得到当前用户的兴趣特征，其中，兴趣特征具体可以包括用户对历史行为对象的操作特征，例如：用户观赏、和/或搜索、和/或点击、和/或关注、和/或收藏 过某一历史内容；并以上述兴趣特征为维度建立用户的兴趣特征向量，并利用上述兴趣特征向量计算其他用户与当前用户的相似度，确定相似度大于第一阈值的用户为当前用户的相似用户，根据当前用户的相似用户的推荐内容生成第二推荐内容；
在本发明的一种应用示例中，假设根据用户甲的生态历史行为数据建立了兴趣特征向量1，获取区别于当前用户的其他用户的兴趣特征向量i，其中，i可以为区别于当前用户的其他用户的标识；并计算兴趣特征向量i与兴趣特征向量1的余弦值，确定该余弦值即为与当前用户的相似度，根据上述相似度确定的用户甲的相似用户为用户乙和用户丁，则获取用户乙的推荐内容1和用户丁的推荐内容2，并合并推荐内容1和推荐内容2为推荐内容。
步骤203中，可以根据上述用户的生态历史行为数据中的行为对象，获取与上述行为对象相关联的推荐内容，并根据上述相关联的推荐内容生成第三推荐内容。
在本发明的一种应用示例中，假设用户使用视频播放软件观看了一部名为“红高粱”的电影，那么上述历史行为数据中的行为对象即可以为“红高粱”，可以获取到与“红高粱”相关的音乐，如插曲“酒神曲”、“妹妹你大胆的往前走”等，还可以继续根据这些插曲等又可以找到相关的歌手、创作人员或其他音乐专辑，从而获得更多的关联音乐，以根据上述音乐生成第三推荐内容。
在本发明的另一种应用示例中，用户通过电子书软件阅读了小说“士兵突击”，那么便可以根据生态历史行为数据中记录下来的小说名称“士兵突击”获取到同名小说改编的电视剧，进而找到电视剧里的片头、片尾及插曲，甚至同一演员参演的其他影视剧，得到一些关联的音乐，以根据上述音乐生成第三推荐内容。
在本发明的再一种应用示例中，用户通过浏览器软件浏览了一些网站，留下了多个URL(Uniform Resoure Locator，统一资源定位器)历史记录，那么便可以根据生态历史行为数据中记录下来的这些网站的URL去获取相应网页上的背景音乐作为关联音乐，以根据上述音乐生成第三推 荐内容。
在本发明的又一种应用示例中，用户还操作了一款游戏软件，那么便可以根据生态历史行为数据中记录下来该游戏的名称等获取到游戏中的相关配乐，甚至是获取到改编的同名动画片等的配乐，以根据上述音乐生成第三推荐内容。
以上示例性的列举了根据生态历史行为数据中获取相关联的推荐内容的方式。在发明实施例中，对于生态历史行为数据所记录的内容以及获取相关联音乐的方式，都可以根据具体场景而定，本公开实施例并不对生态历史行为数据以及获取相关联音乐的方式进行限制。
需要说明的是，在根据用户的生态历史行为数据生成推荐内容前，还可以对用户的生态历史行为数据进行过滤，以过滤掉不符合用户喜好的生态历史行为数据。例如，在生态历史行为数据中记录的用户观看某电影的时长过短(3min)时，可以认为用户并不喜欢该电影，故可以将其过滤掉，可以理解，本发明实施例对于具体的过滤规则不加以限制。
本实际应用中，上述根据所述第一推荐内容、所述第二推荐内容和所述第三推荐内容中的至少一种，生成至少一项推荐内容的过程具体可以包括：确定第一推荐内容、或者第二推荐内容、或者第三推荐内容为推荐内容；或者上述三个推荐内容任意组合合并以生成推荐内容。
在本发明的一种可选实施例中，上述根据上述第一推荐内容、上述第二推荐内容和上述第三推荐内容中的至少一种，生成至少一项推荐内容的步骤，具体可以包括：
按预置比例选取上述第一推荐内容、和/或上述第二推荐内容、和/或上述第三推荐内容，以得到至少一项推荐内容。
在本发明的一种应用示例1中，假设本发明实施例中第一预置比例为百分之二十，第二预置比例为百分之二十，第三预置比例为百分之六十，则本发明实施例具体可以为：按百分之二十的比例获取第一推荐内容的推荐内容，按百分之二十的比例获取第二推荐内容的推荐内容，按百分之六十的比例获取第三推荐内容的推荐内容，将上述获取的三部分推荐内容进行合并，以得到推荐内容；
在本发明的一种应用示例2中，假设本发明实施例中第一预置比例为百分之二十，第二预置比例为百分之八十，则本发明实施例具体可以为：按百分之二十的比例获取第一推荐内容的推荐内容，按百分之八十获取第二推荐内容的推荐内容，若上述两部分的推荐内容数量不足，则使用第三推荐内容进行补充。
本发明的一种可选实施例中，上述预置比例可由本领域技术人员依据实际应用需求确定，例如：若本领域技术人员认为第一推荐内容的准确率更高一些，则可以将第一推荐内容对应的比例设置的相对高一点。
在本发明的另一种可选实施例中，还可以根据用户对推荐内容的行为数据确定上述预置比例，例如：可以分别对用户针对第一推荐内容、第二推荐内容、第三推荐内容中的推荐内容的浏览行为或者收听行为进行统计，并根据上述统计结果确定第一推荐内容、第二推荐内容、第三推荐内容中的推荐内容分别占用户浏览或收听的总推荐内容的比例，以将该比例做为当前的预置比例。
可以理解，本领域技术人员依据实际应用需求确定预置比例的方式一和根据用户对推荐内容的行为数据确定上述预置比例的方式二可以组合使用，例如，在初始时可以采用方式一确定预置比例，而随着用户对于推荐内容的行为数据的积累，可以不断地采用方式二对当前的预置比例进行调整等，可以理解，本发明实施例对于预置比例的具体确定方式不加以限制。
方法实施例三
参照图3，示出了本发明一种推荐方法实施例三的步骤流程图，具体可以包括：
步骤301、根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据具体可以包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
步骤302、提取所述推荐内容的推荐内容特征；
步骤303、将所述推荐内容的推荐内容特征、和/或用户特征、和/或用户与历史内容的交互特征输入因式分子机FM(Factorization Machines，因式分子机)中，由所述FM模型输出所述用户对所述推荐内容的喜爱度；其中，通过对所述生态历史行为数据进行分析，以得到所述用户特征及所述用户与历史内容的交互特征；
步骤304、根据所述FM模型输出的所述用户对所述推荐内容的喜爱度对所述推荐内容进行排序；
步骤305、将根据所述用户对所述推荐内容的喜爱度进行排序后的推荐内容推荐给所述用户。
相对于方法实施例一，本发明实施例增加了步骤302至步骤304，以对推荐内容按用户的喜爱度进行排序，并通过步骤305将排序后的推荐内容推荐给用户，以使最符合用户喜好的推荐内容排在最前面的位置，以为用户提供更好的体验。
本发明实施例中，推荐内容特征具体可以包括推荐内容的标签等各种属性(例如：90后、摇滚、欧美等等属性)，例如：一首名字为“双节棍”的歌，其推荐内容特征具体可以包括：90后、饶舌、中国风、大陆等等特征；用户特征具体可以为用户的画像特征等等；用户与历史内容的交互特征具体可以包括：用户对历史内容进行的点击、和/或收藏、和/或红心关注等操作。
本发明实施例可以将上述推荐内容的推荐内容特征、和/或用户特征，和/或用户与历史内容的交互特征输入因式分子机FM模型中，由上述FM模型根据上述多个向量维度进行计算并输出上述用户对所述推荐内容的喜爱度，并比较用户对推荐内容的喜爱度的大小，进而可以根据用户对推荐内容的喜爱度由大到小的顺序对上述推荐内容进行排序。
本发明一种可选实施例中，上述FM模型具体可以为：
Figure PCTCN2016089244-appb-000001
其中，上述u可以代表当前用户的标识；i可以代表当前推荐内容的标识；d可以代表综合特征，其中，综合特征具体可以包括如下特征的至 少一项：用户特征、历史内容特征(推荐内容特征)及用户与历史内容的交互特征，上述u、i、d可以作为自变量，参与上述FM模型的运算；y可以代表预测结果，也即当前用户对推荐内容的喜爱度；x可以代表训练样例(推荐内容)；W0可以代表全局偏置因子；Wu可以代表用户特征偏置因子；Wi可以代表推荐内容特征偏置因子，Wd可以代表综合特征参数因子；Vu，f、Vi，f可以代表用户和推荐内容之间的交互因子；Vu，f、Vd，f可以代表用户和综合特征之间的交互因子。
在本发明的一种可选实施例中，本发明实施例具体还可以包括如下步骤，以训练上述FM模型：
步骤S1、从所述用户的生态历史行为数据中提取综合特征；其中，所述综合特征具体可以包括如下特征的至少一项：用户特征、历史内容特征及用户与历史内容的交互特征；
步骤S2、将所述综合特征融合至FM模型中，以训练所述FM模型。
本发明实施例中，上述历史内容特征可以为从生态历史行为数据中获取的历史内容的特征，上述历史内容用于表示用户操作过的内容，具体可以包括：用户观赏、和/或搜索、和/或点击、和/或关注、和/或收藏过的内容。
本发明实施例中，利用从用户的生态历史行为数据中提取中的综合特征对FM模型进行训练，以得到能够根据用户的喜好而对推荐内容进行喜爱度的预测的模型公式。
综上，本发明实施例提供的一种推荐方法，可以根据提取的综合特征训练FM模型，以根据训练出的FM模型对本发明实施例根据生态历史行为数据而生成的推荐内容进行排序，以得到最优排序的推荐内容并推荐给用户，由于本发明实施例中FM模型可以根据从生态历史行为数据中分析获得的用户特征和用户与历史内容的交互特征，及从推荐内容中分析得到的推荐内容特征等多个特征向量，也即基于多个向量维度对用户对推荐内容的喜爱度进行预测，因此预测的用户对推荐内容的喜爱度更为精准，进而根据上述喜爱度进行排序，能够获得最优的排序结果。
需要说明的是，对于方法实施例，为了简单描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本申请实施例并不受所 描述的动作顺序的限制，因为依据本申请实施例，某些步骤可以采用其他顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于可选实施例，所涉及的动作并不一定是本申请实施例所必须的。
装置实施例一
参照图4，示出了本发明一种推荐装置实施例一的结构示意图，具体可以包括：生成单元401及推荐单元402；其中，
上述生成单元401，可以用于根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
推荐单元402，可以用于将所述推荐内容推荐给所述用户；
本发明的一种可选实施例中，上述推荐装置具体还可以包括：
候补单元，用于判断所述推荐内容的数目是否小于第一阈值，若所述推荐内容的数目小于所述第一阈值，则获取侯补推荐内容对所述推荐内容进行补充；其中，所述侯补推荐内容用于表示推荐给所有用户的公共推荐内容。
本发明的一种可选实施例中，本发明实施例具体还可以包括：
推荐理由单元，可以用于将所述推荐内容的推荐理由推荐给所述用户。
装置实施例二
参照图5，示出了本发明一种推荐装置实施例二的结构示意图，具体可以包括：生成单元501及推荐单元502；其中，
上述生成单元501，可以用于根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及 用户在至少两个终端上的至少一个应用中的历史行为数据；
推荐单元502，可以用于将所述推荐内容推荐给所述用户；
其中，上述生成单元501具体可以包括：
第一生成子单元5011，可以用于根据用户的生态历史行为数据计算所述用户的画像特征，并根据所述画像特征生成第一推荐内容；和/或
第二生成子单元5012，可以用于根据所述用户的生态历史行为数据计算所述用户的相似用户，并根据所述相似用户的推荐内容生成第二推荐内容；和/或
第三生成子单元5013，可以用于根据所述用户的生态历史行为数据中的行为对象，获取与所述行为对象相关联的推荐内容，并根据所述相关联的推荐内容生成第三推荐内容；
生成子单元5014，可以用于根据所述第一推荐内容、所述第二推荐内容和所述第三推荐内容中的至少一种，生成至少一项推荐内容。
在本发明的一种可选实施例中，上述生成推荐内容子单元5014，具体可以包括：
获取模块，可以用于按预置比例选取所述第一推荐内容、和/或所述第二推荐内容、和/或所述第三推荐内容，以得到至少一项推荐内容。
装置实施例三
参照图6，示出了本发明一种推荐装置实施例三的结构示意图，具体可以包括：生成单元601、第一提取单元602、计算单元603、排序单元604及推荐单元605；其中，
上述生成单元601，可以用于根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
第一提取单元602，可以用于提取所述推荐内容的推荐内容特征；
计算单元603，可以用于将所述推荐内容的推荐内容特征、和/或用户特征，和/或用户与历史内容的交互特征输入因式分子机FM模型中，由所述FM模型输出所述用户对所述推荐内容的喜爱度；其中，通过对所述生态历史行为数据进行分析，以得到所述用户特征及所述用户与历史内容的交互特征；
排序单元604，可以用于根据所述FM模型输出的所述用户对所述推荐内容的喜爱度对所述推荐内容进行排序；
推荐单元605，可以用于将所述推荐内容推荐给所述用户；
则上述推荐单元605，具体可以包括：
推荐子单元6051，可以用于将根据所述用户对所述推荐内容的喜爱度进行排序后的所述推荐内容推荐给所述用户。
装置实施例四
参照图7，示出了本发明一种推荐装置实施例四的结构示意图，具体可以包括：生成单元701、第一提取单元702、计算单元703、排序单元704、第二提取单元705、训练单元706及推荐单元707；其中，
上述生成单元701，可以用于根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
第一提取单元702，可以用于提取所述推荐内容的推荐内容特征；
计算单元703，可以用于将所述推荐内容的推荐内容特征、和/或用户特征，和/或用户与历史内容的交互特征输入因式分子机FM模型中，由所述FM模型输出所述用户对所述推荐内容的喜爱度；其中，通过对所述生态历史行为数据进行分析，以得到所述用户特征及所述用户与历史内容的交互特征；
排序单元704，可以用于根据所述FM模型输出的所述用户对所述推荐内容的喜爱度对所述推荐内容进行排序；
第二提取单元705，可以用于从所述用户的生态历史行为数据中提取综合特征；其中，所述综合特征包括如下特征的至少一项：用户特征、历史内容特征及用户与历史内容的交互特征；
训练单元706，可以用于将所述综合特征融合至FM模型中，以训练所述FM模型；
推荐单元707，可以用于将所述推荐内容推荐给所述用户；
则上述推荐单元707，具体可以包括：
推荐子单元7071，可以用于将根据所述用户对所述推荐内容的喜爱度进行排序后的所述推荐内容推荐给所述用户。
对于装置实施例而言，由于其与方法实施例基本相似，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
以上所描述的装置实施例仅仅是示意性的，其中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模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下，即可以理解并实施。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到各实施方式可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件。基于这样的理解，上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如ROM/RAM、磁碟、光盘等，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)执行各个实施例或者实施例的某些部分所述的方法。
例如，图8示出了可以实现根据本发明的智能终端。该智能终端传统上包括处理器810和以存储器820形式的计算机程序产品或者计算机可读介质。存储器820可以是诸如闪存、EEPROM(电可擦除可编程只读存储器)、EPROM、硬盘或者ROM之类的电子存储器。存储器820具有用于 执行上述方法中的任何方法步骤的程序代码831的存储空间830。例如，用于程序代码的存储空间830可以包括分别用于实现上面的方法中的各种步骤的各个程序代码831。这些程序代码可以从一个或者多个计算机程序产品中读出或者写入到这一个或者多个计算机程序产品中。这些计算机程序产品包括诸如硬盘，紧致盘(CD)、存储卡或者软盘之类的程序代码载体。这样的计算机程序产品通常为如参考图9所述的便携式或者固定存储单元。该存储单元可以具有与图8的智能终端中的存储器820类似布置的存储段、存储空间等。程序代码可以例如以适当形式进行压缩。通常，存储单元包括计算机可读代码831’，即可以由例如诸如810之类的处理器读取的代码，这些代码当由智能终端运行时，导致该智能终端执行上面所描述的方法中的各个步骤。
最后应说明的是：以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (16)

  1. 一种推荐方法，其特征在于，所述方法包括：
    根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
    将所述推荐内容推荐给所述用户。
  2. 根据权利要求1所述的推荐方法，其特征在于，所述根据用户的生态历史行为数据生成至少一项推荐内容的步骤，包括：
    根据用户的生态历史行为数据计算所述用户的画像特征，并根据所述画像特征生成第一推荐内容；和/或
    根据所述用户的生态历史行为数据计算所述用户的相似用户，并根据所述相似用户的推荐内容生成第二推荐内容；和/或
    根据所述用户的生态历史行为数据中的行为对象，获取与所述行为对象相关联的推荐内容，并根据所述相关联的推荐内容生成第三推荐内容；
    根据所述第一推荐内容、所述第二推荐内容和所述第三推荐内容中的至少一种，生成至少一项推荐内容。
  3. 根据权利要求1所述的推荐方法，其特征在于，所述方法还包括：
    提取所述推荐内容的推荐内容特征；
    将所述推荐内容的推荐内容特征、和/或用户特征，和/或用户与历史内容的交互特征输入因式分子机FM模型中，由所述FM模型输出所述用户对所述推荐内容的喜爱度；其中，通过对所述生态历史行为数据进行分析，以得到所述用户特征及所述用户与历史内容的交互特征；
    根据所述FM模型输出的所述用户对所述推荐内容的喜爱度对所述推荐内容进行排序；
    则所述将所述推荐内容推荐给所述用户的步骤，包括：
    将根据所述用户对所述推荐内容的喜爱度进行排序后的推荐内容推荐给所述用户。
  4. 根据权利要求3所述的推荐方法，其特征在于，所述方法还包括：
    从所述用户的生态历史行为数据中提取综合特征；其中，所述综合特征包括如下特征的至少一项：用户特征、历史内容特征及用户与历史内容的交互特征；
    将所述综合特征融合至FM模型中，以训练所述FM模型。
  5. 根据权利要求2所述的方法，其特征在于，所述根据所述第一推荐内容、所述第二推荐内容和所述第三推荐内容中的至少一种，生成至少一项推荐内容的步骤，包括：
    按预置比例选取所述第一推荐内容、和/或所述第二推荐内容、和/或所述第三推荐内容，以得到至少一项推荐内容。
  6. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    判断所述推荐内容的数目是否小于第一阈值，若所述推荐内容的数目小于所述第一阈值，则获取侯补推荐内容进行补充；其中，所述侯补推荐内容用于表示推荐给所有用户的公共推荐内容。
  7. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    将所述推荐内容的推荐理由推荐给所述用户。
  8. 一种推荐装置，其特征在于，包括：
    生成单元，用于根据用户的生态历史行为数据生成至少一项推荐内容；所述用户的生态历史行为数据包括如下历史行为数据中的至少一种：用户在至少一个终端上的至少两个应用中的历史行为数据、及用户在至少两个终端上的至少一个应用中的历史行为数据；
    推荐单元，用于将所述推荐内容推荐给所述用户。
  9. 根据权利要求8所述的推荐装置，其特征在于，所述生成单元，包括：
    第一生成子单元，用于根据用户的生态历史行为数据计算所述用户的画像特征，并根据所述画像特征生成第一推荐内容；和/或
    第二生成子单元，用于根据所述用户的生态历史行为数据计算所述用户的相似用户，并根据所述相似用户的推荐内容生成第二推荐内容；和/或
    第三生成子单元，用于根据所述用户的生态历史行为数据中的行为对象，获取与所述行为对象相关联的推荐内容，并根据所述相关联的推荐内容生成第三推荐内容；及
    生成推荐内容子单元，用于根据所述第一推荐内容、所述第二推荐内容和所述第三推荐内容中的至少一种，生成至少一项推荐内容。
  10. 根据权利要求8所述的推荐装置，其特征在于，所述装置还包括：
    第一提取单元，用于提取所述推荐内容的推荐内容特征；
    计算单元，用于将所述推荐内容的推荐内容特征、和/或用户特征，和/或用户与历史内容的交互特征输入因式分子机FM模型中，由所述FM模型输出所述用户对所述推荐内容的喜爱度；其中，通过对所述生态历史行为数据进行分析，以得到所述用户特征及所述用户与历史内容的交互特征；
    排序单元，用于根据所述FM模型输出的所述用户对所述推荐内容的喜爱度对所述推荐内容进行排序；
    则所述推荐单元，包括：
    推荐子单元，用于将根据所述用户对所述推荐内容的喜爱度进行排序后的所述推荐内容推荐给所述用户。
  11. 根据权利要求10所述的推荐装置，其特征在于，所述装置还包括：
    第二提取单元，用于从所述用户的生态历史行为数据中提取综合特征；其中，所述综合特征包括如下特征的至少一项：用户特征、历史内容特征及用户与历史内容的交互特征；
    训练单元，用于将所述综合特征融合至FM模型中，以训练所述FM模型。
  12. 根据权利要求9所述的推荐装置，其特征在于，所述生成推荐内容 子单元，包括：
    获取模块，用于按预置比例选取所述第一推荐内容、和/或所述第二推荐内容、和/或所述第三推荐内容，以得到至少一项推荐内容。
  13. 根据权利要求8所述的推荐装置，其特征在于，所述装置还包括：
    候补单元，用于判断所述推荐内容的数目是否小于第一阈值，若所述推荐内容的数目小于所述第一阈值，则获取侯补推荐内容对所述推荐内容进行补充；其中，所述侯补推荐内容用于表示推荐给所有用户的公共推荐内容。
  14. 根据权利要求8所述的推荐装置，其特征在于，所述装置还包括：
    推荐理由单元，用于将所述推荐内容的推荐理由推荐给所述用户。
  15. 一种计算机程序，包括计算机可读代码，当所述计算机可读代码在智能终端上运行时，导致所述智能终端执行根据权利要求1-7中的任一个所述的推荐方法。
  16. 一种计算机可读介质，其中存储了如权利要求15所述的计算机程序。
PCT/CN2016/089244 2015-12-09 2016-07-07 一种推荐方法和装置 Ceased WO2017096877A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/248,497 US20170169349A1 (en) 2015-12-09 2016-08-26 Recommending method and electronic device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510908328.8A CN105975472A (zh) 2015-12-09 2015-12-09 一种推荐方法和装置
CN201510908328.8 2015-12-09

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/248,497 Continuation US20170169349A1 (en) 2015-12-09 2016-08-26 Recommending method and electronic device

Publications (1)

Publication Number Publication Date
WO2017096877A1 true WO2017096877A1 (zh) 2017-06-15

Family

ID=56988247

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/089244 Ceased WO2017096877A1 (zh) 2015-12-09 2016-07-07 一种推荐方法和装置

Country Status (3)

Country Link
US (1) US20170169349A1 (zh)
CN (1) CN105975472A (zh)
WO (1) WO2017096877A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114968246A (zh) * 2022-08-01 2022-08-30 深圳市明源云科技有限公司 数据分析组件生成方法、设备及计算机可读存储介质

Families Citing this family (40)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106504019A (zh) * 2016-10-31 2017-03-15 深圳前海弘稼科技有限公司 一种植物推荐方法及装置
CN106557560A (zh) * 2016-11-11 2017-04-05 天翼爱音乐文化科技有限公司 基于用户兴趣的层次音乐推荐方法
CN106776892A (zh) * 2016-11-30 2017-05-31 北京红马传媒文化发展有限公司 基于音乐平台数据评估音乐作品网络关注数据方法及系统
CN108255893B (zh) * 2016-12-29 2021-03-30 北京国双科技有限公司 个性化对象推荐方法和装置
CN106649842A (zh) * 2016-12-30 2017-05-10 上海博泰悦臻电子设备制造有限公司 一种基于融合数据的交叉推荐方法、系统及一种车机
CN106850780A (zh) * 2017-01-16 2017-06-13 北京奇虎科技有限公司 系统级的应用程序信息推荐方法、装置和移动终端
US10609453B2 (en) 2017-02-21 2020-03-31 The Directv Group, Inc. Customized recommendations of multimedia content streams
CN107809485A (zh) * 2017-10-31 2018-03-16 广州云移信息科技有限公司 一种信息推荐方法及终端
CN107918658B (zh) * 2017-11-20 2021-05-07 金蝶软件（中国）有限公司 一种商机生成方法及系统
CN108011941B (zh) * 2017-11-29 2019-07-12 Oppo广东移动通信有限公司 内容推送方法、装置、服务器以及存储介质
CN110019163A (zh) * 2017-12-05 2019-07-16 北京京东尚科信息技术有限公司 对象特征的预测、推荐的方法、系统、设备和存储介质
CN108563321A (zh) * 2018-01-02 2018-09-21 联想(北京)有限公司 信息处理方法和电子设备
CN108960988A (zh) * 2018-06-28 2018-12-07 北京金山安全软件有限公司 一种个性化壁纸推荐方法、装置、终端设备及存储介质
CN109064091B (zh) * 2018-07-13 2021-09-17 天津五八到家科技有限公司 资源确定、资源处理方法及装置
CN108984752B (zh) * 2018-07-17 2021-06-04 华北理工大学 一种图书馆专业书籍智能推荐方法
CN109063163B (zh) * 2018-08-14 2022-12-02 腾讯科技（深圳）有限公司 一种音乐推荐的方法、装置、终端设备和介质
CN109033441A (zh) * 2018-08-16 2018-12-18 安徽大尺度网络传媒有限公司 一种基于大数据分析的推送方法及装置
CN109492128B (zh) * 2018-10-30 2020-01-21 北京字节跳动网络技术有限公司 用于生成模型的方法和装置
CN109408729B (zh) * 2018-12-05 2022-02-08 广州市百果园信息技术有限公司 推荐物料确定方法、装置、存储介质和计算机设备
CN109948057B (zh) * 2019-03-21 2022-03-01 北京地平线机器人技术研发有限公司 感兴趣内容推送方法、装置及电子设备和介质
CN110046303B (zh) * 2019-04-09 2022-05-17 有光创新（北京）信息技术有限公司 一种基于需求匹配平台实现的信息推荐方法及装置
CN110222233B (zh) * 2019-06-14 2021-01-15 北京达佳互联信息技术有限公司 视频推荐方法、装置、服务器及存储介质
CN110457590B (zh) * 2019-06-25 2021-08-27 华院计算技术(上海)股份有限公司 基于小数据输入的智能用户画像方法
CN110413165B (zh) * 2019-06-26 2021-07-16 五八有限公司 图标配置方法、装置、电子设备和计算机可读介质
CN110287421A (zh) * 2019-06-28 2019-09-27 北京金山安全软件有限公司 一种信息内容推荐方法、装置及电子设备
CN110569429B (zh) * 2019-08-08 2023-11-24 创新先进技术有限公司 一种内容选择模型的生成方法、装置和设备
CN110910199B (zh) * 2019-10-16 2024-05-28 中国平安人寿保险股份有限公司 项目信息的排序方法、装置、计算机设备及存储介质
CN113139122B (zh) * 2020-01-20 2024-11-22 阿里巴巴集团控股有限公司 信息推荐方法、系统及设备
CN111439268B (zh) * 2020-03-31 2023-03-14 重庆长安汽车股份有限公司 一种主动提供个性化服务的方法、装置、云端服务器及汽车
CN111899047A (zh) * 2020-07-14 2020-11-06 拉扎斯网络科技（上海）有限公司 资源推荐方法、装置、计算机设备及计算机可读存储介质
CN112035743B (zh) * 2020-08-28 2021-10-15 腾讯科技（深圳）有限公司 数据推荐方法、装置、计算机设备以及存储介质
CN114372193A (zh) * 2020-10-15 2022-04-19 上海倍增智能科技有限公司 一种用户画像精准投放系统及投放方法
CN112163165B (zh) * 2020-10-21 2024-05-17 腾讯科技（深圳）有限公司 信息推荐方法、装置、设备及计算机可读存储介质
CN113010946B (zh) * 2021-02-26 2024-01-23 深圳市万翼数字技术有限公司 数据分析方法、电子设备及相关产品
CN115114513B (zh) * 2021-03-22 2025-12-19 腾讯科技（深圳）有限公司 基于人工智能的推荐方法、装置及电子设备
CN113592535B (zh) * 2021-06-30 2024-04-16 北京新氧科技有限公司 一种广告推荐方法、装置、电子设备和存储介质
CN115705379A (zh) * 2021-08-05 2023-02-17 中移(成都)信息通信科技有限公司 智能推荐方法及装置、设备、存储介质
CN113742593B (zh) * 2021-09-15 2025-02-25 北京沃东天骏信息技术有限公司 用于推送信息的方法和装置
CN114780842B (zh) * 2022-04-20 2022-12-13 北京字跳网络技术有限公司 一种数据处理方法、装置、设备及存储介质
CN116842229A (zh) * 2023-05-22 2023-10-03 北京百度网讯科技有限公司 关联对象推荐方法、装置、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102073717A (zh) * 2011-01-07 2011-05-25 南京大学 一种面向垂直电子商务网站的首页推荐方法
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
CN103136253A (zh) * 2011-11-30 2013-06-05 腾讯科技（深圳）有限公司 获取信息的方法和装置
CN104750789A (zh) * 2015-03-12 2015-07-01 百度在线网络技术（北京）有限公司 标签的推荐方法及装置
CN105095343A (zh) * 2015-05-28 2015-11-25 百度在线网络技术（北京）有限公司 信息处理方法、信息显示方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN102073717A (zh) * 2011-01-07 2011-05-25 南京大学 一种面向垂直电子商务网站的首页推荐方法
CN103136253A (zh) * 2011-11-30 2013-06-05 腾讯科技（深圳）有限公司 获取信息的方法和装置
CN104750789A (zh) * 2015-03-12 2015-07-01 百度在线网络技术（北京）有限公司 标签的推荐方法及装置
CN105095343A (zh) * 2015-05-28 2015-11-25 百度在线网络技术（北京）有限公司 信息处理方法、信息显示方法及装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114968246A (zh) * 2022-08-01 2022-08-30 深圳市明源云科技有限公司 数据分析组件生成方法、设备及计算机可读存储介质

Also Published As

Publication number Publication date
CN105975472A (zh) 2016-09-28
US20170169349A1 (en) 2017-06-15

Similar Documents

Publication Publication Date Title
WO2017096877A1 (zh) 一种推荐方法和装置
US10650861B2 (en) Video summarization and collaboration systems and methods
Sun et al. Ranking domain-specific highlights by analyzing edited videos
Tso-Sutter et al. Tag-aware recommender systems by fusion of collaborative filtering algorithms
KR101816113B1 (ko) 컴퓨터 실행 방법, 시스템 및 컴퓨터 판독 가능 매체
CN104756512B (zh) 媒体概括
CN108197330B (zh) 基于社交平台的数据挖掘方法及装置
US9235853B2 (en) Method for recommending musical entities to a user
US11973819B1 (en) Image recommendations for thumbnails for online media items based on user activity
JP2021103543A (ja) ライブストリームコンテンツを推奨するための機械学習の使用
EP3690674A1 (en) Method for recommending video content
Garcia del Molino et al. Phd-gifs: personalized highlight detection for automatic gif creation
CN109511015B (zh) 多媒体资源推荐方法、装置、存储介质及设备
CN104486649B (zh) 视频内容评级方法及装置
CN109165302A (zh) 多媒体文件推荐方法及装置
Cho et al. AMPS: Predicting popularity of short-form videos using multi-modal attention mechanisms in social media marketing environments
WO2015101155A1 (zh) 向用户推荐信息的方法
CN114845149B (zh) 视频片段的剪辑方法、视频推荐方法、装置、设备及介质
CN103207917B (zh) 标注多媒体内容的方法、生成推荐内容的方法及系统
KR20200049193A (ko) 콘텐츠 추천 방법 및 이를 지원하는 서비스 장치
Shah Multimodal-based multimedia analysis, retrieval, and services in support of social media applications
CN110888896A (zh) 数据搜寻方法及其数据搜寻系统
Matsumoto et al. Music video recommendation based on link prediction considering local and global structures of a network
Sun et al. Ranking highlights in personal videos by analyzing edited videos
CN118747230B (zh) 音频副本检测方法及装置、设备、存储介质、程序产品

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: 16872077

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16872077

Country of ref document: EP

Kind code of ref document: A1