WO2017201976A1 - 主题推荐方法以及装置 - Google Patents

主题推荐方法以及装置 Download PDF

Info

Publication number
WO2017201976A1
WO2017201976A1 PCT/CN2016/105167 CN2016105167W WO2017201976A1 WO 2017201976 A1 WO2017201976 A1 WO 2017201976A1 CN 2016105167 W CN2016105167 W CN 2016105167W WO 2017201976 A1 WO2017201976 A1 WO 2017201976A1
Authority
WO
WIPO (PCT)
Prior art keywords
objects
recommended
target
target user
user
Prior art date
Application number
PCT/CN2016/105167
Other languages
English (en)
French (fr)
Inventor
刘志容
唐睿明
董振华
何秀强
曹国祥
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
Publication of WO2017201976A1 publication Critical patent/WO2017201976A1/zh
Priority to US16/198,704 priority Critical patent/US11830033B2/en

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
    • G06F16/95—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
    • G06Q30/0255—Targeted advertisements based on user history
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • G06F16/285—Clustering or classification
    • 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
    • G06F16/951—Indexing; Web crawling techniques
    • 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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Definitions

  • the present invention relates to the field of terminals, and in particular, to a subject recommendation method and apparatus.
  • the present invention provides a subject recommendation method and apparatus capable of recommending a recommendation object to a user in the form of a topic.
  • the present invention provides a subject recommendation method.
  • the computing device collects a historical operation behavior of the sample user for the M objects, and predicts, according to the historical operation behavior of each of the M objects, the target user to each of the M objects according to the sample user Favorite value.
  • M is a positive integer greater than or equal to 1.
  • the computing device also collects category classification data of the N recommended objects, and classifies the N recommended objects according to the category classification data of the N recommended objects, thereby obtaining X topics.
  • the N recommended objects are part or all of the M objects.
  • Each of the X topics includes at least one of the N recommended objects.
  • X is a positive integer greater than or equal to 1
  • N is a positive integer greater than or equal to 1
  • N ⁇ M is a positive integer greater than or equal to
  • the target user Calculating, by the target user, a preference value of the target user for each of the X topics according to a preference value of the recommended object included in each of the X topics, and the target
  • the theme is pushed to the target user.
  • the target theme is the subject of the X users whose preference value is greater than a recommendation threshold.
  • the subject recommendation method proposed by the present invention first classifies the recommended objects into multiple topics, and then, the preference value of the theme is greater than the threshold. The theme is pushed to the target user.
  • the above method can push a plurality of recommended objects of a topic of interest to the target user, so that the target user can have multiple choices in the same topic. Moreover, more than one topic can be pushed to the target user, so that the target user can select among multiple topics, thereby truly attracting the interest of the target user and bringing benefits.
  • the computing device determines, according to the historical operation behavior of the sample user for each of the M objects, the sample user to the M The preference data for each object in the object.
  • the computing device trains the prediction model based on the feature data of the sample user, the preference data of the sample user for each of the M objects, the context feature, and the feature data of each of the M objects.
  • the computing device inputs the feature data of the target user and the feature data of each of the M objects into the trained prediction model to predict the target user to each of the M objects The preference value.
  • the computing device constructs the sample user and the M devices according to a history operation behavior of the sample user for each of the M objects An interaction diagram between the objects, and then calculating similarities between the M objects according to the interaction diagram between the sample user and the M objects. And determining, according to the historical operation behavior of each of the n operation objects by the target user, the target user to the n operations The preference value of each operation object in the object.
  • the operation object is an object in which the target user has performed a history operation behavior among the M objects.
  • n is a positive integer greater than or equal to 1, n ⁇ M.
  • the computing device predicts the target user to the M objects according to the preference value of the target user for each of the n operation objects and the similarity between the M objects The preference value of each object.
  • the computing device performs clustering according to the category distinguishing data of the N recommended objects, thereby obtaining X topics. Therefore, the computing device can automatically classify similar recommended objects under the same topic without manual intervention.
  • the computing device presets the X preset topics and the category preset conditions corresponding to each of the X topics. Different topics correspond to different classification preset conditions. The computing device determines whether the category distinguishing data of each of the N recommended objects meets a category preset condition corresponding to any one of the X topics, and matches any one of the X topics The recommended objects of the classification preset conditions corresponding to the theme are classified into the corresponding topics. Therefore, the computing device can automatically set the classification preset conditions of the theme as needed to filter out the customized theme.
  • the present invention provides a subject recommendation device, the device comprising a prediction module, a classification module, a calculation module, and a recommendation module.
  • the prediction module is configured to collect a historical operation behavior of the sample user for the M objects, and predict, according to the historical operation behavior of each of the M objects, the target user in the M objects The preference value of each object.
  • M is a positive integer greater than or equal to 1.
  • the classification module is configured to collect category classification data of the N recommended objects, and classify the N recommended objects according to the category classification data of the N recommended objects, thereby obtaining X topics.
  • the N recommended objects are part or all of the M objects.
  • Each of the X topics includes at least one of the N recommended objects.
  • X is a positive integer greater than or equal to 1
  • N is a positive integer greater than or equal to 1
  • N ⁇ M The calculating module is configured to calculate, according to a preference value of the recommended object included in each of the X topics, the target user's preference value for each of the X topics .
  • the recommendation module is configured to push the target topic to the target user.
  • the target theme is the subject of the X users whose preference value is greater than a recommendation threshold.
  • the subject recommendation method proposed by the present invention first classifies the recommended objects into multiple topics, and then, the preference value of the theme is greater than the threshold. The theme is pushed to the target user.
  • the above method can push a plurality of recommended objects of a topic of interest to the target user, so that the target user can have multiple choices in the same topic. Moreover, more than one topic can be pushed to the target user, so that the target user can select among multiple topics, thereby truly attracting the interest of the target user and bringing benefits.
  • the prediction module includes a determining unit, a training unit, and a prediction unit.
  • the determining unit is configured to determine preference data of the sample user for each of the M objects according to a historical operation behavior of the sample user for each of the M objects.
  • the training unit is configured to perform a prediction model on the prediction model according to the feature data of the sample user, the preference data of the sample user for each of the M objects, the context feature, and the feature data of each of the M objects. training.
  • the prediction unit is configured to input feature data of the target user and feature data of each of the M objects into a trained prediction model to predict the target user to the M objects The preference value of each object.
  • the prediction module includes a construction unit, a calculation unit, a determination unit, and a prediction unit.
  • the building unit is configured to construct an interaction diagram between the sample user and the M objects according to historical operation behavior of the sample user for each of the M objects.
  • the calculating unit is configured to calculate a similarity between the M objects according to an interaction diagram between the sample user and the M objects.
  • the determination sheet The element is configured to determine, according to the historical operation behavior of each of the n operation objects by the target user, a preference value of the target user to each of the n operation objects.
  • the operation object is an object in which the target user has performed a history operation behavior among the M objects.
  • n is a positive integer greater than or equal to 1, n ⁇ M.
  • the prediction unit is configured to predict, according to the preference value of each of the n operation objects by the target user, and the similarity between the M objects, the target user The preference value of each of the M objects.
  • the classification module includes a clustering unit.
  • the clustering unit is configured to perform clustering according to the category distinguishing data of the N recommended objects, thereby obtaining X topics. Therefore, the computing device can automatically classify similar recommended objects under the same topic without manual intervention.
  • the classification module includes a preset unit and a belonging unit.
  • the preset unit is configured to preset the X preset topics and the category preset conditions corresponding to each of the X topics. Different topics correspond to different classification preset conditions.
  • the categorizing unit is configured to determine whether the category distinguishing data of each of the N recommended objects meets a categorization preset condition corresponding to any one of the X topics, and the The recommended objects of the classification preset conditions corresponding to any one of the X themes are classified into the corresponding topics. Therefore, the computing device can automatically set the classification preset conditions of the theme as needed to filter out the customized theme.
  • the present invention provides a computing device.
  • the computing device includes a storage unit, a communication interface, and a processor coupled to the storage unit and communication interface.
  • the storage unit is configured to store an instruction
  • the processor is configured to execute the instruction
  • the communication interface is configured to transmit data with a target user.
  • the present invention provides a computer readable storage medium, the computer readable storage
  • the medium stores program code for the subject recommendation performed by the computing device.
  • the program code includes instructions for performing the method in the first aspect.
  • FIG. 1 is a flowchart of a method for recommending a topic according to an embodiment of the present invention
  • FIG. 2 is a flow chart showing a specific step of step 110 in the subject recommendation method shown in FIG. 1;
  • step 3 is a flow chart showing another specific step of step 110 in the subject recommendation method shown in FIG. 1;
  • FIG. 4 is an interaction diagram between a sample user and an object in a topic recommendation method according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a display table in a topic recommendation method according to an embodiment of the present invention.
  • FIG. 6 is a schematic diagram of a tag cloud in a topic recommendation method according to an embodiment of the present invention.
  • FIG. 7 is a schematic structural diagram of a theme recommendation apparatus according to an embodiment of the present disclosure.
  • FIG. 8 is a schematic structural diagram of another subject recommendation apparatus according to an embodiment of the present invention.
  • FIG. 9 is a schematic structural diagram of still another subject recommendation apparatus according to an embodiment of the present invention.
  • FIG. 10 is a schematic structural diagram of a computing device provided by an implementation of the present invention.
  • the computing device can be a desktop computer, a notebook computer, or a tablet computer.
  • the computing device can also have a CPU (Central Processing Unit), an MCU (Micro Control Unit), a DSP (Digital Signal Processor, etc.), such as a Tablet Personal Computer, a smart phone, and the like. Digital Signal Processor) or a device for executing a processor such as a control circuit or a logic circuit.
  • CPU Central Processing Unit
  • MCU Micro Control Unit
  • DSP Digital Signal Processor
  • a device for executing a processor such as a control circuit or a logic circuit.
  • FIG. 1 is a flowchart of a method for recommending a topic according to an embodiment of the present invention.
  • the execution subject of the subject recommendation method of the present implementation is a computing device, and the subject recommendation method of the present implementation includes the following steps:
  • the computing device collects a historical operation behavior of the sample user for the M objects, and predicts a preference value of the target user for each of the M objects according to the historical operation behavior of the sample user for each of the M objects.
  • the object may be an application.
  • the application may be Uber, facebook, Myspace, Twitter, and MSN (Microsoft Service Network), and the like.
  • the object may also be an article, a video, a song, or the like, which is not specifically limited by the present invention.
  • a sample user is a user who has historically manipulated the object and was selected as a sample.
  • the historical user behaviors of the sample user include, but are not limited to, browsing, clicking, downloading, using, paying, uninstalling, and evaluating.
  • the historical operation behavior of the sample user on the object may include the historical operation behavior of the target user on the object, or may not include the historical operation behavior of the target user on the object.
  • predicting the target user's preference value for each of the M objects according to the historical operation behavior of the sample object for each of the M objects specifically includes the following steps:
  • the computing device determines, according to a historical operation behavior of the sample user for each of the M objects, the sample user's preference data for each of the M objects.
  • M is a positive integer greater than or equal to 1.
  • the preference data can be a value between the intervals [0, 1]. When the value is 1, it indicates that the sample user has the highest preference for the object. When the value is 0, the sample user has the lowest preference for the object.
  • the computing device determines the preference data of the sample user for each of the M objects based on the historical operational behavior of the sample user for each of the M objects.
  • the A object is one of the M objects
  • the computing device may pre-set the correspondence between the historical operation behavior of the sample object by the sample user and the preference data of the sample user to the A object, and then, according to the sample user pair.
  • the historical operation behavior of the A object and the above correspondence determine the preference data of the sample user for the A object.
  • Presetting the correspondence between the historical operation behavior of the sample object by the sample user and the preference data of the sample user to the A object is: if the sample user pays for the A object, the preference data of the sample user to the A object is 1, If the sample user downloads the A object, the sample user's preference data for the A object is 0.8. If the sample user browses the A object, the sample user's preference data for the A object is 0.2. Therefore, once the sample user has paid for the A object, it can be determined that the sample user's preference data for the A object is 1.
  • sample user browses, downloads, and pays for the A object at the same time, it may be determined that the sample user has the highest preference data for the A object, or that the sample user's preference data for the A object is the sum of the three, etc. Wait.
  • the computing device trains the prediction model according to the feature data of the sample user, the sample user's preference data for each of the M objects, the context feature, and the feature data of each of the M objects.
  • the feature data of the sample user is data capable of reflecting the characteristics of the sample user.
  • the feature data of the sample user may include, but is not limited to, the gender, age, place of residence, and the like of the sample user.
  • the gender of the sample user can be represented by “0” and “1”
  • "0” is represented as female
  • "1" is expressed as male.
  • the age of the sample user can be directly expressed by the age of the sample user
  • the residence of the sample user can be represented by a code, for example, "1” for New York, "2" for Washington, and the like.
  • a contextual feature is data that is capable of reflecting the characteristics of the environment, such as weather and the like.
  • the feature data of the object is data capable of reflecting the characteristics of the sample user.
  • the feature data of the object may include, but is not limited to, a category of the object (shopping class or game class, etc.) and the like.
  • the category of the object can be represented by code, for example, "1" for the shopping class, "2" for the game class, and so on.
  • the predictive model can be a Logistic Regression model, a decision tree, and a matrix decomposition.
  • the feature data of the sample user, the sample user's preference data for each of the M objects, the context feature, and the feature data of each of the M objects are trained on the prediction model.
  • the sample user includes sample user A 1 , sample user A 2 , ...; M objects include: object a 1 , object a 2 , .
  • the characteristic data samples user A 1, the context feature, object a characteristic data 1 and a sample user A 1 on the object a preference data 1 as one of the training data on the prediction model training; characteristic data of the sample user of A 1, context feature, object a characteristic data 2, and a sample user a 1 on the object a preference data 2 as a training data prediction model training; the feature data samples user of a 2, contextual features, objects a feature data 2 and
  • the sample user A 2 trains the prediction model on the preference data of the object a 2 as a training data. After a large amount of data training, the predictive model learns a priori knowledge of the user's preference value for objects with certain feature data.
  • the computing device inputs the feature data of the target user and the feature data of each of the M objects to the trained prediction model to predict a preference value of the target user for each of the M objects.
  • the preference value can be a value between the intervals [0, 1]. When the value is 1, it indicates that the target user has the highest preference for the object. When the value is 0, the target user has the lowest preference for the object.
  • the user who already has some feature data in the model has prior knowledge of the preference value of the object with certain feature data, so when inputting the feature data of the target user and each of the M objects After the feature data of the object, it is possible to predict the preference value of the target user for each of the M objects.
  • the target user is the target user u
  • the M objects include the object a 1 , the object a 2 .
  • the prediction model can predict the preference value of the target user u to the object a 1 ; when the feature data and the object of the target user u are input to the prediction model After the feature data of a 2 , the prediction model can predict the preference value of the target user u to the object a 2 .
  • predicting the target user's preference value for each of the M objects according to the historical operation behavior of the sample object for each of the M objects specifically includes the following steps:
  • the computing device constructs an interaction graph between the sample user and the M objects according to the historical operation behavior of the sample user for each of the M objects.
  • the interaction diagram between the sample user and the M objects can be constructed as follows, as shown in FIG. 4, a black dot in the figure is represented as a sample user, and a white point in the figure is represented as M objects.
  • An object in the figure, an edge between the black point and the white point is represented as a sample user corresponding to the black point having a historical operation behavior for the object corresponding to the white point.
  • the object corresponding to the white point has a different historical operation behavior, and the edge is artificially given different weights. For example, when the historical operation behavior is paid, the weight is 1; when the historical operation behavior is download, the weight is 0.8, and when the historical operation behavior is browsing, the weight is 0.2 and so on.
  • the computing device calculates the similarity between the M objects according to the interaction diagram between the sample user and the M objects.
  • the computing device can calculate the similarity between the two objects according to the SimRank algorithm.
  • the SimRank algorithm is characterized in that when the input is an interaction diagram (the interaction diagram is composed of multiple points and the edges between two points), the output is the similarity between all the points in the interaction diagram.
  • the SimRank algorithm is a mature algorithm. Please refer to the relevant data for specific algorithms.
  • the computing device uses the interaction graph between the sample user and the M objects as the input of the SimRank algorithm, the output is the similarity between all the points in the interaction graph, that is, including the similarity between the sample users.
  • the similarity between the sample user and the M objects and the similarity between the M objects.
  • the computing device selects the similarity between the M objects from the output.
  • the similarity between the two objects can be a value between the intervals [0, 1], when the value is 1, it means that the two objects are similar. The highest degree, when the value is 0, it means that the two objects have the lowest similarity.
  • the computing device determines, according to a historical operation behavior of the target user for each of the n operation objects, a preference value of the target user for each of the n operation objects.
  • the operation object is an object in which the target user has performed historical operation behavior among the M objects, and n is a positive integer greater than or equal to 1, n ⁇ M.
  • the computing device may pre-set the relationship between the historical operation behavior of the target object by the target user and the preference value of the target user to the operation object.
  • the target user is the target user u
  • the operation object a is one of the n operation objects. If the target user u has paid for the operation object a, the target user u has a preference value of 1 for the operation object a; if the target user u once downloaded the operation object a, the target user u has a preference value for the operation object a. If it is 0.8, if the target user u has browsed the operation object a, the target user u has a preference value of 0.2 for the operation object a and the like.
  • the computing device determines a preference value of the target user for each of the n operation objects according to the historical operation behavior of the target user for each of the n operation objects. For example, when n operation objects include Uber, facebook, and google, the target user's preference value for each of the n operation objects may be determined as: if the target user has paid for Uber, the target user pair Uber's preference value is determined to be 1; if the target user has downloaded facebook, the target user's preference value for facebook is determined to be 0.8. If the target user has browsed google, the target user's preference value for google is determined to be 0.2.
  • the target user only when the historical operation behavior of the target user is different for the n operation objects, the target user has different preference values for each of the n operation objects, but in actual applications, when browsing times, payment When the amount, the frequency of use, and the like are different, the target user may also have different preference values for each of the n operation objects.
  • the computing device predicts a preference value of the target user for each of the M objects according to a preference value of the target user for each of the n operation objects and a similarity between the M objects.
  • the computing device takes the preference value of each of the n operation objects and the similarity between the M objects as the input of the formula (1) to calculate the target user for each of the n operation objects.
  • the preference value of the operands is:
  • Sim(u,a) g(SimRank(a 1 ,a),...,SimRank(a k ,a),...,SimRank(a n ,a),w(u,a 1 ),. ..,w(u,a k ),...,w(u,a n ))
  • Sim(u, a) represents the preference value of the target user u for the object a
  • SimRank(a k , a) represents the similarity between the operation object a k and the object a
  • w(u, a k ) represents the target user.
  • u is a preference value of the operation object a k
  • a k is represented as any one of n operation objects
  • n is the number of the operation objects, 0 ⁇ k ⁇ n
  • the function g is any aggregation function, for example, the aggregation function is Weighted average function and so on.
  • the formula (1) can also be expressed as:
  • the following example illustrates how the computing device predicts the target user's preference for the object Hailo based on the operating objects Uber, facebook, and google.
  • the target user has a preference value of 1 for Uber
  • the target user has a preference value of 0.8 for Facebook
  • the target user has a preference value of 0.2 for google.
  • the similarity between Uber and Hailo is 1, and the similarity between facebook and Hailo is 0.5.
  • the computing device collects the category distinguishing data of the N recommended objects, and classifies the N recommended objects according to the category distinguishing data of the N recommended objects, thereby obtaining X topics.
  • each of the X topics includes at least one of the N recommended objects, X is a positive integer greater than or equal to 1, N is a positive integer greater than or equal to 1, and N ⁇ M.
  • the N recommended objects are some or all of the M objects.
  • N of the M objects may be selected as the recommended objects at random, or N of the M objects may be selected as recommended according to certain conditions.
  • Object In a specific embodiment, among the M objects, the target user whose preference value is greater than the selection threshold is selected as the recommended object; or, according to the target user's preference value, the M objects are sorted from high to low, and Among the M objects, the object whose serial number is the top N is selected as the recommended object.
  • the selection threshold may be manually set according to requirements. When the selection threshold is set higher, the number of recommended objects is smaller, and vice versa, the number of recommended objects is larger.
  • the value of N can be a positive integer, and the value of N can also be set manually according to needs. When the value of N is larger, the number of recommended objects is larger, and vice versa, the number of recommended objects is smaller.
  • Topics can be artificially defined, for example, topics can be artificially defined as shopping, sports, finance, entertainment, most needed, most popular, up to date, and so on.
  • the theme includes at least a first level theme and a second level theme, wherein the first level theme is a superior theme of the second level theme, and each of the first level topics includes at least one second level theme.
  • the first level theme is "entertainment”
  • the second level theme is "video star” and "singer star” and so on.
  • the computing device performs clustering based on the category discrimination data of the N recommended objects, thereby obtaining X topics. Specifically, the computing device outputs the category discrimination data of the N recommended objects as the input of the clustering algorithm, and outputs the classification case of the N recommended objects.
  • the category distinguishing data of the recommended objects a 1 to a 8 is used as an input of the clustering algorithm, and the output is that the objects a 1 to a 3 are aggregated into the same category, and the objects a 4 and a 7 are aggregated to the same In one category, objects a 5 , a 6 , and a 8 are aggregated into the same category.
  • the computing device determines the subject matter of the category based on the tags of the recommended objects that are aggregated into the same category.
  • the labels of the objects a 5 , a 6 , and a 8 are financial, financial, and financial, respectively, and the subject of the category in which the objects a 5 , a 6 , and a 8 are located may be determined as financial.
  • the category distinguishing data of the recommended object may be a category and a label of the recommended object, and the like.
  • the clustering algorithm may be a K-means clustering algorithm, a hierarchical clustering algorithm, a K-MEDOIDS algorithm, a CLARANS algorithm, a BIRCH algorithm, a CURE algorithm, a CHAMELEON algorithm, and the like.
  • the computing device presets X topics and a category preset condition corresponding to each of the X topics, wherein the different topics correspond to different category preset conditions.
  • the computing device may preset the X topics as “most needed”, “most popular”, “latest”, etc., respectively, and the corresponding category preset condition of the theme “most needed” is the most frequently used and It is the object of the top N in which the target user has the highest preference value for the object; the corresponding pre-set condition of the "most popular” is that everyone prefers to download and is the top N of the target user's favorite value to the object.
  • the object in the "newest" can be defined as the object with the shortest shelf time and the top N in the target user's favorite value of the object.
  • the computing device determines whether the category distinguishing data of each of the N recommended objects meets the category preset condition corresponding to any one of the X topics, and matches the category corresponding to any one of the X topics.
  • the recommended objects of the condition are classified into the corresponding topics.
  • the category distinguishing data of the recommended object may be the number of times of using the recommended object, the number of times of downloading, the time of being downloaded, and the like.
  • the computing device calculates, according to the preference value of the recommended object included in each of the X topics, the target user's preference value for each of the X topics.
  • the favorite value of the recommended object included in each of the X topics is taken as the input of the formula (2), and is output as the favorite value of the target user for each of the X themes.
  • Score(u, t) represents the preference value of the target user u for the theme t
  • Sim(u, a j ) represents the preference value of the target user u for the recommended object a j
  • a j represents any one of the recommended objects in the topic t
  • m is the number of recommended objects in the subject t, 0 ⁇ j ⁇ m
  • the function F is any aggregate function.
  • the formula (2) can be expressed as:
  • Score(u,t) (Sim(u,a 1 )+...+Sim(u,a j )+...+Sim(u,a m ))/m.
  • the target user has a preference value of 0.8 for a 1
  • the target user has a preference value of 0.6 for a 2
  • the target user has a preference value of 0.4 for a 3 .
  • the computing device pushes the target topic to the target user.
  • the target theme is the subject of the X topics, and the target user's preference value is greater than the recommended threshold.
  • the recommended threshold can be set manually according to needs. When the recommended threshold is set higher, the number of recommended topics is smaller, and vice versa, the more the recommended topics are.
  • the computing device pushes the target topic to the target user in the manner shown in FIG.
  • the display table 500 shown in FIG. 5 includes a name 510 of the target theme, an icon 520 of the recommended object in the target theme, and a theme description 530 of the target theme.
  • the topic description 530 describes the characteristics of the target topic or the reason for recommending the target topic, and the like.
  • the name 510 of the target theme is placed above the display list 500 in a capitalized manner.
  • the icons 520 of the plurality of recommended objects in the target theme 510 are disposed in a side-by-side manner under the name 510 of the target theme.
  • the topic description 530 of the target topic 510 is set under the icons 520 of the plurality of recommended objects in a distinct font.
  • the computing device pushes the target topic to the target user in the manner of the tag cloud shown in FIG.
  • the tag cloud 600 shown in FIG. 6 includes a name 610 of the target topic and a tag 620 of the recommended object in the target topic.
  • the name 610 of the target topic is displayed in the tag cloud 600 in a capitalized manner, and the tag 620 of the recommended object in the target topic is distributedly displayed in the tag cloud 600 with a font size smaller than the name 610 of the target theme.
  • the display by the tag cloud 600 can facilitate simultaneous display of multiple themes, and the display mode is vivid, which can attract users' interest and generate considerable profits.
  • FIG. 7 is a schematic structural diagram of a theme recommendation apparatus according to an embodiment of the present invention.
  • the subject recommendation device 700 of the embodiment includes a prediction module 710, a classification module 720, a calculation module 730, and a recommendation module 740.
  • the prediction module 710 is configured to collect a historical operation behavior of the sample user for the M objects, and predict, according to the historical operation behavior of each of the M objects, the target user to each of the M objects The preference value of the object, where M is a positive integer greater than or equal to 1.
  • the classification module 720 is configured to collect category classification data of the N recommended objects, and classify the N recommended objects according to the category classification data of the N recommended objects, thereby obtaining X topics, where the X items are Each of the topics includes at least one of the N recommended objects, wherein the N recommended objects are part or all of the M objects, and X is a positive integer greater than or equal to 1. , N is a positive integer greater than or equal to 1, and N ⁇ M.
  • the calculating module 730 is configured to calculate a preference value of the target user for each of the X topics according to a preference value of the recommended object included in each of the X topics.
  • the recommendation module 740 is configured to push the target topic to the target user, wherein the target topic is the subject of the X topics whose preference value is greater than a recommendation threshold.
  • the historical operational behavior includes at least one of the following operational behaviors: browsing, clicking, downloading, using, paying, uninstalling, and evaluating.
  • the recommendation module 740 is configured to push the target topic to the target user in a tag cloud manner, where the tag cloud includes a name of the target topic and a label of the recommended object in the target topic.
  • the recommendation module 740 is configured to push the target topic to the target user in a manner of displaying a table, where the display table includes a name of the target topic and an icon of the recommended object in the target theme. .
  • the theme includes at least a first level theme and a second level theme, wherein the first level theme includes at least one second level theme.
  • FIG. 8 is a schematic structural diagram of another subject recommendation apparatus according to an embodiment of the present invention.
  • the theme recommendation apparatus of the embodiment is optimized for the prediction module 710 and the classification module 720 of the theme recommendation apparatus shown in FIG. And got it.
  • the prediction module 710 includes a determining unit 711, a training unit 713, and a prediction unit 715.
  • the classification module 720 includes a clustering unit 721.
  • the determining unit 711 is configured to determine preference data of the sample user for each of the M objects according to a historical operation behavior of the sample user for each of the M objects.
  • the training unit 713 is configured to perform a prediction model according to the feature data of the sample user, the preference data of the sample user for each of the M objects, the context feature, and the feature data of each of the M objects. Train.
  • the prediction unit 715 is configured to input feature data of the target user and feature data of each of the M objects into a trained prediction model to predict the target user to be in the M objects. The preference value of each object.
  • the clustering unit 721 is configured to perform clustering according to the category distinguishing data of the N recommended objects, thereby obtaining X topics.
  • FIG. 9 is a schematic structural diagram of still another subject recommendation apparatus according to an embodiment of the present invention.
  • the theme recommendation apparatus of this embodiment is optimized for the theme recommendation apparatus shown in FIG. 7.
  • the subject recommendation device also includes a selection module 750.
  • the selecting module 750 is configured to select, as the recommended object, an object whose preference value of the target user is greater than a selection threshold, or a high-to-low target of the M object according to a preference value of the target user. Sorting is performed, and an object whose serial number is the top N among the M objects is selected as a recommended object.
  • the prediction module 710 includes a construction unit 712, a calculation unit 714, a determination unit 716, and a prediction unit 718. among them,
  • the building unit 712 is configured to construct an interaction diagram between the sample user and the M objects according to historical operation behavior of the sample user for each of the M objects.
  • the calculating unit 714 is configured to calculate the similarity between the M objects according to the interaction diagram between the sample user and the M objects.
  • the determining unit 716 is configured to determine a preference value of the target user for each of the n operation objects according to a history operation behavior of each of the n operation objects by the target user, where the operation
  • the object is an object in which the target user has performed historical operation behavior among the M objects, and n is a positive integer greater than or equal to 1, n ⁇ M.
  • the prediction unit 718 is configured to predict, according to the preference value of each of the n operation objects by the target user, and the similarity between the M objects, the target user to the M objects The preference value of each object in .
  • the classification module 720 includes a preset unit 722 and a classification unit 724. among them,
  • the preset unit 722 is configured to preset the X presets and the classified preset conditions corresponding to each of the X themes, where the different themes correspond to different classified preset conditions.
  • the merging unit 724 is configured to determine whether the category distinguishing data of each of the N recommended objects meets the category preset condition corresponding to any one of the X topics, and the X subject is met.
  • the recommended object of the classification preset condition corresponding to any one of the topics belongs to the corresponding topic.
  • FIG. 10 is a schematic structural diagram of a computing device according to an embodiment of the present invention.
  • the computing device 1000 of the present implementation includes a processor 1010, a user interface 1020, an input and output unit, a communication unit 1060, and a memory 1070.
  • the processor 1010 is a computing center of the computing device 1000 and has powerful computing capabilities, and is capable of performing various operations at high speed.
  • the processor 1010 may be a central processing unit (CPU), a digital signal processing (DSP) device, a micro control unit (MCU), or a logic circuit for performing control or operations.
  • the processor 1010 may be a single core processor or a multi-core processor.
  • the User Interface (UI) 1020 is a medium for interaction and information exchange between the system and the user, and implements a conversion between an internal form of information and a human acceptable form.
  • the user interface 1020 can include three types: a command interface, a program interface, and a graphic interface.
  • the input and output unit may include a touch unit 1040 and other input devices 1050.
  • the touch unit 1040 is also referred to as a touch display or a touch pad, and can collect touch operations on or near the user (such as a user using a finger, a stylus, or the like, any suitable object or accessory on the touch-sensitive surface or touch sensitive Operation near the surface), and drive the corresponding connecting device according to a preset program.
  • the touch unit 1040 can include two parts: a touch detection device and a touch controller.
  • the touch detection device detects the touch orientation of the user, and detects a signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts the touch information into contact coordinates, and sends the touch information.
  • the processor 1010 is provided and can receive commands from the processor 1010 and execute them.
  • touch-sensitive surfaces can be implemented in a variety of types, including resistive, capacitive, infrared, and surface acoustic waves.
  • the input-output unit can also include other input devices 1050.
  • other input devices 1050 may include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control buttons, switch buttons, etc.), trackballs, mice, joysticks, and the like.
  • the communication unit 1060 can communicate by using wireless technologies such as cellular technology, Bluetooth technology, zibee technology, WLAN, etc., and the communication unit 1060 can also communicate by using wired technologies such as UART technology, print port technology, USB technology, and wired network transmission technology.
  • wireless technologies such as cellular technology, Bluetooth technology, zibee technology, WLAN, etc.
  • wired technologies such as UART technology, print port technology, USB technology, and wired network transmission technology.
  • the storage unit 1070 may include an internal memory as well as an external memory.
  • the internal memory is used to store instructions and data currently running the program and to exchange information directly with the processor 1010, which is the primary source of processing data for the processor 1010.
  • the internal memory can include read only memory as well as random access memory.
  • the external memory has large storage capacity and low price, but the storage speed is slow. It is used to store a large number of programs, data and intermediate results that are temporarily unused, and can exchange information with the internal memory in batches when needed. External memory can only exchange information with internal memory and cannot be accessed directly by other components of the computer system. Commonly used external memories are disks, tapes, discs, and the like.
  • the processor 1010 executes the one or more programs that include instructions for performing the following operations:
  • M is a positive integer greater than or equal to 1;
  • each of the X topics includes at least one of the N recommended objects, wherein the N recommended objects are part or all of the M objects, X is a positive integer greater than or equal to 1, and N is greater than or a positive integer equal to 1, and, N ⁇ M;
  • the processor 1010 is further configured to determine preference data of the sample user for each of the M objects according to historical operation behavior of each of the M objects by the sample user; a prediction model according to the feature data of the sample user, the preference data of the sample user for each of the M objects, the context feature, and the feature data of each of the M objects Performing training; inputting feature data of the target user and feature data of each of the M objects to a trained prediction model to predict the target user to each of the M objects Favorite value.
  • the processor 1010 is further configured to construct an interaction diagram between the sample user and the M objects according to historical operation behavior of each of the M objects by the sample user;
  • An interaction diagram between the sample user and the M objects calculates a similarity between the M objects; determining, according to the historical operation behavior of each of the n operation objects by the target user a preference value of the target user for each of the n operation objects, wherein the operation object is an object of the M objects, the target user has performed a historical operation behavior, and n is greater than or equal to 1 a positive integer, n ⁇ M; predicting the target user to the said user based on the preference value of each of the n operational objects and the similarity between the M objects The preference value of each of the M objects.
  • the historical operational behavior includes at least one of the following operational behaviors: browsing, clicking, downloading, using, paying, uninstalling, and evaluating.
  • the processor 1010 is further configured to perform clustering according to the category distinguishing data of the N recommended objects, thereby obtaining X topics.
  • the processor 1010 is further configured to pre-set the X preset topics and the category preset conditions corresponding to each of the X topics, where different topics correspond to different classification preset conditions; Determining whether the category distinguishing data of each of the N recommended objects meets the category preset condition corresponding to any one of the X topics, and matching the category corresponding to any one of the X topics The recommended objects of the preset conditions are classified into the corresponding topics.
  • the processor 1010 is further configured to: push the target topic to the target user in a manner of displaying a table, where the display table includes a name of the target topic and a recommended object in the target topic. icon.
  • the processor 1010 is further configured to: push the target topic to the tag cloud a target user, wherein the tag cloud includes a name of the target topic and a tag of the recommended object in the target topic.
  • the theme includes at least a first level theme and a second level theme, wherein the first level theme includes at least one second level theme.
  • the processor 1010 is further configured to: select, among the M objects, an object whose preference value is greater than a selection threshold as a recommended object; or select the M object according to a preference value of the target user.
  • the sorting is performed from high to low, and the object whose serial number is the top N among the M objects is selected as the recommended object.
  • the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Databases & Information Systems (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种主题推荐方法以及装置。其中，所述方法包括：采集样本用户对M个对象的历史操作行为，并根据样本用户对M个对象中的每个对象的历史操作行为预测目标用户对M个对象中的每个对象的喜好值（110）；采集N个推荐对象的类别区分数据，并根据N个推荐对象的类别区分数据将N个推荐对象进行类别区分，从而得到X个主题（120），X个主题中的每个主题均包括N个推荐对象中的至少一个推荐对象，N个推荐对象为M个对象中的部分或者全部；根据目标用户对X个主题中的每个主题中包括的推荐对象的喜好值计算得到目标用户对X个主题中的每个主题的喜好值（130）；将目标主题推送给目标用户（140），目标主题为X个主题中的，目标用户的喜好值大于推荐阈值的主题。

Description

主题推荐方法以及装置 技术领域
本发明涉及终端领域，尤其涉及一种主题推荐方法以及装置。
背景技术
随着应用市场的发展，应用供应商之间的竞争也越来越激烈。在国内，华为应用市场、百度应用市场、360应用市场和豌豆荚应用市场等犹如雨后春笋般出现；在国外，App Store，Google Play等占据了非大陆用户的应用市场的巨大份额。为了抢占应用市场，应用供应商通常通过各种方法将自己的应用推荐给用户，以吸引用户使用自己的应用。
但是，在实践中发现，现有的推荐方式只是简单地估计用户可能感兴趣的几个应用并推荐给用户，难以真正引起用户的兴趣并带来收益。
发明内容
本发明提供了一种主题推荐方法以及装置，能够以主题的形式将推荐对象推荐给用户。
第一方面，本发明提供了一种主题推荐方法。计算设备采集样本用户对M个对象的历史操作行为，并根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值。M为大于或者等于1的正整数。计算设备还采集N个推荐对象的类别区分数据，并根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题。所述N个推荐对象为所述M个对象中的部分或者全部。所述X个主题中的每个主题均包括所述N个推荐对象中的至少一个推荐对象。 X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M。计算设备根据所述目标用户对所述X个主题中的每个主题中包括的推荐对象的喜好值计算得到所述目标用户对所述X个主题中的每个主题的喜好值，并将目标主题推送给所述目标用户。所述目标主题为所述X个主题中的，所述目标用户的喜好值大于推荐阈值的主题。
不同于现有技术只是简单地估计用户可能感兴趣的几个应用并推荐给用户，本发明提出的主题推荐方法先将推荐对象分类成多个主题，然后，将对主题的喜好值大于阈值的主题推送给目标用户。上述方法能够将目标用户感兴趣的主题的多个推荐对象进行推送，使得目标用户在同一主题中可以有多种选择。而且，可以向目标用户推送不止一个主题，使得目标用户能够在多个主题中进行选择，从而真正引起目标用户的兴趣并带来收益。
结合第一方面，第一方面的第一种可能的实施方式中，计算设备根据所述样本用户对所述M个对象中的每个对象的历史操作行为确定所述样本用户对所述M个对象中的每个对象的喜好数据。计算设备再根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型进行训练。最后，计算设备将所述目标用户的特征数据和所述M个对象中的每个对象的特征数据输入到训练好的预测模型以预测所述目标用户对所述M个对象中的每个对象的喜好值。
结合第一方面，第一方面的第二种可能的实施方式中，计算设备根据所述样本用户对所述M个对象中的每个对象的历史操作行为构建所述样本用户和所述M个对象之间的交互图，再根据所述样本用户和所述M个对象之间的交互图计算所述M个对象两两之间的相似性。然后，根据所述目标用户对n个操作对象中的每个操作对象的历史操作行为确定所述目标用户对所述n个操 作对象中的每个操作对象的喜好值。所述操作对象为所述M个对象中，所述目标用户曾经进行了历史操作行为的对象。n为大于或者等于1的正整数，n≤M。最后，计算设备根据所述目标用户对所述n个操作对象中的每个操作对象的喜好值以及所述M个对象两两之间的相似性预测所述目标用户对所述M个对象中的每个对象的喜好值。
结合第一方面任意一种实施方式，第一方面的第四种可能的实施方式中，计算设备根据所述N个推荐对象的类别区分数据进行聚类，从而得到X个主题。因此，计算设备能够将相类似的推荐对象自动划分到同一个主题下，不需要人工进行干预。
结合第一方面任意一种实施方式，第一方面的第五种可能的实施方式中，计算设备预先设置所述X个主题以及所述X个主题中的每个主题对应的分类预设条件。不同的主题对应不同的分类预设条件。计算设备判断所述N个推荐对象中的每个推荐对象的类别区分数据是否符合所述X个主题中的任意一个主题对应的分类预设条件，并将符合所述X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。因此，计算设备能够按照需要自主设定主题的分类预设条件以筛选出自定义的主题。
第二方面，本发明提供了一种主题推荐装置，所述装置包括预测模块、分类模块、计算模块以及推荐模块。所述预测模块用于采集样本用户对M个对象的历史操作行为，并根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值。M为大于或者等于1的正整数。所述分类模块用于采集N个推荐对象的类别区分数据，并根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题。所述N个推荐对象为所述M个对象中的部分或者全部。所述X个主题中的每个主题均包括所述N个推荐对象中的至少一个推荐对象。 X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M。所述计算模块用于根据所述目标用户对所述X个主题中的每个主题中包括的推荐对象的喜好值计算得到所述目标用户对所述X个主题中的每个主题的喜好值。并且，所述推荐模块用于将目标主题推送给所述目标用户。所述目标主题为所述X个主题中的，所述目标用户的喜好值大于推荐阈值的主题。
不同于现有技术只是简单地估计用户可能感兴趣的几个应用并推荐给用户，本发明提出的主题推荐方法先将推荐对象分类成多个主题，然后，将对主题的喜好值大于阈值的主题推送给目标用户。上述方法能够将目标用户感兴趣的主题的多个推荐对象进行推送，使得目标用户在同一主题中可以有多种选择。而且，可以向目标用户推送不止一个主题，使得目标用户能够在多个主题中进行选择，从而真正引起目标用户的兴趣并带来收益。
结合第二方面，第二方面的第一种可能的实施方式中，所述预测模块包括确定单元、训练单元以及预测单元。所述确定单元用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为确定所述样本用户对所述M个对象中的每个对象的喜好数据。所述训练单元用于根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型进行训练。最后，所述预测单元用于将所述目标用户的特征数据和所述M个对象中的每个对象的特征数据输入到训练好的预测模型以预测所述目标用户对所述M个对象中的每个对象的喜好值。
结合第二方面，第二方面的第二种可能的实施方式中，所述预测模块包括构建单元、计算单元、确定单元以及预测单元。所述构建单元用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为构建所述样本用户和所述M个对象之间的交互图。所述计算单元用于根据所述样本用户和所述M个对象之间的交互图计算所述M个对象两两之间的相似性。然后，所述确定单 元用于根据所述目标用户对n个操作对象中的每个操作对象的历史操作行为确定所述目标用户对所述n个操作对象中的每个操作对象的喜好值。所述操作对象为所述M个对象中，所述目标用户曾经进行了历史操作行为的对象。n为大于或者等于1的正整数，n≤M。最后，所述预测单元用于根据所述目标用户对所述n个操作对象中的每个操作对象的喜好值以及所述M个对象两两之间的相似性预测所述目标用户对所述M个对象中的每个对象的喜好值。
结合第二方面任意一种实施方式，第二方面的第四种可能的实施方式中，所述分类模块包括聚类单元。所述聚类单元用于根据所述N个推荐对象的类别区分数据进行聚类，从而得到X个主题。因此，计算设备能够将相类似的推荐对象自动划分到同一个主题下，不需要人工进行干预。
结合第二方面任意一种实施方式，第二方面的第五种可能的实施方式中，所述分类模块包括预设单元以及归入单元。所述预设单元用于预先设置所述X个主题以及所述X个主题中的每个主题对应的分类预设条件。不同的主题对应不同的分类预设条件。然后，所述归入单元用于判断所述N个推荐对象中的每个推荐对象的类别区分数据是否符合所述X个主题中的任意一个主题对应的分类预设条件，并将符合所述X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。因此，计算设备能够按照需要自主设定主题的分类预设条件以筛选出自定义的主题。
第三方面，本发明提供了一种计算设备。所述计算设备包括存储单元、通信接口及与所述存储单元和通信接口耦合的处理器。其中，所述存储单元用于存储指令，所述处理器用于执行所述指令，所述通信接口用于与目标用户传输数据。当所述处理器在执行所述指令时，可根据所述指令执行在第一方面中的方法。
第四方面，本发明提供了一种计算机可读存储介质，所述计算机可读存储 介质存储了计算设备所执行的用于主题推荐的程序代码。所述程序代码包括用于执行在第一方面中的方法的指令。
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简单地介绍，显而易见地，下面描述中的附图仅仅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是本发明实施例提供的一种主题推荐方法的流程图；
图2是图1所示的主题推荐方法中的步骤110的一种具体步骤的流程图；
图3是图1所示的主题推荐方法中的步骤110的另一种具体步骤的流程图；
图4是本发明实施例提供的一种主题推荐方法中的样本用户和对象之间的交互图；
图5是本发明实施例提供的一种主题推荐方法中的展示表的示意图；
图6是本发明实施例提供的一种主题推荐方法中的标签云的示意图；
图7是本发明实施例提供的一种主题推荐装置的结构示意图；
图8是本发明实施例提供的另一种主题推荐装置的结构示意图；
图9是本发明实施例提供的又一种主题推荐装置的结构示意图；
图10是本发明实施提供的一种计算设备的结构示意图。
具体实施方式
下面结合附图和实施例进行说明。
计算设备可以是台式电脑(computer)、笔记本电脑(notebook)、平板电 脑(Tablet Personal Computer)、智能手机(Smart phone)等等，计算设备还可以是具有CPU(中央处理单元，Central Processing Unit)、MCU(微控制单元，Microcontroller Unit)、DSP(数字信号处理器，Digital Signal Processor)或用于执行控制或运算的逻辑电路等处理器的设备。
请参阅图1，图1是本发明实施例提供的一种主题推荐方法的流程图。本实施的主题推荐方法的执行主体为计算设备，本实施的主题推荐方法包括如下步骤：
110：计算设备采集样本用户对M个对象的历史操作行为，并根据样本用户对M个对象中的每个对象的历史操作行为预测目标用户对M个对象中的每个对象的喜好值。
对象(Item)可以是应用，例如，应用可以是Uber、facebook、Myspace、twitter以及MSN(Microsoft Service Network)等等，此外，对象还可以是文章、视频、歌曲等等，本发明不作具体限定。样本用户是曾经对对象有历史操作行为并被选取作为样本的用户。其中，样本用户对对象的历史操作行为包括但不限于浏览、点击、下载、使用、付费、卸载和评价。样本用户对对象的历史操作行为可以包括目标用户对对象的历史操作行为，也可以不包括目标用户对对象的历史操作行为。
在一具体的实施例中，如图2所示，根据样本用户对M个对象中的每个对象的历史操作行为预测目标用户对M个对象中的每个对象的喜好值具体包括如下步骤：
111：计算设备根据样本用户对M个对象中的每个对象的历史操作行为确定样本用户对M个对象中的每个对象的喜好数据。其中，M为大于或者等于1的正整数。
喜好数据可以是区间[0,1]之间的一个数值，当数值为1时，表示样本用户对对象的喜好程度最高，当数值为0时，表示样本用户对对象的喜好程度最低。
下面举例说明计算设备如何根据样本用户对M个对象中的每个对象的历史操作行为确定样本用户对M个对象中的每个对象的喜好数据。例如，A对象为M个对象中的其中一个对象，计算设备可以预先设定样本用户对A对象的历史操作行为与样本用户对A对象的喜好数据之间的对应关系，然后，根据样本用户对A对象的历史操作行为和上述对应关系确定样本用户对A对象的喜好数据。预先设定样本用户对A对象的历史操作行为与样本用户对A对象的喜好数据之间的对应关系为：如果样本用户对A对象进行了付费，则样本用户对A对象的喜好数据为1，如果样本用户对A对象进行了下载，则样本用户对A对象的喜好数据为0.8，如果样本用户对A对象进行了浏览，则样本用户对A对象的喜好数据为0.2。所以，一旦样本用户对A对象进行了付费，就可以确定样本用户对A对象的喜好数据为1。或者，一旦样本用户同时对A对象进行了浏览、下载和付费，则可以确定样本用户对A对象的喜好数据为最高的一个，或者，确定样本用户对A对象的喜好数据为三者之和等等。
113：计算设备根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型进行训练。
样本用户的特征数据是能够反映样本用户的特点的数据，例如，样本用户的特征数据可以包括但不限于样本用户的性别、年龄以及居住地等等。简单地，样本用户的性别可以用“0”和“1”进行表示，“0”表示为女性，“1”表示为男性。样本用户的年龄可以直接用样本用户的年龄来表示，样本用户的居住地可以用代码进行表示，例如，“1”表示纽约，“2”表示华盛顿等等。
上下文特征是能够反应环境特点的数据，例如，天气等等。
对象的特征数据是能够反应样本用户的特点的数据，例如，对象的特征数据可以包括但不限于对象的类别(购物类或者游戏类等等)等等。简单地，对象的类别可以用代码进行表示，例如，“1”表示购物类，“2”表示游戏类等等。
预测模型可以是逻辑回归(Logistic Regression)模型、决策树和矩阵分解等等。
训练时，将样本用户的特征数据、样本用户对M个对象中的每个对象的喜好数据、上下文特征和M个对象中的每个对象的特征数据对预测模型进行训练。例如，样本用户包括样本用户A1，样本用户A2，……；M个对象包括：对象a1，对象a2，……。将样本用户A1的特征数据，上下文特征，对象a1的特征数据以及样本用户A1对对象a1的喜好数据作为其中一条训练数据对预测模型进行训练；将样本用户A1的特征数据，上下文特征，对象a2的特征数据以及样本用户A1对对象a2的喜好数据作为一条训练数据对预测模型进行训练；将样本用户A2的特征数据，上下文特征，对象a2的特征数据以及样本用户A2对对象a2的喜好数据作为一条训练数据对预测模型进行训练……。经过大量数据进行训练后，预测模型学习到了具有某些特征数据的用户对具有某些特征数据的对象的喜好值的先验知识。
115：计算设备将目标用户的特征数据和M个对象中的每个对象的特征数据输入到训练好的预测模型以预测目标用户对M个对象中的每个对象的喜好值。
喜好值可以是区间[0,1]之间的一个数值，当数值为1时，表示目标用户对对象的喜好程度最高，当数值为0时，表示目标用户对对象的喜好程度最低。
对于训练好的预测模型，模型中已经具有某些特征数据的用户对具有某些特征数据的对象的喜好值的先验知识，所以，当输入目标用户的特征数据和M个对象中的每个对象的特征数据后，就可以预测到目标用户对M个对象中的每个对象的喜好值。举例说明，目标用户为目标用户u，M个对象包括对象a1，对象a2……。当向预测模型输入目标用户u的特征数据和对象a1的特征数据后，预测模型就可以预测出目标用户u对对象a1的喜好值；当向预测模型输入目标用户u的特征数据和对象a2的特征数据后，预测模型就可以预测出目标用户u 对对象a2的喜好值……。
在一具体的实施例中，如图3所示，根据样本用户对M个对象中的每个对象的历史操作行为预测目标用户对M个对象中的每个对象的喜好值具体包括如下步骤：
112：计算设备根据样本用户对M个对象中的每个对象的历史操作行为构建样本用户和M个对象之间的交互图。
例如，样本用户和M个对象之间的交互图可以根据如下方式进行构建，如图4所示，图中的一个黑点表示为一个样本用户，图中的一个白点表示为M个对象中的一个对象，图中连接黑点和白点之间的一条边表示为该黑点对应的样本用户对该白点对应的对象具有历史操作行为。此外，还可以根据该黑点对应的样本用户对该白点对应的对象具有历史操作行为的不同，人为为这条边赋予不同的权值。例如，当历史操作行为为付费时，赋予权值为1；当历史操作行为为下载时，赋予权值为0.8，当历史操作行为为浏览时，赋予权值为0.2等等。
114：计算设备根据样本用户和M个对象之间的交互图计算M个对象两两之间的相似性。
在一具体的实施方式中，计算设备可以根据SimRank算法计算出M个对象两两之间的相似性。其中，SimRank算法的特点为：当输入为交互图(交互图由多个点和两点之间的边构成)时，输出为交互图中所有点两两之间的相似性。SimRank算法是一种成熟的算法，具体的算法请参阅有关资料。计算设备将样本用户和M个对象之间的交互图作为SimRank算法的输入时，输出为交互图中的所有点两两之间的相似性，即，包括样本用户两两之间的相似性，样本用户和M个对象两两之间的相似性以及M个对象两两之间的相似性。计算设备从输出中选择出M个对象两两之间的相似性。其中，M个对象两两之间的相似性可以是区间[0,1]之间的一个数值，当数值为1时，表示两个对象相似 度最高，当数值为0时，表示两个对象的相似度最低。
可以理解的是，计算M个对象两两之间的相似性的算法可以有多种，例如，PPR(personal page rank)算法等等，此处SimRank算法只是一种举例而非限定。
116：计算设备根据目标用户对n个操作对象中的每个操作对象的历史操作行为确定目标用户对n个操作对象中的每个操作对象的喜好值。其中，操作对象为M个对象中，目标用户曾经进行了历史操作行为的对象，n为大于或者等于1的正整数，n≤M。
计算设备可以预先设置目标用户对操作对象的历史操作行为和目标用户对操作对象的喜好值之间的关系，例如，目标用户为目标用户u，操作对象a是n个操作对象中的一个操作对象，如果目标用户u曾经对操作对象a进行付费，则目标用户u对操作对象a的喜好值为1；如果目标用户u曾经对操作对象a进行下载，则目标用户u对操作对象a的喜好值为0.8，如果目标用户u曾经对操作对象a进行浏览，则目标用户u对操作对象a的喜好值为0.2等等。然后，计算设备根据目标用户对n个操作对象中的每个操作对象的历史操作行为确定目标用户对n个操作对象中的每个操作对象的喜好值。举例来说，当n个操作对象包括Uber、facebook和google时，目标用户对n个操作对象中的每个操作对象的喜好值可以确定为：如果目标用户曾经对Uber进行付费，则目标用户对Uber的喜好值确定为1；如果目标用户曾经对facebook进行下载，则目标用户对facebook的喜好值确定为0.8，如果目标用户曾经对google进行浏览，则目标用户对google的喜好值确定为0.2。上述仅以目标用户对n个操作对象的历史操作行为不同时，则目标用户对n个操作对象中的每个操作对象的喜好值不同为例，但是，在实际应用中，当浏览次数、付费金额、使用频率等不同时，也会导致目标用户对n个操作对象中的每个操作对象的喜好值不同。
118：计算设备根据目标用户对n个操作对象中的每个操作对象的喜好值以及M个对象两两之间的相似性预测目标用户对M个对象中的每个对象的喜好值。
计算设备将目标用户对n个操作对象中的每个操作对象的喜好值以及M个对象两两之间的相似性作为公式(1)的输入，以计算目标用户对n个操作对象中的每个操作对象的喜好值。其中，公式(1)为：
Sim(u,a)＝g(SimRank(a1,a),...,SimRank(ak,a),...,SimRank(an,a),w(u,a1),...,w(u,ak),...,w(u,an))
                                                   公式(1)
其中，Sim(u,a)表示目标用户u对对象a的喜好值，SimRank(ak,a)表示操作对象ak与对象a之间的相似性，w(u,ak)表示目标用户u对操作对象ak的喜好值，ak表示为n个操作对象中的任意一个，n为所述操作对象的数量，0<k≤n，函数g为任意一个聚合函数，例如聚合函数为加权平均函数等等。当函数g为加权平均函数时，公式(1)也可以表示为：
Figure PCTCN2016105167-appb-000001
下面举例进行说明计算设备是如何根据操作对象Uber、facebook以及google预测目标用户对对象Hailo的喜好值的。假设目标用户对Uber的喜好值为1，目标用户对facebook的喜好值为0.8，目标用户对google的喜好值为0.2，而，Uber和Hailo的相似性为1，facebook和Hailo的相似性为0.5，google和Hailo的相似性为0.2，则根据加权平均函数计算可知：目标用户对对象Hailo的喜好值＝(1*1+0.8*0.5+0.2*0.2)/3≈0.85。
120：计算设备采集N个推荐对象的类别区分数据，并根据N个推荐对象的类别区分数据将N个推荐对象进行类别区分，从而得到X个主题。其中，X个主题中的每个主题均包括N个推荐对象中的至少一个推荐对象，X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M。
N个推荐对象为M个对象中的部分或者全部。当N个推荐对象为M个对象中的部分时，可以随意从M个对象中选择其中的N个对象作为推荐对象，也可以根据一定的条件从M个对象中选择其中的N个对象作为推荐对象。在一具体的实施例中，将M个对象中，目标用户的喜好值大于选择阈值的对象选择为推荐对象；或者，按照目标用户的喜好值对M个对象从高至低进行排序，并将M个对象中序号为前N的对象选择为推荐对象。其中，选择阈值可以根据需要人为进行设置，当选择阈值设置得越高，则推荐对象的数量越少，反之，则推荐对象的数量越多。N的数值可以是正整数，并且N的数值同样可以根据需要人为进行设置，当N的数值越大时，则推荐对象的数量越多，反之，则推荐对象的数量越少。
主题可以是人为定义的，例如：主题可以人为定义为购物、体育、金融、娱乐、最需要的，最流行的、最新的等等。在一具体的实施例中，主题至少包括第一级主题以及第二级主题，其中，第一级主题为第二级主题的上级主题，每个第一级主题包括至少一个第二级主题。例如，第一级主题为“娱乐”，第二级主题为“影视明星”以及“歌手明星”等等。
在一具体的实施例中，计算设备根据N个推荐对象的类别区分数据进行聚类，从而得到X个主题。具体地，计算设备将N个推荐对象的类别区分数据作为聚类算法的输入，则输出为N个推荐对象的分类情况。举例进行说明，将推荐对象a1至a8的类别区分数据作为聚类算法的输入，则输出为对象a1至a3被聚合到同一个类别中，对象a4和a7被聚合到同一个类别中，对象a5、a6和a8被聚合到同一个类别中。计算设备根据被聚合到同一类别的推荐对象的标签确定该类别的主题。例如，对象a5、a6和a8的标签分别为金融、金融和财经，则对象a5、a6和a8所在类别的主题可以确定为金融。其中，推荐对象的类别区分数据可以为推荐对象的类别和标签等等。聚类算法可以是K-means聚类算法、hierarchical聚类算法、K-MEDOIDS算法、CLARANS算法、BIRCH 算法、CURE算法以及CHAMELEON算法等等。
在一具体的实施例中，计算设备预先设置X个主题以及X个主题中的每个主题对应的分类预设条件，其中，不同的主题对应不同的分类预设条件。例如，计算设备可以预先设置X个主题分别为“最需要的”、“最流行的”、“最新的”等等，则主题“最需要的”对应的分类预设条件为使用最频繁的且是在目标用户对对象的喜好值最高的前N个中的对象；“最流行的”对应的分类预设条件为大家都比较喜欢下载且是在目标用户对对象的喜好值最高的前N个中的对象；“最新的”可以定义为上架时间较短的且是在目标用户对对象的喜好值最高的前N个中的对象。计算设备判断N个推荐对象中的每个推荐对象的类别区分数据是否符合X个主题中的任意一个主题对应的分类预设条件，并将符合X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。其中，推荐对象的类别区分数据可以为推荐对象的使用次数、被下载次数和上架时间等等。
130：计算设备根据目标用户对X个主题中的每个主题中包括的推荐对象的喜好值计算得到目标用户对X个主题中的每个主题的喜好值。
将目标用户对X个主题中的每个主题中包括的推荐对象的喜好值作为公式(2)的输入，则输出为目标用户对X个主题中的每个主题中的喜好值。
Score(u,t)＝F(Sim(u,a1),...,Sim(u,aj),...,Sim(u,am))      公式(2)
其中，Score(u,t)表示目标用户u对主题t的喜好值，Sim(u,aj)表示目标用户u对推荐对象aj的喜好值，aj表示主题t中的任意一个推荐对象，m为表示主题t中的推荐对象的数量，0<j≤m，函数F为任意一个聚合函数，例如聚合函数为平均函数时，则公式(2)可以表示为：
Score(u,t)＝(Sim(u,a1)+...+Sim(u,aj)+...+Sim(u,am))/m。
举例进行说明，如果主题t中包括推荐对象a1至a3，目标用户对a1的喜好值为0.8，目标用户对a2的喜好值为0.6，目标用户对a3的喜好值为0.4，则 目标用户对主题t的喜好值＝(0.8+0.6+0.4)/3＝0.6。
140：计算设备将目标主题推送给目标用户。其中，目标主题为X个主题中的，目标用户的喜好值大于推荐阈值的主题。
推荐阈值可以根据需要人为进行设置，当推荐阈值设置得越高，则推荐的主题的数量越少，反之，则推荐的主题的数量越多。
在一具体的实施例中，计算设备将目标主题以如图5所示的展示表的方式推送给所述目标用户。其中，如图5所示的展示表500包括目标主题的名称510、目标主题中的推荐对象的图标520以及目标主题的主题描述530。其中，主题描述530对目标主题的特点或者推荐目标主题的理由等等进行描述。目标主题的名称510以大写居中的方式设置在展示表500的上方。目标主题510中的多个推荐对象的图标520以并列的方式设置在目标主题的名称510之下。目标主题510的主题描述530以明显的字体设置在多个推荐对象的图标520之下。通过展示表这种方式进行显示，能够清楚地表示出目标主题510的名称、目标主题510包括哪些推荐对象以及目标主题的主题描述530，吸引用户的兴趣并产生可观的收益。
在一具体的实施例中，计算设备将目标主题以图6所示的标签云的方式推送给目标用户。图6所示的标签云600包括目标主题的名称610以及目标主题中的推荐对象的标签620。在标签云600中，目标主题的名称610以大写的方式显示在标签云600中，目标主题中的推荐对象的标签620以比目标主题的名称610小的字号分布式显示在标签云600中。通过标签云600这种方式进行显示，能够便于对多个主题同时进行显示，而且，展示方式生动，能够吸引用户的兴趣并产生可观的收益。
请参阅图7，图7是本发明实施例提供的一种主题推荐装置的结构示意图。本实施例的主题推荐装置700包括：预测模块710、分类模块720、计算模块730以及推荐模块740。
预测模块710用于采集样本用户对M个对象的历史操作行为，并根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值，其中，M为大于或者等于1的正整数。
分类模块720用于采集N个推荐对象的类别区分数据，并根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题，其中，所述X个主题中的每个主题均包括所述N个推荐对象中的至少一个推荐对象，其中，所述N个推荐对象为所述M个对象中的部分或者全部，X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M。
计算模块730用于根据所述目标用户对所述X个主题中的每个主题中包括的推荐对象的喜好值计算得到所述目标用户对所述X个主题中的每个主题的喜好值。
推荐模块740用于将目标主题推送给所述目标用户，其中，所述目标主题为所述X个主题中的，所述目标用户的喜好值大于推荐阈值的主题。
可选地，所述历史操作行为包括以下至少一种操作行为：浏览、点击、下载、使用、付费、卸载和评价。
可选地，推荐模块740用于将所述目标主题以标签云的方式推送给所述目标用户，其中，所述标签云包括所述目标主题的名称以及所述目标主题中的推荐对象的标签
可选地，推荐模块740用于将所述目标主题以展示表的方式推送给所述目标用户，其中，所述展示表包括所述目标主题的名称以及所述目标主题中的推荐对象的图标。
可选地，所述主题至少包括第一级主题以及第二级主题，其中，所述第一级主题包括至少一个第二级主题。
可理解的是，主题推荐装置的具体执行步骤还可以参考前述方法实施例的内容，此处不再赘述。
请参阅图8，图8是本发明实施例提供的另一种主题推荐装置的结构示意图.本实施例的主题推荐装置是对图7所示的主题推荐装置的预测模块710、分类模块720优化而得到的。其中，所述预测模块710包括确定单元711、训练单元713以及预测单元715。所述分类模块720包括聚类单元721。
所述确定单元711用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为确定所述样本用户对所述M个对象中的每个对象的喜好数据。
所述训练单元713用于根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型进行训练。
所述预测单元715用于将所述目标用户的特征数据和所述M个对象中的每个对象的特征数据输入到训练好的预测模型以预测所述目标用户对所述M个对象中的每个对象的喜好值。
所述聚类单元721用于根据所述N个推荐对象的类别区分数据进行聚类，从而得到X个主题。
可理解的是，主题推荐装置的具体执行步骤还可以参考前述方法实施例的内容，此处不再赘述。
请参阅图9，图9是本发明实施例提供的又一种主题推荐装置的结构示意图.本实施例的主题推荐装置是对图7所示的主题推荐装置优化而得到的。
所述主题推荐装置还包括选择模块750。所述选择模块750用于将所述M个对象中，所述目标用户的喜好值大于选择阈值的对象选择为推荐对象；或者，按照目标用户的喜好值对所述M个对象从高至低进行排序，并将所述M个对象中序号为前N的对象选择为推荐对象。
所述预测模块710包括构建单元712、计算单元714、确定单元716以及预测单元718。其中，
构建单元712用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为构建所述样本用户和所述M个对象之间的交互图。
计算单元714用于根据所述样本用户和所述M个对象之间的交互图计算所述M个对象两两之间的相似性。
确定单元716用于根据所述目标用户对n个操作对象中的每个操作对象的历史操作行为确定所述目标用户对所述n个操作对象中的每个操作对象的喜好值，所述操作对象为所述M个对象中，所述目标用户曾经进行了历史操作行为的对象，n为大于或者等于1的正整数，n≤M。
预测单元718用于根据所述目标用户对所述n个操作对象中的每个操作对象的喜好值以及所述M个对象两两之间的相似性预测所述目标用户对所述M个对象中的每个对象的喜好值。
所述分类模块720包括预设单元722以及归入单元724。其中，
预设单元722用于预先设置所述X个主题以及所述X个主题中的每个主题对应的分类预设条件，其中，不同的主题对应不同的分类预设条件。
归入单元724用于判断所述N个推荐对象中的每个推荐对象的类别区分数据是否符合所述X个主题中的任意一个主题对应的分类预设条件，并将符合所述X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。
可理解的是，主题推荐装置的具体执行步骤还可以参考前述方法实施例的内容，此处不再赘述。
请参阅图10，图10是本发明实施例提供的一种计算设备的结构示意图。 本实施的计算设备1000包括：处理器1010、用户接口1020、输入输出单元、通讯单元1060以及存储器1070。
处理器1010是计算设备1000的运算中心，具有强大的运算能力，能够高速地进行各种运算。处理器1010可以是中央处理单元(Central Processing Unit，CPU)、数字信号处理(Digital Signal Process，DSP)器，微控制单元(MCU，Microcontroller Unit)或用于执行控制或运算的逻辑电路等。处理器1010可以是单核处理器，也可以是多核处理器。
用户接口(User Interface，UI)1020是系统和用户之间进行交互和信息交换的媒介，它实现信息的内部形式与人类可以接受形式之间的转换。用户接口1020可以包括命令接口、程序接口、图形接口三种。
输入输出单元可包括触控单元1040以及其他输入设备1050。触控单元1040也称为触摸显示屏或者触控板，可收集用户在其上或附近的触摸操作(比如用户使用手指、触笔等任何适合的物体或附件在触敏表面上或在触敏表面附近的操作)，并根据预先设定的程式驱动相应的连接装置。可选的，触控单元1040可包括触摸检测装置和触摸控制器两个部分。其中，触摸检测装置检测用户的触摸方位，并检测触摸操作带来的信号，将信号传送给触摸控制器；触摸控制器从触摸检测装置上接收触摸信息，并将它转换成触点坐标，再送给处理器1010，并能接收处理器1010发来的命令并加以执行。此外，可以采用电阻式、电容式、红外线以及表面声波等多种类型实现触敏表面。除了触敏表面，输入输出单元还可以包括其他输入设备1050。具体地，其他输入设备1050可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆等中的一种或多种。
通讯单元1060可以采用蜂窝技术、蓝牙技术、zibee技术、WLAN等无线技术进行通信，通讯单元1060还可以采用UART技术、打印口技术、USB技术和有线网络传输技术等等有线技术进行通信。
存储单元1070可以包括内部存储器以及外部存储器。内部存储器用来存放当前运行程序的指令和数据，并直接与处理器1010交换信息，是处理器1010处理数据的主要来源。内部存储器可以包括只读存储器以及随机存储器。外部存储器储存容量大，价格低，但储存速度慢，用于存放大量暂时不用的程序，数据和中间结果，需要时，可成批的与内部存储器进行信息交换。外部存储器只能与内部存储器交换信息，不能被计算机系统的其他部件直接访问。常用的外部存储器有磁盘，磁带，光盘等。
处理器1010执行述一个或者一个以上程序包含用于进行以下操作的指令：
采集样本用户对M个对象的历史操作行为，并根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值，其中，M为大于或者等于1的正整数；
采集N个推荐对象的类别区分数据，并根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题，其中，所述X个主题中的每个主题均包括所述N个推荐对象中的至少一个推荐对象，其中，所述N个推荐对象为所述M个对象中的部分或者全部，X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M；
根据所述目标用户对所述X个主题中的每个主题中包括的推荐对象的喜好值计算得到所述目标用户对所述X个主题中的每个主题的喜好值；
将目标主题推送给所述目标用户，其中，所述目标主题为所述X个主题中的，所述目标用户的喜好值大于推荐阈值的主题。
可选地，处理器1010还用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为确定所述样本用户对所述M个对象中的每个对象的喜好数据；根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型 进行训练；将所述目标用户的特征数据和所述M个对象中的每个对象的特征数据输入到训练好的预测模型以预测所述目标用户对所述M个对象中的每个对象的喜好值。
可选地，处理器1010还用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为构建所述样本用户和所述M个对象之间的交互图；根据所述样本用户和所述M个对象之间的交互图计算所述M个对象两两之间的相似性；根据所述目标用户对n个操作对象中的每个操作对象的历史操作行为确定所述目标用户对所述n个操作对象中的每个操作对象的喜好值，所述操作对象为所述M个对象中，所述目标用户曾经进行了历史操作行为的对象，n为大于或者等于1的正整数，n≤M；根据所述目标用户对所述n个操作对象中的每个操作对象的喜好值以及所述M个对象两两之间的相似性预测所述目标用户对所述M个对象中的每个对象的喜好值。
可选地，所述历史操作行为包括以下至少一种操作行为：浏览、点击、下载、使用、付费、卸载和评价。
可选地，处理器1010还用于根据所述N个推荐对象的类别区分数据进行聚类，从而得到X个主题。
可选地，处理器1010还用于预先设置所述X个主题以及所述X个主题中的每个主题对应的分类预设条件，其中，不同的主题对应不同的分类预设条件；判断所述N个推荐对象中的每个推荐对象的类别区分数据是否符合所述X个主题中的任意一个主题对应的分类预设条件，并将符合所述X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。
可选地，处理器1010还用于将所述目标主题以展示表的方式推送给所述目标用户，其中，所述展示表包括所述目标主题的名称以及所述目标主题中的推荐对象的图标。
可选地，处理器1010还用于将所述目标主题以标签云的方式推送给所述 目标用户，其中，所述标签云包括所述目标主题的名称以及所述目标主题中的推荐对象的标签。
可选地，所述主题至少包括第一级主题以及第二级主题，其中，所述第一级主题包括至少一个第二级主题。
可选地，处理器1010还用于将所述M个对象中，所述目标用户的喜好值大于选择阈值的对象选择为推荐对象；或者，按照目标用户的喜好值对所述M个对象从高至低进行排序，并将所述M个对象中序号为前N的对象选择为推荐对象。
可以理解的是，计算设备1000具体执行步骤还可以参考前述方法实施例的内容，此处不再赘述。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存储记忆体(Random Access Memory，RAM)等。

Claims (20)

  1. 一种主题推荐方法，其特征在于，包括：
    采集样本用户对M个对象的历史操作行为，并根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值，其中，M为大于或者等于1的正整数；
    采集N个推荐对象的类别区分数据，并根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题，其中，所述X个主题中的每个主题均包括所述N个推荐对象中的至少一个推荐对象，其中，所述N个推荐对象为所述M个对象中的部分或者全部，X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M；
    根据所述目标用户对所述X个主题中的每个主题中包括的推荐对象的喜好值计算得到所述目标用户对所述X个主题中的每个主题的喜好值；
    将目标主题推送给所述目标用户，其中，所述目标主题为所述X个主题中的，所述目标用户的喜好值大于推荐阈值的主题。
  2. 根据权利要求1所述的方法，其特征在于，根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值具体包括：
    根据所述样本用户对所述M个对象中的每个对象的历史操作行为确定所述样本用户对所述M个对象中的每个对象的喜好数据；
    根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型进行训练；
    将所述目标用户的特征数据和所述M个对象中的每个对象的特征数据输入到训练好的预测模型以预测所述目标用户对所述M个对象中的每个对象的 喜好值。
  3. 根据权利要求1所述的方法，其特征在于，根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值具体包括：
    根据所述样本用户对所述M个对象中的每个对象的历史操作行为构建所述样本用户和所述M个对象之间的交互图；
    根据所述样本用户和所述M个对象之间的交互图计算所述M个对象两两之间的相似性；
    根据所述目标用户对n个操作对象中的每个操作对象的历史操作行为确定所述目标用户对所述n个操作对象中的每个操作对象的喜好值，所述操作对象为所述M个对象中，所述目标用户曾经进行了历史操作行为的对象，n为大于或者等于1的正整数，n≤M；
    根据所述目标用户对所述n个操作对象中的每个操作对象的喜好值以及所述M个对象两两之间的相似性预测所述目标用户对所述M个对象中的每个对象的喜好值。
  4. 根据权利要求1至3任一权利要求所述的方法，其特征在于，所述历史操作行为包括以下至少一种操作行为：浏览、点击、下载、使用、付费、卸载和评价。
  5. 根据权利要求1至4任一权利要求所述的方法，其特征在于，根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题具体包括：
    根据所述N个推荐对象的类别区分数据进行聚类，从而得到X个主题。
  6. 根据权利要求1至4任一权利要求所述的方法，其特征在于，根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得 到X个主题具体包括：
    预先设置所述X个主题以及所述X个主题中的每个主题对应的分类预设条件，其中，不同的主题对应不同的分类预设条件；
    判断所述N个推荐对象中的每个推荐对象的类别区分数据是否符合所述X个主题中的任意一个主题对应的分类预设条件，并将符合所述X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。
  7. 根据权利要求1-6任一权利要求所述的方法，其特征在于，将目标主题推送给所述目标用户具体为：
    将所述目标主题以展示表的方式推送给所述目标用户，其中，所述展示表包括所述目标主题的名称以及所述目标主题中的推荐对象的图标。
  8. 根据权利要求1至6任一权利要求所述的方法，其特征在于，将目标主题推送给所述目标用户具体为：
    将所述目标主题以标签云的方式推送给所述目标用户，其中，所述标签云包括所述目标主题的名称以及所述目标主题中的推荐对象的标签。
  9. 根据权利要求1至8任一权利要求所述的方法，其特征在于，所述主题至少包括第一级主题以及第二级主题，其中，所述第一级主题包括至少一个第二级主题。
  10. 根据权利要求1至9任一权利要求所述的方法，其特征在于，所述采集N个推荐对象的类别区分数据之前，所述方法还包括：
    将所述M个对象中，所述目标用户的喜好值大于选择阈值的对象选择为推荐对象；或者，
    按照目标用户的喜好值对所述M个对象从高至低进行排序，并将所述M个对象中序号为前N的对象选择为推荐对象。
  11. 一种主题推荐装置，其特征在于，所述装置包括预测模块、分类模块、 计算模块以及推荐模块，其中，
    所述预测模块用于采集样本用户对M个对象的历史操作行为，并根据所述样本用户对所述M个对象中的每个对象的历史操作行为预测目标用户对所述M个对象中的每个对象的喜好值，其中，M为大于或者等于1的正整数；
    所述分类模块用于采集N个推荐对象的类别区分数据，并根据所述N个推荐对象的类别区分数据将所述N个推荐对象进行类别区分，从而得到X个主题，其中，所述X个主题中的每个主题均包括所述N个推荐对象中的至少一个推荐对象，其中，所述N个推荐对象为所述M个对象中的部分或者全部，X为大于或者等于1的正整数，N为大于或者等于1的正整数，且，N≤M；
    所述计算模块用于根据所述目标用户对所述X个主题中的每个主题中包括的推荐对象的喜好值计算得到所述目标用户对所述X个主题中的每个主题的喜好值；
    所述推荐模块用于将目标主题推送给所述目标用户，其中，所述目标主题为所述X个主题中的，所述目标用户的喜好值大于推荐阈值的主题。
  12. 根据权利要求10所述的装置，其特征在于，所述预测模块包括确定单元、训练单元以及预测单元，其中，
    所述确定单元用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为确定所述样本用户对所述M个对象中的每个对象的喜好数据；
    所述训练单元用于根据样本用户的特征数据、样本用户对所述M个对象中的每个对象的喜好数据、上下文特征和所述M个对象中的每个对象的特征数据对预测模型进行训练；
    所述预测单元用于将所述目标用户的特征数据和所述M个对象中的每个对象的特征数据输入到训练好的预测模型以预测所述目标用户对所述M个对象中的每个对象的喜好值。
  13. 根据权利要求11所述的装置，其特征在于，所述预测模块包括构建单元、计算单元、确定单元以及预测单元，
    所述构建单元用于根据所述样本用户对所述M个对象中的每个对象的历史操作行为构建所述样本用户和所述M个对象之间的交互图；
    所述计算单元用于根据所述样本用户和所述M个对象之间的交互图计算所述M个对象两两之间的相似性；
    所述确定单元用于根据所述目标用户对n个操作对象中的每个操作对象的历史操作行为确定所述目标用户对所述n个操作对象中的每个操作对象的喜好值，所述操作对象为所述M个对象中，所述目标用户曾经进行了历史操作行为的对象，n为大于或者等于1的正整数，n≤M；
    所述预测单元用于根据所述目标用户对所述n个操作对象中的每个操作对象的喜好值以及所述M个对象两两之间的相似性预测所述目标用户对所述M个对象中的每个对象的喜好值。
  14. 根据权利要求11至13任一权利要求所述的装置，其特征在于，所述历史操作行为包括以下至少一种操作行为：浏览、点击、下载、使用、付费、卸载和评价。
  15. 根据权利要求11至14任一权利要求所述的装置，其特征在于，所述分类模块包括聚类单元，
    所述聚类单元用于根据所述N个推荐对象的类别区分数据进行聚类，从而得到X个主题。
  16. 根据权利要求11至14任一权利要求所述的装置，其特征在于，所述分类模块包括预设单元以及归入单元，
    所述预设单元用于预先设置所述X个主题以及所述X个主题中的每个主题对应的分类预设条件，其中，不同的主题对应不同的分类预设条件；
    所述归入单元用于判断所述N个推荐对象中的每个推荐对象的类别区分数据是否符合所述X个主题中的任意一个主题对应的分类预设条件，并将符合所述X个主题中的任意一个主题对应的分类预设条件的推荐对象归入对应的主题。
  17. 根据权利要求11-16任一权利要求所述的装置，其特征在于，所述推荐模块具体用于将所述目标主题以展示表的方式推送给所述目标用户，其中，所述展示表包括所述目标主题的名称以及所述目标主题中的推荐对象的图标。
  18. 根据权利要求11至16任一权利要求所述的装置，其特征在于，所述推荐模块具体用于将所述目标主题以标签云的方式推送给所述目标用户，其中，所述标签云包括所述目标主题的名称以及所述目标主题中的推荐对象的标签。
  19. 根据权利要求11至18任一权利要求所述的装置，其特征在于，所述主题至少包括第一级主题以及第二级主题，其中，所述第一级主题包括至少一个第二级主题。
  20. 根据权利要求11至19任一权利要求所述的装置，其特征在于，所述装置还包括选择模块，
    所述选择模块用于将所述M个对象中，所述目标用户的喜好值大于选择阈值的对象选择为推荐对象；或者，
    所述选择模块用于按照目标用户的喜好值对所述M个对象从高至低进行排序，并将所述M个对象中序号为前N的对象选择为推荐对象。
PCT/CN2016/105167 2016-05-24 2016-11-09 主题推荐方法以及装置 WO2017201976A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/198,704 US11830033B2 (en) 2016-05-24 2018-11-21 Theme recommendation method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610351369.6A CN107423308B (zh) 2016-05-24 2016-05-24 主题推荐方法以及装置
CN201610351369.6 2016-05-24

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/198,704 Continuation US11830033B2 (en) 2016-05-24 2018-11-21 Theme recommendation method and apparatus

Publications (1)

Publication Number Publication Date
WO2017201976A1 true WO2017201976A1 (zh) 2017-11-30

Family

ID=60411063

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/105167 WO2017201976A1 (zh) 2016-05-24 2016-11-09 主题推荐方法以及装置

Country Status (3)

Country Link
US (1) US11830033B2 (zh)
CN (1) CN107423308B (zh)
WO (1) WO2017201976A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110347916A (zh) * 2019-06-11 2019-10-18 北京三快在线科技有限公司 跨场景的项目推荐方法、装置、电子设备及存储介质
CN113312889A (zh) * 2021-06-01 2021-08-27 平安证券股份有限公司 报表处理方法、装置、终端和存储介质

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110309414B (zh) * 2018-03-16 2021-08-13 阿里巴巴（中国）有限公司 内容推荐方法、内容推荐装置和电子设备
CN110322305B (zh) * 2018-03-30 2023-03-24 阿里巴巴集团控股有限公司 数据对象信息提供方法、装置及电子设备
US11120067B2 (en) * 2018-07-17 2021-09-14 International Business Machines Corporation Present controlled heterogeneous digital content to users
CN110059248B (zh) * 2019-03-21 2022-12-13 腾讯科技（深圳）有限公司 一种推荐方法、装置及服务器
CN111080339B (zh) * 2019-11-18 2024-01-30 口口相传(北京)网络技术有限公司 基于场景的类目偏好数据生成方法及装置
CN111506833B (zh) * 2020-06-12 2023-05-02 中国人民大学 基于单源SimRank精确解的好友推荐方法
CN111768130B (zh) * 2020-08-06 2024-05-17 京东科技控股股份有限公司 用户分配方法、装置、电子设备及可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090006368A1 (en) * 2007-06-29 2009-01-01 Microsoft Corporation Automatic Video Recommendation
CN101398839A (zh) * 2008-10-23 2009-04-01 浙江大学 一种有声网页新闻的个性化推送方法
US20090277322A1 (en) * 2008-05-07 2009-11-12 Microsoft Corporation Scalable Music Recommendation by Search
CN105335491A (zh) * 2015-10-20 2016-02-17 杭州东信北邮信息技术有限公司 基于用户点击行为来向用户推荐图书的方法和系统
CN105868372A (zh) * 2016-03-31 2016-08-17 广州酷狗计算机科技有限公司 标签分配方法及装置

Family Cites Families (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6317722B1 (en) * 1998-09-18 2001-11-13 Amazon.Com, Inc. Use of electronic shopping carts to generate personal recommendations
WO2001071459A2 (en) * 2000-03-22 2001-09-27 Unifiedmarket, Inc. Method and system for a network-based securities marketplace
US7610216B1 (en) * 2000-07-13 2009-10-27 Ebay Inc. Method and system for detecting fraud
CA2552264A1 (en) * 2003-07-02 2005-01-13 Mobipay International, S.A. Digital mobile telephone transaction and payment system
US7565370B2 (en) * 2003-08-29 2009-07-21 Oracle International Corporation Support Vector Machines in a relational database management system
JP4543930B2 (ja) * 2005-01-07 2010-09-15 ソニー株式会社 情報処理装置および方法、並びにプログラム
US20070022040A1 (en) * 2005-07-19 2007-01-25 Raz Gordon System and Method for Facilitating Network Based Commerce
US20080077574A1 (en) * 2006-09-22 2008-03-27 John Nicholas Gross Topic Based Recommender System & Methods
US8571607B1 (en) * 2007-04-19 2013-10-29 At&T Mobility Ii Llc Streaming ring tones
US9922119B2 (en) * 2007-11-08 2018-03-20 Entit Software Llc Navigational ranking for focused crawling
CN101216825B (zh) * 2007-12-29 2010-10-27 朱廷劭 标引关键词提取/预测方法
US20100211485A1 (en) * 2009-02-17 2010-08-19 Augustine Nancy L Systems and methods of time period comparisons
US8358753B2 (en) * 2009-12-31 2013-01-22 International Business Machines Corporation Interactive voice response (IVR) cloud user interface
CN102346894B (zh) * 2010-08-03 2017-03-01 阿里巴巴集团控股有限公司 推荐信息的输出方法、系统及服务器
US20120101852A1 (en) * 2010-10-25 2012-04-26 Hartford Fire Insurance Company System and method for determining insurance adjustments based on a life event
US8554703B1 (en) * 2011-08-05 2013-10-08 Google Inc. Anomaly detection
US20130073346A1 (en) * 2011-09-16 2013-03-21 David Chun Identifying companies most closely related to a given company
CN102387207A (zh) * 2011-10-21 2012-03-21 华为技术有限公司 基于用户反馈信息的推送方法和推送系统
CN103116588B (zh) 2011-11-17 2017-07-04 深圳市世纪光速信息技术有限公司 一种个性化推荐方法及系统
CN102831234B (zh) * 2012-08-31 2015-04-22 北京邮电大学 基于新闻内容和主题特征的个性化新闻推荐装置和方法
US20140074828A1 (en) * 2012-09-12 2014-03-13 Myemptyslate, Inc. Systems and methods for cataloging consumer preferences in creative content
US20140280575A1 (en) * 2013-03-14 2014-09-18 Google Inc. Determining activities relevant to users
CN103235824A (zh) * 2013-05-06 2013-08-07 上海河广信息科技有限公司 根据浏览网页确定用户感兴趣的网页文本的方法和系统
WO2015008388A1 (ja) * 2013-07-19 2015-01-22 楽天株式会社 レコメンド情報提示装置、レコメンド情報提示方法及びレコメンド情報提示プログラム
CN104794115A (zh) * 2014-01-16 2015-07-22 腾讯科技（深圳）有限公司 应用推荐方法和系统
CN104915359B (zh) * 2014-03-14 2019-05-28 华为技术有限公司 主题标签推荐方法及装置
EP2996366B1 (en) * 2014-03-19 2020-03-11 Huawei Technologies Co., Ltd. Application recommendation method, system and server
CN105095219B (zh) * 2014-04-23 2019-02-01 华为技术有限公司 微博推荐方法和终端
US10255903B2 (en) * 2014-05-28 2019-04-09 Interactive Intelligence Group, Inc. Method for forming the excitation signal for a glottal pulse model based parametric speech synthesis system
CN105205081B (zh) * 2014-06-27 2019-11-05 华为技术有限公司 物品推荐方法和装置
KR101619748B1 (ko) * 2015-06-18 2016-05-16 주식회사 청암아이티 온톨로지 기반의 스마트 학습 또는 수업 공유 시스템
US20160379288A1 (en) * 2015-06-29 2016-12-29 Wal-Mart Stores, Inc. Integrated Meal Plan Generation and Supply Chain Management
CN105389590B (zh) * 2015-11-05 2020-01-14 Tcl集团股份有限公司 一种视频聚类推荐方法和装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090006368A1 (en) * 2007-06-29 2009-01-01 Microsoft Corporation Automatic Video Recommendation
US20090277322A1 (en) * 2008-05-07 2009-11-12 Microsoft Corporation Scalable Music Recommendation by Search
CN101398839A (zh) * 2008-10-23 2009-04-01 浙江大学 一种有声网页新闻的个性化推送方法
CN105335491A (zh) * 2015-10-20 2016-02-17 杭州东信北邮信息技术有限公司 基于用户点击行为来向用户推荐图书的方法和系统
CN105868372A (zh) * 2016-03-31 2016-08-17 广州酷狗计算机科技有限公司 标签分配方法及装置

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110347916A (zh) * 2019-06-11 2019-10-18 北京三快在线科技有限公司 跨场景的项目推荐方法、装置、电子设备及存储介质
CN113312889A (zh) * 2021-06-01 2021-08-27 平安证券股份有限公司 报表处理方法、装置、终端和存储介质
CN113312889B (zh) * 2021-06-01 2023-11-03 平安证券股份有限公司 报表处理方法、装置、终端和存储介质

Also Published As

Publication number Publication date
CN107423308B (zh) 2020-07-07
CN107423308A (zh) 2017-12-01
US20190087884A1 (en) 2019-03-21
US11830033B2 (en) 2023-11-28

Similar Documents

Publication Publication Date Title
WO2017201976A1 (zh) 主题推荐方法以及装置
US11080340B2 (en) Systems and methods for classifying electronic information using advanced active learning techniques
US11238083B2 (en) Intelligently driven visual interface on mobile devices and tablets based on implicit and explicit user actions
AU2016225947B2 (en) System and method for multimedia document summarization
US11314791B2 (en) Smart match autocomplete system
US9043351B1 (en) Determining search query specificity
WO2022016522A1 (zh) 推荐模型的训练方法、推荐方法、装置及计算机可读介质
Agarwal et al. Statistical methods for recommender systems
US11392659B2 (en) Utilizing machine learning models to generate experience driven search results based on digital canvas gesture inputs
WO2021120875A1 (zh) 搜索方法、装置、终端设备及存储介质
TWI662495B (zh) 推廣資訊的處理方法、裝置及系統
US10915752B2 (en) Computer vision based asset evaluation
US11715151B2 (en) Systems and methods for retraining of machine learned systems
CN115885297A (zh) 可区分用户-项目协同聚类
CN107391540A (zh) 一种小程序展示方法、装置及分类器
CN111143678B (zh) 推荐系统和推荐方法
CN112115710B (zh) 一种行业信息识别方法及装置
US20170293691A1 (en) Identifying Abandonment Using Gesture Movement
WO2018149244A1 (zh) 一种基于目标应用推荐相关联应用的方法和装置
US11886809B1 (en) Identifying templates based on fonts
WO2023138475A1 (zh) 图标管理方法、装置、设备和存储介质
CN114265948A (zh) 图像推送方法和装置
CN112257812A (zh) 一种标注样本确定方法、装置、机器可读介质及设备
CN111475721A (zh) 信息推送方法、装置、设备及存储介质
US20240119484A1 (en) Privacy-preserving rules-based targeting using machine learning

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

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

Ref document number: 16902960

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 16902960

Country of ref document: EP

Kind code of ref document: A1