WO2018014771A1 - Procédé, dispositif et système de poussée d'objet de données - Google Patents

Procédé, dispositif et système de poussée d'objet de données Download PDF

Info

Publication number
WO2018014771A1
WO2018014771A1 PCT/CN2017/092743 CN2017092743W WO2018014771A1 WO 2018014771 A1 WO2018014771 A1 WO 2018014771A1 CN 2017092743 W CN2017092743 W CN 2017092743W WO 2018014771 A1 WO2018014771 A1 WO 2018014771A1
Authority
WO
WIPO (PCT)
Prior art keywords
data object
user
data
feature
feature vector
Prior art date
Application number
PCT/CN2017/092743
Other languages
English (en)
Chinese (zh)
Inventor
倪娜
王晓伟
廖闯
樊志国
Original Assignee
阿里巴巴集团控股有限公司
倪娜
王晓伟
廖闯
樊志国
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 阿里巴巴集团控股有限公司, 倪娜, 王晓伟, 廖闯, 樊志国 filed Critical 阿里巴巴集团控股有限公司
Publication of WO2018014771A1 publication Critical patent/WO2018014771A1/fr

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

Definitions

  • the present application relates to the field of data processing technologies, and in particular, to a data object push method, a data object push device, and a data object push system.
  • Content push is a way to promote specific multimedia content such as advertising content, news content, notification content, audio and video to a specific group of people.
  • the technique of targeting people using user tags is a common technique in the field of advertising and other content delivery.
  • the technology is a tag for users by data such as the user's historical behavior or the attributes of the user.
  • the tag technology based on the user's historical behavior mainly targets the user's browsing, searching, clicking, and historical behaviors such as clicking and converting advertisements or content, and mapping the user to a certain tag. For example, for e-commerce websites, according to the user's search, clicks on products, collections, shopping carts, transactions, and other historical behaviors, the corresponding store customers, new customers, n-day collections of shops / purchases A label for users of store products.
  • the user attribute tag is generally the attribute of the user itself, such as the user's gender, age, region, occupation, and the like. It has nothing to do with the specific store.
  • tags such as population attributes and user behaviors can meet the crowd division and delivery requirements in some scenarios, but cannot accurately predict the user's actual intentions. It is a tag that is not directly related to the conversion target.
  • this user tag-based user content matching strategy is relatively blind in the case of unpredictable future users.
  • the technical problem to be solved by the embodiments of the present application is to provide a method for pushing a data object, and pushing the most relevant data object to the user in combination with the actual intention of the user.
  • the embodiment of the present application further provides a device for pushing a data object and a system for pushing a data object to ensure implementation and application of the foregoing method.
  • the embodiment of the present application discloses a system for pushing a data object, and the system includes:
  • One or more processors are One or more processors;
  • One or more modules the one or more modules being stored in the memory and configured to be executed by the one or more processors, the one or more modules having the following functions:
  • Determining one or more data objects for the access request and determining, from the plurality of data object sets, a target data object set related to the first user;
  • the embodiment of the present application further discloses a method for pushing a data object, where the method includes:
  • Determining one or more data objects for the access request and determining, from the plurality of data object sets, a target data object set related to the first user;
  • the step of determining a target data object set related to the first user from the plurality of data object sets includes:
  • the first N data object sets having the largest actual preference value are selected as the target data object set related to the first user, where N is a positive integer and N is smaller than the number of data object sets.
  • the step of determining the actual preference values of the first user and the plurality of data object sets respectively comprises:
  • the prediction preference value is corrected by using the correlation degree, respectively, to obtain an actual preference value of the first user and the data object set.
  • the step of separately obtaining the predicted preference values of the plurality of data object sets comprises:
  • the set feature vector and the user feature vector are respectively input into a preset prediction model to obtain a predicted preference value of the data object set.
  • the prediction model is established as follows:
  • Modeling is performed according to the sample information, the attribute feature vector, and the visitor feature vector to generate a predictive model.
  • the step of respectively acquiring the set feature vector of the data object set include:
  • the data object includes header information
  • the determining the correlation between the first user and the data object set respectively comprises:
  • the intent word vector and the similarity of each set word vector are respectively calculated as the degree of correlation between the first user and the corresponding data object set.
  • the step of selecting to send the at least one data object to the client from the target data object set and the one or more data objects determined for the access request comprises:
  • the first user is a buyer user
  • the second user is a seller user
  • the data object associated with the second user is an item sold by a seller user
  • the data object set has an association in the same store. A combination of goods of a relationship.
  • the embodiment of the present application further discloses an apparatus for pushing a data object, where the apparatus includes:
  • the access request receiving module is configured to receive a second user-associated data object access request sent by the client corresponding to the first user, where the second user has an associated plurality of data objects, and according to the multiple data a collection of multiple data objects determined by the object;
  • a data object determining module for determining one or more data pairs for the access request Elephant
  • a target data object set determining module configured to determine, from the plurality of data object sets, a target data object set related to the first user
  • a data object sending module configured to select, from the target data object set and the one or more data objects determined for the access request, to send at least one data object to the client.
  • the target data object set determining module comprises:
  • An actual preference determining submodule configured to respectively determine an actual preference value of the first user and the plurality of data object sets
  • a target data object set selection submodule configured to select a top N data object set with the actual actual preference value as a target data object set related to the first user, where N is a positive integer and N is smaller than the data object The number of collections.
  • the actual preference determining submodule comprises:
  • a prediction preference acquiring unit configured to separately acquire prediction preference values of the plurality of data object sets
  • a correlation calculation unit configured to respectively determine a correlation between the first user and the data object set
  • a correcting unit configured to correct the predicted preference value by using the correlation degree for each data object set, to obtain an actual preference value of the first user and the data object set.
  • the predicted preference obtaining unit comprises:
  • a set feature vector obtaining subunit configured to respectively acquire a set feature vector of the data object set
  • a user feature vector obtaining sub-unit configured to acquire a feature value of a user feature of the user of the entire network, and determine a corresponding user feature vector according to the feature value of the user feature;
  • a prediction preference calculation subunit configured to respectively input the set feature vector and the user feature vector into a preset prediction model to obtain a predicted preference value of the data object set.
  • the prediction model is established as follows:
  • Modeling is performed according to the sample information, the attribute feature vector, and the visitor feature vector to generate a predictive model.
  • the set feature vector obtaining subunit is further configured to:
  • the data object includes title information
  • the correlation calculation unit includes:
  • a set word vector calculation subunit configured to separately calculate a set word vector of the data object set according to a preset word vector model
  • An intent word vector calculation subunit configured to acquire a specified number of data objects recently browsed by the first user, and acquire an intent word vector of the first user based on the specified number of data objects;
  • the similarity calculation subunit is configured to separately calculate the similarity of the intent word vector and each set word vector as the correlation between the first user and the corresponding data object set.
  • the data object sending module includes:
  • a data object selection submodule configured to select at least one data object from the target data object set and the one or more data objects determined for the access request
  • a target page generating submodule configured to generate a target page according to the selected data object
  • the target page returns a sub-module for returning the target page to the client.
  • the first user is a buyer user
  • the second user is a seller user
  • the data object associated with the second user is an item sold by a seller user
  • the data object set has an association in the same store. A combination of goods of a relationship.
  • the embodiments of the present application include the following advantages:
  • the second user may preset one or more data object sets, and when receiving the second user-associated data object access request sent by the client corresponding to the first user, determine one or more for the request.
  • Data objects and determining a target data object set associated with the first user from the plurality of data object sets, and selecting from the target data object set and one or more data objects determined for the request.
  • the client sends at least one data object, so that the data object obtained by the client corresponding to the first user is more consistent with the first user preference, and the accurate data object push is implemented.
  • FIG. 1 is a flow chart showing the steps of a method for pushing a data object according to the present application
  • FIG. 1a is a schematic diagram of a target page in an embodiment of a data object push method according to the present application
  • FIG. 2 is a structural block diagram of an apparatus for pushing a data object according to the present application
  • FIG. 3 is a schematic structural diagram of a server according to an embodiment of the present application.
  • FIG. 1 a flow chart of steps of a method for pushing a data object according to the present application is shown. Specifically, the method may include the following steps:
  • Step 101 Receive a data object access request associated with the second user sent by the client corresponding to the first user.
  • the second user has a plurality of associated data objects, and a plurality of data object sets determined according to the plurality of data objects.
  • the user of the embodiment of the present application may include at least a first user and a second user, where the second user is a provider of the data object, and the first user is a pushed party.
  • the second user may have an associated plurality of data objects that may be presented in a presentation page of the designated website and/or presented in a page associated with the second user in the designated website.
  • the information about the page associated with the second user may also be displayed in the presentation page, for example, page identification information and/or page link information of the page associated with the second user, etc., when clicking on the second user association
  • the page associated with the second user may be redirected, and the page associated with the second user may include a plurality of data objects associated with the second user.
  • the data object displayed in the presentation page may include link information, and when the data object is selected, may jump to a page corresponding to the link information.
  • the page corresponding to the link information may be the The details page of the data object or the page associated with the second user.
  • the second user may further combine the associated multiple data objects into a plurality of data object sets, where the data object set may be a set of a plurality of data objects having an association relationship.
  • the second user can associate the data objects I1, I2, I3 in one data object set G1, the data objects I4, I5 in the data object set G2, and associate the data objects I6, I7, I8, I9 in the data.
  • the object collection is in G3.
  • the first user may send a data object access request associated with the second user by using the client corresponding to the first user.
  • the client corresponding to the first user may include a client or a browser client that specifies the website.
  • the first user may issue an access request by using at least one of the following manners:
  • the first user may load a presentation page in the client corresponding to the first user, where the presentation page includes multiple data objects including the data object associated with the second user, and the first user selects the first page in the presentation page.
  • the second user associates the data object, it is determined that the first user has issued an access request for the data object associated with the second user.
  • the first user may also search for the required second user-related data object in the presentation page by searching, and click the second user association when the search is successful.
  • Data object at this point it can be determined that the first user sent An access request for a data object associated with the second user.
  • the first user may also invoke an interface provided by the client corresponding to the first user to transmit a link address of the data object associated with the second user to issue an access request for the data object.
  • the first user may also invoke an interface provided by the client corresponding to the first user to transmit a link address of the page associated with the second user to issue an access request for the data object.
  • the manner in which the client corresponding to the first user sends a data object access request for the second user is only an example of the embodiment of the present application, and the technical person in the field adopts the client corresponding to the first user in other manners. It is possible to issue a data object access request associated with the second user, which is not limited in this embodiment of the present application.
  • the embodiment of the present application may be applied to an e-commerce scenario, where the first user may be a buyer user, the second user may be a seller user, and the second user associated page may be a page where the seller user's store is located;
  • the data object associated with the second user may be an item sold by a seller user, and the data object set may be a combination of items having an associated relationship in the same store.
  • a seller can combine a plurality of related products that are sold in advance to obtain a product combination, for example, a toothbrush, a toothpaste, a cup, or a mobile phone, a mobile phone case, or a mobile phone. Film, headphone cable, charger, and other products related to mobile phones are combined.
  • Step 102 Determine one or more data objects for the access request, and determine, from the plurality of data object sets, a target data object set related to the first user;
  • one or more corresponding to the real-time access intention of the first user may be determined for the access request.
  • Two user-associated data objects Two user-associated data objects.
  • the first user issues an access request by selecting a data object associated with the second user in the presentation page and/or a link address of the data object associated with the second user, Then, according to the access request, the corresponding data object can be directly determined.
  • the page associated with the second user may be determined according to the access request. And obtaining the data object included in the page associated with the second user as the data object corresponding to the access request.
  • the data object set associated with the second user may be determined according to the access request, and Extracting a target data object set from the plurality of data object sets, wherein the target data object set is a most relevant data object set of the first user among the plurality of data object sets.
  • the step of determining a target data object set related to the first user from the plurality of data object sets may include the following sub-steps:
  • Sub-step S11 respectively determining an actual preference value of the first user and the plurality of data object sets
  • the calculation of the actual preference value in the embodiment of the present application needs to consider the access behavior of the first user in real time.
  • the sub-step S11 may further include the following sub-steps:
  • Sub-step S111 respectively acquiring prediction preference values of the plurality of data object sets
  • the predicted preference value may be an offline calculated value, and the calculation of the value may be performed without comprehensive consideration of the user's real-time access behavior, and based on the characteristics of the data object set and the multi-dimensional features in the specified website. The value obtained.
  • the sub-step S111 may further include the following sub-steps:
  • Sub-step S1111 respectively acquiring a set feature vector of the data object set
  • the sub-step S1111 may further include: acquiring feature values and features of attribute features of each data object in the data object set. a vector; summarizing feature values and feature vectors of attribute features of all data objects in the data object set to obtain a set feature vector of the data object set.
  • each data object has an attribute feature that reflects an attribute of the data object.
  • the corresponding attribute feature may include, but is not limited to, the category, price, sales volume, and location of the product. , efficacy, gender distribution, age distribution, etc. of the user who purchased the product.
  • the feature value v j of the attribute feature of the data object (the feature value of the attribute feature of the jth data object) may be first obtained, and the attribute feature vector of the data object is calculated according to the feature value v j
  • the feature value of the attribute feature can be obtained according to the interface specified by the call.
  • the linearly transformed eigenvector (eigenvector) is a non-degenerate vector whose direction is unchanged under the transformation.
  • the calculation method of the feature vector is not limited.
  • one or more set feature vectors of the data object set may be acquired according to attribute attribute values and attribute feature vectors of all data objects included in the data object set. Characterize the set of data objects in real time into the same vector space of the data objects.
  • the set feature vector of the data object set may be calculated by using the following formula (1):
  • the logarithmic function is used to smooth the eigenvalues of the attribute features of the data object. It can be seen that the larger the eigenvalue of the attribute feature of the data object, the data object is The greater the proportion of the data object collection. For example, the larger the sales volume of a product, the greater the proportion of the product in the product combination.
  • Sub-step S1112 acquiring feature values of user features of the entire network user, and determining corresponding user feature vectors according to the feature values of the user features;
  • the entire network user refers to each guest user in the specified website.
  • the user feature may include a behavior feature and/or a population attribute feature, wherein the behavior feature may include, but is not limited to, a user browsing, collecting, and the like of the data object; the demographic attribute feature may include But not limited to: the user's gender, age, occupation, etc.
  • the behavior characteristics of the user may be obtained from the log records of the specified website, and the demographic attribute characteristics of the user are obtained from the user database of the specified website, and the user database records information such as the demographic characteristics of each registered user. .
  • the feature values of a certain demographic attribute and/or behavior characteristic of all users may be obtained first, and the eigenvalues of all users are aggregated to obtain an average eigenvalue, and then according to the average Eigenvalue calculation user feature vector
  • the manner of calculating the feature vector according to the feature value is as described in the foregoing sub-step S1111, and details are not described herein again.
  • Sub-step S1113 respectively inputting the set feature vector and the user feature vector into a preset prediction model to obtain a predicted preference value of the data object set.
  • the preset prediction model may be a model obtained by offline training.
  • the prediction model may be established in the following manner, but it should be understood that the embodiment of the present application is not limited thereto. It is possible for domain technicians to establish predictive models with the same effect in other ways:
  • the calculation method of the visitor feature vector is similar to the calculation method of the user feature vector described above, except that the data of the user feature vector is derived from the access behavior of all visitors of the entire network, and the data of the visitor feature vector is derived from all accesses of the second. The visitor's access behavior for the user's associated page.
  • the visitor feature vector is obtained according to the visitor feature value, and the visitor feature value is a feature value of the visitor feature, and the visitor feature may include a behavior feature and/or a demographic attribute feature, wherein the behavior feature may include, but is not limited to: a visitor pair The behavior of browsing, collecting, and the like of the data objects associated with the two users; the demographic attribute characteristics may include, but are not limited to, the gender, age, occupation, and the like of the visitor.
  • the attribute feature vector of the data object is as described in the above sub-step S1111, and will not be described again here.
  • the access behavior data of all the visitors accessing the associated page of the second user in the associated page may be a visitor accessing the data object in the associated page by the visitor of the associated page of the second user, the access behavior may be expressed as (u, i), where u is the visitor identity of the visited associated page, and i is the data object identifier accessed by the visitor in the associated page.
  • one (u, i) can be used as a sample information.
  • the sample information is a positive sample, otherwise, if u does not perform a specified operation on i, the sample information is a negative sample.
  • the sample is a positive sample. Otherwise, if the user browses an item but does not purchase the item, then the sample Is a negative sample.
  • the attribute feature vector and the visitor feature vector After determining the sample information, the attribute feature vector and the visitor feature vector, it can be modeled as a parameter and modeled by a preset modeling algorithm to obtain a predictive model.
  • the modeling algorithm may be an MPI (Message Passing Interface, which can be used for parallel computing) - GBDT (Gradient Boosting Decision Tree, A machine learning algorithm widely used for classification or regression problems, an iterative decision tree algorithm consisting of multiple decision trees, the conclusions of all trees are added together to make the final answer.
  • the algorithm is nonlinear after modeling. Predictive model.
  • the function of the prediction model is to estimate the probability that a first user will perform a specified operation on a certain data object according to the historical access data of the visitor of the second user's associated page and the attribute characteristics of the data object in the associated page. For example, based on historical behavior data such as in-store visitors and commodities, the probability that a future user will have a purchase relationship for a certain product is estimated.
  • the output user's predicted preference value for each data object set may be expressed as follows:
  • Sub-step S112 determining a correlation degree between the first user and the data object set respectively;
  • the degree of relevance of the first user to each data object set may also be calculated.
  • the sub-step S112 may further include the following sub-steps:
  • Sub-step S1121 respectively calculating a set word vector of the data object set according to a preset word vector model
  • Word Embedding is a way to mathematicalize words in the language.
  • the word vector is used to represent the word vector of each data object.
  • the preset word vector model can be generated in the following manner:
  • Each data object has title information in the specified website, and the title information of each data object in the specified website is used as a corpus, and the corpus is trained by a neural network-based word vector algorithm to obtain a multi-dimensional word vector model, the word vector.
  • the model includes word vectors for multiple data objects.
  • the word vector algorithm based on neural network may include a cyclic neural network algorithm, a recurrent neural network algorithm, etc., and the principle of one of the neural network algorithms is shown below.
  • C(w) represents the word vector corresponding to the data object w.
  • the whole model uses a unique set of word vectors, which exists in the matrix C (matrix of
  • the conversion from w to C(w) is to take a row from the matrix.
  • the first layer of the network is C(wt-n+1),...,C(wt-2),C(wt-1) These n-1 vectors are joined end to end to form a (n-1)m-dimensional vector, denoted by x below.
  • the second layer of the network is calculated directly using d+Hx, and d is an offset term. After that, tanh is used as the activation function.
  • the third layer (output layer) of the network has a total of
  • ⁇ h matrix) is the parameter of the hidden layer to the output layer.
  • ⁇ (n-1)m) The matrix contains the straight edges from the input layer to the output layer (the straight edge is a linear transformation from the input layer directly to the output layer). If you do not need a straight edge, set W to 0.
  • the word vector model can be obtained by the stochastic gradient descent rule.
  • the word vector model After obtaining the word vector model, the word vector model can also be synchronized to the storage for real-time calculation.
  • the set word vector of each data object set can be calculated according to the word vector model and the attribute features of the data object.
  • the set word vector can be calculated using the following formula (2)
  • v j is the eigenvalue of the attribute feature of the jth data object, and is smoothed by a logarithmic function.
  • the word vector for the jth data object is the eigenvalue of the attribute feature of the jth data object, and is smoothed by a logarithmic function.
  • Sub-step S1122 acquiring a specified number of data objects that the first user recently browsed, and acquiring an intent word vector of the first user based on the specified number of data objects;
  • the real-time intention of the user may be predicted in combination with the historical browsing behavior of the first user. Specifically, a specified number of data objects that the first user recently browsed (for example, 10 items that were recently browsed in the day) may be acquired, and a word vector of the specified number of data objects is obtained, according to the word vector of the specified number of data objects. To get the first user's intent word vector.
  • a specified number of data objects that the first user recently browsed for example, 10 items that were recently browsed in the day
  • a word vector of the specified number of data objects is obtained, according to the word vector of the specified number of data objects.
  • the first user's intent word vector may be calculated using the following formula (3):
  • T j is the number of seconds of the data object browsed by the user u from the current time
  • ⁇ is the attenuation coefficient
  • Sub-step S1123 respectively calculating the intent word vector and the similarity of each set word vector as the degree of relevance of the first user and the corresponding data object set.
  • the similarity of the two may be separately calculated, the similarity reflecting the degree of correlation between the first user and the corresponding data object set.
  • the intent word vector and the similarity of each set word vector may be calculated using the following formula (4):
  • Sub-step S113 for each data object set, correct the predicted preference value by using the correlation degree, respectively, to obtain an actual preference value of the first user and the data object set.
  • the first user's intent word vector and each set word vector are obtained.
  • the predicted preference value can be corrected according to the degree of correlation, and the actual preference value of the first user and each data object set is obtained.
  • the actual preference value can be calculated using equation (5) below:
  • ⁇ is the fusion parameter
  • Sub-step S12 selecting the first N data object sets with the actual actual preference value as the target data object set related to the first user, where N is a positive integer and N is smaller than the number of data object sets.
  • the first N data object sets having the largest actual preference value may be used as the target data object set related to the first user.
  • N is a positive integer and N is less than the number of data object sets.
  • Step 103 Select to send at least one data object to the client from the target data object set and the one or more data objects determined for the access request.
  • At least one data object may be selected and sent to the client to display the selected data object in the client.
  • step 103 may further include the following sub-steps:
  • Sub-step S21 selecting at least one data object from the target data object set and the one or more data objects determined for the access request;
  • Sub-step S22 generating a target page according to the selected data object
  • Sub-step S23 returning the target page to the client.
  • the plurality of data objects included in the target data object set and one or more data objects corresponding to the access request may be included.
  • the manner of selecting may be randomly selected or selected according to a preset priority policy, for example, priority selection.
  • Corresponding data objects then select the data objects in the target data collection.
  • a target page may be generated according to the selected data object, and the target page is returned to the client to display the target page in the client.
  • the target page may be displayed in an application app client of the mobile terminal, or the target page may also be displayed in a browser client of the PC or the mobile terminal.
  • the target page 10 displayed by the client a plurality of data objects 20 corresponding to the access request and a plurality of data object sets 30 related to the first user may be included, and each of the data object sets 30 may include a plurality of data objects 20 , so that the content displayed on the target page matches the user's preference.
  • the second user may preset one or more data object sets, and when receiving the second user-associated data object access request sent by the client corresponding to the first user, determine one or more for the request.
  • Data objects and determining a target data object set associated with the first user from the plurality of data object sets, and selecting from the target data object set and one or more data objects determined for the request.
  • the client sends at least one data object, so that the data object obtained by the client corresponding to the first user is more consistent with the first user preference, and the accurate data object push is implemented.
  • FIG. 2 a block diagram of a device embodiment of a data object push of the present application is shown, which may include the following modules:
  • the access request receiving module 201 is configured to receive a second user-associated data object access request sent by the client corresponding to the first user, where the second user has multiple numbers associated with According to the object, and a plurality of data object sets determined according to the plurality of data objects;
  • a data object determining module 202 configured to determine one or more data objects for the access request
  • a target data object set determining module 203 configured to determine, from the plurality of data object sets, a target data object set related to the first user
  • the data object sending module 204 is configured to select to send the at least one data object to the client from the target data object set and the one or more data objects determined for the access request.
  • the target data object set determining module 203 may include the following sub-modules:
  • An actual preference determining submodule configured to respectively determine an actual preference value of the first user and the plurality of data object sets
  • a target data object set selection submodule configured to select a top N data object set with the actual actual preference value as a target data object set related to the first user, where N is a positive integer and N is smaller than the data object The number of collections.
  • the actual preference determining submodule may further include the following units:
  • a prediction preference acquiring unit configured to separately acquire prediction preference values of the plurality of data object sets
  • a correlation calculation unit configured to respectively determine a correlation between the first user and the data object set
  • a correcting unit configured to correct the predicted preference value by using the correlation degree for each data object set, to obtain an actual preference value of the first user and the data object set.
  • the prediction preference acquiring unit may further include the following subunits:
  • a set feature vector obtaining subunit configured to respectively acquire a set feature vector of the data object set
  • a user feature vector obtaining sub-unit configured to acquire a feature value of a user feature of the user of the entire network, and determine a corresponding user feature vector according to the feature value of the user feature;
  • a prediction preference calculation subunit configured to respectively input the set feature vector and the user feature vector into a preset prediction model to obtain a predicted preference value of the data object set.
  • the prediction model is established as follows:
  • Modeling is performed according to the sample information, the attribute feature vector, and the visitor feature vector to generate a predictive model.
  • the set feature vector obtaining subunit is further configured to:
  • the data object includes header information
  • the relevance calculation unit may further include the following subunits:
  • a set word vector calculation subunit configured to separately calculate a set word vector of the data object set according to a preset word vector model
  • An intent word vector calculation subunit configured to acquire a specified number of data objects recently browsed by the first user, and acquire an intent word vector of the first user based on the specified number of data objects;
  • the similarity calculation subunit is configured to separately calculate the similarity of the intent word vector and each set word vector as the correlation between the first user and the corresponding data object set.
  • the data object sending module 204 may include the following sub-modules:
  • a data object selection submodule configured to select at least one data object from the target data object set and the one or more data objects determined for the access request
  • a target page generating submodule configured to generate a target page according to the selected data object
  • the target page returns a sub-module for returning the target page to the client.
  • the first user is a buyer user
  • the second user is a seller user
  • the data object associated with the second user is a product sold by the seller user.
  • the data object set is a combination of items having an associated relationship in the same store.
  • the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.
  • the embodiment of the present application further provides a system for pushing a data object, and the system for pushing the data object may include:
  • One or more processors are One or more processors;
  • One or more modules the one or more modules being stored in the memory and configured to be executed by one or more processors, wherein the one or more modules have the following functions:
  • Determining one or more data objects for the access request and determining, from the plurality of data object sets, a target data object set related to the first user;
  • the one or more modules may have the following functions:
  • the first N data object sets having the largest actual preference value are selected as the target data object set related to the first user, where N is a positive integer and N is smaller than the number of data object sets.
  • the one or more modules may have the following functions:
  • the prediction preference value is corrected by using the correlation degree, respectively, to obtain an actual preference value of the first user and the data object set.
  • the one or more modules may have the following functions:
  • the set feature vector and the user feature vector are respectively input into a preset prediction model to obtain a predicted preference value of the data object set.
  • the one or more modules may have the following functions:
  • Modeling is performed according to the sample information, the attribute feature vector, and the visitor feature vector to generate a predictive model.
  • the one or more modules may have the following functions:
  • the one or more modules may have the following functions:
  • the intent word vector and the similarity of each set word vector are respectively calculated as the degree of correlation between the first user and the corresponding data object set.
  • the one or more modules may have the following functions:
  • the first user is a buyer user
  • the second user is a seller user
  • the data object associated with the second user is an item sold by a seller user
  • the data object set is in the same store.
  • FIG. 3 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • the server 300 can vary considerably depending on configuration or performance, and can include one or more central processing units (CPUs) 322 (eg, one or more processors) and memory 332, one or one
  • the above storage medium 330 storing the application 342 or the data 344 (for example, one or one storage device in Shanghai).
  • the memory 332 and the storage medium 330 may be temporarily stored or persistently stored.
  • the program stored on storage medium 330 may include one or more modules (not shown), each of which may include a series of instruction operations in the server.
  • the central processor 322 can be configured to communicate with the storage medium 330 to perform a series of instruction operations in the storage medium 330 on the server 300.
  • Server 300 may also include one or more power sources 326, one or more wired or wireless network interfaces 350, one or more input and output interfaces 358, one or more keyboards 356, and/or one or more operating systems 341.
  • power sources 326 one or more wired or wireless network interfaces 350
  • input and output interfaces 358 one or more input and output interfaces 358
  • operating systems 341. Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.
  • central processor 322 can execute instructions on server 300 for:
  • Determining one or more data objects for the access request and determining, from the plurality of data object sets, a target data object set related to the first user;
  • the central processor 322 can also execute instructions on the server 300 for:
  • the first N data object sets having the largest actual preference value are selected as the target data object set related to the first user, where N is a positive integer and N is smaller than the number of data object sets.
  • the central processor 322 can also execute instructions on the server 300 for:
  • the prediction preference value is corrected by using the correlation degree, respectively, to obtain an actual preference value of the first user and the data object set.
  • the central processor 322 can also execute instructions on the server 300 for:
  • the set feature vector and the user feature vector are respectively input into a preset prediction model to obtain a predicted preference value of the data object set.
  • the central processor 322 can also execute instructions on the server 300 for:
  • Modeling is performed according to the sample information, the attribute feature vector, and the visitor feature vector to generate a predictive model.
  • the central processor 322 can also execute instructions on the server 300 for:
  • the central processor 322 can also execute instructions on the server 300 for:
  • the intent word vector and the similarity of each set word vector are respectively calculated as the degree of correlation between the first user and the corresponding data object set.
  • the central processor 322 can also execute instructions on the server 300 for:
  • the first user is a buyer user
  • the second user is a seller user
  • the data object associated with the second user is an item sold by a seller user
  • the data object set is in the same store.
  • embodiments of the embodiments of the present application can be provided as a method, apparatus, or computer program product. Therefore, the embodiments of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Moreover, embodiments of the present application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • Embodiments of the present application are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block of the flowchart illustrations and/or FIG.
  • These computer program operating instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing terminal device to produce a machine for execution by a processor of a computer or other programmable data processing terminal device
  • the operational instructions generate means for implementing the functions specified in one or more of the flow or in a block or blocks of the flowchart.
  • the computer program operating instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that operational instructions stored in the computer readable memory produce manufacturing including the operational command device
  • the operation instruction means implements the functions specified in one block or a plurality of blocks of a flow or a flow and/or a block diagram of the flowchart.
  • These computer program operating instructions can also be loaded onto a computer or other programmable data processing terminal device such that a series of operational steps are performed on the computer or other programmable terminal device to produce computer-implemented processing, such that the computer or other programmable terminal
  • the operational instructions executed on the device provide steps for implementing the functions specified in one or more blocks of the flowchart or in a flow or block of the flowchart.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Information Transfer Between Computers (AREA)

Abstract

L'invention concerne un procédé, un dispositif et un système de poussée d'objet de données, le procédé consistant : à recevoir une demande d'accès à un objet de données associée à un second utilisateur et envoyée par un client correspondant à un premier utilisateur (101), le second utilisateur étant associé à des objets de données multiples; à déterminer un ou plusieurs objets de données par rapport à la demande d'accès en fonction d'ensembles d'objets de données multiples déterminés par les objets de données multiples et à déterminer un ensemble d'objets de données cible associé au premier utilisateur à partir des ensembles d'objets de données multiples (102); et à sélectionner d'envoyer au client au moins un objet de données de l'ensemble d'objets de données cible et desdits objets de données déterminés par rapport à la demande d'accès (103). Le procédé, le dispositif et le système peuvent être combinés à l'intention réelle d'un utilisateur et utilisés pour pousser l'objet de données le plus pertinent vers l'utilisateur.
PCT/CN2017/092743 2016-07-21 2017-07-13 Procédé, dispositif et système de poussée d'objet de données WO2018014771A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610581054.0A CN107644036B (zh) 2016-07-21 2016-07-21 一种数据对象推送的方法、装置及系统
CN201610581054.0 2016-07-21

Publications (1)

Publication Number Publication Date
WO2018014771A1 true WO2018014771A1 (fr) 2018-01-25

Family

ID=60991981

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/092743 WO2018014771A1 (fr) 2016-07-21 2017-07-13 Procédé, dispositif et système de poussée d'objet de données

Country Status (3)

Country Link
CN (1) CN107644036B (fr)
TW (1) TW201804400A (fr)
WO (1) WO2018014771A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110517099A (zh) * 2018-05-22 2019-11-29 北京京东尚科信息技术有限公司 用于确定关联供应方的方法和装置
CN110324370B (zh) * 2018-03-29 2023-05-30 北京京东尚科信息技术有限公司 服务端向客户端推送数据的方法和装置

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110109901B (zh) * 2018-02-02 2024-05-24 北京京东尚科信息技术有限公司 筛选目标对象的方法和装置
CN108898067B (zh) * 2018-06-06 2021-04-30 北京京东尚科信息技术有限公司 确定人和物关联度的方法、装置及计算机可读存储介质
CN110147882B (zh) * 2018-09-03 2023-02-10 腾讯科技（深圳）有限公司 神经网络模型的训练方法、人群扩散方法、装置及设备
CN110032860B (zh) * 2018-12-27 2020-07-28 阿里巴巴集团控股有限公司 登录方式的推送、展示方法、装置及设备
CN111724221A (zh) * 2019-03-19 2020-09-29 北京京东尚科信息技术有限公司 确定商品匹配信息的方法、系统、电子设备及存储介质
CN110378749B (zh) * 2019-07-25 2023-09-26 深圳前海微众银行股份有限公司 客户端相似性的评估方法、装置、终端设备及存储介质
CN113379436A (zh) * 2020-03-09 2021-09-10 阿里巴巴集团控股有限公司 信息处理方法、装置、计算设备及介质
CN112000884A (zh) * 2020-08-13 2020-11-27 腾讯音乐娱乐科技（深圳）有限公司 一种用户内容推荐方法及装置、服务器、存储介质
US11977728B1 (en) * 2022-12-22 2024-05-07 Lifetrack Medical Systems Private Ltd. Interface-integrated permissions configuration

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101408960A (zh) * 2007-10-12 2009-04-15 阿里巴巴集团控股有限公司 一种个性化信息推荐方法和装置
CN102340514A (zh) * 2010-07-15 2012-02-01 腾讯科技（北京）有限公司 网络信息推送方法及系统
US20120253917A1 (en) * 2011-03-28 2012-10-04 Samsung Electronics Co., Ltd Method and apparatus for managing and providing user purchase information
CN105450586A (zh) * 2014-07-24 2016-03-30 阿里巴巴集团控股有限公司 信息推送方法、系统、服务器及客户端
CN105590220A (zh) * 2014-10-24 2016-05-18 无锡蓝阳谐波科技有限公司 一种互联网商品的推送展示方法及装置

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101853463A (zh) * 2009-03-30 2010-10-06 北京邮电大学 基于客户特征的协同过滤推荐方法和系统
US8433710B2 (en) * 2010-09-09 2013-04-30 Ebay Inc. Sizing content recommendation system
CN102663627A (zh) * 2012-04-26 2012-09-12 焦点科技股份有限公司 个性化推荐方法
CN103425677B (zh) * 2012-05-18 2016-08-24 阿里巴巴集团控股有限公司 关键词分类模型确定方法、关键词分类方法及装置
CN104866474B (zh) * 2014-02-20 2018-10-09 阿里巴巴集团控股有限公司 个性化数据搜索方法及装置
US9817903B2 (en) * 2014-11-25 2017-11-14 Ebay Inc. Methods and systems for social network based content recommendations

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101408960A (zh) * 2007-10-12 2009-04-15 阿里巴巴集团控股有限公司 一种个性化信息推荐方法和装置
CN102340514A (zh) * 2010-07-15 2012-02-01 腾讯科技（北京）有限公司 网络信息推送方法及系统
US20120253917A1 (en) * 2011-03-28 2012-10-04 Samsung Electronics Co., Ltd Method and apparatus for managing and providing user purchase information
CN105450586A (zh) * 2014-07-24 2016-03-30 阿里巴巴集团控股有限公司 信息推送方法、系统、服务器及客户端
CN105590220A (zh) * 2014-10-24 2016-05-18 无锡蓝阳谐波科技有限公司 一种互联网商品的推送展示方法及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110324370B (zh) * 2018-03-29 2023-05-30 北京京东尚科信息技术有限公司 服务端向客户端推送数据的方法和装置
CN110517099A (zh) * 2018-05-22 2019-11-29 北京京东尚科信息技术有限公司 用于确定关联供应方的方法和装置

Also Published As

Publication number Publication date
TW201804400A (zh) 2018-02-01
CN107644036A (zh) 2018-01-30
CN107644036B (zh) 2021-06-29

Similar Documents

Publication Publication Date Title
WO2018014771A1 (fr) Procédé, dispositif et système de poussée d'objet de données
KR102220273B1 (ko) 상품 추천 방법 및 이를 이용하는 서버
US10783361B2 (en) Predictive analysis of target behaviors utilizing RNN-based user embeddings
US9563705B2 (en) Re-ranking results in a search
JP5897019B2 (ja) 候補製品のリンクリストを判定する方法および装置
WO2018196424A1 (fr) Procédé et appareil de recommandation
US11704376B2 (en) Retrieval of content using link-based search
US9767417B1 (en) Category predictions for user behavior
US9269049B2 (en) Methods, apparatus, and systems for using a reduced attribute vector of panel data to determine an attribute of a user
US9767204B1 (en) Category predictions identifying a search frequency
CN111274501B (zh) 推送信息的方法、系统和非暂时性存储介质
CN110264277B (zh) 由计算设备执行的数据处理方法及装置、介质和计算设备
WO2020221022A1 (fr) Procédé de recommandation d'objet de service
US20150278858A1 (en) Method and system for producing segment indexes
US10474670B1 (en) Category predictions with browse node probabilities
CN111967924A (zh) 商品推荐方法、商品推荐装置、计算机设备和介质
KR101639656B1 (ko) 광고 제공 방법 및 서버 장치
US11210341B1 (en) Weighted behavioral signal association graphing for search engines
US10387934B1 (en) Method medium and system for category prediction for a changed shopping mission
WO2020233228A1 (fr) Procédé et appareil destinés à pousser des informations
US11062371B1 (en) Determine product relevance
CN110110199B (zh) 信息输出方法和装置
JP6865706B2 (ja) 情報処理装置、情報処理方法、及び情報処理プログラム
CN112488355A (zh) 基于图神经网络预测用户评级的方法和装置
JP6007300B1 (ja) 算出装置、算出方法および算出プログラム

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: 17830405

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: 17830405

Country of ref document: EP

Kind code of ref document: A1