CN110162714A - Content delivery method, calculates equipment and computer readable storage medium at device - Google Patents
Content delivery method, calculates equipment and computer readable storage medium at device Download PDFInfo
- Publication number
- CN110162714A CN110162714A CN201910092267.0A CN201910092267A CN110162714A CN 110162714 A CN110162714 A CN 110162714A CN 201910092267 A CN201910092267 A CN 201910092267A CN 110162714 A CN110162714 A CN 110162714A
- Authority
- CN
- China
- Prior art keywords
- data
- content
- known users
- preference
- friend
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003860 storage Methods 0.000 title claims abstract description 28
- 238000002716 delivery method Methods 0.000 title claims abstract description 11
- 238000010801 machine learning Methods 0.000 claims abstract description 61
- 238000000034 method Methods 0.000 claims abstract description 46
- 239000011159 matrix material Substances 0.000 claims description 97
- 239000013598 vector Substances 0.000 claims description 47
- 230000015654 memory Effects 0.000 claims description 20
- 230000004044 response Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 8
- 238000000513 principal component analysis Methods 0.000 claims description 8
- 239000002609 medium Substances 0.000 description 30
- 238000012549 training Methods 0.000 description 19
- 230000006870 function Effects 0.000 description 12
- 238000012545 processing Methods 0.000 description 10
- 230000035606 childbirth Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000000712 assembly Effects 0.000 description 2
- 238000000429 assembly Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000011521 glass Substances 0.000 description 2
- 230000008676 import Effects 0.000 description 2
- 230000006855 networking Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 206010034719 Personality change Diseases 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 239000012120 mounting media Substances 0.000 description 1
- 210000003205 muscle Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 229910052710 silicon Inorganic materials 0.000 description 1
- 239000010703 silicon Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Information Transfer Between Computers (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to content delivery method, content push device, calculate equipment and computer readable storage medium.Content delivery method includes: the individuation data and social networks data for obtaining potential user;Based on acquired individuation data and social networks data, input data associated with the potential user is constructed;The input data is input at least one Machine learning classifiers, wherein each of at least one described Machine learning classifiers are each configured to generate the output data for indicating that whether the preference of the potential user matches with object content based on the input data;And the output data generated depending at least one described Machine learning classifiers, selectively initiate object content described in the client push to the potential user.The matching degree between push content and audient's preference can be improved in this method, so that content delivery is more targetedly and more efficient.
Description
Technical field
The present invention relates to machine learning techniques field, relate in particular to a kind of content delivery method, content push device,
Calculate equipment and computer readable storage medium.
Background technique
Content push based on user property is used widely in internet industry.For example, number of site mentions
The service for being referred to as " guessing that you like " is supplied, wherein when user clicks in application program (" APP ") or the page of browser
After the link of one content (for example, news, video, commodity), other than the page of this content is presented, arrive and this
The link of the similar other content of content is also presented.
Such service is typically based on machine learning techniques, wherein computer according to big data understand user preference and
Predict possible result.However, existing content delivery method is in terms of accuracy rate and validity, there are problems, so that being pushed
Content to user may not be that the user likes.This leads to lower information delivery efficiency.
Summary of the invention
Advantageously, providing a kind of can be improved for user to the discrimination of the preference of object content and therefore improves
The specific aim of content push and the mechanism of accuracy rate.
According to an aspect of the present invention, a kind of content delivery method is provided, comprising: obtain the personalized number of potential user
According to social networks data;Based on acquired individuation data and social networks data, construct related to the potential user
The input data of connection;The input data is input at least one Machine learning classifiers, wherein at least one described machine
Each of Study strategies and methods are each configured to the preference for indicating the potential user is generated based on the input data whether
The output data to match with object content;And the output number generated depending at least one described Machine learning classifiers
According to selectively initiating object content described in the client push to the potential user.
In some embodiments, object content described in the client push selectively initiated to the potential user
It include: that the output data generated in response to each of at least one described Machine learning classifiers indicates the potential use
The preference at family matches with the object content, object content described in the client push of Xiang Suoshu potential user;And response
In the output data that any of at least one described Machine learning classifiers generate indicate the preference of the potential user with
The object content mismatches, not object content described in the client push to the potential user.
In some embodiments, each of at least one described Machine learning classifiers are instructed by following steps
It gets: obtaining the respective individuation data of multiple known users and social networks data;Based on acquired respective individual character
Change data and social networks data, building respectively to the associated corresponding input data of the multiple known users;Offer and institute
The associated respective objects output data of multiple known users is stated, wherein target output data associated with each known users
Indicate whether the preference of the known users matches with the object content;And by the corresponding input data and described corresponding
Target output data inputs each Machine learning classifiers for training.
In some embodiments, it is described building respectively to the associated corresponding input data packet of the multiple known users
Include: for each known users: the individuation data based on the known users generates the first matrix;Based on the known users
Property data and social networks data generate the second matrix, wherein multiple friends of second matrix description known users with
Between the known users in terms of personal attribute between respective similarity and the multiple friend and the known users right
The respective similarity in terms of the personal preference of n content, n is natural number, and wherein the n content is different from the mesh
Mark content;First matrix and second matrix are subjected to physics merging, to form third matrix;And reduce described the
The dimension of three matrixes, to obtain the 4th matrix as input data associated with the known users.
In some embodiments, the individuation data of each known users includes: in m attribute for describe the known users
Respective attributes m attribute value;With indicate the known users preference whether with the corresponding contents phase in the n content
Matched n preference numerical value, m are natural number.The social networks data of each known users include multiple friends of the known users
The respective individuation data of friend, the individuation data of each friend include: the respective attributes in m attribute for describe the friend
M attribute value;With n preference number for indicating that whether the preference of the friend matches with the corresponding contents in the n content
Value.
In some embodiments, the individuation data based on the known users generate the first matrix include: by this
Know that m attribute value vector of the individuation data of user turns to row vector as first matrix.
In some embodiments, the individuation data and social networks data based on the known users generates the second square
Battle array includes: the personalized number of the corresponding m attribute value and the known users from the respective individuation data of the multiple friend
According to m attribute value, export between the multiple friend and the known users the respective similarity in terms of personal attribute;From
N of the individuation data of the corresponding n preference numerical value and known users of the multiple respective individuation data of friend are partially
Good numerical value exports inclined in the individual for each content in the n content between the multiple friend and the known users
The good respective similarity of aspect;By institute derived between the multiple friend and the known users in terms of personal attribute it is respective
Similarity vector turns to the first column vector;By derived from institute between the multiple friend and the known users for the n
Respective similarity vector turns to corresponding second column vector in terms of the personal preference of each content in content;And it will
Each second column vector of first row vector sum is concatenated in the row direction to form second matrix.
In some embodiments, respective in terms of personal attribute between described export the multiple friend and the known users
Similarity include: m of individuation data by m attribute value of the individuation data of each friend Yu the known users
Correspondence numerical value in attribute value is compared;The number for indicating equal comparison result is counted;And described in determining
Similarity of the ratio of number and m between the friend and the known users in terms of personal attribute.
In some embodiments, for the n content between described export the multiple friend and the known users
In each content personal preference in terms of respective similarity include: by n preference number of the individuation data of each friend
Value is compared with the corresponding numerical value in n preference numerical value of the individuation data of the known users;Refer in response to the comparison
Show that the friend and the known users have the identical preference for the same content in the n content, with this by the friend
Know that the similarity between user in terms of the personal preference for the content is set as predetermined value;And in response to the comparison
It indicates that the friend and the known users have the difference preference for the same content in the n content, by the friend and is somebody's turn to do
The personal preference of similarity between known users in terms of to(for) the content is set as zero.
In some embodiments, described to carry out physics to merge including: to be expert at by first matrix and second matrix
First matrix and second matrix are concatenated to form the third matrix on direction.
In some embodiments, the dimension for reducing the third matrix include: to the third matrix carry out it is main at
Analysis.
In some embodiments, the target output data of each known users include indicate the known users preference whether
The numerical value to match with the object content.
In some embodiments, the building input data associated with the potential user includes: potential based on this
The individuation data of user generates the first matrix;Individuation data and social networks data based on the potential user generate second
Matrix, wherein each in terms of personal attribute between the multiple friends and the potential user of second matrix description potential user
From similarity and the multiple friend and the potential user between in terms of the personal preference for n content respective phase
Like degree, n is natural number, and wherein the n content is different from the object content;By first matrix and described second
Matrix carries out physics merging, to form third matrix;And reduce the dimension of the third matrix, using obtain the 4th matrix as
Input data associated with the potential user.
In some embodiments, the individuation data of potential user includes: the phase in m attribute for describe the potential user
Answer m attribute value of attribute;Whether match with the corresponding contents in the n content with the preference for indicating the potential user
N preference numerical value, m is natural number.The social networks data of potential user include that multiple friends of the potential user are respective
Individuation data, the individuation data of each friend include: m attribute of the respective attributes in m attribute for describe the friend
Numerical value;With n preference numerical value for indicating that whether the preference of the friend matches with the corresponding contents in the n content.
In some embodiments, it includes: that this is dived that the individuation data based on the potential user, which generates the first matrix,
Row vector is turned to as first matrix in m attribute value vector of the individuation data of user.
In some embodiments, the individuation data and social networks data based on the potential user generates the second square
Battle array includes: the personalized number of the corresponding m attribute value and the potential user from the respective individuation data of the multiple friend
According to m attribute value, export between the multiple friend and the potential user the respective similarity in terms of personal attribute;From
N of the individuation data of the corresponding n preference numerical value and potential user of the multiple respective individuation data of friend are partially
Good numerical value exports inclined in the individual for each content in the n content between the multiple friend and the potential user
The good respective similarity of aspect;By institute derived between the multiple friend and the potential user in terms of personal attribute it is respective
Similarity vector turns to the first column vector;By derived from institute between the multiple friend and the potential user for the n
Respective similarity vector turns to corresponding second column vector in terms of the personal preference of each content in content;And it will
Each second column vector of first row vector sum is concatenated in the row direction to form second matrix.
In some embodiments, respective in terms of personal attribute between described export the multiple friend and the potential user
Similarity include: m of individuation data by m attribute value of the individuation data of each friend Yu the potential user
Correspondence numerical value in attribute value is compared;The number for indicating equal comparison result is counted;And described in determining
Similarity of the ratio of number and m between the friend and the potential user in terms of personal attribute.
In some embodiments, for the n content between described export the multiple friend and the potential user
In each content personal preference in terms of respective similarity include: by n preference number of the individuation data of each friend
Value is compared with the corresponding numerical value in n preference numerical value of the individuation data of the potential user;Refer in response to the comparison
Show that the friend and the potential user have the identical preference for the same content in the n content, which is dived with this
The similarity in terms of the personal preference for the content is set as predetermined value between users;And in response to the comparison
It indicates that the friend and the potential user have the difference preference for the same content in the n content, by the friend and is somebody's turn to do
The personal preference of similarity between potential user in terms of to(for) the content is set as zero.
In some embodiments, described to carry out physics to merge including: to be expert at by first matrix and second matrix
First matrix and second matrix are concatenated to form the third matrix on direction.
In some embodiments, the dimension for reducing the third matrix include: to the third matrix carry out it is main at
Analysis.
According to another aspect of the present invention, a kind of content push device is provided, comprising: for obtaining of potential user
The device of property data and social networks data;For based on acquired individuation data and social networks data, building with
The device of the associated input data of potential user;For the input data to be input at least one machine learning point
The device of class device, wherein each of at least one described Machine learning classifiers are each configured to based on the input data
Generate the output data for indicating that whether the preference of the potential user matches with object content;And described in being used to depend on extremely
The output data that few Machine learning classifiers generate, is selectively initiated described in the client push to the potential user
The device of object content.
According to another aspect of the present invention, a kind of content push device is provided, comprising: obtain module, be configured to obtain
Take the individuation data and social networks data of potential user;Module is constructed, is configured to based on acquired individuation data
With social networks data, input data associated with the potential user is constructed;Input module is configured to the input
Data are input at least one Machine learning classifiers, wherein each of at least one described Machine learning classifiers quilt
It is configured to generate the output number for indicating whether the preference of the potential user matches with object content based on the input data
According to;And pushing module, it is configured to depend on the output data that at least one described Machine learning classifiers generate, selectivity
Object content described in client push of the ground to the potential user.
According to another aspect of the present invention, a kind of calculating equipment, including memory and processor, the memory are provided
It is configured to store computer program instructions on it, the computer program instructions ought be performed rush on the processor
The processor is set to execute method described in first aspect.
According to another aspect of the present invention, a kind of computer readable storage medium is provided, stores computer program thereon
Instruction, the computer program instructions promote the processor to execute side described in first aspect when performing on a processor
Method.
According to the embodiment being described below, these and other aspects of the invention will be apparent it is clear, and
It will be elucidated with reference to the embodiment being described below.
Detailed description of the invention
Below in conjunction with attached drawing in the description of exemplary embodiment, more details of the invention, feature and advantage quilt
It is open, in the accompanying drawings:
Fig. 1 shows the flow chart of the operation in the training stage according to an embodiment of the present invention for Machine learning classifiers;
Fig. 2A shows the example of the various dimensions of the essential attribute of description user according to an embodiment of the present invention;
Fig. 2 B shows the example of the various dimensions of the interest attribute of description user according to an embodiment of the present invention;
Fig. 2 C shows reflection user and its whether friend likes the exemplary user interface of a certain content;
Fig. 3 shows the instantiation procedure that input data is constructed in the operation of Fig. 1;
Fig. 4 shows the schematically and exemplarily diagram of content delivery method according to an embodiment of the present invention;
Fig. 5 shows the schematically and exemplarily figure for the user interface being pushed at the client of the potential user of object content
Show;
Fig. 6 shows the schematic block diagram of content push device according to an embodiment of the present invention;And
Fig. 7 shows an example system comprising represents one or more systems that various techniques described herein may be implemented
The Example Computing Device of system and/or equipment.
Specific embodiment
Present inventors appreciate that constructing the method for the input data of Machine learning classifiers in classification accuracy side
Generation good result is faced to have a very big impact.Specifically, by being introduced into friend of the user in social networking service (SNS)
The attribute and preference of friend can provide more dimensions to judge whether the preference of user matches with content to be pushed, thus
Improve the classification accuracy of classifier.In addition, carrying out ballot decision using multiple Machine learning classifiers, can be further improved
The matching degree being pushed between content and the preference of user.Based on such opinion, the solution that will be discussed in more detail below is proposed
Certainly scheme.
Before the embodiments of the invention are explained in detail, several terms used herein are defined.
1, Machine learning classifiers.Term " Machine learning classifiers " refers to the machine for being designed to solve classification problem
Learning model.Specifically, the training sample that classifier can use marked good classification is trained, for judge one it is new
Observe classification belonging to sample.The example of Machine learning classifiers includes but is not limited to: neural network is (for example, convolutional Neural net
Network CNN, Recognition with Recurrent Neural Network RNN, etc.) and support vector machines (SVM).
2, content.Term " content " can be generically referred to via the logical of such as internet, Intranet, telecommunication network etc
Believe the information of infrastructure distribution, video, audio, picture, text etc..Content can describe specific object, such as newly
News event, TV guide, film screening, buyer's guide, advertising campaign, discount coupon etc..
3, object content.Term " object content " refers to the content of user to be pushed to.
4, individuation data.Term " individuation data " is data associated with the user, for describing of the user
It is humanized and its for one or more contents personal preference.
5, social networks data.Term " social networks data " is data associated with the user comprising the user exists
The individuation data of one or more friends in social networking service.
6, known users.Term " known users " refers to state that whether its preference matches with object content for machine
It is known user for Study strategies and methods.The individuation data and social networks data of known users are building machine learning point
The basis of the training data of class device.
7, potential user.Term " potential user " refers to state that whether its preference matches with object content for machine
It is unknown user for Study strategies and methods.For Machine learning classifiers, potential user is a new observation sample.
8, client.Term " client " may refer to various types of equipment, such as desktop computer, server
Computer, laptop or netbook computer, mobile device (for example, tablet computer or phablet equipment, honeycomb or
Other radio telephones (for example, smart phone), notepad computers, mobile station), wearable device (for example, glasses, wrist-watch),
Amusement equipment (for example, amusement appliance, be communicably coupled to set-top box, the game machine of display equipment), TV or other displays are set
The application program run in standby, automobile computer etc. or the various types of equipment.
Present invention proposition predicts whether the preference of potential user matches with object content using Machine learning classifiers,
And the object content selectively thus is pushed to potential user.The use of Machine learning classifiers includes the training stage and holds
Row order section, they temporally and spatially can be independent.Training stage relates to the use of the training sample of marked good classification
This is trained classifier, and executes the stage and relate to the use of housebroken classifier and classify to new observation sample.
How Machine learning classifiers in the execution stage are used for content push in order to better understand, introduce machine learning first below
The training stage of classifier.
Fig. 1 shows the flow chart of the method 100 of training machine Study strategies and methods according to an embodiment of the present invention.
At step 110, the respective individuation data of multiple known users and social networks data are obtained.
The individuation data of each known users may include:
M attribute value of the respective attributes in m attribute of the known users is described;With
Indicate the n preference the numerical value whether preference of the known users matches with the corresponding contents in n content.M and n
For natural number.
M attribute value describes the respective attributes in the m attribute of user.As an example, not a limit, m attribute can be with
Including one or more of the following terms:
Location, for example, city, administrative area, the common geographical location in the past period (for example, one month) etc.;
Gender;
Age, for example, specific to how old, affiliated age bracket etc.;
Mobile phone parameters, for example, OS Type (for example, iOS or Android), brand, model, the network class supported
Type etc.;
Marriage and childbirth situation, such as, if it is married, whether bring up child etc.;
Level of education, for example, doctor, master, undergraduate education etc.;
Topic of interest, for example, house property, automobile, education, sport, health, game etc..
Some in personal attribute, which can be, to be collected when user registers to website or social activity APP.It is alternatively or attached
Add ground, some in personal attribute, which can be, to be counted and recorded during user uses browser or social activity APP.There are each
Kind of mode obtains these personal attributes.Fig. 2A shows the example of the basic representation data of wechat user, and Fig. 2 B is shown
The examples of the interest tags data of wechat user.In the case where wechat, it can be obtained from the basic representation data of user
Essential attribute in personal attribute, for example, location, gender, mobile phone parameters, marriage and childbirth situation, level of education etc., and can be from
The various interest tags of user obtain the topic of interest in personal attribute.
In some embodiments, these personal attributes can be already saved in database, and their acquisition relates to
And data are fetched from some storage location (for example, Local or Remote database).In the embodiment being described below, m attribute
Numerical value can be used as such as integer value.For example, male is 1, and female 0 in gender;In mobile phone parameters, Android 1, iOS is
0;In marriage and childbirth situation, unmarried is 0, it is married and without child be 1, it is married and have child be 2;In level of education, undergraduate education
And the above are 1, the following are 0 for undergraduate education;In topic of interest, interested in some topic is 1, and loseing interest in is 0, etc.
Deng.
Whether the preference of n preference numerical value user matches with the corresponding contents in n content.With the individual of user
Attribute is similar, and user can be the personal preference of some content to be collected when user registers to website or social activity APP.
Alternatively or cumulatively, it, which can be, during user is using browser or social activity APP is counted and is recorded.
Still by taking wechat as an example, Fig. 2 C shows the schematic of " circle of friends " of wechat user (calling " user A " in the following text)
User interface, wherein the user A has shared the Positive evaluation to a song " Sa Linna ".This shows that user A likes the content
(that is, song " Sa Linna ").
In embodiments, n preference numerical value can be used as such as binary value.For example, for content x
The preference of (x is natural number, and x≤n), numerical value " 1 " instruction user matches with content x, and numerical value " 0 " indicates user's
Preference and content x are mismatched.
The social networks data of each known users may include the respective personalized number of multiple friends of the known users
According to.Similarly, the individuation data of each friend includes:
M attribute value of the respective attributes in the m attribute of the friend is described;With
Indicate the n preference the numerical value whether preference of the friend matches with the corresponding contents in the n content.
The individuation data of friend can be obtained in a manner of identical with the individuation data of known users described above
It takes.For example, the song " Sa Linna " that friend B, C of user A shares user A with D carries out in user interface shown in fig. 2 C
Comment, wherein friend B and C give Positive evaluation, show that they like this content, and friend D gives passiveness and comments
Valence shows that he does not like this content.In this way, the respective preference of friend B, C and D for the content (" song " Sa Linna " ")
Numerical value can be collected and record.The attribute value of friend can also be with the attribute value with known users described above
It obtains identical mode to be acquired, repeat no more for simplicity.
In embodiment, the social networks of user can be obtained from the server of social APP and its associated database.
For example, for each user, server is safeguarded social networks chain, can be determined from the social networks chain in the case where wechat
Multiple friends of the user.In another example in microbloggingTMIn the case where, for each user, server also safeguards social networks chain,
It is by the user, the user other users of interest and pays close attention to the other users of the user and links together.From such
Social networks chain can position friend of some specific user in social networks, and collect or take in turn from database
Return the individuation data of these friends.
At step 120, based on the acquired respective individuation data of the multiple known users and social networks number
According to, building respectively to the associated corresponding input data of the multiple known users.Fig. 3 shows the instantiation procedure of step 120.
The process is carried out for each user in the multiple known users.
With reference to Fig. 3, at step 121, the individuation data based on the known users generates the first matrix.In some implementations
In example, m attribute value in the individuation data of the known users is quantified as row vector as first matrix.Example
Such as, it is assumed that user A above-mentioned has 8 attribute values in table 1-1 as follows, then the first matrix can be expressed as.Such first matrix features user A from personal factor.
Table 1-1
Gender | Mobile phone parameters | Marriage and childbirth situation | Level of education | Topic 1 | Topic 2 | Topic 3 | Topic 4 |
1 | 0 | 1 | 1 | 0 | 1 | 1 | 0 |
At step 122, individuation data and social networks data based on the known users generate the second matrix, wherein
It is respective similar in terms of personal attribute between multiple friends of second matrix description known users and the known users
It is respective similar in terms of the personal preference for the n content between degree and the multiple friend and the known users
Degree.In embodiment, this can be by following operation (i) to (v) executing.
(i) from the individual character of corresponding the m attribute value and the known users of the respective individuation data of the multiple friend
M attribute value for changing data, exports respective similar in terms of personal attribute between the multiple friend and the known users
Degree.Still by taking user A above-mentioned as an example, it is assumed that there are four friend B, C, D and E for user A tool, they are by following table 1-
Attribute shown in 2 is portrayed.
Table 1-2
Gender | Mobile phone parameters | Marriage and childbirth situation | Level of education | Topic 1 | Topic 2 | Topic 3 | Topic 4 | |
Friend B | 0 | 1 | 1 | 0 | 0 | 0 | 0 | 1 |
Friend C | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 1 |
Friend D | 1 | 0 | 1 | 1 | 0 | 0 | 1 | 1 |
Friend E | 1 | 0 | 1 | 1 | 0 | 1 | 1 | 0 |
In some embodiments, respective similarity can in terms of personal attribute between friend B, C, D and E and the user A
With by the number of same alike result numerical value respective between friend B, C, D and E and the user A and the total m(of personal attribute's numerical value at this
It is ratio 8) in example to measure.For example, for friend B, by 8 attribute values " 01100001 " of friend B respectively with
8 attribute values " 10110110 " of family A are compared.Then, to the equal attribute value between the friend B and the user A
Number counted.Specifically, the number of the equal attribute value between friend B and the user A is counted as 2.Then,
Similarity between friend B and user A in terms of personal attribute is 2/8=0.25.In the same way, friend C, D and E and use
It is 0.5,0.75 and 1 that respective similarity can measure respectively in terms of personal attribute between the A of family.
(ii) from the individual character of corresponding the n preference numerical value and the known users of the respective individuation data of the multiple friend
N preference numerical value for changing data, exports between the multiple friend and the known users for each of described n content
Respective similarity in terms of the personal preference of content.Continue the example of user A, it is assumed that user A and its four friends B, C, D and E
Having the preference as shown in following table 1-3 for n content (in this example, n=5), (1 instruction is liked, and 0 instruction is not
Like):
Table 1-3
Content 1 | Content 2 | Content 3 | Content 4 | Content 5 | |
User A | 1 | 0 | 1 | 0 | 1 |
Friend B | 1 | 1 | 0 | 1 | 0 |
Friend C | 1 | 0 | 1 | 0 | 1 |
Friend D | 1 | 1 | 1 | 1 | 1 |
Friend E | 1 | 0 | 1 | 1 | 0 |
Specifically, by n preference numerical value in the individuation data of each friend in described friend B, C, D and E (at this
In example, n=5) with the correspondence numerical value in n preference numerical value (in this example, n=5) in the individuation data of the user A into
Row compares.For example, for friend B, will indicate its whether like 5 preference numerical value " 11010 " of content 1 to 5 respectively with instruction
The 5 preference numerical value " 10101 " whether user A likes content 1 to 5 are compared.
If the comparison indicates that the friend and the user A have for the identical of the same content in 5 contents
The personal preference of similarity between the friend and the user A in terms of to(for) the content is then set as predetermined value by preference,
Such as 1.For example, for friend B, since both the friend B and the user A have the identical preference numerical value 1 for content 1,
So the similarity between the friend B and the user A in terms of the personal preference for content 1 can be set as 1.
If the comparison indicates that the friend and the user A have the difference for the same content in 5 contents
The personal preference of similarity between the friend and the user A in terms of to(for) the content is then set as zero by preference.For example,
For friend B, due to the friend B likes content 2 and 4 and the user A does not like content 2 and 4, it is possible to by the friend B with
The personal preference of similarity between the user A in terms of to(for) content 2 and 4 is set as 0.Similarly, since the friend B is not liked
Joyous content 3 and 5 and the user A likes content 3 and 5, it is possible to by between the friend B and the user A for content 3 and 5
Personal preference in terms of similarity be set as 0.
In the same way, in the personal preference side for content 1 to 5 between available friend C, D and E and user A
The respective similarity in face.
(iii) by institute derived between the multiple friend and the known users in terms of personal attribute respective similarity
Vector turns to first row vector.In the example of user A, due between friend B, C, D and E and user A in terms of personal attribute
Respective similarity is respectively 0.25,0.5,0.75 and 1, so obtaining the first following column vector:
。
By derived from institute between the multiple friend and the known users interior for each of described n content
Respective similarity vector turns to corresponding second column vector in terms of the personal preference of appearance.In the example of user A, friend
B, respective similarity is respectively 1,1,1 and 1 in terms of the personal preference for content 1 between C, D and E and user A;Friend B,
C, respective similarity is respectively 0,1,0 and 1 in terms of the personal preference for content 2 between D and E and user A;Friend B, C,
Respective similarity is respectively 0,1,1 and 1 in terms of the personal preference for content 3 between D and E and user A;Friend B, C, D
Respective similarity is respectively 0,1,0 and 0 in terms of the personal preference for content 4 between E and user A;Friend B, C, D and
Respective similarity is respectively 0,1,1 and 0 in terms of the personal preference for content 5 between E and user A.Therefore, it obtains as follows
Shown in 5 the second column vectors:
、、、、。
(v) each second column vector of the first row vector sum is concatenated into (concatenate) in the row direction
To form second matrix.In the example of user A, by 5 the second column vector concatenations of first row vector sum, following institute is obtained
The second matrix shown.Such second matrix features user A from social networks dimension.
。
It will be appreciated that operation is not (i) to needing (v) to be sequentially executed with described above.For example, operation is (iii)
It can immediately operate and (i) execute, and operate (iv) can immediately operate and (ii) execute.In another example operation can (ii) grasp
It is performed before making (i).
At step 123, first matrix and second matrix are subjected to physics merging, to form third matrix.
Term " physics merging " merges different from mathematical operation, and can be understood as meaning to concatenate.This is remained about in a Genus Homo
The minutia of similarity in terms of property and the similarity in terms of the personal preference for content.In some embodiments, exist
First matrix and second matrix are concatenated to form the third matrix on line direction.In the example of user A
In, by the first matrixWith the second matrixMerge and obtains as follows the
Three matrixes:
,
Wherein blank position is filled 0.Such third matrix features user from both personal factor and social networks dimension
A。
At step 124, the dimension of the third matrix is reduced, to obtain the 4th matrix as related to the known users
The input data of connection.
Step 124 aims to solve the problem that sparsity problem caused by third matrix is filled 0 blank position, reduces and calculates
Complexity.In some embodiments, the reduction of dimension can be by carrying out principal component analysis (PCA) to the third matrix come real
It is existing.In other embodiments, other Method of Data with Adding Windows can be used.Initial data is transformed to one by linear transformation by PCA
The expression of each dimension linear independence of group, realizes the extraction of the main feature component of data.After the operation of PCA dimensionality reduction, the 4th matrix
With columns identical with third matrix, but the line number of reduction, such as 1 row or 2 rows.The example pseudo-code of PCA dimensionality reduction operation is shown
It is as follows out:
import numpy as np
from sklearn.decomposition import PCA
A=np.array ([third matrix])
Pca=PCA (n_components=1) // dimensionality reduction to a line
pca.fit(a)
print(pca.explained_variance_ratio_)
Print (pca.explained_variance_) // the 4th matrix of output
In this way, available corresponding 4th matrix is as input data for each user in multiple known users.
It constructs as a result, and the associated corresponding input data of multiple known users.These input datas constitute Machine learning classifiers
Training dataset.
For the purpose of the training of Machine learning classifiers, it is also necessary to provide target output data for classifier.Target is defeated
Data are exactly " answer " of classification problem out.In the present context, classifier is for being different from content 1 to n, to be pushed
Object content and be trained to.Therefore, for each known users, target output data is exactly the preference of the known users
The state whether to match with the object content.
Referring back to Fig. 1, at step 130, respective objects output number associated with the multiple known users is provided
According to.In some embodiments, for each user in the multiple known users, provide indicate the user preference whether with
The numerical value that object content matches is as output data associated with the known users.For example, if user likes in target
Hold, then target output data is set to 1;If user does not like object content, target output data is set to 0.?
In some embodiments, indicate target output data that whether preference of known users matches with object content can with above
It is obtained about the same way of the preference numerical value description in individuation data.
In this way, being directed to each known users, the corresponding training data for Machine learning classifiers is obtained (that is, the 4th
Matrix) and corresponding target output data (that is, answer of classification problem).
At step 140, the corresponding input data and the respective objects output data are inputted into the machine learning
Classifier is for training.
In the case where convolutional neural networks, the input data as the 4th matrix can directly input convolutional Neural net
Network, because convolutional neural networks are good at directly processing two-dimensional matrix.Convolutional neural networks are widely used in image procossing, wherein
Two-dimensional pixel data matrix is directly inputted convolutional neural networks as input.In the case where Recognition with Recurrent Neural Network, the 4th square
Data in battle array can be read line by line and the data in every a line are provided in the input layer of Recognition with Recurrent Neural Network
One corresponding input node.In the case where support vector machines, the data in the 4th matrix can be read line by line and
It sequentially concatenates to form an input vector.Support vector machines is trained to construct a hyperplane to carry out input vector
Classify (liking or do not like object content).For other machines learning model, the mapping for inputting data into model can regard feelings
Condition and configure, details are not described herein for simplicity.
Although in the above embodiments, step 130 is described as serially executing with step 110 and 120, this hair
It is bright without being limited thereto.For example, step 130 can be performed in parallel before step 120 or with step 120.
Using the training dataset and corresponding target output data being such as constructed above, Machine learning classifiers can be instructed
Practice hidden between the state whether attribute and social networks and the preference of user for excavating user match with object content
Containing relevance.Due to not only considering the attribute and preference of user oneself, but also the attribute and preference of the friend of user are also contemplated,
Machine learning classifiers are expected to have more strict decision logic, to improve classification accuracy.Housebroken machine learning
Classifier can be used for predicting whether potential user's (that is, new observation sample) except known users likes object content.
Once training obtains Machine learning classifiers, so that it may using trained Machine learning classifiers come in performance objective
Hold the push of potential user.In a specific example, it is assumed that Machine learning classifiers have been directed in target to be pushed
Hold, trained according to the operation of training stage described above, wherein descriptions above 1 to n is respectively inhomogeneity
Other commodity or n discount coupon of service (discount coupon 1,2,3 ..., n), and object content to be pushed is that some is newly shown
Film " XYZ " discount coupon.It is described below with reference to Figure 4 and 5 according to an embodiment of the present invention based at least one machine learning
The content delivery method of classifier.
Fig. 4 shows the schematically and exemplarily diagram of content delivery method 400 according to an embodiment of the present invention.
At step 410, the individuation data and social networks data of potential user (also referred to as " user a ") are obtained.Step
Rapid 410 correspond to the step 110 in the training stage described above.
The individuation data of potential user a also includes:
M attribute value of the respective attributes in the m attribute of the potential user a is described;With
Indicate the n preference the numerical value whether preference of the potential user a matches with the corresponding contents in n content.
It is assumed that potential user a have table 2-1 in 8 attribute values (m=8) and table 2-2 in 5 preference numerical value (n=
5).
Table 2-1
Gender | Mobile phone parameters | Marriage and childbirth situation | Level of education | Topic 1 | Topic 2 | Topic 3 | Topic 4 |
1 | 0 | 1 | 1 | 0 | 1 | 1 | 0 |
Table 2-2
Content 1(discount coupon 1) | Content 2(discount coupon 2) | Content 3(discount coupon 3) | Content 4(discount coupon 4) | Content 5(discount coupon 5) | |
Potential user a | 1 | 0 | 1 | 0 | 1 |
Also, the social networks data of potential user a also include the respective personalization of multiple friends of the potential user a
Data.The individuation data of each friend includes:
M attribute value of the respective attributes in the m attribute of the friend is described;With
Indicate the n preference the numerical value whether preference of the friend matches with the corresponding contents in the n content.
It is assumed that four friends b, c, d and e of potential user a have the preference number in attribute value and table 2-4 in table 2-3
It is worth (in this example, m=8, n=5).
Table 2-3
Gender | Mobile phone parameters | Marriage and childbirth situation | Level of education | Topic 1 | Topic 2 | Topic 3 | Topic 4 | |
Friend b | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 |
Friend c | 1 | 1 | 0 | 0 | 1 | 1 | 0 | 1 |
Friend d | 1 | 0 | 1 | 1 | 0 | 1 | 1 | 1 |
Friend e | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 0 |
Table 2-4
Content 1(discount coupon 1) | Content 2(discount coupon 2) | Content 3(discount coupon 3) | Content 4(discount coupon 4) | Content 5(discount coupon 5) | |
Friend b | 1 | 0 | 0 | 1 | 0 |
Friend c | 0 | 0 | 1 | 0 | 1 |
Friend d | 1 | 1 | 0 | 1 | 1 |
Friend e | 1 | 1 | 1 | 1 | 0 |
At step 420, based on the individuation data and social networks data of the acquired potential user a, building
Input data associated with the potential user a.Step 420 corresponds to the step 120 in the training stage described above, and
And it is described herein referring to step 120.Step 420 may include following operation (1) to (4).
(1) individuation data based on the potential user a generates the first matrix.It is and latent according to the attribute value in table 2-1
It can be expressed as in associated first matrix of user a。
(2) individuation data and social networks data based on the potential user a generate the second matrix.This can pass through behaviour
Make (2-1) to (2-5) Lai Shixian.
(2-1) is from the corresponding m attribute value of the multiple respective individuation data of friend b, c, d and e and this is potential
M attribute value of the individuation data of user a exports between the multiple friend b, c, d and e and the potential user a a
The humanized respective similarity of aspect.Specifically, by the attribute value of each friend in friend b, c, d and e respectively with it is potential
Correspondence numerical value in the attribute value of user a is compared.Then, to the equal attribute between the friend and the potential user a
The number of numerical value is counted.M=8 in this example the number of equal attribute value and m() ratio be confirmed as the friend with
Similarity between potential user a in terms of personal attribute.In this way, in a Genus Homo between friend b, c, d and e and potential user a
Property the respective similarity of aspect can to measure respectively be 0.375,0.25,0.875 and 0.75.
(2-2) is from the corresponding n preference numerical value of the multiple respective individuation data of friend b, c, d and e and this is potential
N preference numerical value of the individuation data of user a, exports between the multiple friend b, c, d and e and the potential user a right
Respective similarity in terms of the personal preference of each content in the n content.Specifically, by described friend b, c, d and e
In each friend individuation data in 5 preference numerical value and the potential user a individuation data in 5 preference numbers
Correspondence numerical value in value is compared.If the comparison indicates that the friend and the potential user a have for 5 contents
In same content identical preference, then by between the friend and the potential user a in terms of the personal preference for the content
Similarity be set as predetermined value, such as 1.If the comparison indicates that the friend and the potential user a have for described 5
The difference preference of same content in a content, then will be inclined in the individual for the content between the friend and the potential user a
The similarity of good aspect is set as zero.
(2-3) by derived from institute between the multiple friend b, c, d and e and the potential user a in terms of personal attribute it is each
From similarity vector turn to the first column vector.According to (2-1), due between friend b, c, d and e and potential user a in individual
Respective similarity is respectively 0.375,0.25,0.875 and 0.75 in terms of attribute, so obtaining the first following column vector:
。
(2-4) by institute derived between the multiple friend b, c, d and e and the potential user a for the n content
In each content personal preference in terms of respective similarity vector turn to corresponding second column vector.According to (2-2),
Between friend b, c, d and e and potential user a in terms of the personal preference for content 1 respective similarity be respectively 1,0,1 and
1;Respective similarity is respectively 1,1,0 in terms of the personal preference for content 2 between friend b, c, d and e and potential user a
With 0;Between friend b, c, d and e and potential user a in terms of the personal preference for content 3 respective similarity be respectively 0,
1,0 and 1;Respective similarity is respectively in terms of the personal preference for content 4 between friend b, c, d and e and potential user a
0,1,0 and 0;Respective similarity is distinguished in terms of the personal preference for content 5 between friend b, c, d and e and potential user a
It is 0,1,1 and 0.Therefore, 5 the second column vectors as follows are obtained:
、、、、。
(2-5) concatenates each second column vector of the first row vector sum in the row direction to form described
Two matrixes.The second matrix shown below is obtained by 5 the second column vector concatenations of first row vector sum for potential user a.
。
It will be appreciated that operation (2-1) does not need to be sequentially executed with described above to (2-5).For example, operation
(2-3) can immediately operate (2-1) execution, and operate (2-4) can immediately operate (2-2) execution.In another example operation (2-
2) it can be performed before operation (2-1).
(3) first matrix and second matrix are subjected to physics merging, to form third matrix.In the embodiment
In, first matrix and second matrix are concatenated to obtain third matrix as follows in the row direction:
,
Wherein blank position is filled 0.
(4) dimension for reducing the third matrix, to obtain the 4th matrix as input associated with the potential user a
Data.
At step 430, the input data associated with the potential user a is input at least one described machine
Device Study strategies and methods.In the example of fig. 4, though it is shown that the classifier of three neural network forms, but this is only example
Property and it is schematical.In other embodiments, the classifier or other types of more or less (such as one) can be used
Classifier.
Method 400 is related to the execution stage for the Machine learning classifiers trained, rather than the training stage, therefore, in step
Target output data is not necessarily at 430.On the contrary, Machine learning classifiers will be generated based on input data indicates the potential user a's
The output data whether preference matches with object content (discount coupon of film " XYZ ").For example, 1 or the value close to 1 output
Data indicate that the potential user a likes the discount coupon of film " XYZ ", and 0 or indicate that this is potential close to the output data of 0 value
User a does not like the discount coupon of film " XYZ ".
At step 440, depending on the output data that at least one described Machine learning classifiers generate, selectively send out
Play the discount coupon to the client push film " XYZ " of the potential user a.Specifically, if at least one described machine
The output data that each of Study strategies and methods generate indicates that the potential user a likes the discount coupon of film " XYZ ", then
To the discount coupon of the client push film " XYZ " of the potential user a, and if at least one described machine learning classification
The output data that any of device generates indicates that the potential user a does not like the discount coupon of film " XYZ ", then not to described
The discount coupon of the client push film " XYZ " of potential user a.Although will be appreciated that in Fig. 4, each classifier it is defeated
It is provided to one " multiplier " progress operation out, but this is exemplary and schematical.In other embodiments, it can adopt
With other logics to determine whether the output data of each Machine learning classifiers indicates preference and the institute of the potential user a
Object content is stated to match.
Fig. 5 shows the exemplary user interface at the client of potential user a.In this example, potential user a is pre-
Survey is to like the discount coupon of film " XYZ ", and the discount coupon is received in his or her client.
Since Machine learning classifiers not only consider the attribute and preference of potential user a oneself in classification, but also examine
Consider the social networks (attribute and preference of his or she friend) of potential user a, therefore it has and more judges dimension, thus
Improve the accuracy of classification.In this way, object content is pushed to the potential user that its preference really matches with it.This is improved
The specific aim and efficiency of content delivery.Particularly, using multiple Machine learning classifiers, due to multiple engineerings
The ballot decision for practising classifier, can be further improved the prediction whether to match with object content for the preference of potential user
Accuracy.
It will be appreciated that 8 personal attributes described in above embodiment are exemplary, and in other embodiments
In any suitable personal attribute can be used.It will be further understood that although content 1 is to n and wait push in above embodiment
Object content be described as discount coupon, but the invention is not restricted to this.In other embodiments, content 1 is to n and to be pushed
Object content can be any kind of content, such as video, audio, picture, text etc..It is retouched more specifically, content can be
State the information of such as media event, TV guide, film screening, buyer's guide, advertising campaign etc..In addition, content 1 to n and
Object content to be pushed does not need type having the same.For example, content 1 can have video, sound to n and object content
Frequently, the different type in picture and text.Moreover, content 1 can be related to different themes to n and object content.For example, they
One or more can be about media event, one or more of which can be about TV programme, and
One or more of which can be about buyer's guide.
Fig. 6 shows the schematic block diagram of content push device 600 according to an embodiment of the present invention.With reference to Fig. 6, content is pushed away
Sending device 600 includes obtaining module 610, building module 620, input module 630 and pushing module 640.
Obtain individuation data and social networks data that module 610 is configured to obtain potential user.Obtain module 610
Operation have been described above and be described in detail about the embodiment of the method illustrated in conjunction with Fig. 4, and in order to succinctly rise
See and is not repeated herein.
Building module 620 is configured to individuation data and social networks number based on the acquired potential user
According to building input data associated with the potential user.The operation of building module 620 has been described above about in conjunction with Fig. 4
The embodiment of the method illustrated is described in detail, and is not repeated herein for simplicity.
Input module 630 is configured to the input data associated with the potential user being input at least one
Machine learning classifiers.The operation of input module 630 has been described above to be carried out about the embodiment of the method illustrated in conjunction with Fig. 4
Detailed description, and be for simplicity not repeated herein.
Pushing module 640 is configured to depend on the output data that at least one described Machine learning classifiers generate, choosing
Selecting property ground object content described in client push to the potential user.The operation of pushing module 640 have been described above about
The embodiment of the method illustrated in conjunction with Fig. 4 is described in detail, and is not repeated herein for simplicity.
It can be by soft it will be appreciated that obtaining module 610, building module 620, input module 630 and pushing module 640
Part, firmware, hardware or combinations thereof are realized, as will be described further below.
Fig. 7 shows an example system 700 comprising represent may be implemented one of various techniques described herein or
It the Example Computing Device 710 of multiple systems and/or equipment, network 740 and is communicated via network 740 with equipment 710 is calculated
Multiple client 750.
Network 740 can be a variety of different networks, including internet, Local Area Network, telephone network, Intranet, its
His public and/or proprietary network, a combination thereof etc..
Client 750 can be various types of equipment, such as desktop computer, server computer, notebook
Computer or netbook computer, mobile device are (for example, tablet computer or phablet equipment, honeycomb or other radio telephones
(for example, smart phone), notepad computers, mobile station), wearable device (for example, glasses, wrist-watch), amusement equipment (example
Such as, amusement appliance, be communicably coupled to set-top box, the game machine of display equipment), TV or other display equipment, automobile calculate
The application program run on machine etc. or the various types of equipment.
Calculating equipment 710 can be server or any other suitable calculating equipment or the calculating of such as service provider
System, range from the wholly-owned source device with a large amount of memories and processor resource to limited memory and/or from
Manage the low resource device of resource.In some embodiments, meter can be taken above for the content push device 600 of Fig. 6 description
Calculate the form of equipment 710.
Example Computing Device 710 as shown includes the processing system 711 being coupled with each other, one or more computers
Readable medium 712 and one or more I/O Interfaces 713.Although being not shown, calculating equipment 710 can also include being
Bus of uniting or other data and order conveyer system, various assemblies are coupled to each other.System bus may include different bus
Any one or combination of structure, the bus structures such as memory bus or Memory Controller, peripheral bus, general string
Row bus, and/or processor or local bus using any one of various bus architectures.It is contemplated that various other show
Example, such as control and data line.
Processing system 711 represents the function that one or more operations are executed using hardware.Therefore, processing system 711 is schemed
It is shown as including the hardware element 714 that can be configured to processor, functional block etc..This may include being realized within hardware as dedicated
Integrated circuit or the other logical devices formed using one or more semiconductors.Hardware element 714 is not by the material that it is formed
Or in which the limitation of the processing mechanism used.For example, processor can be by (multiple) semiconductor and/or transistor (for example, electronics
Integrated circuit (IC)) composition.In such context, processor-executable instruction can be electronically-executable instruction.
Computer-readable medium 712 is illustrated as including storage/memory 715.715 table of storage/memory
Show memory/memory capacity associated with one or more computer-readable mediums.Storage/memory 715 can wrap
Include Volatile media (such as random-access memory (ram)) and/or non-volatile media (such as read-only memory (ROM), sudden strain of a muscle
It deposits, CD, disk etc.).Storage/memory 715 may include mounting medium (for example, RAM, ROM, Fixed disk drive
Device etc.) and removable medium (for example, flash memory, removable hard disk drive, CD etc.).Computer-readable medium 712 can be with
By be described further below it is various other in a manner of configured.
One or more input/output interfaces 713 represent allow users to calculate equipment 710 key in order and information and
Also allow to present information to user using various input-output apparatus and/or be sent to the function of other assemblies or equipment.It is defeated
The example for entering equipment includes keyboard, cursor control device (for example, mouse), microphone (for example, inputting for voice), scanning
Instrument, touch function (for example, capacitive or other sensors for being configured as detection physical touch), camera are (for example, can use can
See or sightless wavelength (such as infrared frequency) will not to be related to the motion detection touched be gesture), network interface card, receiver etc..
The example of output equipment includes display equipment (for example, monitor or projector), loudspeaker, printer, haptic response apparatus, net
Card, transmitter etc..
Calculating equipment 710 further includes Content push strategy 716.Content push strategy 716 can be used as calculation procedure instruction
It is stored in storage/memory 715.Content push strategy 716 can be together with processing system 711 and I/O interface 713
Realize the repertoire of the modules about Fig. 6 content push device 600 described.
It herein can be in hardware and software element or the general various technologies of described in the text up and down of program module.Generally, this
A little modules include routines performing specific tasks or implementing specific abstract data types, programs, objects, element, component, data knot
Structure etc..Term as used herein " module ", " function " and " component " typicallys represent software, firmware, hardware or combinations thereof.Herein
The technology of description be characterized in it is platform-independent, it is meant that these technologies can be flat in the various calculating with various processors
It is realized on platform.
The realization of described module and technology can store on some form of computer-readable medium or across certain
The computer-readable medium transmission of kind form.Computer-readable medium may include various Jie that can be accessed by calculating equipment 710
Matter.As an example, not a limit, computer-readable medium may include " computer readable storage medium " and " computer-readable letter
Number medium ".
With simple signal transmission, carrier wave or signal itself on the contrary, " computer readable storage medium " is to refer to persistently
The medium and/or equipment of storage information and/or tangible storage device.Therefore, computer readable storage medium refers to non-signal
Bearing medium.Computer readable storage medium include such as volatile and non-volatile, removable and irremovable medium and/or
To be suitable for storage information (such as computer readable instructions, data structure, program module, logic element/circuit or other numbers
According to) the hardware of storage equipment etc realized of method or technique.The example of computer readable storage medium may include but not
It is limited to RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical storages
Device, hard disk, cassette tape, tape, disk storage device or other magnetic storage apparatus or other storage equipment, tangible medium
Or the product suitable for storing expectation information and can be accessed by computer.
" computer-readable signal media ", which refers to be configured as such as sending an instruction to via network, calculates equipment 710
The signal bearing medium of hardware.Signal media typically can by computer readable instructions, data structure, program module or other
Data are embodied in such as modulated data signal of carrier wave, data-signal or other transmission mechanisms.Signal media further includes any
Information transmitting medium.Term " modulated data signal " refer to the information in signal is encoded in this way being arranged or
Change the signal of one or more of its feature.As an example, not a limit, communication media includes such as cable network or straight
The wireless medium of the wired medium of line and such as sound, RF, infrared and other wireless medium in succession.
As previously mentioned, hardware element 714 and computer-readable medium 712 represent the instruction realized in the form of hardware, module,
Programming device logic and/or immobilising device logic, can be used to implement technique described herein in some embodiments
At least some aspects.Hardware element may include integrated circuit or system on chip, specific integrated circuit (ASIC), field-programmable
The component of gate array (FPGA), Complex Programmable Logic Devices (CPLD) and other realizations or other hardware devices in silicon.
In this context, hardware element can be used as instruction, module and/or the logic for executing and being embodied by hardware element and be defined
Program task processing equipment, and for store be used for execution instruction hardware device, for example, previously described calculating
Machine readable storage medium storing program for executing.
Combination above-mentioned can be used for realizing various techniques described herein and module.It therefore, can be by software, hardware
Or program module and other program modules are embodied as on some form of computer readable storage medium and/or by one or more
The one or more instructions and/or logic that a hardware element 714 embodies.Calculating equipment 710 can be configured as realization and software
And/or the corresponding specific instruction of hardware module and/or function.Thus, for example by using the computer-readable of processing system
Storage medium and/or hardware element 714 at least partly can realize that be embodied as module can be by calculating equipment with hardware
710 modules executed as software.Instruction and/or function can be by one or more products (for example, one or more calculate sets
Standby 710 and/or processing system 711) can be performed/can operate to realize the techniques described herein, module and example.
Technique described herein can be supported by these various configurations of calculating equipment 710, and be not limited to this paper institute
The specific example of the technology of description.Calculate equipment 710 function can also by using distributed system, such as pass through following institute
The platform 730 stated entirely or partly is realized on " cloud " 720.
Cloud 720 includes and/or representative is used for the platform 730 of resource 732.The hardware of 730 abstract cloud 720 of platform is (for example, clothes
Be engaged in device) and software resource bottom function.Resource 732 may include executing calculating on far from the server for calculating equipment 710
The application and/or data that can be used when machine processing.Resource 732 can also include by internet and/or passing through such as honeycomb
Or the service that the subscriber network of Wi-Fi network provides.
Platform 730 can be connect with abstract resource and function with that will calculate equipment 710 with other calculating equipment.Platform 730 is also
It can be used for the classification of abstract resource to provide the corresponding water of the demand for the resource 732 realized via platform 730 encountered
Flat classification.Therefore, in interconnection equipment embodiment, the realization of functions described herein can be distributed in whole system 700.
For example, function can be realized partly on calculating equipment 710 and through the platform 730 of the function of abstract cloud 720.
By research attached drawing, disclosure and appended claims, those skilled in the art are in the required guarantor of practice
When the theme of shield, it is to be understood that and realize the modification for the disclosed embodiments.In detail in the claims, word " comprising " is not
Exclude other elements or step, and indefinite article "a" or "an" be not excluded for it is multiple.It is wanted in mutually different appurtenance
The only fact that certain measures are described in asking does not indicate that the combination of these measures cannot be used to make a profit.
Claims (16)
1. a kind of content delivery method, comprising:
Obtain the individuation data and social networks data of potential user;
Based on acquired individuation data and social networks data, input data associated with the potential user is constructed;
The input data is input at least one Machine learning classifiers, wherein at least one described Machine learning classifiers
Each of be each configured to generate based on the input data preference for indicating the potential user whether with object content
The output data to match;And
Depending on the output data that at least one described Machine learning classifiers generate, selectively initiate to the potential user
Client push described in object content.
2. the method as described in claim 1, wherein the client push institute selectively initiated to the potential user
Stating object content includes:
The output data generated in response to each of at least one described Machine learning classifiers indicates the potential use
The preference at family matches with the object content, object content described in the client push of Xiang Suoshu potential user;And
The potential user is indicated in response to the output data that any of at least one described Machine learning classifiers generate
Preference and the object content mismatch, not object content described in the client push to the potential user.
3. the method as described in claim 1, wherein each of at least one described Machine learning classifiers by with
Lower step is trained to obtain:
Obtain the respective individuation data of multiple known users and social networks data;
Based on acquired respective individuation data and social networks data, construct related to the multiple known users respectively
The corresponding input data of connection;
Respective objects output data associated with the multiple known users is provided, wherein associated with each known users
Target output data indicates whether the preference of the known users matches with the object content;And
The corresponding input data and the respective objects output data are inputted into each Machine learning classifiers to be used to instruct
Practice.
4. method as claimed in claim 3, wherein the building is associated corresponding defeated to the multiple known users respectively
Entering data includes:
For each known users:
Individuation data based on the known users generates the first matrix;
Individuation data and social networks data based on the known users generate the second matrix, wherein second matrix description
Between the multiple friends and the known users of the known users in terms of personal attribute respective similarity and the multiple friend
Respective similarity, n are natural number in terms of the personal preference for n content between friend and the known users, and wherein
The n content is different from the object content;
First matrix and second matrix are subjected to physics merging, to form third matrix;And
The dimension for reducing the third matrix, to obtain the 4th matrix as input data associated with the known users.
5. method as claimed in claim 4,
Wherein the individuation data of each known users includes:
M attribute value of the respective attributes in m attribute of the known users is described;With
Indicate that the n preference the numerical value whether preference of the known users matches with the corresponding contents in the n content, m are
Natural number, and
Wherein the social networks data of each known users include the respective individuation data of multiple friends of the known users, often
The individuation data of a friend includes:
M attribute value of the respective attributes in the m attribute of the friend is described;With
Indicate the n preference the numerical value whether preference of the friend matches with the corresponding contents in the n content.
6. method as claimed in claim 5, wherein the individuation data based on the known users generates the first matrix packet
It includes: m attribute value vector of the individuation data of the known users is turned into row vector as first matrix.
7. method as claimed in claim 5, wherein the individuation data and social networks data based on the known users
Generating the second matrix includes:
From the individuation data of the corresponding m attribute value and known users of the respective individuation data of the multiple friend
M attribute value exports between the multiple friend and the known users the respective similarity in terms of personal attribute;
From the individuation data of the corresponding n preference numerical value and known users of the respective individuation data of the multiple friend
N preference numerical value exports between the multiple friend and the known users in for each content in the n content
Respective similarity in terms of people's preference;
By respective similarity vector turns in terms of personal attribute between the multiple friend and the known users derived from institute
First column vector;
By derived between the multiple friend and the known users in the individual for each content in the n content
Respective similarity vector turns to corresponding second column vector in terms of preference;And
Each second column vector of the first row vector sum is concatenated in the row direction to form second matrix.
8. the method for claim 7, wherein in a Genus Homo between described export the multiple friend and the known users
Property the respective similarity of aspect include:
By m attribute value of m attribute value of the individuation data of each friend and the individuation data of the known users
In correspondence numerical value be compared;
The number for indicating equal comparison result is counted;And
Determine similarity of the ratio of the number and m between the friend and the known users in terms of personal attribute.
9. the method for claim 7, wherein for institute between described export the multiple friend and the known users
Respective similarity includes: in terms of stating the personal preference of each content in n content
By n preference numerical value of n preference numerical value of the individuation data of each friend and the individuation data of the known users
In correspondence numerical value be compared;
Indicate that the friend and the known users have for the identical of the same content in the n content in response to the comparison
The personal preference of similarity between the friend and the known users in terms of to(for) the content is set as predetermined number by preference
Value;And
Indicate that the friend and the known users have the difference for the same content in the n content in response to the comparison
The personal preference of similarity between the friend and the known users in terms of to(for) the content is set as zero by preference.
10. the method for claim 7, wherein described carry out physics merging for first matrix and second matrix
It include: in the row direction to concatenate first matrix and second matrix to form the third matrix.
11. method as claimed in claim 4, wherein the dimension for reducing the third matrix includes: to the third square
Battle array carries out principal component analysis.
12. method as claimed in claim 3, wherein the target output data of each known users includes indicating the known users
The numerical value that whether matches with the object content of preference.
13. a kind of content push device, comprising:
For obtaining the individuation data of potential user and the device of social networks data;
For constructing input number associated with the potential user based on acquired individuation data and social networks data
According to device;
For the input data to be input to the device of at least one Machine learning classifiers, wherein at least one described machine
Each of Study strategies and methods are each configured to the preference for indicating the potential user is generated based on the input data whether
The output data to match with object content;And
The output data generated for depending at least one described Machine learning classifiers, is selectively initiated to described potential
The device of object content described in the client push of user.
14. a kind of content push device, comprising:
Module is obtained, is configured to obtain the individuation data and social networks data of potential user;
Module is constructed, is configured to based on acquired individuation data and social networks data, building and the potential user
Associated input data;
Input module is configured to the input data being input at least one Machine learning classifiers, wherein it is described at least
Each of one Machine learning classifiers, which are each configured to generate based on the input data, indicates the potential user's
The output data whether preference matches with object content;And
Pushing module is configured to depend on the output data that at least one described Machine learning classifiers generate, selectively
Object content described in client push to the potential user.
15. a kind of calculating equipment, including memory and processor, the memory are configured to store computer program on it
Instruction, the computer program instructions promote the processor perform claim to require in 1-12 when executing on the processor
Described in any item methods.
16. a kind of computer readable storage medium stores computer program instructions thereon, the computer program instructions, which are worked as, to be located
The processor perform claim is promoted to require method described in any one of 1-12 when executing on reason device.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910092267.0A CN110162714B (en) | 2019-01-30 | 2019-01-30 | Content pushing method, device, computing equipment and computer readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910092267.0A CN110162714B (en) | 2019-01-30 | 2019-01-30 | Content pushing method, device, computing equipment and computer readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110162714A true CN110162714A (en) | 2019-08-23 |
CN110162714B CN110162714B (en) | 2023-11-14 |
Family
ID=67644811
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910092267.0A Active CN110162714B (en) | 2019-01-30 | 2019-01-30 | Content pushing method, device, computing equipment and computer readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110162714B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112820416A (en) * | 2021-02-26 | 2021-05-18 | 重庆市公共卫生医疗救治中心 | Major infectious disease queue data typing method, typing model and electronic equipment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106776928A (en) * | 2016-12-01 | 2017-05-31 | 重庆大学 | Recommend method in position based on internal memory Computational frame, fusion social environment and space-time data |
CN108711075A (en) * | 2018-05-22 | 2018-10-26 | 阿里巴巴集团控股有限公司 | A kind of Products Show method and apparatus |
CN109087138A (en) * | 2018-07-26 | 2018-12-25 | 北京京东金融科技控股有限公司 | Data processing method and system, computer system and readable storage medium storing program for executing |
-
2019
- 2019-01-30 CN CN201910092267.0A patent/CN110162714B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106776928A (en) * | 2016-12-01 | 2017-05-31 | 重庆大学 | Recommend method in position based on internal memory Computational frame, fusion social environment and space-time data |
CN108711075A (en) * | 2018-05-22 | 2018-10-26 | 阿里巴巴集团控股有限公司 | A kind of Products Show method and apparatus |
CN109087138A (en) * | 2018-07-26 | 2018-12-25 | 北京京东金融科技控股有限公司 | Data processing method and system, computer system and readable storage medium storing program for executing |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112820416A (en) * | 2021-02-26 | 2021-05-18 | 重庆市公共卫生医疗救治中心 | Major infectious disease queue data typing method, typing model and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN110162714B (en) | 2023-11-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11803730B2 (en) | Webinterface presentation using artificial neural networks | |
US10922609B2 (en) | Semi-supervised learning via deep label propagation | |
US10354184B1 (en) | Joint modeling of user behavior | |
Bari et al. | Predictive analytics for dummies | |
US20180174070A1 (en) | Personalizing User Experiences With Electronic Content Based on User Representations Learned from Application Usage Data | |
CN109800885A (en) | It is determined for the rule of black box machine learning model | |
US10984245B1 (en) | Convolutional neural network based on groupwise convolution for efficient video analysis | |
US20160155063A1 (en) | Iterative Classifier Training on Online Social Networks | |
CN110020188A (en) | Recommended based on the Global Vector of implicit interactions and profile data | |
EP4180991A1 (en) | Neural network distillation method and apparatus | |
CN115917535A (en) | Recommendation model training method, recommendation device and computer readable medium | |
JP2019113943A (en) | Information providing apparatus, information providing method, and program | |
WO2024002167A1 (en) | Operation prediction method and related apparatus | |
Khan et al. | Comparative analysis on Facebook post interaction using DNN, ELM and LSTM | |
WO2023050143A1 (en) | Recommendation model training method and apparatus | |
US20210350202A1 (en) | Methods and systems of automatic creation of user personas | |
US20190205702A1 (en) | System and method for recommending features for content presentations | |
US20220277205A1 (en) | Automated machine learning to generate recommendations for websites or applications | |
CN112650942A (en) | Product recommendation method, device, computer system and computer-readable storage medium | |
CN110162714A (en) | Content delivery method, calculates equipment and computer readable storage medium at device | |
Mengle et al. | Mastering machine learning on Aws: advanced machine learning in Python using SageMaker, Apache Spark, and TensorFlow | |
CN112862538A (en) | Method, apparatus, electronic device, and medium for predicting user preference | |
Motte | Mathematical models for large populations, behavioral economics, and targeted advertising | |
JP7464346B2 (en) | Object attribute expression generation model capable of generating object attribute expressions, object attribute estimation device and method | |
CN117390295B (en) | Method and device for recommending objects based on mask module |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |