CN108304432A - Information push processing method, information push processing unit and storage medium - Google Patents
Information push processing method, information push processing unit and storage medium Download PDFInfo
- Publication number
- CN108304432A CN108304432A CN201710647371.2A CN201710647371A CN108304432A CN 108304432 A CN108304432 A CN 108304432A CN 201710647371 A CN201710647371 A CN 201710647371A CN 108304432 A CN108304432 A CN 108304432A
- Authority
- CN
- China
- Prior art keywords
- user
- terminal
- information
- classification
- behavioral data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of information push processing method, including：The behavioral data for obtaining the operation of detection user in terminal and being formed；Clustering processing is carried out according to the behavioral data for belonging to the terminal, obtains operating the operating characteristics in different dimensions in the terminal from choosing in cluster result；The feature clustered to the terminal is predicted according to disaggregated model, obtains the classification results of active user in the terminal；The replacement of user in terminal is judged according to the classification results of active user in terminal；According to the replacement of user in terminal, realize that adaptation replaces the information push of the corresponding classification results of user.The invention also discloses a kind of information push processing unit and storage mediums.
Description
Technical field
The present invention relates to the communication technology more particularly to a kind of information push processing method, information push processing unit and deposit
Storage media.
Background technology
The development of internet especially mobile Internet becomes the approach to become more and more important for obtaining information.In order to promote use
Family obtains the efficiency of information, and the relevant technologies provide information advancing technique, usually, by calculating the preference of user and to user
Send the information for meeting preference.
The relevant technologies are by identifying different terminals (such as smart mobile phone, tablet computer), calculating the preference of user and lead to
It crosses various modes and sends the information for meeting user preference to terminal, saved the operation that user searches for information, improve acquisition letter
The efficiency of breath.
However, scheme of the relevant technologies based on terminal pushed information, always the user of using terminal will not switch this
The hypothesis of sample is pushed into row information calculating user preference, it is clear that in multi-user's usage scenario that present terminal becomes increasingly abundant
In, it cannot achieve the accurate push of information.
Invention content
A kind of information push processing method of offer of the embodiment of the present invention, information push processing unit and storage medium；It can
It is applicable in the accurate push that information is realized in the usage scenario of the different user of terminal.
What the technical solution of the embodiment of the present invention was realized in：
The embodiment of the present invention provides a kind of information push processing method, including：
The behavioral data for obtaining the operation of detection user in terminal and being formed；
Clustering processing is carried out according to the behavioral data for belonging to the terminal, is chosen from cluster result and obtains the terminal
Operating characteristics of the middle operation in different dimensions；
The feature clustered to the terminal is predicted according to disaggregated model, obtains active user in the terminal
Classification results；
The replacement of user in terminal is judged according to the classification results of active user in terminal；
According to the replacement of user in terminal, realize that adaptation replaces the information push of the corresponding classification results of user.
The embodiment of the present invention also provides a kind of information push processing unit, including：
Data capture unit, the behavioral data formed for obtaining the operation for detecting user in terminal；
Characteristics determining unit, for carrying out clustering processing according to the behavioral data for belonging to the terminal, from cluster result
Middle selection obtains operating the operating characteristics in different dimensions in the terminal；
Taxon, the feature for being clustered to the terminal are predicted according to disaggregated model, obtain the end
The classification results of active user in end；
Judging unit, the replacement for judging user in terminal according to the classification results of active user in terminal；Push is single
Member realizes that adaptation replaces the information push of the corresponding classification results of user for the replacement according to user in terminal.
In said program, the data capture unit detects that the terminal meets timing condition specifically for obtaining to work as
When, the behavioral data of formation acquired before timing condition meets；Alternatively,
The terminal is obtained when generating the potential event that characterization user replaces, formation acquires before generating potential event
Behavioral data；Alternatively,
When detecting the terminal and being in specific information push scene, formation meets corresponding information push scene characteristic
The behavioral data of operation and formation.
In said program, the judging unit is specifically used for the classification as active user in the terminal, with the terminal
The distance of the history classification of middle user exceeds distance threshold, judges that user replaces in terminal；
Alternatively, the classification of active user is identical twice in succession in the terminal, and gone through with user in the terminal
When history classification is different, judge that user replaces in terminal.
In said program, further include：
Recognition unit, when for obtaining behavioral data from multiple terminals, according to the hardware of the behavioral data carrying obtained
Mark, the terminal of the obtained behavioral data ownership of identification.
In said program, the push unit, specifically for when user replaces in the terminal,
According to the classification results for replacing user, the information that stereotactic conditions in server meet corresponding classification results is inquired；
Inquired information is pushed to the terminal.
In said program, the push unit, specifically for when user replaces in the terminal,
According to the classification results for replacing user, inquires stereotactic conditions in the terminal and meet the information of corresponding classification results simultaneously
It presents.
The embodiment of the present invention also provides a kind of storage medium, is stored thereon with executable program, at the executable code
Reason device realizes information push processing method provided in an embodiment of the present invention when executing.
The embodiment of the present invention also provides a kind of information push processing unit, including：Processor and for store can locate
The memory of the executable program run on reason device realizes that the present invention is real when the processor is for running the executable program
The information push processing method of example offer is provided.
The embodiment of the present invention has the advantages that：
1) according to the behavioral data extraction operation feature of record user's operation, prediction user's classification, due to operating characteristics energy
The difference of enough intuitive reflection different classifications users operationally, therefore precise classification can be carried out to user based on disaggregated model；
2) the case where being replaced according to user in same terminal has and adaptively realizes information push, overcome the relevant technologies
In based on the condition that terminal user will not convert mechanical pushed information defect, be applicable in terminal use in single user and
The different scenes that multi-user uses are obviously improved the precision that information pushes in same terminal.
Description of the drawings
Fig. 1-1 is an optional configuration diagram of information push provided in an embodiment of the present invention；
Fig. 1-2 is another optional configuration diagram of information push provided in an embodiment of the present invention；
Fig. 1-3 is another optional configuration diagram of information push provided in an embodiment of the present invention；
Fig. 2 is an optional software/hardware structural schematic diagram of information push processing unit provided in an embodiment of the present invention；
Fig. 3-1 is the optional processing flow schematic diagram one of information push processing method provided in an embodiment of the present invention；
Fig. 3-2 is the optional schematic diagram that the sample of behavioral data provided in an embodiment of the present invention is clustered；
Fig. 4 is the processing flow schematic diagram of trained decision-tree model provided in an embodiment of the present invention；
Fig. 5 is the processing flow schematic diagram provided in an embodiment of the present invention for calculating information gain；
Fig. 6 is decision tree schematic diagram provided in an embodiment of the present invention；
Fig. 7 is the optional processing flow schematic diagram two of information push processing method of the embodiment of the present invention；
Fig. 8 is the optional processing flow schematic diagram three of information push processing method of the embodiment of the present invention；
Fig. 9 is the optional processing flow schematic diagram four of information push processing method of the embodiment of the present invention；
Figure 10 optional configuration diagram provided in an embodiment of the present invention launched for advertisement；
Figure 11 is an optional configuration diagram provided in an embodiment of the present invention for news push；
Figure 12 is the composed structure schematic diagram that information of the embodiment of the present invention pushes processing unit；
Figure 13 is the schematic diagram that the relevant technologies are directed to that same terminal is pushed into row information；
Figure 14 is the schematic diagram that the embodiment of the present invention is directed to that same terminal is pushed into row information.
Specific implementation mode
The present invention is further described in detail below with reference to the accompanying drawings and embodiments.It should be appreciated that described herein
Specific embodiment is only used to explain the present invention, is not intended to limit the present invention.
Before the present invention will be described in further detail, to involved in the embodiment of the present invention noun and term say
Bright, noun and term involved in the embodiment of the present invention are suitable for following explanation.
1) decision tree：Decision tree (Decision Tree) is a tree construction (can be binary tree or non-binary trees).Its
Each nonleaf node indicates that the test in a feature, each branch represent output of this feature in some codomain, and every
A leaf node stores a classification.The process that decision is carried out using decision tree is exactly to be tested in item to be sorted since root node
Corresponding feature, and output branch is selected according to its value, until reaching leaf node, using the classification of leaf node storage as determining
Plan result.
2)ID3：A kind of algorithm of leaf node is selected in decision tree, core concept is exactly with information gain measures characteristic
Selection, the maximum feature of information gain is into line splitting after selection division.
3) comentropy：The size of information content, comentropy herein use the feature structure of a dimension with referring in disaggregated model
When making class condition, information content that the information content of disaggregated model is calculated based on ad hoc fashion.
4) information gain：The difference of the comentropy of former and later two categorizing systems is exactly information gain, and comentropy is information content
Quantizating index.
The embodiment of the present invention provides information push processing method, information push processing unit and storage medium, Ke Yili
Solution, the scene that can be suitable for the push of various types information typically such as news push and advertisement pushing, are certainly not limited to
This, the push of various types information suitable for internet.
For information pushes the implementation of processing unit, in alternative embodiment of the present invention, information push processing dress is provided
Setting can be in end side and server side while the mode implemented, and corresponding storage medium can be arranged in end side and server
Side, the corresponding processing completed in end side and server side push processing unit in end side and server with reference to information
The implementation of side illustrates.
As an example, information push processing unit is implemented in end side and server side, and Fig. 1-1 shows the present invention
One optional configuration diagram of the information push that embodiment provides, is equipped with for realizing specific application purpose in terminal
Client, such as news client, videoconference client and social network client, client are collected user and are grasped in the terminal
The behavioral data of work simultaneously reports to server；Server predicts the active user's of client according to disaggregated model and behavioral data
Classification results obtain from database and meet the information of classification results and be pushed to client.
As another example, information pushes processing unit and implements in end side and server side, and Fig. 1-2 shows this hair
Another the optional configuration diagram for the information push that bright embodiment provides is equipped with for realizing specific application mesh in terminal
Client, such as news client, videoconference client and social network client, client collect user in the terminal into
The behavioral data of row operation, and according to disaggregated model and the classification results of the active user of behavioral data prediction client, and will
Classification results report to server by network；Server obtains the information for meeting classification results according to classification results from database
Push to client.
As another example, information push processing unit is implemented in end side, and Fig. 1-3 shows that the embodiment of the present invention carries
Another optional configuration diagram of the information push of confession, is equipped with for realizing the client of specific application purpose in terminal
End, such as news client, videoconference client and social network client, client collect what user was operated in the terminal
Behavioral data, and according to disaggregated model and the classification results of the active user of behavioral data prediction client, according to classification results
The information for meeting classification results is obtained from database in client presentation.
For the software/hardware structure that information pushes processing unit 10, referring to Fig. 2, including：Hardware layer, driving layer, operation
System layer and application layer.However, it will be understood by those of skill in the art that can be according to reality for information push processing unit 10
It applies needs to be arranged compared with the more components of Fig. 2, or omission setting unit component is needed according to implementation.
The hardware layer of information push processing unit 10 includes processor 161, input/output interface 163, memory 164 with
And network interface 162, component can be through system bus connection communications.
Processor 161 may be used central processing unit (CPU), microprocessor (MCU, Microcontroller Unit),
Application-specific integrated circuit (ASIC, Application Specific Integrated Circuit) or logic programmable gate array
(FPGA, Field-Programmable Gate Array) is realized.
Input/output interface 163 may be used such as display screen, touch screen, loud speaker input/output device and realize.
The realization of the non-volatile memory mediums such as flash memory, hard disk, CD may be used in memory 164, can also use double
The volatile storage mediums such as rate (DDR, Double Data Rate) dynamic buffering are realized, wherein being stored with to execute above-mentioned wide
Accuse the executable instruction of effect analysis method.
Network interface 162 is to the offer of the processor 161 external data memory 164 that for example strange land is arranged based on network transmission
Agreement (TCP, Transfer Control Protocol)/User Data Protocol (UDP, User Datagram Protocol)
Access ability.
Driving layer include for for operating system 166 identify hardware layer and with the middleware 165 of each assembly communication of hardware layer,
Such as can be the set of the driver of each component for hardware layer.
Operating system 166 is used to provide the graphical interfaces of user oriented (such as advertiser and ad system operation maintenance personnel), behaviour
Making system 166 supports user such as to be grasped to the software environment of above equipment to the control embodiment of the present invention of equipment via graphical interfaces
Make system type, version does not limit, such as can be (SuSE) Linux OS and UNIX operating system etc..
Application layer include for realizing the application program 167 of information push processing method provided in an embodiment of the present invention, when
So, it can also include other applications 168.
So far, the information involved in the embodiment of the present invention is described according to its function push processing unit (server
Or terminal), based on the network architecture shown in Fig. 1-1, the information shown in Fig. 1-2 push processing unit illustrative view of functional configuration with
And Fig. 2 shows information push processing unit software/hardware structure, continue to information push provided in an embodiment of the present invention handle
Scheme illustrate.
In the following, the configuration diagram of the information push in conjunction with shown in Fig. 1-1 pushes information provided in an embodiment of the present invention
The scheme of processing illustrates, and Fig. 3-1 shows that the optional flow of information push processing method provided in an embodiment of the present invention is shown
It is intended to one, will be illustrated according to each step.
The operation of step S101, terminal detection user form behavioral data.
In an alternative embodiment, the example of behavioral data is：Operation, time and environment etc.；Wherein, operation refers to for example
Action type, such as use shortcut key, use mouse gestures；Time includes：At the beginning of operation and the end time；Environment is
Refer to the environment that operation is formed, such as the primary function interface of client or operating system.
The embodiment of the present invention detects user's operation and forms behavioral data with different dimensions, such as detection terminal operating system
In the global user's operation that receives and the behavioral data that is formed；Or detect user's operation in specific type client
And the behavioral data formed.Terminal records the behavioral data of user from global dimension, both including user to the primary of operating system
The behavioral data that the operation of function interface is formed also includes the behavioral data that user is directed to client record；In this way, passing through acquisition
The operating characteristics of user's more various dimensions, not only can guarantee the validity of the behavioral data in a detection cycle, but also can improve basis
Operating characteristics determine the accuracy of class of subscriber；Improve the accuracy of information push.Terminal is recorded from specific type client
The behavioral data of user can be directed to the operation of specific type client based on user, to judge to use specific type client
The user at end classifies, and is classified come pushed information according to the user for specific type client, can improve determining for information push
Tropism.
In an alternative embodiment, the behavioral data for detecting user's operation in terminal and being formed can also be following scheme,
Information first residing for detection terminal pushes scene, determines feature associated with detected information push scene, then examine
Survey the behavioral data for meeting associated feature.Here, the letter residing for terminal can be determined according to the classification of the client of terminal operating
Breath push scene, correspondingly, the different corresponding information classification demands of information push scene is also different.
For example when news client (such as Tencent's news), in the terminal is in operating status, information pushes scene
The as scene of news push；So, the corresponding information of news push scene, which is classified, may include：Sports news, entertainment news,
Financial and economic news and military news；It is characterized as correspondingly, information classification is corresponding：Sport, amusement, finance and economics, military affairs.Therefore, eventually
The operation for subject contents such as sport, amusement, finance and economics, military affairs in news client is held, and based on the operation detected, there is needle
To the formation behavioral data of property；It is respectively formed the behavioral data for sport, amusement, finance and economics, military affairs.Terminal is directed to acquisition again
Behavioral data detection meet the behavioral data of associated feature；Such as it can be scheduled for storing behavioral data in terminal
Storage region detects the behavioral data for meeting linked character.
For another example when music client end (such as QQ music) in the terminal is in operating status, the information residing for terminal pushes away
It is the scene of music push to send scene；So, the corresponding information of music push scene, which is classified, may include：Classic class music,
Popular class music；Correspondingly, information classification is corresponding to be characterized as classical music, pop music.Terminal is directed to the behavior number of acquisition again
Meet the behavioral data of associated feature according to detection；For example, can be in the scheduled memory block for storing behavioral data of terminal
The behavioral data of domain detection service linked character.
By taking the application scenarios of advertisement as an example, the arbitrary client (such as Taobao, Tencent's news, the QQ that are installed in terminal
Music, U.S. group's take-away etc.) in the process of running, it can classify from the angle of the product involved by advertisement, such as from quick
The consumer goods, electronic product, clothes, food angle classify；Correspondingly, classification it is corresponding be characterized as consumption, electronic product,
Clothes, food；Terminal detects the behavioral data for meeting associated feature in the behavioral data of acquisition again, for example, can be at end
Hold the scheduled behavioral data for storing the storage region detection service linked character of behavioral data.
Step S102, terminal record detected behavioral data, server are reported when meeting report condition.
In alternative embodiment of the present invention, it may include the following two kinds scheme that terminal to server, which reports behavioral data, the
One is terminals when detecting that itself meets timing condition, itself collected behavioral data is reported to server；Here, fixed
When condition be that the period of behavioral data is reported according to the terminal to server that is flexibly arranged of actual needs, such as 10 seconds, 1 minute；
When terminal is according to the timing condition of setting, periodically reports behavioral data, the behavioral data reported every time only includes terminal
Collected behavioral data after once reporting operation before reporting operation recently apart from this.
Here, the type of timing condition can be global unified value, to ensure that server obtains the timeliness of behavioral data
Property.The type of timing condition can also be distinguished according to the priority (significance level) of push terminal, such as apple terminal
Timing is less than the timing of android terminal, and the timing of intelligent mobile phone terminal is less than the timing of tablet computer；It so, it is possible to ensure service
Device obtains the behavioral data of emphasis push-type terminal immediately.
In alternative embodiment of the present invention, terminal is reported when generating the potential event that characterization user replaces to server
Behavioral data；Wherein, the potential event of characterization user replacement may include：The switching of application program or operating system, which log in, to be used
The switching etc. at family；Wherein, the switching of application program refers to that the application program that terminal is currently running is changed to another by one.This
When, the behavioral data that server reports includes generating collected behavior number between the potential event that characterization user replaces twice
According to；In this way, flexibly reporting corresponding behavioral data, Neng Goubao to server according to the operational state change of itself by terminal
The promptness that server obtains behavioral data is demonstrate,proved, so that server is in time according to behavioral data, realizes that adaptation replaces user's phase
The information of classification results is answered to push.
Obtained from above two server the behavioral data of terminal specific implementation process can be seen that behavior data by
Terminal reports to server；Can by the client in terminal report mechanism realize server obtain terminal behavior number
According to.
By server to for the scene of terminal user's advertisement, after can integrating on the client for accessing advertisement
The Software Development Kit (Software Development Kit, SDK) of platform server, client is by running SDK
To complete client to the acquisition of the behavioral data of user and report.Therefore, the SDK for accessing advertisement background server is integrated
All clients can be with advertisement.
Step S103, server according to belong to terminal behavioral data carry out clustering processing, according to different clusters from
The operating characteristics of middle extraction different dimensions.
In alternative embodiment of the present invention, cluster is carried out as difference according to the tightness degree contacted between the sample of behavioral data
Grouping be not excluded in the embodiment of the present invention using various for selected characteristic in the grouping of each cluster as operating characteristics
Clustering algorithm carries out clustering processing, such as hierarchical clustering, K mean cluster and neural network clustering etc. to the sample of behavioral data.
Fig. 3-2 shows one that the sample of behavioral data provided in an embodiment of the present invention is clustered optional signal
Figure, by taking hierarchical clustering as an example, according to each sample of behavioral data sample space distribution, will apart from (such as Euclidean distance,
Exhausted angle value distance etc.) similar in sample be aggregated to different groupings, realize that 1) the distance between different grouping maximizes；2) each
The distance between sample maximizes in grouping, and sample in same grouping is made to have a homogeney as high as possible, and different points
Should then have heterogeneity as high as possible between group.
Each grouping for cluster extracts a character subset in the whole features having from the sample of grouping, then
With interpretational criteria, (criterion of one character subset quality degree of evaluation, can be different under different application scenarios, such as can
Whether uniformly judged according to the distribution of feature) this feature subset to be evaluated, the result of evaluation is compared with stopping criterion,
If meeting stopping criterion just to stop, otherwise continuing to generate next group of character subset, continue Feature Selection.
By taking behavioral data is detection user's operation obtains in a browser as an example, clustered according to behavioral data and in cluster
Following operating characteristics are chosen in different grouping：Shortcut key, mouse gestures, keyboard input, multiwindow, browser setting, webpage
The operating characteristics of several disjoint different dimensions such as browsing, interest Web page classifying.
Step S104, server predicts the operating characteristics that terminal clusters according to disaggregated model, obtains terminal
The classification results of middle active user.
In alternative embodiment of the present invention, server needs disaggregated model of the training for carrying out user's classification in advance, will
, can be according to the built-in processing logic of disaggregated model after operating characteristics input disaggregated model, the operating characteristics inputted correspond to
Classification results.The disaggregated model pre-established can be when meeting timing condition based on terminal, formation before timing condition
The sample data of acquisition is built；When can also be based on the potential event for generating characterization user's replacement in terminal, formation is being produced
The sample data structure acquired before raw potential event；When can also be that being in specific information based on terminal pushes scene, formed
Sample data structure.Correspondingly, when being predicted the behavioral data of the active user of acquisition using disaggregated model, it is available
Disaggregated model corresponding with the type of the behavioral data of active user is predicted.
For example, when the behavioral data of the active user of acquisition is that terminal is in specific information push scene, just
It is in the sample data that specific information push scene is formed using terminal, train classification models are simultaneously predicted, this differentiation
The mode that information push scene is predicted takes full advantage of the difference of operation of the different user in same information pushes scene
Property, it can realize and precise classification is carried out to different user in an information pushes scene.
In alternative embodiment of the present invention, when disaggregated model uses classification-tree method, server is according to classification-tree method
Including each dimensional characteristics priority descending, successively with the corresponding class condition of the operating characteristics of each dimension, to what is obtained
The operating characteristics of respective dimensions carry out classification judgement, until, meet the class condition of a feature in priority descending, will be met
Feature it is corresponding classify be determined as the classification results of active user in terminal.
Server need to train the decision-tree model for carrying out user's classification in advance；In one embodiment, training decision tree
The flow diagram of model, as shown in figure 4, including the following steps：
Step S1041 obtains the candidate feature for the different dimensions that each user's sample has in user's sample set.
Here, user's sample set is the behavioral data collection of user, such as the candidate feature of different dimensions includes：Whether use
Whether shortcut key uses mouse gestures, whether changes browser configuration.
Step S1042, when calculating classifies to user in user's sample set with the candidate feature of each dimension, corresponding letter
Cease gain.
Decision tree theory based on ID3 algorithms below, the candidate feature based on each dimension in user's sample set to using
Classify at family.
Realization process is：First, the behavioural habits of multiple users are recorded, multiple users have different occupations, age, property
Not, the differences such as hobby, to acquire comprehensively and accurately behavioral data.It is included whether using quick using the behavioural habits of browser
Key, whether using mouse gestures, whether change browser setting, text input speed, whether open multiwindow, webpage be averaged it is clear
It lookes at duration, interest Web page classifying etc., by the feature of these clear-cut individualities, can be used for describing a user, while can also
For distinguishing user.
Secondly, decision tree modeling is carried out using ID3 algorithms；Here, it is situated between first to the principle for carrying out decision tree modeling
It continues：If the value of a stochastic variable X be x={ x1, x2, x3...nx }, each probability got be respectively p1, p2,
P3...pn }, then the self-information of X is defined as
As soon as that is, the situation of change of variable may be more, then the information content that it is carried is bigger.For classification
For system, classification C is variable, its value is c1, c2, c3...cn, and the probability that each classification occurs is P respectively
(c1),P(c2),p(c3)...p(cn)；Wherein, n is the sum of classification；At this point, the entropy of categorizing system can be expressed as
Based on the definition of above- mentioned information entropy, information gain is briefly described below.Information gain is to be directed to a spy
For sign, i.e., for a feature t, it is respectively how many, the difference of the two that system, which has information content when this feature and no this feature,
Value is exactly the information content that this feature is brought to decision tree, i.e. information gain.
It should be noted that only carrying out model training so that ID3 algorithms build decision-tree model as an example here, other structures are determined
Other training patterns of the algorithm of plan tree-model and for identification terminal user are suitable for the embodiment of the present invention.
In one embodiment, when calculating classifies to user in user's sample set with the candidate feature of each dimension,
The specific implementation process of corresponding information gain, as shown in figure 5, including the following steps：
Step S10421 calculates the initial information entropy based on priori classification result in user's sample set.
Step S10422 constructs different class conditions with the operating characteristics of any dimension, calculates item in different categories
When part is classified for sample of users in sample set, the corresponding reference information entropy of decision-tree model.
Step S10423 calculates comentropy of the difference of reference information entropy and initial information entropy as respective dimensions feature.
For example, the feature got and corresponding user classification, as shown in table 1 below：
Table 1
Five user data are shared in table 1, user's classification results include two classes；Wherein, there are three A classes user and two are non-
A class users；Also, ratio of the A classes user in total user is 3/5, and ratio of the non-A classes user in total user is 2/5.Root
The comentropy that current system can be calculated according to formula H (x) is-(2/5*log22/5+3/5*log23/5)=0.1591+
0.1331=0.2922.
Whether the use of shortcut key to be standard come for dividing user's classification, three users use shortcut key, using fast
It is A class users there are two user, it is 2/3 that A class users, which account for using the ratio of shortcut key user, in three users of prompt key；One
User is non-A classes user, and it is 1/3 that non-A classes user, which accounts for using the ratio of shortcut key user,；So, continue to apply mechanically formula meter above
Obtained comentropy is：-(1/3*log21/3+2/3*log22*3)=0.2764.
For two users without using shortcut key, one is A class users, and one is non-A classes user, A classes user and B classes
User is 1/2 in the ratio of the user without using shortcut key.So, continue to apply mechanically the comentropy that formula above is calculated
For：-(1/2*log21/2+1/2*log21/2)=0.301.
For whether using this feature of shortcut key, user's ratio using shortcut key is 3/5, corresponding comentropy
It is 0.2764；It is 2/5 without using user's ratio of shortcut key, corresponding comentropy is 0.301.Therefore, if use shortcut key
The corresponding comentropy of feature be：3/5*0.2746+2/5*0.301=0.28516.Therefore, if use the information of shortcut key
Gain subtracts the system information entropy using shortcut key tagsort, i.e. 0.2922-0.28516=equal to initial system information entropy
0.00704.The value indicates by whether great change can be brought to whole system information content by carrying out user's classification using shortcut key
Change, the bigger explanation characteristic of value is better for user's classifying quality, and accuracy is higher.Whether same method calculates uses mouse
Gesture and the information gain for whether changing browser configuration.
Step S1043, the highest candidate feature of information gain for selecting predetermined quantity are the corresponding feature of respective dimensions, and
The priority descending of the operating characteristics of respective dimensions is formed according to the descending of information gain.
In one embodiment, server is based on above-mentioned calculating, due to whether changing the corresponding comentropy of browser configuration>It is
It is no to use the corresponding comentropy of shortcut key>Whether the corresponding comentropy of mouse gestures is used；Therefore,
The priority descending of operating characteristics for forming respective dimensions is：Whether browser configuration is changed, whether using quick
Whether key uses mouse gestures.
When size according to information gain establishes decision-tree model, information gain is bigger, indicates that the information gain is corresponding
Feature is stronger for the certainty of categorised decision, and this feature is determined as root node, and then, the size according to information gain is to phase
The feature answered once arranges downwards, forms decision tree as shown in FIG. 6.The decision tree be the prediction that is formed according to sample data not
Carry out the model of data, therefore, the sample data of acquisition is more, corresponding feature is more, the decision tree scale of formation is huger；Phase
It answers, decision tree prediction result in large scale is more accurate.
When being classified to user based on above-mentioned decision tree, the root node whether user has decision tree is detected first and is corresponded to
Feature, i.e., user whether change browser configuration, if user change browser configuration, it is determined that user be A class users, knot
Line journey.If user does not change browser and configures, further detect whether user uses shortcut key, if user uses soon
Prompt key, it is determined that user is A class users, and flow terminates.If user does not use shortcut key, further detect whether user makes
With mouse gestures, if user uses mouse gestures, it is determined that user is A class users, and flow terminates.If user does not use mouse
Mark gesture, it is determined that user is non-A classes user.
Step S105, server judge the replacement of user in terminal according to the classification results of terminal active user.
In alternative embodiment of the present invention, server judges the classification of active user in terminal, is gone through with user in terminal
The distance of history classification exceeds distance threshold, judges that user replaces in terminal.Wherein, the distance of classification refers to according to user class
The distance of distance in user's classification chart of other similarity arrangement form, user's classification is bigger, illustrates the uneven class size of user
It is bigger；The history classification of user can be that the last user classifies.Therefore, a kind of specific implementation mode is exactly server
The classification for judging active user in terminal, in terminal user it is nearest one classification at a distance from exceed distance threshold when, judge
Terminal user replaces.
In another alternative embodiment of the present invention, when server judges that the classification of active user is identical twice in succession, also,
Identical user's classification judges that user replaces in terminal from when the history classification of user is different in terminal twice；Here, eventually
The history classification of user refers to the last user's classification in end.
In the embodiment of the present invention, when judging the replacement of user in terminal according to the classification results of terminal active user, adopt
With fault tolerant mechanism, the accuracy classified to user is provided.
Step S106 realizes that adaptation replaces the information push of the corresponding classification results of user according to the replacement of user in terminal.
In an alternate embodiment of the present invention, when server determines that the user in terminal replaces, according to replacement user
Classification results, inquiry server stereotactic conditions meet the information of corresponding classification results, and the information inquired is pushed to terminal,
So that the information of push is presented on the display interface of terminal.
It is the optional processing flow schematic diagram two of the information push processing method of the embodiment of the present invention referring to Fig. 7；This hair
The method of the method for bright embodiment as shown in figure 3 is similar, the difference is that, further include before step S102：
Step S107 when server obtains behavioral data from multiple terminals, is carried hard according to the behavioral data obtained
Part identifies, the terminal of the obtained behavioral data ownership of identification.
For example, when server obtains behavioral data from multiple terminals, according to the hardware of the behavioral data carrying obtained
Mark, as GUID identifies the terminal that obtained behavioral data belongs to.
Correspondingly, when executing step S105, the information that adaptation is replaced the corresponding classification results of user by server pushes to
Corresponding terminal.
The flow diagram three of information push processing method provided in an embodiment of the present invention, as shown in figure 8, applied to eventually
End, includes the following steps：
Step S201, terminal obtain the behavioral data for itself detecting user's operation and being formed.
User's operation involved by the embodiment of the present invention and the behavioral data that is formed can be that user is directed to terminal itself
The behavioral data of operation and formation, such as whether using shortcut key, whether using mouse gestures, based on the text input speed of keyboard
The data of equal operations；Involved user's operation and the behavioral data that is formed can also be user for being installed on the visitor severed
The operation data at family end, such as：Whether open multiwindow, whether change browser setting, webpage average browsing duration, interest net
The data of the operations such as page classification.
In alternative embodiment of the present invention, the first is terminal when detecting that itself meets timing condition, records itself and adopts
The behavioral data collected；Here, timing condition is that the terminal to server being flexibly arranged according to actual needs reports behavioral data
Period, such as 10 seconds, 1 minute；When terminal is according to the timing condition of setting, periodically reports behavioral data, remember every time
The behavioral data of record only includes terminal collected behavior after once recording operation before this record operation recently
Data.
Here, the type of timing condition can be global unified value, to ensure the timeliness of the behavioral data of record.It is fixed
When condition type can also be distinguished according to the priority (significance level) of push terminal, such as the timing of apple terminal is small
In the timing of android terminal, the timing of intelligent mobile phone terminal is less than the timing of tablet computer；It so, it is possible to ensure that server is instant
Obtain the behavioral data of emphasis push-type terminal.
In alternative embodiment of the present invention, terminal records behavioral data when generating the potential event that characterization user replaces；
Wherein, the potential event of characterization user replacement may include：The switching of application program or the switching of operating system login user
Deng；Wherein, the switching of application program refers to that the application program that terminal is currently running is changed to another by one.At this point, service
The behavioral data of device record includes generating collected behavioral data between the potential event that characterization user replaces twice；In this way,
When generating the potential event that characterization user replaces according to itself by terminal, corresponding behavioral data is just recorded, can ensure end
End obtains the behavioral data that user's replacement may occur in time.
In an alternative embodiment, the behavioral data that terminal obtains itself detection user's operation and formed can also include such as
The third lower scheme, i.e. terminal detect the information push scene residing for itself first, and determine and pushed with detected information
The associated feature of scene, then detection meets the behavioral data of associated feature in itself.Here, terminal can be transported according to itself
The classification of capable client determines that the information residing for itself pushes scene, correspondingly, the different corresponding letters of information push scene
It is also different to cease classification demand.
For example when news client (such as Tencent's news), in the terminal is in operating status, information pushes scene
The as scene of news push；So, the corresponding information of news push scene, which is classified, may include：Sports news, entertainment news,
Financial and economic news and military news；It is characterized as correspondingly, information classification is corresponding：Sport, amusement, finance and economics, military affairs.Therefore, eventually
The operation for subject contents such as sport, amusement, finance and economics, military affairs in news client is held, and based on the operation detected, there is needle
To the formation behavioral data of property；It is respectively formed the behavioral data for sport, amusement, finance and economics, military affairs.Terminal is directed to acquisition again
Behavioral data detection meet the behavioral data of associated feature；For example, can be scheduled for storing behavioral data in terminal
Storage region detection meet the behavioral data of linked character.
For another example when music client end (such as QQ music) in the terminal is in operating status, the information residing for terminal pushes away
It is the scene of music push to send scene；So, the corresponding information of music push scene, which is classified, may include：Classic class music,
Popular class music；Correspondingly, information classification is corresponding to be characterized as classical music, pop music.Terminal is directed to the behavior number of acquisition again
Meet the behavioral data of associated feature according to detection；For example, can be in the scheduled memory block for storing behavioral data of terminal
The behavioral data of domain detection service linked character.
By taking the application scenarios of advertisement as an example, the arbitrary client (such as Taobao, Tencent's news, the QQ that are installed in terminal
Music, U.S. group's take-away etc.) in the process of running, it can classify from the angle of the product involved by advertisement, such as from quick
The consumer goods, electronic product, clothes, food angle classify；Correspondingly, classification it is corresponding be characterized as consumption, electronic product,
Clothes, food；Terminal detects the behavioral data for meeting associated feature in the behavioral data of acquisition again, for example, can be at end
Hold the scheduled behavioral data for storing the storage region detection service linked character of behavioral data.
Step S202, terminal carry out clustering processing according to the behavioral data of itself, obtain operating in different dimensions in terminal
Operating characteristics.
In alternative embodiment of the present invention, cluster is carried out as not according to the tightness degree contacted between the sample of behavioral data
With grouping, operating characteristics are used as selected characteristic in the grouping of each cluster, are not excluded in the embodiment of the present invention using respectively
Kind clustering algorithm carries out clustering processing, such as hierarchical clustering, K mean cluster and neural network clustering to the sample of behavioral data
Deng.
Step S203, the feature that terminal-pair terminal clusters are predicted according to disaggregated model, are obtained current in terminal
The classification results of user.
In alternative embodiment of the present invention, the operation that terminal executes in the step is serviced with above-described embodiment step S104
The operating process that device executes is identical, is only the replacement of executive agent, which is not described herein again.
Step S204, terminal judge the replacement of user according to the classification results of active user.
In alternative embodiment of the present invention, terminal judge the classification of active user and the history classification of user in terminal away from
From beyond distance threshold, judge that user replaces.Wherein, the distance of classification refers to arranging shape according to the similarity of class of subscriber
At user's classification chart in distance, user classification distance it is bigger, illustrate that the uneven class size of user is bigger；The history of user point
Class can be that the last user classifies.Therefore, a kind of specific implementation mode is exactly that terminal judges the classification of active user,
When exceeding distance threshold at a distance from the last classification with user in terminal, judge that terminal user replaces.
In another alternative embodiment of the present invention, when terminal judges that the classification of active user is identical twice in succession, also, two
Secondary identical user's classification judges that user replaces in terminal from when the history classification of user is different in terminal；Here, terminal
The history classification of middle user refers to the last user's classification.
Step S205, terminal obtain the information for being adapted to corresponding classification results according to the replacement of user.
In alternative embodiment of the present invention, when terminal determines that user replaces, according to the classification results for replacing user, look into
It askes stereotactic conditions and meets the information of corresponding classification results, and the information inquired is presented on to the display interface of itself.
Here, the specific implementation process that terminal inquiry stereotactic conditions meet the information of corresponding classification results can be：Terminal
Request is sent to background server locally retrieving corresponding information or terminal, to obtain corresponding information.
For example, when the information that stereotactic conditions meet corresponding classification results is classical music, terminal can be deposited at itself
Retrieval character is classic music in the music file of storage；When the information that stereotactic conditions meet corresponding classification results is current events and important news
When, terminal sends to background server and asks, to obtain newest current events and important news.
Fig. 9 shows the optional flow diagram four of information push processing method provided in an embodiment of the present invention, by basis
Each step illustrates.
Step S301 to step S304 is identical with the operation that step S201 is executed to step S204.
The classification results of active user are sent to server by step S305, terminal.
Step S306, server obtain the information for being adapted to corresponding classification results according to the replacement of user.
For example, server, according to the classification results for replacing user, inquiry server stereotactic conditions meet corresponding classification results
Information, and the information that inquires is pushed to terminal, so that the information of push is presented on the display interface of terminal.
Below by taking advertisement pushing as an example, the flow of information push processing method of the embodiment of the present invention is illustrated.Figure 10
For an optional network architecture schematic diagram of information push processing method of the embodiment of the present invention, the present embodiments relate to installations
There are the terminal and server of client；Terminal includes：Smart mobile phone, tablet computer, car-mounted terminal and fixed terminal (desktop computer)
Deng terminal of the embodiment of the present invention can be any one in terminal 21, terminal 22, terminal 23 and terminal 24 shown in Fig. 1-1
Or it is multiple, server includes at least any one in server 11 to server 1n.In embodiments of the present invention, advertiser to
Server uploads advertisement, and sets fixed condition corresponding with advertisement.SDK is integrated in the client of terminal, client passes through
Operation SDK can complete acquisition of the client to the behavioral data of user, and the data after acquisition are reported to advertisement backstage
Server.Advertisement background server carries out clustering processing to the behavioral data of user, and predicts the classification results of user.After advertisement
Platform server pull meets the advertisement of classification results, and according to the ordering strategies such as bid ranking to meet the advertisements of classification results into
Row sequence.The dispensing end of advertisement background server launches the advertisement in sequencing queue according to Sort Priority, i.e., will row
Advertisement in sequence queue pushes to client according to priority and presents.
Below by taking news push as an example, the flow of information push processing method of the embodiment of the present invention is illustrated.Client
It collects the behavioral data that user is operated in the terminal and reports to server in end；Server is according to disaggregated model and behavior number
It is predicted that the classification results of the active user of client.Classification results of the server from multiple network platforms crawl and active user
The news to match；Server is ranked up the news of crawl according to strategies such as timeliness, clicking rates；It again will be in sequencing queue
News according to Sort Priority push to client present.
Based on explanation above-mentioned, it is possible to understand that the letter of the application program 167 of the information push processing function of Fig. 2 is realized on ground
The composed structure of breath push processing unit 100 as shown in figure 12 below illustrates the function of each unit.
Data capture unit 101, the behavioral data formed for obtaining the operation for detecting user in terminal；
Characteristics determining unit 102 obtains grasping in terminal for carrying out clustering processing according to the behavioral data for belonging to terminal
Make the operating characteristics in different dimensions；
Taxon 103, for being predicted according to disaggregated model the feature that terminal clusters, obtain in terminal when
The classification results of preceding user；
Judging unit 104, the replacement for judging user in terminal according to the classification results of active user in terminal；Push
Unit 105 realizes that adaptation replaces the information push of the corresponding classification results of user for the replacement according to user in terminal.
In a specific embodiment, data capture unit 101 are detecting that it is fixed that terminal meets specifically for obtaining terminal
When condition when, formation timing condition meet before acquire behavioral data, alternatively, obtain terminal generate characterization user replace
Potential event when, the behavioral data of formation acquired before generating potential event.
In a specific embodiment, data capture unit 101 are specifically used for the information residing for detection terminal and push field
Scape determines feature associated with corresponding push scene；Obtain corresponding terminal in detection meet associated feature operation and shape
At behavioral data.
In a specific embodiment, taxon 103, specifically for each dimensional characteristics for including according to classification-tree method
Priority descending, it is special to the operation of the respective dimensions obtained successively with the corresponding class condition of the operating characteristics of each dimension
Sign carries out classification judgement, until, meet the class condition of a feature in priority descending, by the corresponding classification of the feature met
It is determined as the classification results of active user in terminal.
In a specific embodiment, information push processing unit further includes：Sequencing unit 106, in user's sample
Concentrate the candidate feature for obtaining the different dimensions that each user's sample has；It calculates with the candidate feature of each dimension to user's sample set
When middle user classifies, corresponding information gain；
The highest candidate feature of information gain of selected predetermined quantity is the corresponding feature of respective dimensions, and is increased according to information
The descending of benefit forms the priority descending of the operating characteristics of respective dimensions.
In a specific embodiment, sequencing unit 106 are specifically used for calculating in user's sample set based on priori classification knot
The initial information entropy of fruit；Different class conditions is constructed with the operating characteristics of any dimension, calculates condition needle in different categories
When classifying to sample of users in sample set, corresponding reference information entropy；Calculate the difference of reference information entropy and initial information entropy
It is worth the comentropy as respective dimensions feature.
In a specific embodiment, information push processing unit further includes：Recognition unit 107 is used for from multiple terminals
When obtaining behavioral data, according to the hardware identifier that the behavioral data obtained carries, the obtained behavioral data ownership of identification
Terminal.
In a specific embodiment, push unit 105, specifically for when user replaces in terminal, according to more
It uses the classification results at family instead, inquires the information that stereotactic conditions in server meet corresponding classification results；It is inquired to terminal push
The information arrived.
In a specific embodiment, push unit 105, specifically for when user replaces in terminal,
According to the classification results for replacing user, inquires stereotactic conditions in terminal and meet the information of corresponding classification results and be in
It is existing.
In a specific embodiment, the judging unit 104 is specifically used for point as active user in the terminal
Exceed distance threshold at a distance from the history classification of user in class, with the terminal, judges that user replaces in terminal；Alternatively,
When the classification of active user in the terminal is identical twice in succession, and from when the history classification of user is different in the terminal,
Judge that user replaces in terminal.
It should be noted that：The information push processing unit that above-described embodiment provides into row information when pushing processing, only
With the division progress of above-mentioned each program module for example, in practical application, can as needed and by above-mentioned processing distribution by
Different program modules is completed, i.e., the internal structure of device is divided into different program modules, described above complete to complete
Portion or part are handled.In addition, information push processing unit and information push processing method embodiment that above-described embodiment provides
Belong to same design, specific implementation process refers to embodiment of the method, and which is not described herein again.
In the exemplary embodiment, the embodiment of the present invention additionally provides a kind of computer readable storage medium, such as including
The memory of executable program, the processor that above-mentioned executable program can be pushed processing unit by information executes, aforementioned to complete
Method and step.Computer readable storage medium can be FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic table
The memories such as face memory, CD or CD-ROM；Can also be including one of above-mentioned memory or that arbitrarily combines various set
It is standby, such as mobile phone, computer, tablet device, personal digital assistant, server.
The embodiment of the present invention also provides a kind of information push processing unit, including：Processor and for store can locate
The memory of the executable program run on reason device,
Wherein, processor is for when running executable program, executing：
The behavioral data for obtaining the operation of detection user in terminal and being formed；
Clustering processing is carried out according to the behavioral data for belonging to terminal, it is special in the operation of different dimensions to obtain operation in terminal
Sign；
The feature that terminal clusters is predicted according to disaggregated model, obtains the classification knot of active user in terminal
Fruit；
The replacement of user in terminal is judged according to the classification results of active user in terminal；
According to the replacement of user in terminal, realize that adaptation replaces the information push of the corresponding classification results of user.
When processor is additionally operable to operation executable program, execute：
Terminal is obtained when detecting that terminal meets timing condition, the behavior number of formation acquired before timing condition meets
According to, alternatively,
Terminal is obtained when generating the potential event that characterization user replaces, the row of formation acquired before generating potential event
For data.
When processor is additionally operable to operation executable program, execute：
Information residing for detection terminal pushes scene, determines feature associated with corresponding push scene；
It obtains and detects the operation for meeting associated feature in corresponding terminal and the behavioral data formed.
When processor is additionally operable to operation executable program, execute：
According to the priority descending for each dimensional characteristics that classification-tree method includes, corresponded to successively with the operating characteristics of each dimension
Class condition, classification judgement is carried out to the operating characteristics of the respective dimensions obtained, until,
The corresponding classification of the feature met is determined as in terminal by the class condition for meeting a feature in priority descending
The classification results of active user.
When processor is additionally operable to operation executable program, execute：
The candidate feature for the different dimensions that each user's sample has is obtained in user's sample set；
When calculating classifies to user in user's sample set with the candidate feature of each dimension, corresponding information gain；
The highest candidate feature of information gain of selected predetermined quantity is the corresponding feature of respective dimensions, and is increased according to information
The descending of benefit forms the priority descending of the operating characteristics of respective dimensions.
When processor is additionally operable to operation executable program, execute：
Calculate the initial information entropy based on priori classification result in user's sample set；
Different class conditions is constructed with the operating characteristics of any dimension, the condition in different categories that calculates is directed to sample set
When middle sample of users is classified, corresponding reference information entropy；
Calculate comentropy of the difference of reference information entropy and initial information entropy as respective dimensions feature.
When processor is additionally operable to operation executable program, execute：
When obtaining behavioral data from multiple terminals,
According to the hardware identifier that the behavioral data obtained carries, the terminal of the obtained behavioral data ownership of identification.
When processor is additionally operable to operation executable program, execute：
When user replaces in terminal, according to the classification results for replacing user, inquires stereotactic conditions in server and accord with
Close the information of corresponding classification results；Inquired information is pushed to terminal.
When processor is additionally operable to operation executable program, execute：
When user replaces in terminal, according to the classification results for replacing user, inquires stereotactic conditions in terminal and meet
The accordingly information of classification results and presentation.
When processor is additionally operable to operation executable program, execute：
When in the classification of active user in the terminal, with the terminal at a distance from the history classification of user beyond apart from threshold
Value, judges that user replaces in terminal；
Alternatively, the classification of active user is identical twice in succession in the terminal, and gone through with user in the terminal
When history classification is different, judge that user replaces in terminal.
The embodiment of the present invention also provides a kind of storage medium, is stored thereon with executable program, at the executable code
When managing device operation, execute：
The behavioral data for obtaining the operation of detection user in terminal and being formed；
Clustering processing is carried out according to the behavioral data for belonging to terminal, it is special in the operation of different dimensions to obtain operation in terminal
Sign；
The feature that terminal clusters is predicted according to disaggregated model, obtains the classification knot of active user in terminal
Fruit；
The replacement of user in terminal is judged according to the classification results of active user in terminal；
According to the replacement of user in terminal, realize that adaptation replaces the information push of the corresponding classification results of user.
When executable code processor is run, also execute：
Terminal is obtained when detecting that terminal meets timing condition, the behavior number of formation acquired before timing condition meets
According to, alternatively, terminal is obtained when generating the potential event that characterization user replaces, the row of formation acquired before generating potential event
For data.
When executable code processor is run, also execute：
Information residing for detection terminal pushes scene, determines feature associated with corresponding push scene；
It obtains and detects the operation for meeting associated feature in corresponding terminal and the behavioral data formed.
When executable code processor is run, also execute：
According to the priority descending for each dimensional characteristics that classification-tree method includes, corresponded to successively with the operating characteristics of each dimension
Class condition, classification judgement is carried out to the operating characteristics of the respective dimensions obtained, until,
The corresponding classification of the feature met is determined as in terminal by the class condition for meeting a feature in priority descending
The classification results of active user.
When executable code processor is run, also execute：
The candidate feature for the different dimensions that each user's sample has is obtained in user's sample set；
When calculating classifies to user in user's sample set with the candidate feature of each dimension, corresponding information gain；
The highest candidate feature of information gain of selected predetermined quantity is the corresponding feature of respective dimensions, and is increased according to information
The descending of benefit forms the priority descending of the operating characteristics of respective dimensions.
When executable code processor is run, also execute：
Calculate the initial information entropy based on priori classification result in user's sample set；
Different class conditions is constructed with the operating characteristics of any dimension, the condition in different categories that calculates is directed to sample set
When middle sample of users is classified, corresponding reference information entropy；
Calculate comentropy of the difference of reference information entropy and initial information entropy as respective dimensions feature.
When executable code processor is run, also execute：
When obtaining behavioral data from multiple terminals,
According to the hardware identifier that the behavioral data obtained carries, the terminal of the obtained behavioral data ownership of identification.
When executable code processor is run, also execute：
When user replaces in terminal,
According to the classification results for replacing user, the information that stereotactic conditions in server meet corresponding classification results is inquired；
Inquired information is pushed to terminal.
When executable code processor is run, also execute：
When user replaces in terminal,
According to the classification results for replacing user, information and presentation that stereotactic conditions in terminal meet corresponding classification results are inquired
The present invention.
Explanation based on above-described embodiment, it can be understood as in the related technology, pushed into row information for same terminal
Schematic diagram, as shown in figure 13, due to cannot clearly distinguish A, B, C class user using the terminal, and A, B, C three classes user distinguish
With different use features；Therefore, terminal or server are only capable of corresponding to A, B, C class user for the terminal presentation/push
Three kinds of information using feature；In presenting or pushing three kinds of information process using feature, accidentally presentation/push is certainly existed
The case where information or more presentation/pushed informations.If A classes user is in using terminal, and presentation/push is electronic product
Information belongs to the case where accidentally presentation/pushed information；The case where to avoid accidentally presentation/pushed information, three kinds of presentation/push may be selected
Using the information of feature, that is, the case where belonging to more presentation/pushed informations.In this way, not only reducing the effect and net of information popularization
Network resource utilization, and increase operation cost.
Using the above embodiment of the present invention information push processing method, it can judge that terminal is used according to the behavioral data of user
The replacement at family, and according to the replacement of terminal user, realize that adaptation replaces the information push of the corresponding classification results of user；Such as Figure 14 institutes
Show, for same terminal, active user's classification of distinguishing terminal can be specified, so as to according to active user's classification accurately push with
The information of class of subscriber adaptation.
In conclusion the embodiment of the present invention has the following technical effect that
1) the behavioral data extraction operation feature formed by the operation to user, the operating characteristics based on different dimensions
It predicts user's classification, according to being pushed into row information for user's classification suitability, can targetedly be carried out based on class of subscriber
Information pushes, and improves information push accuracy and network resource utilization.
2) by recording the behavioral data of user from global dimension, the operating characteristics of user's more various dimensions can be obtained, both
It can guarantee the validity of the behavioral data in a detection cycle, and the accuracy for determining class of subscriber can be improved, further carry
The accuracy of high information push.
3) by recording the behavioral data of user from specific type client, to judge the use using specific type client
Family is classified, and is classified come pushed information according to the user for specific type client, can be improved the directionality of information push.
4) terminal just obtains the behavior number of user when meeting timing condition or generating the potential event of characterization user's replacement
According to, can ensure the timeliness of behavioral data, further improve information push timeliness.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (15)
1. a kind of information push processing method, which is characterized in that including：
The behavioral data for obtaining the operation of detection user in terminal and being formed；
Clustering processing is carried out according to the behavioral data for belonging to the terminal, obtains grasping in the terminal from choosing in cluster result
Make the operating characteristics in different dimensions；
Selected obtained feature is predicted according to disaggregated model, obtains the classification results of active user in the terminal；
The replacement of user in terminal is judged according to the classification results of active user in terminal；
According to the replacement of user in terminal, realize that adaptation replaces the information push of the corresponding classification results of user.
2. the method as described in claim 1, which is characterized in that the acquisition detects the operation of user in the terminal and formed
Behavioral data, including：
Behavioral data is obtained at least one in the following ways：
It obtains when detecting that the terminal meets timing condition, the behavioral data of formation acquired before timing condition meets；
It obtains when generating the potential event that characterization user replaces in the terminal, formation acquires before generating potential event
Behavioral data；
Detect the terminal be in specific information push scene when, formation meet corresponding information push scene characteristic operation
And the behavioral data formed.
3. the method as described in claim 1, which is characterized in that the feature clustered to the terminal is according to classification mould
Type is predicted, the classification results of active user in the terminal are obtained, including：
When the disaggregated model uses classification-tree method,
According to the priority descending for each dimensional characteristics that the classification-tree method includes, successively with the operating characteristics of each dimension
Corresponding class condition carries out classification judgement to the operating characteristics of the respective dimensions obtained, until,
The corresponding classification of the feature met is determined as described by the class condition for meeting any feature in the priority descending
The classification results of active user in terminal.
4. the method as described in claim 1, which is characterized in that further include：
When the disaggregated model is classification-tree method,
The candidate feature for the different dimensions that each user's sample has is obtained in user's sample set；
When calculating classifies to user in user's sample set with the candidate feature of each dimension, the classification-tree method
Corresponding information gain；
The highest candidate feature of information gain of selected predetermined quantity is the corresponding feature of respective dimensions, and according to information gain
Descending forms the priority descending of the operating characteristics of respective dimensions.
5. method as claimed in claim 4, which is characterized in that the calculating is with the candidate feature of each dimension to the use
When user classifies in the sample set of family, the corresponding information gain of the classification-tree method, including：
Calculate the initial information entropy based on priori classification result in user's sample set；
Different class conditions is constructed with the operating characteristics of any dimension, calculates and institute is directed to the different class condition
When stating that sample of users is classified in sample set, corresponding reference information entropy；
Calculate information gain of the difference of the reference information entropy and the initial information entropy as respective dimensions feature.
6. the method as described in claim 1, which is characterized in that the classification results according to active user in terminal judge eventually
The replacement of user in end, including：
When in the classification of active user in the terminal, with the terminal user history classification at a distance from exceed distance threshold,
Judge that user replaces in terminal；
Alternatively, dividing when the classification of active user in the terminal is identical twice in succession, and with the history of user in the terminal
When class difference, judge that user replaces in terminal.
7. the method as described in claim 1 to 6, which is characterized in that further include：
When obtaining behavioral data from multiple terminals,
According to the hardware identifier that the behavioral data obtained carries, the different terminals of the obtained behavioral data ownership of identification.
8. the method as described in claim 1 to 6, which is characterized in that described to realize that adaptation replaces the corresponding classification results of user
Information pushes, including：
When user replaces in the terminal,
According to the classification results for replacing user, inquiry stereotactic conditions meet the information of corresponding classification results；
Before user replaces herein in the terminal, inquired information is pushed to the terminal.
9. the method as described in claim 1 to 6, which is characterized in that described to realize that adaptation replaces the corresponding classification results of user
Information pushes, including：
When user replaces in the terminal,
According to the classification results for replacing user, inquires stereotactic conditions in the terminal and meet the information of corresponding classification results and be in
It is existing.
10. a kind of information pushes processing unit, which is characterized in that including：
Data capture unit, the behavioral data formed for obtaining the operation for detecting user in terminal；
Characteristics determining unit is selected for carrying out clustering processing according to the behavioral data for belonging to the terminal from cluster result
Obtain the operating characteristics operated in the terminal in different dimensions；
Taxon obtains currently using in the terminal for predicting selected obtained feature according to disaggregated model
The classification results at family；
Judging unit, the replacement for judging user in terminal according to the classification results of active user in terminal；
Push unit realizes that adaptation replaces the information push of the corresponding classification results of user for the replacement according to user in terminal.
11. information as claimed in claim 10 pushes processing unit, which is characterized in that
The taxon, specifically for when the disaggregated model uses classification-tree method, including according to classification-tree method
The priority descending of each dimensional characteristics, successively with the corresponding class condition of operating characteristics of each dimension, to the phase obtained
The operating characteristics of dimension are answered to carry out classification judgement, until,
The corresponding classification of the feature met is determined as described by the class condition for meeting any feature in the priority descending
The classification results of active user in terminal.
12. information as claimed in claim 10 pushes processing unit, which is characterized in that further include：
Sequencing unit, for when the disaggregated model is classification-tree method, obtaining each user's sample tool in user's sample set
The candidate feature of some different dimensions；
When calculating classifies to user in user's sample set with the candidate feature of each dimension, the classification-tree method
Corresponding information gain；
The highest candidate feature of information gain of selected predetermined quantity is the corresponding feature of respective dimensions, and according to information gain
Descending forms the priority descending of the operating characteristics of respective dimensions.
13. information as claimed in claim 12 pushes processing unit, which is characterized in that
The sequencing unit is specifically used for calculating the initial information entropy based on priori classification result in user's sample set；
Different class conditions is constructed with the operating characteristics of any dimension, calculates and institute is directed to the different class condition
When stating that sample of users is classified in sample set, corresponding reference information entropy；
Calculate information gain of the difference of the reference information entropy and the initial information entropy as respective dimensions feature.
14. a kind of information pushes processing unit, which is characterized in that including：
Memory, for storing executable program；
Processor when executable program for executing memory storage, realizes claim 1 to 9 any one of them
Information push processing method.
15. a kind of storage medium, which is characterized in that it is stored with executable program, when the executable code processor executes,
Realize claim 1 to 9 any one of them information push processing method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710647371.2A CN108304432B (en) | 2017-08-01 | 2017-08-01 | Information push processing method, information push processing device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710647371.2A CN108304432B (en) | 2017-08-01 | 2017-08-01 | Information push processing method, information push processing device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108304432A true CN108304432A (en) | 2018-07-20 |
| CN108304432B CN108304432B (en) | 2021-09-07 |
Family
ID=62872582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710647371.2A Active CN108304432B (en) | 2017-08-01 | 2017-08-01 | Information push processing method, information push processing device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108304432B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101658A (en) * | 2018-08-31 | 2018-12-28 | 优视科技新加坡有限公司 | Information search method, device and equipment/terminal/server |
| CN109509028A (en) * | 2018-11-15 | 2019-03-22 | 北京奇虎科技有限公司 | A kind of advertisement placement method and device, storage medium, computer equipment |
| CN109561424A (en) * | 2018-11-30 | 2019-04-02 | 维沃移动通信（深圳）有限公司 | A kind of Data Identification generation method and mobile terminal |
| CN110248217A (en) * | 2019-07-08 | 2019-09-17 | 中国联合网络通信集团有限公司 | A kind of user data synchronization method and device |
| CN113260989A (en) * | 2018-12-19 | 2021-08-13 | 奥迪股份公司 | Method and computer program for receiving, managing and outputting user-related data files of a plurality of different data types on a user interface of a device, and device for storing and operating a computer program |
| CN115187344A (en) * | 2022-09-13 | 2022-10-14 | 南通久拓智能装备有限公司 | Big data-based user preference analysis and identification method |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH10247198A (en) * | 1997-03-05 | 1998-09-14 | Nippon Telegr & Teleph Corp <Ntt> | Taste sorting method and device |
| US20050177846A1 (en) * | 2003-11-21 | 2005-08-11 | Canon Kabushiki Kaisha | Contents extraction method, contents extraction apparatus, contents information display method and apparatus |
| CN101339562A (en) * | 2008-08-15 | 2009-01-07 | 北京航空航天大学 | Portal personalized recommendation service system introducing into interest model feedback and update mechanism |
| US20110302032A1 (en) * | 2010-06-08 | 2011-12-08 | Kazuo Ishii | Content recommendation device and content recommendation method |
| CN103516588A (en) * | 2012-06-30 | 2014-01-15 | 北京神州泰岳软件股份有限公司 | Method and system of background processing of client-side |
| CN104008184A (en) * | 2014-06-10 | 2014-08-27 | 百度在线网络技术（北京）有限公司 | Method and device for pushing information |
| CN104933075A (en) * | 2014-03-20 | 2015-09-23 | 百度在线网络技术（北京）有限公司 | User attribute predicting platform and method |
| CN105005593A (en) * | 2015-06-30 | 2015-10-28 | 北京奇艺世纪科技有限公司 | Scenario identification method and apparatus for multi-user shared device |
| CN105404680A (en) * | 2015-11-25 | 2016-03-16 | 百度在线网络技术（北京）有限公司 | Searching recommendation method and apparatus |
| CN106131703A (en) * | 2016-06-28 | 2016-11-16 | 青岛海信传媒网络技术有限公司 | A kind of method of video recommendations and terminal |
-
2017
- 2017-08-01 CN CN201710647371.2A patent/CN108304432B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH10247198A (en) * | 1997-03-05 | 1998-09-14 | Nippon Telegr & Teleph Corp <Ntt> | Taste sorting method and device |
| US20050177846A1 (en) * | 2003-11-21 | 2005-08-11 | Canon Kabushiki Kaisha | Contents extraction method, contents extraction apparatus, contents information display method and apparatus |
| CN101339562A (en) * | 2008-08-15 | 2009-01-07 | 北京航空航天大学 | Portal personalized recommendation service system introducing into interest model feedback and update mechanism |
| US20110302032A1 (en) * | 2010-06-08 | 2011-12-08 | Kazuo Ishii | Content recommendation device and content recommendation method |
| CN103516588A (en) * | 2012-06-30 | 2014-01-15 | 北京神州泰岳软件股份有限公司 | Method and system of background processing of client-side |
| CN104933075A (en) * | 2014-03-20 | 2015-09-23 | 百度在线网络技术（北京）有限公司 | User attribute predicting platform and method |
| CN104008184A (en) * | 2014-06-10 | 2014-08-27 | 百度在线网络技术（北京）有限公司 | Method and device for pushing information |
| CN105005593A (en) * | 2015-06-30 | 2015-10-28 | 北京奇艺世纪科技有限公司 | Scenario identification method and apparatus for multi-user shared device |
| CN105404680A (en) * | 2015-11-25 | 2016-03-16 | 百度在线网络技术（北京）有限公司 | Searching recommendation method and apparatus |
| CN106131703A (en) * | 2016-06-28 | 2016-11-16 | 青岛海信传媒网络技术有限公司 | A kind of method of video recommendations and terminal |
Non-Patent Citations (2)
| Title |
|---|
| AMIN JAVARI等: "Accurate and Novel Recommendations: An Algorithm Based on Popularity Forecasting", 《ACM TRANSACTIONS ON INTELLIGENT SYSTEMS AND TECHNOLOGY》 * |
| 黄皓璇等: "基于用户兴趣变化的 Slope One 协同过滤推荐算法", 《工业控制计算机》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101658A (en) * | 2018-08-31 | 2018-12-28 | 优视科技新加坡有限公司 | Information search method, device and equipment/terminal/server |
| CN109509028A (en) * | 2018-11-15 | 2019-03-22 | 北京奇虎科技有限公司 | A kind of advertisement placement method and device, storage medium, computer equipment |
| CN109561424A (en) * | 2018-11-30 | 2019-04-02 | 维沃移动通信（深圳）有限公司 | A kind of Data Identification generation method and mobile terminal |
| CN109561424B (en) * | 2018-11-30 | 2021-08-27 | 维沃移动通信（深圳）有限公司 | Data identifier generation method and mobile terminal |
| CN113260989A (en) * | 2018-12-19 | 2021-08-13 | 奥迪股份公司 | Method and computer program for receiving, managing and outputting user-related data files of a plurality of different data types on a user interface of a device, and device for storing and operating a computer program |
| CN110248217A (en) * | 2019-07-08 | 2019-09-17 | 中国联合网络通信集团有限公司 | A kind of user data synchronization method and device |
| CN110248217B (en) * | 2019-07-08 | 2022-04-22 | 中国联合网络通信集团有限公司 | User data synchronization method and device |
| CN115187344A (en) * | 2022-09-13 | 2022-10-14 | 南通久拓智能装备有限公司 | Big data-based user preference analysis and identification method |
| CN115187344B (en) * | 2022-09-13 | 2022-12-09 | 南通久拓智能装备有限公司 | Big data-based user preference analysis and identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108304432B (en) | 2021-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10878550B2 (en) | Utilizing deep learning to rate attributes of digital images | |
| CN108304432A (en) | Information push processing method, information push processing unit and storage medium | |
| US11037060B2 (en) | Analyzing sequence data using neural networks | |
| Tennant et al. | Scalable real-time classification of data streams with concept drift | |
| Aggarwal et al. | Event detection in social streams | |
| Chen et al. | Personalized QoS-aware web service recommendation and visualization | |
| CN110909182A (en) | Multimedia resource searching method and device, computer equipment and storage medium | |
| US10587709B1 (en) | Determining session intent | |
| WO2013189296A1 (en) | Method and system for processing recommended target software | |
| CN103117903A (en) | Internet surfing unusual flow detection method and device | |
| CN108664515B (en) | A kind of searching method and device, electronic equipment | |
| US20210073669A1 (en) | Generating training data for machine-learning models | |
| Ko et al. | Network prediction with traffic gradient classification using convolutional neural networks | |
| TWI544348B (en) | Open-ended detection and categorization of word clusters in text data | |
| Roy et al. | Social multimedia signals | |
| WO2022148108A1 (en) | Systems, devices and methods for distributed hierarchical video analysis | |
| Keyvanpour | A survey on community detection methods based on the nature of social networks | |
| CN116823410B (en) | Data processing method, object processing method, recommending method and computing device | |
| CN116484105B (en) | Service processing method, device, computer equipment, storage medium and program product | |
| Ying et al. | FrauDetector+ An Incremental Graph-Mining Approach for Efficient Fraudulent Phone Call Detection | |
| CN112925964A (en) | Big data acquisition method based on cloud computing service and big data acquisition service system | |
| Capdevila et al. | Mining urban events from the tweet stream through a probabilistic mixture model | |
| Keneshloo et al. | Predicting the shape and peak time of news article views | |
| CN116091133A (en) | Target object attribute identification method, device and storage medium | |
| Rozario et al. | Community detection in social network using temporal data |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221117 Address after: 1402, Floor 14, Block A, Haina Baichuan Headquarters Building, No. 6, Baoxing Road, Haibin Community, Xin'an Street, Bao'an District, Shenzhen, Guangdong 518133 Patentee after: Shenzhen Yayue Technology Co.,Ltd. Address before: 518000 Tencent Building, No. 1 High-tech Zone, Nanshan District, Shenzhen City, Guangdong Province, 35 Floors Patentee before: TENCENT TECHNOLOGY (SHENZHEN) Co.,Ltd. |