Summary of the invention
In order to solve above technical matters, the invention provides the off-grid method and apparatus of a kind of predictive user.
The invention discloses the off-grid method of a kind of predictive user, comprising:
The distribution moments configuration website list of elements is given the registered user of login, and the initial value of each element is set;
Write down all operations of said registered user;
Described registered user's operation is gathered, calculate the average active value of user;
Judge whether the average active value of a plurality of different users is downward trend in the predetermined amount of time, if, confirm as from network users, if not, then return the registered user that the step distribution moments configuration website list of elements is given login, the initial value of each element is set.
Preferably, the average active value computing method of said user are: establishing the average active value of user is H
d, wherein d is the date, establishing ∑ is the element value sum, establishes T for using the mean value algorithm to the valuation that all users' ∑ calculates, and establishes n and operates number for the user, then
Preferably, the initial value that also comprises the element in the website list of elements that resets.
Preferably, the average active value of described calculating user comprises: network is the server distribution of many clusters, needs first combined data, and then calculates.
Preferably; The matrix form website list of elements of stating comprises horizontal row and longitudinal row; The said user behavior of laterally classifying as is classified, and described user behavior classification comprises: browse, deliver, comment on, share, delete, described at least one website function sets classification of vertically classifying as; Described website function sets classification comprises: strange thing, photograph album, daily record, good friend, BLOG.
The off-grid device of a kind of predictive user is used for said method, comprising:
List of elements allocation units are used for the registered user that the distribution moments configuration website list of elements is given login, set the initial value of each element value;
The user operation records unit links to each other with described list of elements allocation units, is used to write down all operations of said registered user;
Average active value computing unit links to each other with described user operation records unit, is used for described registered user's operation is gathered, and calculates average active value of user and preservation;
User behavior trend judging unit links to each other with described average active value computing unit, judges whether the average active value of a plurality of different users is downward trend in the predetermined amount of time.
Preferably, said average active value computing unit adopts following formula to calculate the average active value H of user
d: wherein d is the date, and establishing ∑ is the element value sum, establishes T for using the mean value algorithm to the valuation that all users' ∑ calculates, and establishes n and operates number for the user, then
Preferably, also comprise the element value reset unit that is arranged between described average active value computing unit and the downtrending judging unit, the initial value of the element of the website list of elements that is used to reset.
Preferably, the average active value of described calculating user comprises: network is the server distribution of many clusters, needs first combined data, and then calculates.
Preferably; The matrix form website list of elements comprises horizontal row and longitudinal row; The described user behavior of laterally classifying as is classified; Described user behavior classification comprises: browses, delivers, comments on, shares, deletes, and described at least one website function sets classification of vertically classifying as, described website function sets classification comprises: strange thing, photograph album, daily record, good friend, BLOG.
The off-grid method and apparatus of a kind of predictive user of embodiment of the present invention; Have following beneficial technical effects: the off-grid method of a kind of predictive user of the present invention, propose a kind of user behavior and excavate mode, cooperate corresponding algorithm; Calculate user's benchmark liveness; From the trend of user's liveness, foresee the off-grid possibility of user in advance, make things convenient for the operation personnel can make its corresponding measures.
Embodiment
By specifying technology contents of the present invention, structural attitude, realized purpose and effect, give explanation below in conjunction with embodiment and conjunction with figs. are detailed.
The objective of the invention is to propose a kind of user behavior and excavate mode, cooperate corresponding algorithm, calculate user's benchmark liveness,, foresee the off-grid possibility of user in advance, make the operation personnel can make its corresponding measures from the trend of user's liveness.
The present invention has used existing Kalman filtering algorithm to calculate the mean value of each unit every day, operating limit Algorithm Analysis user behavior trend then.Can certainly go out the mean value and the analysis user behavior trend of each unit every day with other algorithm computation.
Kalman filtering: a representative instance of Kalman filtering from one group limited, comprise noise, the observation sequence (having deviation) of object space is doped the coordinate and the speed of the position of object.
Kalman filtering algorithm can have good noise reduction, improves the accuracy of the average active value of user that calculates of the present invention, and the user's trend comparison that makes the present invention obtain is accurate.
In order to solve above technical matters, the invention provides the off-grid method and apparatus of a kind of predictive user.
Embodiment 1, sees also Fig. 1, the invention discloses the off-grid method of a kind of predictive user, comprising:
S1. the distribution moments configuration website list of elements is given the registered user of login, and the initial value of each element is set;
The matrix form website list of elements of present embodiment comprises horizontal row and longitudinal row; Laterally classify the user behavior classification as; The user behavior classification comprises: browse, deliver, comment on, share, delete; Vertically classify at least one website function sets classification as, website function sets classification comprises: strange thing, photograph album, daily record, good friend, BLOG.
Concrete implementation at first be set at a user behavior matrix to all operations of a social network sites; Longitudinal row is classified with user behavior (browse, deliver, upgrade); Laterally be (strange thing, photograph album) classification with one or more website same nature function set; Each element is represented the common factor of active user's behavior and website function, and is as shown in Figure 2, and initial value is 0; As shown in Figure 3, the user whenever does an operation, and the element value of respective operations is from increasing 1.
S2. write down all operations of said registered user;
Present embodiment can be used the form record registered user's of log record operation, also can use other common type record group side users' operation.
S3. described registered user's operation is gathered, calculate the average active value of user;
The concrete all operations to the user all carries out log record, and gather the daily record of user's operation to statistical server every day, and statistical server filters and adds up daily record every day, finally calculates user's active value.
S4. judge whether the average active value of a plurality of different users is downward trend in the predetermined amount of time, if, confirm as from net, if not, then return S1.
A kind of computing method of the average active value of calculating user of the present invention are: establishing the average active value of user is H
d, wherein d is the date, establishing ∑ is the element value sum, establishes T for using the mean value algorithm to the valuation that all users' ∑ calculates, and establishes n and operates number for the user, then
Wherein the mean value algorithm specifically is a Kalman filtering algorithm.
Because each is variant for each professional user capture and solicit operation amount, therefore be set at a unit to the boundary element set of functionalities or functionalities set and user behavior, the inner element value sum in unit is a ∑; Use Kalman filtering algorithm that all users' ∑ is calculated valuation T; Active user's ∑ is active user's a active index with the ratio of T value, and each cell-average active index of establishing user same day is A, the trend of statistics active user A value in a period of time; Need to use filtering algorithm to filter out the situation that the user does not reach the standard grade once in a while; If taper off shape, then can be judged as, and need notice operation personnel to make related measure from net purpose user.
Embodiment 2
As shown in Figure 4, present embodiment and embodiment 1 difference also comprise
S31: the element value initial value in the website list of elements that resets.Resetting when can be morning every day, also can be that other times reset.
A website element allocation list is set on the backstage, and each element can be provided with horizontal classification and vertically classification, and the user lands the foreground every day; Automatically for it distributes an interim website list of elements, with array mode record, the initial value of each element is 0; Be buffered on the server, each user had and only had an element array every day, each step operation that the user carries out; Upgrade the value of this element, morning every day to the same day all user's data unify to preserve, and empty buffer memory; To the server distribution of many clusters, need first combined data, and then calculate.Set a set time, data are put in order and calculated.
Like Fig. 3, as an example of the present invention, Fig. 3 is the local module element in paradise, supposes that a is that strange thing is professional; B is that photograph album is professional, and A is for browsing behavior, and B is an operation behavior, and promptly corresponding M21 is for browsing strange thing; M22 is the comment strange thing, and M23 is for sharing strange thing, and M31 is an Album for glancing over pictures, and M32 is a uploading pictures; M33 supposes that for revising front cover first day aA zone TaA value is 10, and aB zone TaB value is 5, and bA zone TbA value is 8; BB zone TbB value is 3, and certain user's ∑ aA is 6, ∑ aB is 1, ∑ bA is 3, and ∑ bB is 0, and then first day active index of user does
Second day aA zone TaA value is 13, and aB zone TaB value is 4, and the regional TbA value of bA is 6, and the regional TbB value of bB is 4, and this user's ∑ aA is 6, ∑ aB is 1, ∑ bA is 3, and ∑ bB is 0, and then first day active index of user does
So analogize,, can judge that this user is for being about to from network users when this user's active index is when continuing downtrending.
The present invention can shift to an earlier date the off-grid tendency of predictive user, and can draw what of user that each local module participates in and some movable actual effect through data.What the present invention studied is user's behavior trend, increases perhaps to delete certain local module, can not bring negative influence to real data, the many more agendas trends that can show the user more of local module.
Embodiment 4, see also Fig. 5, the off-grid device of a kind of predictive user, are used to realize above-mentioned method, comprising:
List of elements allocation units 10, user operation records unit 20, average active value computing unit 30, downtrending judging unit 40, element value reset unit.
List of elements allocation units 10 are used for the registered user that the distribution moments configuration website list of elements is given login, and the initial value of each element is set;
The matrix form website list of elements of present embodiment comprises horizontal row and longitudinal row; Laterally classify the user behavior classification as; The user behavior classification comprises: browse, deliver, comment on, share, delete; Vertically classify at least one website function sets classification as, website function sets classification comprises: strange thing, photograph album, daily record, good friend, BLOG.
Concrete implementation at first be set at a user behavior matrix to all operations of a social network sites; Longitudinal row is classified with user behavior (browse, deliver, upgrade); Laterally be (strange thing, photograph album) classification with one or more website same nature function set; Each element is represented the common factor of active user's behavior and website function, and is as shown in Figure 2, and initial value is 0; As shown in Figure 3, the user whenever does an operation, and the element value of respective operations is from increasing 1.
User operation records unit 20 links to each other with list of elements allocation units 10, is used to write down all operations of said registered user; Present embodiment can be used the form record registered user's of log record operation, also can use other common type record group side users' operation.
Average active value computing unit 30 links to each other with user operation records unit 20, is used for described registered user's operation is gathered, and calculates the average active value of user;
The concrete all operations to the user all carries out log record, and gather the daily record of user's operation to statistical server every day, and statistical server filters and adds up daily record every day, finally calculates user's active value;
Downtrending judging unit 40 links to each other with average active value computing unit 30, is used to judge whether the average active value of a plurality of different users is downward trend in the predetermined amount of time, if, confirm as from net, if not, then return list of elements allocation units 10.
A kind of computing method of the average active value of calculating user of the present invention are: establishing the average active value of user is H
d, wherein d is the date, establishing ∑ is the element value sum, establishes T for using the mean value algorithm to the valuation that all users' ∑ calculates, and establishes n and operates number for the user, then
Wherein the mean value algorithm specifically is a Kalman filtering algorithm.
Because each is variant for each professional user capture and solicit operation amount, therefore be set at a unit to the boundary element set of functionalities or functionalities set and user behavior, the inner element value sum in unit is a ∑; Use Kalman filtering algorithm that all users' ∑ is calculated valuation T; Active user's ∑ is active user's a active index with the ratio of T value, and each cell-average active index of establishing user same day is A, the trend of statistics active user A value in a period of time; Need to use filtering algorithm to filter out the situation that the user does not reach the standard grade once in a while; If taper off shape, then can be judged as, and need notice operation personnel to make related measure from net purpose user.
As shown in Figure 4, present embodiment and embodiment 1 difference also comprise the element value initial value in the website list of elements that resets.Resetting when can be morning every day, also can be that other times reset.
A website element allocation list is set on the backstage, and each element can be provided with horizontal classification and vertically classification, and the user lands the foreground every day; Automatically for it distributes an interim website list of elements, with array mode record, the initial value of each element is 0; Be buffered on the server, each user had and only had an element array every day, each step operation that the user carries out; Upgrade the value of this element, morning every day to the same day all user's data unify to preserve, and empty buffer memory; To the server distribution of many clusters, need first combined data, and then calculate.Set a set time, data are put in order and calculated.
Embodiment preferably also comprises being provided with element value reset unit (among the figure for illustrating), is used for the element value in the timing reset website list of elements; Can be to carry out reset operation morning every day; The average active value of described calculating user comprises: if network is the server distribution of many clusters, needs first combined data, and then calculate; The described user behavior of laterally classifying as is classified; Comprise: browse, deliver, comment on, share, delete, described at least one website function sets classification of vertically classifying as comprises: strange thing, photograph album, daily record, good friend, BLOG.
The off-grid method and apparatus of a kind of predictive user of embodiment of the present invention has following beneficial technical effects:
The present invention proposes a kind of user behavior and excavates mode, cooperates certain algorithm, calculates user's benchmark liveness, from the trend of user's liveness, foresees the off-grid possibility of user in advance, makes the operation personnel can make its corresponding measures.
Combine accompanying drawing that embodiments of the invention are described above; But the present invention is not limited to above-mentioned embodiment, and above-mentioned embodiment only is schematically, rather than restrictive; Those of ordinary skill in the art is under enlightenment of the present invention; Not breaking away under the scope situation that aim of the present invention and claim protect, also can make a lot of forms, these all belong within the protection of the present invention.