CN107818473A - A kind of method and device for judging loyal user - Google Patents
A kind of method and device for judging loyal user Download PDFInfo
- Publication number
- CN107818473A CN107818473A CN201610821820.6A CN201610821820A CN107818473A CN 107818473 A CN107818473 A CN 107818473A CN 201610821820 A CN201610821820 A CN 201610821820A CN 107818473 A CN107818473 A CN 107818473A
- Authority
- CN
- China
- Prior art keywords
- user
- time
- time series
- loyal
- series
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
Landscapes
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Telephonic Communication Services (AREA)
Abstract
The application is related to data statistics technical field, discloses a kind of method and device for judging loyal user, accurately judges the loyal user for specifying viewing mode to realize.This method is：According to user to the historical data caused by setting application mode using behavior, extract very first time arrangement set, the second time series set with stationarity feature and the 3rd time series set with non-stationary feature in very first time arrangement set are determined, the user in the second time series set is defined as the first loyal user；For each time series in the 3rd time series set, the result of the growth trend using characteristic value is obtained, and according to the result got, selects the second loyal user in the 3rd time series set.
Description
Technical field
The application is related to data statistics technical field, more particularly to a kind of method and device for judging loyal user.
Background technology
At present, with the variation of application mode, user is to the occupation mode of a certain product or to a certain viewing mode
Often can not be constant all the time.Analysis user contributes to producer to be directed to product in industry the loyalty of application-specific mode
Property is improved.For example, as user watches, program mode is more and more, and the viewing mode of live telecast can not turn into limit
User processed watches the sole mode of program, and each TV industry needs to continue to optimize the service quality of live telecast, to attract more
The loyal user of more live telecasts.
However, the recognition methods at present on the loyal user of application-specific mode is fewer, therefore, how accurate judgement
The loyal user for going out certain application mode is urgent problem to be solved.
The content of the invention
The embodiment of the present application provides a kind of method and device for judging loyal user, accurately judges to specify to realize
The loyal user of application mode.
The concrete technical scheme that the embodiment of the present application provides is as follows：
A kind of method for judging loyal user, including：
According to user to setting application mode caused by apply behavior historical data, extract very first time arrangement set,
Several time serieses are included in the very first time arrangement set, wherein, included in a time series according to one
Several generated caused by user using the historical data of behavior apply characteristic value；Determine the very first time arrangement set
In the second time series set with stationarity feature and the 3rd time series set with non-stationary feature；By described in
User corresponding to second time series set is defined as the first loyal user of the setting application mode；During for the described 3rd
Between each time series in arrangement set, obtain the result of the growth trend using characteristic value, and according to getting
As a result, the second loyal user of the setting application mode is selected from user corresponding to the 3rd time series set.
So, the recall rate of algorithm is improved, is capable of detecting when more loyal users so that judge the side of loyal user
More precisely, judged result provides definite foundation to set the data analysis of viewing mode, obtains it method closer to truth
To more preferable improvement and development.
Optionally, the result of the growth trend using characteristic value is obtained, including：
The each time series included for the 3rd time series set, calculating two neighboring application characteristic value is in
The frequency n of growth trend；And
It is in increase to calculate two neighboring application characteristic value in each time series that the 3rd time series set includes
The number summation N of trend；
Result using n and N ratio n/N as the growth trend using characteristic value；N and n is natural number, and n≤
N。
Optionally, according to the result got, the second loyal user is selected, including：
When the ratio meets following condition for the moment, the user of time series corresponding to the ratio is defined as described
Second loyal user：
The ratio is in the first setting range；Or
The ratio in the second setting range, and the latter half in time series corresponding to the ratio include should
Showed a rising trend with characteristic value；
Wherein, the value in first setting range is all higher than the value in second setting range.
Optionally, determine that there is the second time series set of stationarity feature, bag in the very first time arrangement set
Include：
Unit root ADF inspections are carried out to the time series in the very first time arrangement set using the model equation of setting
Test, judge have stationarity special in the very first time arrangement set according to whether ADF verification characteristics statistic refuses null hypothesis
4th time series set of sign；And
Judge that there is stationarity in the very first time arrangement set using auto-correlation coefficient figure and PARCOR coefficients figure
5th time series set of feature；
When choosing the common factor in the 4th time series set and the 5th time series set as described second
Between arrangement set.
Optionally, according to a user to applying the historical data of behavior, extraction described the caused by setting application mode
A time series in one time arrangement set, including：
According to a user to the historical data caused by the setting application mode using behavior, extracted for the first sub- time
Sequence and the second Time Sub-series, comprising being generated according to caused by a user using behavior in first Time Sub-series
Several the first subcharacter values, comprising being generated according to caused by a user using behavior in second Time Sub-series
Several the second subcharacter values；
According to the correlation of first Time Sub-series and second Time Sub-series, the first Time Sub-series are determined
The first fitting coefficient and second Time Sub-series the second fitting coefficient；
Based on first fitting coefficient and second fitting coefficient, by first Time Sub-series and described second
Time Sub-series are fitted to one time series.
Optionally, the described first loyal user and the second loyal user are the loyal use of the setting application mode
Family, then it is determined that after the first loyal user and the second loyal user, in addition to：
Analyze the quantity of the loyal user of the quantity of the loyal user and the setting application mode of historical record
Variation tendency；
The result of the variation tendency obtained according to analysis, analyze Service Quality of the setting application mode to user
Amount；
When the service quality is less than the first given threshold, warning information is issued.
Optionally, the described first loyal user and the second loyal user are the loyal use of the setting application mode
Family, then it is determined that after the first loyal user and the second loyal user, in addition to：
Recommend the loyal user application mode related to the setting application mode.
Optionally, the described first loyal user and the second loyal user are the loyal use of the setting application mode
Family, methods described also include：
According to the result of determination of the loyal user, the possibility that each user is lost in for the setting application mode is estimated
Property size；
When the possibility size that user is lost in for the setting application mode is higher than the second given threshold, early warning is issued
Information.
Optionally, the setting application mode includes：Live viewing mode, the viewing mode of carousel, the viewing of program request
Mode, or the occupation mode to specific products.
The embodiment of the present application additionally provides a kind of device for judging loyal user, including：
Extraction unit, for, to the historical data caused by setting application mode using behavior, extracting first according to user
Time series set, several time serieses are included in the very first time arrangement set, wherein, in a time series
Comprising according to caused by a user apply behavior historical data generated several apply characteristic value；
Determining unit, for the second time series collection for determining that there is stationarity feature in the very first time arrangement set
Close and have the 3rd time series set of non-stationary feature；User corresponding to the second time series set is defined as
First loyal user of the setting application mode；
Acquiring unit, for applying characteristic value for each time series in the 3rd time series set, acquisition
Growth trend result；
Selecting unit, the result got for basis, from user corresponding to the 3rd time series set
Select the second loyal user of the setting application mode；
Wherein, the described first loyal user and the second loyal user are the loyal use of the setting viewing mode
Family.
So, the recall rate of algorithm is improved, is capable of detecting when more loyal users so that judge the side of loyal user
More precisely, judged result provides definite foundation to set the data analysis of viewing mode, obtains it method closer to truth
To more preferable improvement and development.
Optionally, the acquiring unit is specifically used for：
The each time series included for the 3rd time series set, calculating two neighboring application characteristic value is in
The frequency n of growth trend；And
It is in increase to calculate two neighboring application characteristic value in each time series that the 3rd time series set includes
The number summation N of trend；
Result using n and N ratio n/N as the growth trend using characteristic value；N and n is natural number, and n≤
N。
Optionally, the selecting unit is specifically used for：
When the ratio meets following condition for the moment, the user of time series corresponding to the ratio is defined as described
Second loyal user：
The ratio is in the first setting range；Or
The ratio in the second setting range, and the latter half in time series corresponding to the ratio include should
Showed a rising trend with characteristic value；
Wherein, the value in first setting range is all higher than the value in second setting range.
Optionally, the determining unit is specifically used for：
Unit root ADF inspections are carried out to the time series in the very first time arrangement set using the model equation of setting
Test, judge have stationarity special in the very first time arrangement set according to whether ADF verification characteristics statistic refuses null hypothesis
4th time series set of sign；And
Judge that there is stationarity in the very first time arrangement set using auto-correlation coefficient figure and PARCOR coefficients figure
5th time series set of feature；
When choosing the common factor in the 4th time series set and the 5th time series set as described second
Between arrangement set.
Optionally, the extraction unit is used for：
According to a user to the historical data caused by the setting application mode using behavior, extracted for the first sub- time
Sequence and the second Time Sub-series, comprising being generated according to caused by a user using behavior in first Time Sub-series
Several the first subcharacter values, comprising being generated according to caused by a user using behavior in second Time Sub-series
Several the second subcharacter values；
According to the correlation of first Time Sub-series and second Time Sub-series, the first Time Sub-series are determined
The first fitting coefficient and second Time Sub-series the second fitting coefficient；
Based on first fitting coefficient and second fitting coefficient, by first Time Sub-series and described second
Time Sub-series are fitted to one time series.
Optionally, the described first loyal user and the second loyal user are the loyal use of the setting application mode
Family, described device also include prewarning unit, are used for：
Analyze the quantity of the loyal user of the quantity of the loyal user and the setting application mode of historical record
Variation tendency；
The result of the variation tendency obtained according to analysis, analyze Service Quality of the setting application mode to user
Amount；
When the service quality is less than the first given threshold, warning information is issued.
Optionally, the described first loyal user and the second loyal user are the loyal use of the setting application mode
Family, described device also include recommendation unit, are used for：
Recommend the loyal user application mode related to the setting application mode.
Optionally, the described first loyal user and the second loyal user are the loyal use of the setting application mode
Family, described device also include prewarning unit, are used for：
According to the result of determination of the loyal user, the possibility that each user is lost in for the setting application mode is estimated
Property size；
When the possibility size that user is lost in for the setting application mode is higher than the second given threshold, early warning is issued
Information.
Optionally, the setting application mode includes：Live viewing mode, the viewing mode of carousel, the viewing of program request
Mode, or the occupation mode to specific products.
Brief description of the drawings
Fig. 1 is the method flow diagram that consumer loyalty degree is judged in the embodiment of the present application；
Fig. 2 is the method flow diagram that live telecast consumer loyalty degree is judged in the embodiment of the present application；
Fig. 3 a and Fig. 3 b are auto-correlation function and partial autocorrelation function figure in the embodiment of the present application；
Fig. 4 a, Fig. 4 b and the change trend curve that Fig. 4 c are time series user characteristics value in the embodiment of the present application；
Fig. 5 is the structure drawing of device that consumer loyalty degree is judged in the embodiment of the present application.
Embodiment
In order that the purpose, technical scheme and advantage of the application are clearer, the application is made below in conjunction with accompanying drawing into
One step it is described in detail, it is clear that described embodiment is only some embodiments of the present application, rather than whole implementation
Example.Based on the embodiment in the application, what those of ordinary skill in the art were obtained under the premise of creative work is not made
All other embodiment, belong to the scope of the application protection.
The problem of loyal user for that can not judge application-specific mode in the prior art, the embodiment of the present application provides
A kind of method and device for judging loyal user, the steady user that not only will determine that out are defined as loyal user, will also utilized
Heuristics further filters out loyal user from non-stationary user, improves the recall rate of algorithm, is capable of detecting when more
Loyal user so that judge the method for loyal user more precisely, judged result is closer to truth, to set viewing mode
Data analysis definite foundation is provided, it is preferably improved and is developed.For example, the viewing mode for live telecast
For, according to the statistics to consumer loyalty degree, either slowly reduced when die-offing occurs in the quantity of loyal user in a period of time,
Early warning can be sent, the service to live telecast user carries out certain feedback.
The method that the embodiment of the present application provides can apply to the application mode of any setting, for example, live viewing side
Formula, the viewing mode of carousel, the viewing mode of program request, or the occupation mode to specific products.
The method provided below in conjunction with the accompanying drawings the embodiment of the present application elaborates.
As shown in figure 1, the embodiment of the present application offer judges that the method flow of loyal user is as follows.
Step 101：According to user to the historical data caused by setting application mode using behavior, very first time sequence is extracted
Row are gathered, and several time serieses are included in very first time arrangement set, wherein, included in a time series according to a use
Several generated caused by family using the historical data of behavior apply characteristic value.
According to a user to the historical data caused by setting application mode using behavior, the very first time sequence is extracted
A time series in row set, including：
According to a user to the historical data caused by the setting application mode using behavior, extracted for the first sub- time
Sequence and the second Time Sub-series, comprising being generated according to caused by a user using behavior in first Time Sub-series
Several the first subcharacter values, comprising being generated according to caused by a user using behavior in second Time Sub-series
Several the second subcharacter values；
According to the correlation of first Time Sub-series and second Time Sub-series, the first Time Sub-series are determined
The first fitting coefficient and second Time Sub-series the second fitting coefficient；
Based on first fitting coefficient and second fitting coefficient, by first Time Sub-series and described second
Time Sub-series are fitted to one time series.
Step 102：Determine the second time series set in very first time arrangement set with stationarity feature and with
3rd time series set of non-stationary feature；
User corresponding to the second time series set is defined as to the first loyal user of the setting application mode；
Specifically, unit root ADF is carried out to the time series in very first time arrangement set using the model equation of setting
Examine, judge that there is stationarity feature in very first time arrangement set according to whether ADF verification characteristics statistic refuses null hypothesis
The 4th time series set；And judge very first time arrangement set using auto-correlation coefficient figure and PARCOR coefficients figure
In have stationarity feature the 5th time series set；
The common factor in the 4th time series set and the 5th time series set is chosen as the second time series set.
Step 103：For each time series in the 3rd time series set, the growth trend using characteristic value is obtained
Result, and according to the result got, the setting application is selected from user corresponding to the 3rd time series set
Second loyal user of mode；
Wherein, the first loyal user and the second loyal user are the loyal user for setting viewing mode.
Specifically, it is in increase to calculate two neighboring application characteristic value in each time series that the 3rd time series set includes
The number summation N of long trend；
Any time sequence included for the 3rd time series set, two neighboring application characteristic value is calculated in growth
The frequency n of trend, and calculate n and N ratio n/N；N, the equal natural numbers of n, n≤N, 0 ＜ n/N ＜ 1.
When ratio meets following condition for the moment, the user of time series corresponding to ratio is defined as into the second loyalty uses
Family：
Ratio is in the first setting range；Or
Ratio is in the second setting range, and the application that the latter half in time series corresponding to the ratio includes is special
Value indicative shows a rising trend；
Wherein, the value in the first setting range is all higher than the value in the second setting range.
It is determined that after the first loyal user and the second loyal user, it can also analyze the loyal user's
The variation tendency of the quantity of the loyal user of quantity and the setting application mode of historical record；According to obtaining analysis
The result of variation tendency, analyze service quality of the setting application mode to user；Set in the service quality less than first
When determining threshold value, warning information is issued.
It is determined that after the first loyal user and the second loyal user, also the loyal user is being recommended and institute
State the related application mode of setting application mode.
According to the result of determination of the loyal user, the possibility that each user is lost in for the setting application mode is estimated
Property size；When the possibility size that user is lost in for the setting application mode is higher than the second given threshold, early warning is issued
Information.
The above method is described in further detail with reference to specific application scenarios.Assuming that the application side of above-mentioned setting
Formula is the viewing mode of live telecast.The definition of loyal user can have two kinds：First, with stable live telecast viewing row
For user；Second, in live telecast and program request product, the user of the viewing mode of live telecast is more likely to.
Judge that the method flow of live telecast loyalty user is as shown in Figure 2.
Step 201：Historical data of the user to viewing behavior caused by live telecast viewing mode is obtained, and to history number
According to being pre-processed.
Step 202：Pretreated data are carried out with feature extraction, generation includes the time series of user characteristics；
Step 203：The time series extracted is handled, obtains new time series.
One user, which browses a live television programs, can produce a viewing record, and historical data includes several use
Family is browsing a plurality of record caused by live television programs.To ensure viewing record to judging that live telecast loyalty user has
Effect property, data of the duration beyond setting range will be browsed and be set to invalid data.For example, choose on January 10th, 2016 to April 10
The viewing record of the live telecast of 5925 users of totally 98 days.According to viewing record, choose viewing duration [60s,
14400s] corresponding to viewing be recorded as effectively viewing record, outside this scope viewing record remove；To what is screened
Effectively viewing record is counted, and obtains time series X corresponding to the total duration that user watches weeklyi(t)={ x (t1),...,x
(tj) | j≤14,1≤i≤n }, and time series F corresponding to the total degree watched weekly of useri(t)={ f (t1),...,f
(tj)|f≤14,1≤i≤n}。x(t1) represent total duration of the user in viewing live telecast in the 1st week, f (t1) represent user the
The total degree of viewing live telecast in 1 week.Xi(t) time series of i-th of user in the viewing duration formation of 14 weeks is represented；Fi(t)
Represent time series of i-th of user in the viewing number formation of 14 weeks.
To the user characteristics time series F of existing bidimensionaliAnd X (t)i(t) it is normalized, avoids data dimension
Influence, all features are put under a referential and are compared.
By calculating the correlation between the viewing duration of user and viewing number, it is 0.6570 to draw coefficient correlation, is shown
Correlation is write, so can not be more serious by the loyalty of the stronger signature analysis user of two kinds of correlations, information overlap.According to this system
Meter result characterizes the loyalty of user by two feature fittings are a kind of feature.Can be according to existing Heuristics and business
Demand obtains new time series to extract feature：
Such as：P=0.3*f+0.7*x.
Step 204：Steady user is judged according to the auto-correlation of time series and partial autocorrelation figure, and, according to unit
Root is examined and identifies steady user.
Step 205：The steady user that the two methods of step 204 are judged takes common factor, and the user in common factor is as loyal
Sincere user.
It is existing judge time series stationarity method in, example one, according to auto-correlation coefficient and PARCOR coefficients come
Judge the stationarity of time series.The autocorrelogram and partial autocorrelation figure of stationary sequence are not that hangover is exactly truncation.Truncation is exactly
Coefficient is 0 after certain rank, and hangover is exactly the trend of a decay, but coefficient value is not all 0.With the increase of exponent number,
Sample autocorrelation function declines and tends to 0, and stationary sequence is faster than what non-stationary series declined.This method can pass through statistics
Method judge the stationarity of time series, but need it is artificially defined occur trailing or the exponent number of truncation is how many when, really
Sequence of fixing time has stationarity.Relatively stable time series is either taken, this method has certain artificial subjectivity.
If the situation of slow-decay or periodic damping occurs in sequence auto-correlation function after zero averaging, illustrate that the sequence can
Can there are certain trend or cyclic swing characteristic, that is, show non-stationary.
Example two, the stationarity for judging according to unit root test time series.Unit root test passes through three following moulds
Type is judged：
Model one：Without constant mean, neutral p ranks autoregressive process：xt=φ1xt-1+...+φpxt-p+εt
Model two：There are constant mean, neutral p ranks autoregressive process：xt=μ+φ1xt-1+...+φpxt-p+εt
Model three：There are constant mean, the p rank autoregressive process of linear trend：xt=μ+β t+ φ1xt-1+...+φpxt-p
+εt
Wherein, t is time variable, and β t are trend term, represents certain trend that time series changes over time, μ is normal
It is several, εtFor residual error item, φ is regression coefficient, and x is time series, and the hypothesis of inspection is both for H1:δ ＜ 0, examine H0:δ=
0, i.e., in the presence of a unit root.When time series is stationary sequence, refusal null hypothesis is examined.If that is, former time series
In the absence of unit root, inspection receives null hypothesis, i.e., the time series is non-stationary series.
Time series and the power of time Relationship can only be judged using auto-correlation function and partial autocorrelation function；Unit
The root method of inspection can have certain mathematical meaning according to the stationarity of hypothesis testing time series；Unit root test with from
The method that related partial autocorrelation function combines can improve the accuracy of stationarity judgement.
Specifically, unit root test is carried out to new time series using the model equation of selection, because live telecast is used
The above-mentioned new time series that the viewing behavior at family is formed does not have linear relationship, and preference pattern two is tested here, foundation
Characteristic statistic differentiates whether it shows smooth performance, if the result examined examines H0 for refusal null hypothesis:δ=0, it is believed that on
State new time series and unit root is not present, corresponding user is steady user.
Utilize auto-correlation function and partial autocorrelation function figure, the statistical property of the above-mentioned new time series of analysis, if above-mentioned
New time series does not show certain correlation as shown in Figure 3 a and Figure 3 b shows, when step value is 3, auto-correlation system
Number shows certain hangover, and PARCOR coefficients show truncation, the relation of the user and time be not it is particularly evident,
Then show that above-mentioned new time series has stationarity.
When judgement of the user Jing Guo both the above method, stationarity is shown as, then the user sees for live telecast
See the loyal user of mode.
But unit root test is applied to live telecast consumer loyalty degree with the method that auto-correlation partial autocorrelation function is combined
Differentiation on, that loyal user loses be present, i.e., now also to have part be to belong to loyal to time series table for the user of non-stationary
Sincere user.For this problem, the embodiment of the present application determines whether non-based on experience on the basis of above-mentioned integrated approach
Loyal user in steady user.
Step 206：Used for non-stationary user experience indicate that judging.
The loyal user that steady user watches mode for live telecast is verified as in step 205, because the user sees weekly
The duration seen is not in that big ups and downs are more steady, illustrates that the user has certain viewing custom；Due to above two method
The loyalty of steady user can only be verified, so further being sentenced using following statistical knowledge to non-stationary user
It is disconnected.
Some users can be analyzed according to historical data and watch indication information, for example, the history obtained according to step 201
Data can obtain following indication information：
Viewing is calculated according to formula (1) and records ratio of user of all numbers more than 7 weeks in all users, is, for example,
70.31%；The difference of latter all characteristic values and the last week characteristic value, observes each use in being calculated adjacent two weeks according to formula (2)
The trend of growth or the reduction of the adjacent characteristic value of each two in time series corresponding to family, when calculating corresponding to all users
Between two neighboring characteristic value shows a rising trend in sequence total degree, the time sequence according to corresponding to formula (3) calculates each user
Obtained multiple ratios are taken average by the ratio of the number that two neighboring characteristic value shows a rising trend in row and above-mentioned total degree,
For example, average is 48.74%；The number and the ratio of total degree that two neighboring characteristic value shows a rising trend are more than 80% user
Number accounts for the ratio of all number of users, for example, 10.9%；The number and total degree that two neighboring characteristic value shows a rising trend
Number of users of the ratio more than 50% account for the ratios of all number of users, for example, 33.33%, two neighboring characteristic value is in increasing
The number of users of the number of long trend and the ratio of total degree more than 20% accounts for the ratio of all number of users, is, for example,
91.05%.Difference sum in whole time series between each two neighboring characteristic value is calculated according to formula (4).
The above-mentioned formula (1) being related to is as follows to formula (4).
week_timei'=week_timei+1-week_timei (2)
Formula (1) is arrived in formula (4), and week indicates all numbers of viewing record, and week_time represents characteristic value weekly,
Week_time' represents the difference of the characteristic value of continuous two weeks, and the frequency n that per represents continuous two weeks characteristic values to increase accounts for total time
Number m ratio, sum represent the summation of each adjacent two weeks characteristic value differences.
According to above indication information, it may be determined that judge the concrete mode of loyal user.
1) when the number that two neighboring characteristic value shows a rising trend in time series corresponding to user accounts for the ratio of total degree
When in the range of [0.8,1.0], user is defined as loyal user.
In this case, user watches behavior generally ascendant trend.
2) when the number that two neighboring characteristic value shows a rising trend in time series corresponding to user accounts for the ratio of total degree
[0.5,0.8) in the range of when, if the viewing characteristic value that the latter half in time series includes shows a rising trend, by user
It is defined as loyal user, a kind of situation is as shown in fig. 4 a；If the viewing characteristic value that the latter half in time series includes is in drop
Low tendency, then user is defined as non-loyal user, as shown in Figure 4 b, Fig. 4 b illustrate user gradually to live telecast to a kind of situation
Viewing form loses interest.Viewing characteristic value is that one kind applies characteristic value.
3) or even when the number that two neighboring characteristic value shows a rising trend in time series corresponding to user accounts for total degree
Ratio [0.2,0.5) when, it is if the viewing characteristic value that the latter half in time series includes shows a rising trend, user is true
It is set to loyal user.Here the latter half requirement defined close to end point, a kind of possible situation as illustrated in fig. 4 c, such as
The final stage time of user has continuous three weeks user characteristics values to be in the trend risen, illustrates that user is experiencing several weeks
Afterwards, progressively live telecast viewing mode is generated and greatly watches interest, then user is confirmed as loyal user.
4) when the number that two neighboring characteristic value shows a rising trend in time series corresponding to user accounts for the ratio of total degree
[0.0,0.2) in the range of when, show user watch live telecast interest have a declining tendency, the certain customers are very likely
It is lost in, user is defined as non-loyal user.
In above-described embodiment, the proportion in each mode is the empirical value determined according to the indication information of acquisition, can
To determine different proportions according to different indication informations, renewal can be optimized at any time, and can be true according to indication information
The fixed different scheme for judging loyal user.
So far, method introduction of the embodiment of the present application to judging loyal user finishes, with it, according to analysis user
Loyalty, further analyze the possibility of customer loss, the possibility of loyal customer loss is smaller；By analyzing a period of time
The situation of change of loyal number of users, to be fed back to live telecast service quality, and occurring die-offing as loyal user or
Person is when slowly reduction, and back services are made with early warning, further searches for the reason for number of users reduces in time；Pass through analysis
Loyal user, accurate play targetedly can be made to the user with certain viewing behavior and recommended.
Based on same inventive concept, as shown in fig.5, the embodiment of the present application additionally provides a kind of dress for judging loyal user
Put 500, including：Extraction unit 501, determining unit 502, acquiring unit 503 and selecting unit 504.
Extraction unit 501, for according to user to applying the historical data of behavior, extraction the caused by setting application mode
One time arrangement set, several time serieses are included in the very first time arrangement set, wherein, a time series
In comprising according to caused by a user using behavior historical data generated several apply characteristic value；
Determining unit 502, for determine extraction unit 501 extract very first time arrangement set in there is stationarity feature
The second time series set and with non-stationary feature the 3rd time series set；By the second time series set
Corresponding user is defined as the first loyal user of the setting application mode；
Acquiring unit 503, for each time sequence in the 3rd time series set that is determined for determining unit 502
Row, obtain the result of the growth trend using characteristic value；
Selecting unit 504, for the result got according to acquiring unit 503, from the 3rd time series set pair
The second loyal user of the setting application mode is selected in the user answered；
Wherein, the first loyal user and the second loyal user are the loyal user for setting viewing mode.
Optionally, acquiring unit 503 is specifically used for：
The each time series included for the 3rd time series set, calculating two neighboring application characteristic value is in
The frequency n of growth trend；And
It is in increase to calculate two neighboring application characteristic value in each time series that the 3rd time series set includes
The number summation N of trend；
Result using n and N ratio n/N as the growth trend using characteristic value；N and n is natural number, and n≤
N。
Optionally, selecting unit 504 is specifically used for：
When ratio meets following condition for the moment, the user of time series corresponding to ratio is defined as into the second loyalty uses
Family：
Ratio is in the first setting range；Or
Ratio is in the second setting range, and the n two neighboring application characteristic value collection to show a rising trend corresponding to ratio
In at the rear portion of time series；
Wherein, the value in the first setting range is all higher than the value in the second setting range.
Optionally, determining unit 502 is specifically used for：
Unit root ADF inspections, root are carried out to the time series in very first time arrangement set using the model equation of setting
When according to ADF verification characteristics statistic whether refusing that there is the 4th of stationarity feature in null hypothesis judgement very first time arrangement set
Between arrangement set；And
Judge that there is stationarity feature in very first time arrangement set using auto-correlation coefficient figure and PARCOR coefficients figure
The 5th time series set；
The common factor in the 4th time series set and the 5th time series set is chosen as the second time series set.
Optionally, extraction unit 501 is used for：According to a user to applying behavior caused by the setting application mode
Historical data, the first Time Sub-series and the second Time Sub-series are extracted, are included in first Time Sub-series according to one
Several the first subcharacter values generated using behavior caused by user, include according to one in second Time Sub-series
Several the second subcharacter values generated caused by user using behavior；
According to the correlation of first Time Sub-series and second Time Sub-series, the first Time Sub-series are determined
The first fitting coefficient and second Time Sub-series the second fitting coefficient；
Based on first fitting coefficient and second fitting coefficient, by first Time Sub-series and described second
Time Sub-series are fitted to one time series.
Optionally, device 500 also includes prewarning unit 505, is used for：
Analyze the quantity of the loyal user of the quantity of the loyal user and the setting application mode of historical record
Variation tendency；
The result of the variation tendency obtained according to analysis, analyze Service Quality of the setting application mode to user
Amount；
When the service quality is less than the first given threshold, warning information is issued.
Optionally, device 500 also includes recommendation unit 506, is used for：
Recommend the loyal user application mode related to the setting application mode.
Optionally, prewarning unit 505 is additionally operable to：
According to the result of determination of the loyal user, the possibility that each user is lost in for the setting application mode is estimated
Property size；
When the possibility size that user is lost in for the setting application mode is higher than the second given threshold, early warning is issued
Information.
Optionally, setting application mode includes：Live viewing mode, the viewing mode of carousel, the viewing mode of program request,
Or the occupation mode to specific products.
It should be understood by those skilled in the art that, embodiments herein can be provided as method, system or computer program
Product.Therefore, the application can use the reality in terms of complete hardware embodiment, complete software embodiment or combination software and hardware
Apply the form of example.Moreover, the application can use the computer for wherein including computer usable program code in one or more
The computer program production that usable storage medium is implemented on (including but is not limited to magnetic disk storage, CD-ROM, optical memory etc.)
The form of product.
The application is with reference to the flow according to the method for the embodiment of the present application, equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that can be by every first-class in computer program instructions implementation process figure and/or block diagram
Journey and/or the flow in square frame and flow chart and/or block diagram and/or the combination of square frame.These computer programs can be provided
The processors of all-purpose computer, special-purpose computer, Embedded Processor or other programmable data processing devices is instructed to produce
A raw machine so that produced by the instruction of computer or the computing device of other programmable data processing devices for real
The device for the function of being specified in present one flow of flow chart or one square frame of multiple flows and/or block diagram or multiple square frames.
These computer program instructions, which may be alternatively stored in, can guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works so that the instruction being stored in the computer-readable memory, which produces, to be included referring to
Make the manufacture of device, the command device realize in one flow of flow chart or multiple flows and/or one square frame of block diagram or
The function of being specified in multiple square frames.
These computer program instructions can be also loaded into computer or other programmable data processing devices so that counted
Series of operation steps is performed on calculation machine or other programmable devices to produce computer implemented processing, so as in computer or
The instruction performed on other programmable devices is provided for realizing in one flow of flow chart or multiple flows and/or block diagram one
The step of function of being specified in individual square frame or multiple square frames.
Although having been described for the preferred embodiment of the application, those skilled in the art once know basic creation
Property concept, then can make other change and modification to these embodiments.So appended claims be intended to be construed to include it is excellent
Select embodiment and fall into having altered and changing for the application scope.
Obviously, those skilled in the art can carry out various changes and modification without departing from this Shen to the embodiment of the present application
Please embodiment spirit and scope.So, if these modifications and variations of the embodiment of the present application belong to the application claim
And its within the scope of equivalent technologies, then the application is also intended to comprising including these changes and modification.
Claims (16)
- A kind of 1. method for judging loyal user, it is characterised in that including：It is described to the historical data caused by setting application mode using behavior, extraction very first time arrangement set according to user Several time serieses are included in very first time arrangement set, wherein, included in a time series according to a user The caused historical data using behavior generated several apply characteristic value；Determine the second time series set in the very first time arrangement set with stationarity feature and with non-stationary 3rd time series set of feature；User corresponding to the second time series set is defined as to the first loyal user of the setting application mode；For each time series in the 3rd time series set, the result of the growth trend of characteristic value is applied in acquisition, And according to the result got, the setting application mode is selected from user corresponding to the 3rd time series set The second loyal user.
- 2. the method as described in claim 1, it is characterised in that the result of the growth trend using characteristic value is obtained, including：The each time series included for the 3rd time series set, two neighboring application characteristic value is calculated in growth The frequency n of trend；AndTwo neighboring application characteristic value in each time series that the 3rd time series set includes is calculated to show a rising trend Number summation N；Result using n and N ratio n/N as the growth trend using characteristic value；N and n is natural number, n≤N.
- 3. method as claimed in claim 2, it is characterised in that according to the result got, select second loyalty User, including：When the ratio meet following condition for the moment, the user of time series corresponding to the ratio is defined as described second Loyal user：The ratio is in the first setting range；OrThe ratio is in the second setting range, and the application that the latter half in time series corresponding to the ratio includes is special Value indicative shows a rising trend；Wherein, the value in first setting range is all higher than the value in second setting range.
- 4. the method as described in claim any one of 1-3, it is characterised in that determine have in the very first time arrangement set Second time series set of stationarity feature, including：Unit root ADF inspections, root are carried out to the time series in the very first time arrangement set using the model equation of setting Whether refuse that null hypothesis judges to have in the very first time arrangement set stationarity feature according to ADF verification characteristics statistic the Four time series set；AndJudge that there is stationarity feature in the very first time arrangement set using auto-correlation coefficient figure and PARCOR coefficients figure The 5th time series set；The common factor in the 4th time series set and the 5th time series set is chosen as the second time sequence Row set.
- 5. the method as described in claim any one of 1-4, it is characterised in that produced according to a user to setting application mode The historical data using behavior, extract a time series in the very first time arrangement set, including：According to a user to the historical data caused by the setting application mode using behavior, the first Time Sub-series are extracted With the second Time Sub-series, apply what behavior was generated according to caused by a user if being included in first Time Sub-series Dry the first subcharacter value, if including what is generated according to caused by a user using behavior in second Time Sub-series Dry the second subcharacter value；According to the correlation of first Time Sub-series and second Time Sub-series, the of the first Time Sub-series are determined Second fitting coefficient of one fitting coefficient and second Time Sub-series；Based on first fitting coefficient and second fitting coefficient, by first Time Sub-series and second period of the day from 11 p.m. to 1 a.m Between sequence fit be one time series.
- 6. the method as described in claim any one of 1-5, it is characterised in that the first loyal user and second loyalty User is the loyal user of the setting application mode, then it is determined that the first loyal user and the second loyal user Afterwards, in addition to：Analyze the change of the quantity of the loyal user of the quantity of the loyal user and the setting application mode of historical record Trend；The result of the variation tendency obtained according to analysis, analyze service quality of the setting application mode to user；When the service quality is less than the first given threshold, warning information is issued.
- 7. the method as described in claim any one of 1-6, it is characterised in that the first loyal user and second loyalty User is the loyal user of the setting application mode, then it is determined that the first loyal user and the second loyal user Afterwards, in addition to：Recommend the loyal user application mode related to the setting application mode.
- 8. the method as described in claim any one of 1-7, it is characterised in that the first loyal user and second loyalty User is the loyal user of the setting application mode, and methods described also includes：According to the result of determination of the loyal user, estimate that the possibility that each user is lost in for the setting application mode is big It is small；When the possibility size that user is lost in for the setting application mode is higher than the second given threshold, issue early warning letter Breath.
- 9. the method as described in claim any one of 1-8, it is characterised in that the setting application mode includes：Live sight See mode, the viewing mode of carousel, the viewing mode of program request, or the occupation mode to specific products.
- A kind of 10. device for judging loyal user, it is characterised in that including：Extraction unit, for, to the historical data caused by setting application mode using behavior, extracting the very first time according to user Arrangement set, several time serieses are included in the very first time arrangement set, wherein, included in a time series Several generated according to caused by a user using the historical data of behavior apply characteristic value；Determining unit, for determine in the very first time arrangement set with stationarity feature the second time series set and The 3rd time series set with non-stationary feature；User corresponding to the second time series set is defined as described Set the first loyal user of application mode；Acquiring unit, for for each time series in the 3rd time series set, obtaining the increasing using characteristic value The result of long trend；Selecting unit, for according to the result got, being selected from user corresponding to the 3rd time series set Second loyal user of the setting application mode.
- 11. device as claimed in claim 10, it is characterised in that the acquiring unit is specifically used for：The each time series included for the 3rd time series set, two neighboring application characteristic value is calculated in growth The frequency n of trend；AndTwo neighboring application characteristic value in each time series that the 3rd time series set includes is calculated to show a rising trend Number summation N；Result using n and N ratio n/N as the growth trend using characteristic value；N and n is natural number, n≤N.
- 12. device as claimed in claim 11, it is characterised in that the selecting unit is specifically used for：When the ratio meet following condition for the moment, the user of time series corresponding to the ratio is defined as described second Loyal user：The ratio is in the first setting range；OrThe ratio is in the second setting range, and the application that the latter half in time series corresponding to the ratio includes is special Value indicative shows a rising trend；Wherein, the value in first setting range is all higher than the value in second setting range.
- 13. the device as described in claim any one of 10-12, it is characterised in that the extraction unit is used for：According to a user to the historical data caused by the setting application mode using behavior, the first Time Sub-series are extracted With the second Time Sub-series, apply what behavior was generated according to caused by a user if being included in first Time Sub-series Dry the first subcharacter value, if including what is generated according to caused by a user using behavior in second Time Sub-series Dry the second subcharacter value；According to the correlation of first Time Sub-series and second Time Sub-series, the of the first Time Sub-series are determined Second fitting coefficient of one fitting coefficient and second Time Sub-series；Based on first fitting coefficient and second fitting coefficient, by first Time Sub-series and second period of the day from 11 p.m. to 1 a.m Between sequence fit be one time series.
- 14. the device as described in claim any one of 10-13, it is characterised in that the first loyal user and described second Loyal user is the loyal user of the setting application mode, and described device also includes prewarning unit, is used for：Analyze the change of the quantity of the loyal user of the quantity of the loyal user and the setting application mode of historical record Trend；The result of the variation tendency obtained according to analysis, analyze service quality of the setting application mode to user；When the service quality is less than the first given threshold, warning information is issued.
- 15. the device as described in claim any one of 10-14, it is characterised in that the first loyal user and described second Loyal user is the loyal user of the setting application mode, and described device also includes recommendation unit, is used for：Recommend the loyal user application mode related to the setting application mode.
- 16. the device as described in claim any one of 10-15, it is characterised in that the first loyal user and described second Loyal user is the loyal user of the setting application mode, and described device also includes prewarning unit, is used for：According to the result of determination of the loyal user, estimate that the possibility that each user is lost in for the setting application mode is big It is small；When the possibility size that user is lost in for the setting application mode is higher than the second given threshold, issue early warning letter Breath.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610821820.6A CN107818473A (en) | 2016-09-13 | 2016-09-13 | A kind of method and device for judging loyal user |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610821820.6A CN107818473A (en) | 2016-09-13 | 2016-09-13 | A kind of method and device for judging loyal user |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107818473A true CN107818473A (en) | 2018-03-20 |
Family
ID=61601060
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610821820.6A Pending CN107818473A (en) | 2016-09-13 | 2016-09-13 | A kind of method and device for judging loyal user |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107818473A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583937A (en) * | 2018-10-26 | 2019-04-05 | 平安科技（深圳）有限公司 | A kind of Products Show method and apparatus |
| CN112163614A (en) * | 2020-09-24 | 2021-01-01 | 广州虎牙信息科技有限公司 | Anchor classification method and device, electronic equipment and storage medium |
| CN113259446A (en) * | 2021-05-25 | 2021-08-13 | 平安科技（深圳）有限公司 | APP message sending method, device, equipment and medium based on user loyalty |
| CN114022202A (en) * | 2021-11-03 | 2022-02-08 | 中南大学 | User loss prediction method and system based on deep learning |
-
2016
- 2016-09-13 CN CN201610821820.6A patent/CN107818473A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583937A (en) * | 2018-10-26 | 2019-04-05 | 平安科技（深圳）有限公司 | A kind of Products Show method and apparatus |
| CN112163614A (en) * | 2020-09-24 | 2021-01-01 | 广州虎牙信息科技有限公司 | Anchor classification method and device, electronic equipment and storage medium |
| CN113259446A (en) * | 2021-05-25 | 2021-08-13 | 平安科技（深圳）有限公司 | APP message sending method, device, equipment and medium based on user loyalty |
| CN113259446B (en) * | 2021-05-25 | 2022-10-14 | 平安科技（深圳）有限公司 | APP message sending method, device, equipment and medium based on user loyalty |
| CN114022202A (en) * | 2021-11-03 | 2022-02-08 | 中南大学 | User loss prediction method and system based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107515825B (en) | Fluency testing method and device, storage medium and terminal | |
| CN107818473A (en) | A kind of method and device for judging loyal user | |
| US20190087737A1 (en) | Anomaly detection and automated analysis in systems based on fully masked weighted directed | |
| CN109496334A (en) | For assessing the device and method of voice quality | |
| EP3279806A1 (en) | Data processing method and apparatus | |
| CN110019943A (en) | Video recommendation method, device, electronic equipment and storage medium | |
| CN108475381A (en) | The method and apparatus of performance for media content directly predicted | |
| CN106327230B (en) | Abnormal user detection method and equipment | |
| US20190087248A1 (en) | Anomaly detection and automated analysis using weighted directed graphs | |
| CN106599047A (en) | Information pushing method and device | |
| CN110472742A (en) | A kind of model variable determines method, device and equipment | |
| CN112633708B (en) | Mechanical equipment fault detection method and device, medium and electronic equipment | |
| CN104252540B (en) | A kind of game configuration method for pushing graded based on computer performance | |
| US11449781B2 (en) | Plant abnormality prediction system and method | |
| CN108269118A (en) | A kind of method and apparatus of data analysis | |
| CN116542747A (en) | Product recommendation method and device, storage medium and electronic equipment | |
| KR101551879B1 (en) | A Realization of Injurious moving picture filtering system and method with Data pruning and Likelihood Estimation of Gaussian Mixture Model | |
| CN106326098A (en) | Method and device for testing page perceived performance | |
| US20190196445A1 (en) | Method and system for sensing fine changes in processing/equipment measurement data | |
| CN115311001A (en) | Method and system for predicting user change tendency based on multiple voting algorithm | |
| CN106326097A (en) | Method and device for testing page perceived performance | |
| CN113961803A (en) | Shooting recommendation method and device and computer-readable storage medium | |
| Seufert et al. | One shot crowdtesting: approaching the extremes of crowdsourced subjective quality testing | |
| CN111428982A (en) | High-quality photovoltaic customer evaluation and screening method based on big data | |
| CN112685610A (en) | False registration account identification method and related device |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180320 |
|
| WD01 | Invention patent application deemed withdrawn after publication |