CN106056241A - Control method and device for predicting user behavior based on large data - Google Patents

Abstract

The invention provides a control method for predicting a user behavior based on large data. The control method is used in classification analysis based on a user behavior large-data set so that a behavior expectation index I is predicted and calculated. The control method comprises the following steps: a, obtaining a user behavior variable y; b, on the basis of a function f<i>(u, y, [phi] ... ...), obtaining characteristic parameters [beta]<1>, [beta]<2> ... [beta]<p> through calculation; c, on the basis of a linear combination formula u=[beta]<1>*1+[beta]<2>*2+...[beta]<p>*p, obtaining an expectation value u through calculation; and d, on the basis of a formula I=g(u), obtaining the behavior expectation index I through calculation. A corresponding control device is further provided. Client indexes aiming at a user risk, a purchase preference, renewal of insurance, insurance adding and other aspects can be designed through the method and device, a specific number of data is preferably selected from thousands of user data, a refined prediction model is constructed, and visual evaluation is performed on a user's future behavior trend on the basis of the prediction model.

Description

A kind of control method based on big data prediction user behavior and device

Technical field

The present invention relates to big data analysis field, the biggest data analysis in the application of insurance industry, more particularly to A kind of control method based on big data prediction user behavior and corresponding device.

Background technology

Along with the arrival of big data age, increasing industry started big data analysis application to its day to client In often management work, such as insurance industry, telecommunications industry, disappearing conduct industry etc. soon, these industries all start based on big data and process Analyze the behavioral pattern of visitor user, it might even be possible to the future trend of prediction user behavior, excavate potential user's resource.In tradition In insurance industry field, data analyst is typically by actuarial technical limit spacing product sales volume and consumer's risk index, and they are more Ground carries out quantitative Analysis and qualitative analysis to data in a dimension, it is impossible to data during the operation of reflection reality on the whole Variation and complexity.The insurance products that insurance company provides a user with at present is the most all overall point according to historical data Analysing and calculate acquisition, these products pass through the behavioral data assessment acquisition one in same index of the integrated survey a large number of users puts down Average, and make an insurance products that can be applicable to most of user based on this, this copes with shifting events by sticking to a fundamental principle Although insurance products can meet the work requirements of the daily extending user of business personnel to a certain extent, but be unfavorable for business personnel for User follows up.

Many times, business personnel is intended to can to continue after user is insured follow-up, by analyze user insured after The user that different rows is characterized by behavioral data carries out sorting out and divides, in order to follow-up provide personalized continuous for dissimilar user Protect service, such as, in conjunction with user's current period insured information summary analysis user compensation risk, add guarantor's numerical value such as probability, by difference The user area of behavior characteristics separates, and relates to corresponding index to build forecast model, and as a example by vehicle insurance product, insurance company is permissible In conjunction with user property (such as age, sex, marital status, occupation, education background etc.) and the behavior characteristics in user's a period of time (such as driving situation, record violating the regulations etc.) assess different user compensation risk after purchase vehicle insurance, or predict this client Have purchased the follow-up purchase intention after vehicle insurance for several times, thus based on user's base of the analysis modeling of behavior under same index On plinth so that business personnel can by data model user behavior had a visual understanding and based on model tendency to user after Continue as there being an accurate prediction.

In present stage, the extraordinary method of neither one can solve the problem that problem mentioned above.In most cases, business Member can only not provide by the comprehensive insurance products analyzing acquisition immobilized substance of all customer data is supplied to user A kind of method based on big data prediction user behavior effectively so that business personnel can be by the user being characterized different rows Carry out classification, Reasons, relate to Index Establishment corresponding data model and move towards, to same type user to obtain following behavior of client The behavior after insurance products of buying is analyzed prediction.

Summary of the invention

In order to overcome, prior art does not provide big data prediction user's future behaviour based on user behavior trend Technical problem, it is an object of the invention to provide a kind of control method based on big data prediction user behavior and corresponding device.

According to an aspect of the present invention, it is provided that a kind of control method based on big data prediction user behavior, it is used for Carry out classification analysis based on the conjunction of user behavior large data sets and expect index I with prediction calculating behavior, comprise the steps:

A. user agenda variable y is obtained；

B. based on function f_i(u, y, φ ...) calculates and obtains characteristic parameter β₁、β₂...β_p, wherein said φ etc. is for removing Other distributed constants outside first moment；Described i represents the i-th user；Described f_i(u, y, φ ...) is the density function of distribution, It is for representing the probability of occurrence of i-th user's agenda；Described p represents the characteristic behavior quantity of user；

C. calculate based on linear combination and obtain users' expectation u:

U=β₁x₁+β₂x₂+......β_px_p, wherein, X={x₁、x2、x₃...x_pRepresent user characteristics Activity Index variable Collection, x_kRepresent a user characteristics Activity Index, described β₁、β₂...β_pIt is characterized parameter；

D. calculate based on formula I=g (u) and obtain described behavior expectation index I.

Preferably, described characteristic behavior quantity p calculates based on equation below and obtains:

$\underset{1}{\overset{p}{\&Sigma;}}{\mathrm{\&lambda;}}_i\&GreaterEqual;\mathrm{\&theta;}$

Wherein, described θ represents first threshold, and it is used for determining described users' expectation u characteristic of correspondence behavior quantity p； Described λ_iRepresent the user behavior index variable x captured based on user characteristics Activity Index₁、x₂、x₃...x_kQuantity of information, its In 0 ＜ p ＜ k.

It is preferably based on equation below and calculates described quantity of information λ_i:

${\mathrm{\&lambda;}}_i=\frac{\underset{1}{\overset{j}{\&Sigma;}}{(x_i-z_j)}^2}{k};$

Wherein said λ_iRepresent described main constituent z_iThe user behavior index variable x captured₁、x₂、x₃...x_kInformation Amount.

According to another aspect of the present invention, also provide for a kind of control device based on big data prediction user behavior, its Index I is expected with prediction calculating behavior for carrying out classification analysis based on the conjunction of user behavior large data sets, including:

First acquisition device, it is used for obtaining user agenda variable y；

First processing means, it is for based on function f_i(u, y, φ ...) calculates and obtains characteristic parameter β₁、β₂...β_p, Wherein said φ etc. are other distributed constants in addition to first moment；Described i represents the i-th user；Described f_i(u, y, φ ...) For the density function of distribution, it is for representing the probability of occurrence of i-th user's agenda；Described p represents the characteristic behavior of user Quantity；

Second processing means, it is for based on linear combination calculating acquisition users' expectation u:

U=β₁x₁+β₂x₂+......β_px_p, wherein, X={x₁、x₂、x₃...x_pRepresent user characteristics Activity Index variable Collection, x_kRepresent a user characteristics Activity Index, described β₁、β₂...β_pIt is characterized parameter；

3rd processing means, it obtains described behavior expectation index I for calculating based on formula I=g (u).

The technology contents provided by the present invention, can efficiently solve cannot be based on big data prediction particular user behavior Technical problem, thus give data analysis side bring good Consumer's Experience, improve work efficiency and data analysis level.

Accompanying drawing explanation

By the detailed description non-limiting example made with reference to the following drawings of reading, other features of the present invention, Purpose and advantage will become more apparent upon:

Fig. 1 illustrates according to the first embodiment of the present invention, a kind of control method based on big data prediction user behavior Flow chart；

Fig. 2 illustrates that the user described based on Poisson distribution function is actual according to a specific embodiment of the present invention Behavior curve；

Fig. 3 illustrates another detailed description of the invention according to the present invention, the prediction mould that Behavior-based control expectation index obtains Type；

Fig. 4 illustrates the another embodiment according to the present invention, the distribution situation of client's predictive value；

Fig. 5 illustrates that the user described based on gamma distribution function is actual according to a specific embodiment of the present invention Behavior curve；

Fig. 6 illustrates according to the second embodiment of the present invention, a kind of control device based on big data prediction user behavior Structure chart；And

Fig. 7 illustrates the structural representation of a typical application scenarios according to the present invention.

Detailed description of the invention

In order to preferably make technical scheme show clearly, below in conjunction with the accompanying drawings the present invention is made into one Step explanation.

It will be appreciated by those skilled in the art that in order to solve prior art cannot by analyze user behavior data predict user The technical problem of future trend, technical program of the present invention lies in providing a kind of controlling party based on big data prediction user behavior The user that method is characterized to be embodied as different rows sets up the solution of forecast model.The technical scheme provided by the present invention, Business personnel can for consumer's risk, buy preference, continue insurance, add the aspect such as guarantor design client's index, from thousands of user Data preferably go out certain amount of data, builds the forecast model of refine and based on this forecast model, user's future behaviour walked Gesture carries out visual assessment.Specifically, in a preferred embodiment of the invention, acquisition next period use is calculated by generalized linear regression method Family behavior expected value u, the user area that then different rows is characterized by design behavior expectation index I=g (u) separates and sets up with u For independent variable, I is the forecast model of dependent variable.

Fig. 1 illustrates according to the first embodiment of the present invention, a kind of control method based on big data prediction user behavior Flow chart, its for based on user behavior large data sets close carry out classification analysis with prediction calculating behavior expectation index I.Specifically Ground, in the present embodiment, is first carried out step S101, obtains user agenda variable y.More specifically, described user is real Behavior variable y in border is user's person community set, and it is for characterizing the build-in attribute of user, as the age, sex, marital status, Occupation, education background etc..Further, described user agenda variable y collects acquisition based on to periodically following the tracks of of user. Preferably, described user agenda variable y also includes the declaration form purchased with user that user produced within the current period insured time period More related row behavior set, such as, if user have purchased a vehicle insurance declaration form, the most corresponding user's agenda the current period Variable y includes user's mileage number of driving within declaration form cycle current period, traffic accident treatment situation, record violating the regulations etc..

Perform subsequently into step S102, based on function f_i(u, y, φ ...) calculates and obtains characteristic parameter β₁、β₂... β_p.Specifically, described φ etc. are other distributed constants in addition to first moment；Described i represents the i-th user；Described f_i(u, y, φ ...) it is the density function being distributed, it is for representing the probability of occurrence of i-th user's agenda；Described p represents user's Characteristic behavior quantity.More specifically, the density function of described distribution includes normal distyribution function, Poisson distribution function or gal Horse distribution function etc..Further, described characteristic parameter β₁、β₂...β_pFor the coefficient of users' expectation u in step S103, its For described users' expectation u is adjusted.In a preference, user's agenda variable that business personnel is observed The probability that y occurs is based on following Poisson distribution function representation:

$P(y=k)=\frac{{\mathrm{\&lambda;}}^k}{k!}{e}^{-\mathrm{\&lambda;}},k=0,1,\mathrm{....}$

Wherein, described k represents the actual occurrence number of described user agenda variable y, in the described λ representation unit time The average originating rate of user agenda variable y, in described Poisson distribution function is used for describing the unit interval, described user is actual The number of times of behavior occurrences, as in figure 2 it is shown, each point on described curve i.e. represents the user behavior change that actual observation arrives The probability P that amount occurs, described Poisson distribution function difference based on λ can obtain a plurality of curve, under observation independence assumption, The probability that all user agenda variable y observed occur isTake described probability f maximum i.e. f_maxTime correspondence u ' parameter beta₁、β₂...β_pAs in described step S103 calculate users' expectation u time coefficient.

Next step S103 is performed, based on linear combination u=β₁x₁+β₂x₂+......β_px_pCalculate and obtain user Expected value u.Specifically, described x₁、x₂、x₃...x_pCollectively constitute a set X={x₁、x₂、x₃...x_p, it is used for representing user Characteristic behavior index variable collection, x_pRepresent pth the characteristic behavior index of user.More specifically, described β₁、β₂...β_pFor institute State the characteristic parameter calculating acquisition in step S102.Further, described users' expectation u is the expectation of next period user behavior Value, it is for characterizing user's user behavior being likely to occur within the new declaration form cycle.In a preference, described user's phase Prestige value u is based on described β_pValue can form a set U, described set U and obtain after including linearly calculating based on different β Users' expectation u, substitutes into the formula f in above-mentioned steps S102 successively by the numerical value in described set U_i(u, y, φ ...) i.e. Can obtain the distribution function curve of correspondence, user's agenda that the summit of the most described distribution function curve i.e. actual observation arrives becomes The design factor β of u during the maximum probability value that amount occurs₁、β₂...β_pIt is the characteristic parameter of described users' expectation u.

Finally perform step S104, calculate based on formula I=g (u) and obtain described behavior expectation index I.Specifically, described Behavior expectation index I represents user's behavior prediction situation within declaration form time period next period.More specifically, described behavior expectation Index I is a connectivity function, and it has the feature that dullness can be micro-.Further, described behavior expectation index I based on described not Distribution density function f with type_iAnd there is the connectivity function of correspondence.Further, index I structure is expected based on described behavior Building the forecast model of correspondence, described forecast model represents based on linear model.Preferably, described forecast model is based on flat square Coordinate system represents, wherein, described users' expectation u is as the x-axis on the corresponding described plane right-angle coordinate of independent variable, described row For expectation index I as the y-axis on the corresponding described plane right-angle coordinate of dependent variable.In a preference, the described behavior phase Hoping index I=log (u), it is applicable to situation when described users' expectation u submits to Poisson distribution or negative binomial distribution, Obtain forecast model as shown in Figure 3 eventually, from forecast model shown in Fig. 3 it can be seen that by technical scheme mould of the present invention The prediction compensation situation claim data actual with user intending obtaining is identical and (more enters one because resolution restriction cannot show Step ground coincide situation), in a preference, damage as a example by danger declaration form compensates amount of money prediction by vehicle insurance business car, business personnel is permissible According to predictive value client being divided into 10 deciles as shown in Figure 4, wherein, (prediction is compensated and actual is compensated all for the 10th decile The highest one group) it is demarcated as excessive risk user, those skilled in the art can compensate according to the prediction that forecast model calculates As risk warning indexes, and formulate suitable business policy according to this index.

Further, described forecast model is additionally based upon three-dimensional system of coordinate and represents, those skilled in the art can also change More embodiments, do not repeat them here.

It will be appreciated by those skilled in the art that technical scheme of the present invention is based preferably on generalized linear regression method and analyzes described The users' expectation u dependence to described behavior expectation index I, wherein, described behavior expectation index I is continuous and abundant light Sliding function, users' expectation u of described different user is separate, and each numeric type users' expectation u and behavior phase The relation of prestige value book I is closely related with the distribution density function type that described users' expectation u is obeyed, and passes through above-mentioned steps S101 calculates the prediction trend graph of user in a certain index obtained of deriving to step S104, and business personnel can be to having difference The user of behavior characteristics sorts out, and provides personalized service for particular type of user, and this is that prior art does not uses Technical scheme, it is possible to optimize the data analysis capabilities of business personnel greatly, improves work efficiency.

Further, in described step S102, described square represents that described user agenda variable y is at different exponent numbers Distance zero point and/or the distance at center, it is preferable that described φ=E ((y-c)^K), wherein, k ＞ 1, c >=0, such as, as k=2 and During c=0, φ=E (y²) representing the second order zero point square of described user agenda variable y, it is actual that it is used for weighing described user The average of behavior variable y or numeral expectation, it will be appreciated by those skilled in the art that described function f_i(u, y, φ ... φ described in) Deng other distribution functions in addition to first moment, can be by above-mentioned φ=E ((y-c)^K) parameter c in formula and parameter k compose Give different numerical value to obtain, thus weighed the different aspect spy of described user agenda variable y by the square of different rank Levy.

In a change case of the present embodiment, described distribution function f_i(u, y, φ ...) divides based on following normal state Cloth function representation:

$F\left(y\right)=\frac{1}{\mathrm{\&sigma;}\sqrt{2\mathrm{\&pi;}}}{e}^{\frac{-{(y-\mathrm{\&mu;})}^2}{2{\mathrm{\&sigma;}}^2}}$

Wherein, described-∞ ＜ y ＜+∞, and-∞ ＜ μ ＜+∞, σ is parameter, and described F function is used for representing described user Agenda variable y obeys the normal distribution that parameter is (μ, σ 2), is denoted as y～N (μ, σ 2).Specifically, described μ and σ is based on institute State users' expectation u and distributed constant φ is determined.

Further, described distribution function f_i(u, y, φ ...) is additionally based upon following gamma distribution function and represents:

$F\left(y\right)=\frac{y^{(a-1)}{\mathrm{\&lambda;}}^a{e}^{-\mathrm{\&lambda;}y}}{\mathrm{\&Gamma;}\left(y\right)},(y>0)$

Wherein, described a represents that form parameter, described b represent scale parameter, is based on described users' expectation u and distribution Parameter phi determines.Specifically, as y ＜ 0, described F (y)=0, as shown in Figure 5.

It will be appreciated by those skilled in the art that compared with the Poisson distribution function employed in described step S102, this change case Described normal distyribution function or gamma distribution function are for describing the probability that described user agenda variable y is continuous numerical value Distribution situation.

In a change case of the present embodiment, described in described step S102, characteristic behavior quantity p is based on equation below Calculate and obtain:

$\underset{1}{\overset{p}{\&Sigma;}}{\mathrm{\&lambda;}}_i\&GreaterEqual;\mathrm{\&theta;}$

Wherein, described θ represents first threshold, and it is used for determining described users' expectation u characteristic of correspondence behavior quantity p； Described λ_iRepresent the user behavior variable index x captured based on user characteristics Activity Index₁、x₂、x₃...x_kQuantity of information, its In 0 ＜ p ＜ k.

Further, described quantity of information λ_iCalculate based on equation below and obtain:

${\mathrm{\&lambda;}}_i=\frac{\underset{1}{\overset{j}{\&Sigma;}}{(x_i-z_j)}^2}{k};$

Wherein said λ_iRepresent described main constituent z_iThe user behavior index variable x captured₁、x₂、x₃...x_kInformation Amount.Preferably, described main constituent z_iBased on the linear combination acquisition to described user behavior index variable.

Preferably, by setting described first threshold θ, original user Activity Index variable is carried out preferably, and then build essence The forecast model of refining.Preferably, p the variable preferably going out to include most quantity of information from k user behavior index variable is made Subsequent calculations is carried out for the element in described user characteristics Activity Index variables set X.In a preference, described first threshold θ represents based on percentage ratio, and its numerical value is preferably 80%, i.e. when the element institute in described user characteristics Activity Index variables set X The quantity of information sum represented accounts for the ratio of described user behavior index variable gross information content when reaching 80%, it is believed that described variables set Number of elements in X has reached requirement, and the quantity of information that in described variables set X, element is comprised can be for representing described user The gross information content of Activity Index variable, the number of elements in the most described variables set X is described step S102 and described step User characteristics behavior quantity p in rapid S103.

In another change case of the present embodiment, described behavior expectation index I=g (u)=u^-1, it is preferably adapted for Described users' expectation u submits to situation during gamma distribution.

Further, described behavior expectation indexIt is preferably adapted for described user and expects index u Submit to situation during binomial distribution.

It will be appreciated by those skilled in the art that technical scheme first with technology such as PCAs from more User behavior index variable x₁、x₂、x₃...x_kThe less user characteristics Activity Index variable containing enough information of middle acquisition Collection X, thus calculate further with other data mining according to the element in these user characteristics Activity Index variables set X simplified Method sets up terse and accurate model, and technical scheme of the present invention has enough accuracies in terms of customer segmentation, at mould Type training, model deployment phase have enough rapidity, ease for use.

Fig. 6 illustrates according to the second embodiment of the present invention, a kind of control device based on big data prediction user behavior Structure chart, its for based on user behavior large data sets close carry out classification analysis with prediction calculating behavior expectation index I, specifically Ground, in the present embodiment, described control device 4 includes the first acquisition device 41, and it is used for obtaining user agenda variable y； First processing means 42, it is for based on function f_i(u, y, φ ...) calculates and obtains characteristic parameter β₁、β₂...β_p；At second Reason device 43, it obtains for expected value u for calculating based on linear combination；3rd processing means 44, its for based on Formula I=g (u) calculates and obtains described behavior expectation index I.Wherein, described φ etc. are other distributed constants in addition to first moment； Described i represents the i-th user；Described f_i(u, y, φ ...) is the density function of distribution, and it is for representing the i-th actual row of user For probability of occurrence；Described p represents the characteristic behavior quantity of user.Preferably, described users' expectation u is based on formula u=β₁x₁ +β₂x₂+......β_px_pCalculate and obtain, wherein, X={x₁、x₂、x₃...x_pRepresent user characteristics Activity Index variables set, x_kTable Show a user characteristics Activity Index, described β₁、β₂...β_pIt is characterized parameter.

Further, described user characteristics behavior quantity p is based on formulaCalculate and obtain, wherein said parameter Concrete meaning and acquisition pattern those skilled in the art are referred to technical scheme described in above-mentioned embodiment illustrated in fig. 1 and obtain, Do not repeat them here.

In a preference, first described control device 4 calls described first acquisition device 41 and obtains based on to user Periodically follow the tracks of and collect the large data sets of the user behavior composition obtained and close, and it is actual therefrom to refine the user needed for this is analyzed Behavior variable y；Described first processing means 42 is based on Poisson distribution functionMeter Calculate the probability of occurrence obtaining above-mentioned user agenda variable y, obtain corresponding probability curve, each parameter in described distribution function Implication those skilled in the art be referred to step S102 described in above-mentioned embodiment illustrated in fig. 1, do not repeat them here；Enter one Step ground, described second processing means 43 is based on linear combination u=β₁x₁+β₂x₂+......β_px_pCalculate and obtain user's expectation Value u, wherein, described x₁、x₂、x₃...x_pCollectively constitute a set X={x₁、x₂、x₃...x_p, it is used for representing user characteristics row For index variable collection, described x_pRepresent pth the characteristic behavior index of user, by described design factor β₁、β₂...β_p's Different values obtain the set U of corresponding users' expectations u composition, thus the numerical value in described set U substitutes into described the successively Poisson distribution function formula in one processing means 42 obtains the distribution function curve of correspondence, and the described hump of final acquisition is i.e. Users' expectation u during user's agenda variable y probability of occurrence maximum that actual observation arrives, and by this users' expectation u pair The design factor β answered₁、β₂...β_pCharacteristic parameter as described users' expectation u；Then, described 3rd processing means 43 base Apply mechanically formula I=log (u) in described users' expectation u and obtain the behavior expectation index I being used for characterizing forecast model.

Further, described user expects that index u obeys Poisson distribution；Described control device 4 is based on generalized linear regression Method analyzes the described users' expectation u dependence to described behavior expectation index I.It will be appreciated by those skilled in the art that this enforcement Technical scheme described in example is by the Collaboration of each module in described control device 4, it is achieved substantial amounts of user's agenda become Amount y refine is distribution function based on probability the process being based ultimately upon data with existing generation prediction curve so that business personnel's energy Enough obtain user's future trend substantially according to user's existing data simple, intuitive, such as settle a claim risk, continuation of insurance purpose etc., This is the technical scheme that prior art does not uses, it is possible to alleviate the workload of business personnel to a great extent, optimizes business personnel Mode of operation so that business personnel can formulate a follow-up service side the most reasonable, more targeted for different user Case.

In a change case of the present embodiment, described first processing means 42 is additionally based upon normal distyribution functionCalculate and obtain described characteristic parameter β 1, β₂...β_p, wherein, described-∞ ＜ y ＜+∞, and- ∞ ＜ μ ＜+∞, σ is parameter, and described F function is used for representing that described user agenda variable y for (μ, σ 2) is just obeying parameter State is distributed, and is denoted as y～N (μ, σ 2), the most such as, described distribution function f_i(u, y, φ ...) is additionally based upon gamma distribution functionRepresenting, wherein, described a represents that form parameter, described b represent scale parameter, It is based on described users' expectation u and distributed constant φ determines, it will be appreciated by those skilled in the art that distribution function described in the present embodiment For drawing the probability of occurrence curve of described user agenda variable y, by users' expectation u correspondence under different numerical value is divided Characteristic of correspondence parameter beta when the drafting of cloth curve obtains described user's agenda occurrences maximum probability₁、β₂...β_p, from And obtain forecast model I=g (u) of correspondence.

In another change case of the present embodiment, it is additionally based upon formula I=g (u)=u^-1Calculate and obtain described behavior expectation Index, it will be appreciated by those skilled in the art that the 3rd processing means 43 described in above-described embodiment obtains row based on formula I=log (u) Technical scheme for expectation index I is preferably adapted for user's agenda occurrences probability that the first processing means 42 obtains Situation based on Poisson distribution arrangement, technical scheme described in this change case is then preferably adapted for described user's agenda variable Probability of occurrence carries out the situation arranged based on gamma distribution, correspondingly, obeys equally in users' expectation u described in this change case It is distributed in gamma.

Further, described 3rd processing means 44 is additionally based upon formulaCalculate and obtain described behavior expectation Index I, it will be appreciated by those skilled in the art that the user's agenda occurrences probability base when described first processing means 42 obtains When binomial distribution, described 3rd processing means 44 is based preferably on formula described in this change case and calculates the behavior phase obtaining correspondence Hope index I and generate corresponding prediction curve, correspondingly, expecting that index u submits to two equally user described in this change case Situation during item distribution.

Fig. 7 illustrates the structural representation of a typical application scenarios according to the present invention, wherein modeling and above-mentioned Fig. 6 Control device described in described embodiment to communicate.Specifically, in the present embodiment, described modeling includes the actual row of user For variable y, it represents user's personal community set over a period to come for refine；Distribution function f_i(u, y, φ ...), its for based on distribution curve obtain user's agenda variable y probability of occurrence the highest time characteristic of correspondence parameter β₁、β₂...β_p；Behavior expectation index I, it obtains user within declaration form time period next period for calculating based on formula I=g (u) Behavior prediction situation, and the forecast model building correspondence intuitively obtains the future behaviour tendency of user.

In one specifically application scenarios, described modeling is by a series of market survey and combines data derivation It is user agenda variable y by ten hundreds of user's primitive behavior data abstraction refines, is then based on Poisson distribution function Depict the probability of occurrence of user's agenda variable of correspondence, obtain corresponding a plurality of probability by change users' expectation and divide Cloth curve, thus by users' expectation corresponding to (hump) when wherein user's agenda variable y probability of occurrence is the highest As users' expectation u required for the technical program, wherein, described users' expectation u is based on formula u=β₁x₁+β₂x₂ +......β_px_pObtaining, described modeling is by substituting into different β₁、β₂...β_pAnd described x₁、x₂、x₃...x_pObtain Different users' expectations, and then obtain above-mentioned a plurality of probability distribution curve, and after described users' expectation u determines, described Modeling calculates based on formula I=log (u) and obtains the behavior expectation index I that user is corresponding, and generates the prediction mould of correspondence Type.Preferably, users' expectation u described in this application scene submits to Poisson distribution or negative binomial distribution situation.Preferably, described Forecast model represents based on linear model.Preferably, described forecast model represents based on plane right-angle coordinate.Further, Described forecast model is additionally based upon three-dimensional system of coordinate and represents, those skilled in the art can change according to actual needs and more enforcement Example, does not repeats them here.

In another application scenarios, described modeling is additionally based upon Binomial Distributing Function and calculates the described characteristic parameter of acquisition β₁、β₂...β_p, correspondingly, described behavior expectation index I is based on formulaCalculate and obtain, people in the art Member understands, in this application scene, described user's agenda occurrences probability and described user expect index u preferably The rule obeying binomial distribution is illustrated on corresponding plane right-angle coordinate.

Further, described modeling is additionally based upon gamma distribution function and calculates the described characteristic parameter β of acquisition₁、β₂...β_p, Correspondingly, described behavior expectation index I is based on formula I=g (u)=u^-1Calculate and obtain, it will be appreciated by those skilled in the art that should With in scene, described user's agenda occurrences probability and described user expect that index u preferably obeys gamma distribution Rule be illustrated on corresponding plane right-angle coordinate.

In another application scenarios, described modeling is additionally based upon normal distyribution function and calculates the described characteristic parameter of acquisition β₁、β₂...β_p, this can obtain forecast model of the present invention equally, it will be appreciated by those skilled in the art that of the present invention different Distribution function is for providing, for different user agenda variable y, the probability curve met the most, and described modeling exists Running is applied mechanically different distribution functions the most successively existing user agenda variable y is fitted, therefrom Screening and the described distribution function the properest for agenda variable y probability of occurrence are as technical scheme of the present invention Preferred distribution function is modeled.

Above the specific embodiment of the present invention is described.It is to be appreciated that the invention is not limited in above-mentioned Particular implementation, those skilled in the art can make various deformation or amendment within the scope of the claims, this not shadow Ring the flesh and blood of the present invention.

Claims (4)

1. a control method based on big data prediction user behavior, it is returned for closing based on user behavior large data sets Alanysis is with prediction calculating behavior expectation index I, it is characterised in that comprise the steps:

A. user agenda variable y is obtained；

B. based on function f_i(u, y, φ ...) calculates and obtains characteristic parameter β₁、β₂...β_p, wherein said φ etc. is except first moment Other outer distributed constants；Described i represents the i-th user；Described f_i(u, y, φ ...) is the density function of distribution, and it is used for Represent the probability of occurrence of i-th user's agenda；Described p represents the characteristic behavior quantity of user；

C. calculate based on linear combination and obtain users' expectation u:

U=β₁x₁+β₂x₂+.....β_px_p, wherein, X={x₁、x₂、x₃...x_pRepresent user characteristics Activity Index variables set, x_kTable Show a user characteristics Activity Index, described β₁、β₂...β_pIt is characterized parameter；

D. calculate based on formula I=g (u) and obtain described behavior expectation index I.

Control method the most according to claim 1, it is characterised in that described characteristic behavior quantity p is based on equation below meter Calculate and obtain:

$\underset{1}{\overset{p}{\&Sigma;}}{\mathrm{\&lambda;}}_i\&GreaterEqual;\mathrm{\&theta;}$

Wherein, described θ represents first threshold, and it is used for determining described users' expectation u characteristic of correspondence behavior quantity p；Described λ_i Represent the user behavior index variable x captured based on user characteristics Activity Index₁、x₂、x₃...x_kQuantity of information, wherein 0 ＜ p ＜ k.

Control method the most according to claim 2, it is characterised in that calculate described quantity of information λ based on equation below_i:

${\mathrm{\&lambda;}}_i=\frac{\underset{1}{\overset{j}{\&Sigma;}}{(x_i-z_j)}^2}{k};$

Wherein said λ_iRepresent described main constituent z_iThe user behavior index variable x captured₁、x₂、x₃...x_kQuantity of information.

4. a control device based on big data prediction user behavior, it is returned for closing based on user behavior large data sets Alanysis is with prediction calculating behavior expectation index I, it is characterised in that including:

First acquisition device, it is used for obtaining user agenda variable y；

First processing means, it is for based on function f_i(u, y, φ ...) calculates and obtains characteristic parameter β₁、β₂...β_p, wherein Described φ etc. are other distributed constants in addition to first moment；Described i represents the i-th user；Described f_i(u, y, φ ...) is for dividing The density function of cloth, it is for representing the probability of occurrence of i-th user's agenda；Described p represents the characteristic behavior quantity of user；

Second processing means, it obtains users' expectation u:u=β for calculating based on linear combination₁x₁+β₂x₂+.....β_px_p, wherein, X={x₁、x₂、x₃...x_pRepresent user characteristics Activity Index variables set, x_kRepresent a user characteristics Activity Index, Described β₁、β₂...β_pIt is characterized parameter；

3rd processing means, it obtains described behavior expectation index I for calculating based on formula I=g (u).