Summary of the invention
To solve the problems, such as to mention in above-mentioned background technique, the present invention provide a kind of method for precisely formulating marketing strategy,Equipment and readable medium, wherein the method for precisely formulating marketing strategy includes the following steps:
Based on existing business datum building user basic label, collect user behavior data predict user behavior andDemand tendency；
According to basic label, behavior and the demand tendency of user, user behavior and product marketing movement are subjected to whole processProcessing is broken up, scattered marketing module is formed；
It is intervened in each marketing module by a variety of marketing modes, formulates accurately marketing program for user.
Further, the basic stamp methods of the building user are as follows:
The label system or portrait system that enterprise or individual are completed based on existing business datum, create basic tag library；And by business datum sample vector, Continuous optimization training is carried out by machine learning classification model.
Further, the whole process break up processing method it is as follows:
User behavior and marketing actions be accurately divided into scattered marketing module based on user experience map, and byThe result of big data analysis is flexibly intervened in each marketing module using a variety of Rich Medias and marketing mode.
Further, the behavior of the user and the prediction technique of demand tendency are as follows；
Build operation platform, collect user's operation process and action trail data, according to LSTM neural network algorithm toFamily behavior is estimated and the prediction of demand tendency.
Further, further includes: automatic to carry out the popularization of Product Experience process after the behavior and demand tendency of predicting user.
Further, the mode that the Product Experience process is promoted includes following one or more:
Automation pushes away text, mail, short message, using push, advertisement change, pop-up, product mix, product difference price.
Another accurate equipment for formulating marketing strategy of the present invention, the equipment includes processor, memory, the placeReason device is connected with the memory by bus, and machine readable code is stored in the memory, and the processor execution is depositedMachine readable code in reservoir is to execute as above any method.
The present invention also provides a kind of computer readable storage medium, computer program generation is stored on the storage mediumCode, when the computer program code is computer-executed to execute as above any method.
The accurate method for formulating marketing strategy provided by the invention, by the basic label of user and predicts user'sBehavior and demand tendency combination whole process break up processing, and the marketing side of quickly response and personalization can be provided for userCase, precise positioning user demand improve marketing accuracy, while the cost of operation and outgoing call personnel are greatly reduced.
When it is implemented, as shown in Figure 1, the method for precisely formulating marketing strategy, includes the following steps:
Based on existing business datum building user basic label, collect user behavior data predict user behavior andDemand tendency；In the step, the method for building up of basic label is by existing business datum mainly by YLZ (easily connection mode evidence) numberAccording to middle offer support, the business datum based on platform in data completes enterprise to personal label system and portrait system；ForExisting business data establishing label library, by enterprise's sample vector；By machine learning classification model carry out depth model intoRow training；
Specifically, such as user A, into this process after the start recording user action trail, with well-established userLabel system beats truth labels to user A, then calculates the weight of each label of user A, in this, as trainedThe input layer of neural network, and with cosine similarity, this user is grouped into some cluster, user's portrait of user A is formed.
The application of label system and portrait system and implementation steps are as follows:
1, classified according to factors such as the attribute of user, demand, product preference and values to user, arranged to modelUseful label forms portrait label system:
Essential characteristic: gender, age bracket, rural migrant worker's mark, difficult personnel, the place where his residence is registered, educational background, personal identification etc.；
Social characteristic: sons and daughters' number, old man's number, registered permanent residence property, family attribute, current Employment, place of working, length of service etc.；
Consumption feature: payment category, expense category etc.；
Social security feature: insurance kind type, the payment time limit, disconnected guarantor's duration, a account remaining sum etc..
2, after the fact that extract user in data warehouse label, text digging is carried out to initial data using TF-IDF algorithmPick, calculates the weighted value of each label of each user.User tag weight=time decaying × user behavior number × TF-IDFCalculate label weight.Wherein:
Time decaying: the certain behaviors of user are constantly weakened by time effects, time of the act away from now remoter, the behavior toMeaning for family is current is smaller.FormulaT is size of the Time To Event apart from current time；
User behavior number: user tag weight daily counts, and user day and the behavior number that the label generates are more,Influence of the label to user is bigger；Example, someone consumes 2 times in certain day, then counts 2；
TF-IDF (word frequency × reverse document-frequency) model calculates label weight: each label is to the importance of user and shouldLabel product of importance in tag population obtains the objective weight value of each label.Word frequency refers to some given wordThe frequency that language occurs in this document.That is total word number size in the number count (w, d) and document d that word w occurs in document d(d) ratio.Reverse document-frequency refers to the IDF of a certain particular words, can be by general act number divided by including the wordFile number, then take logarithm to obtain obtained quotient.That is total number of documents n and occurred number of files docs (w, the D) ratio of word wLogarithm.
3, using neural network model, using user tag as input layer, using different insured schemes as output layer,Hidden layer is 2 layers, constructs full Connection Neural Network, clusters to user.
4, the similarity between two users is calculated using cosine similarity, and then user is clustered.User is generated to drawAs system.
The method of the behavior for predicting user and demand tendency are as follows: collect user's operation process and action trail numberAccording in order to support the perception of entire operation behavior, we have built operation platform, and operation platform has recorded the behavior rail of userMark, activity participate in situation, official documents and correspondence is read and the details of product purchase, have carried out from user's use habit assistant furtherIt data analysis and deduces, using LSTM neural network algorithm, accurately to hold difference of user under the conditions of sometime a certainDemand.Wherein, the specifically used method of LSTM neural network algorithm is as follows:
It using the portrait of user as input data g (z), inputs gate and uses f (Zi), be used herein as sigmoid function workFor activation primitive, the probability for opening door is g (z) * f (Zi), is obtained under the probability that input gate is opened, the value of input.Then,Memory cell will record instantly input value plus a preceding input value and the probability for forgetting door is multiplied by, see whether to forget it is previousSecondary record [c '=g (z) * f (Zi)+cf (Zf)], user behavior is estimated and requirement forecasting is as output vector, is carried out multipairThe LSTM of polymorphic type is predicted
According to basic label, behavior and the demand tendency of user, user behavior and product marketing movement are subjected to whole processProcessing is broken up, scattered marketing module is formed；In the step, feeds back, producing in conjunction with the basic label of user, behavior and demand tendencyProduct experience process on using automation push away text, mail, short message, using push, advertisement change, pop-up, product mix, product differenceThe modes such as price carry out more three-dimensional more accurate promotion to user.The marketing process of product is based on user experience mapMode carried out the processing that whole process is broken up, user behavior and marketing actions are accurately divided into scattered marketing module, andIt by the result of big data analysis, is flexibly intervened in each marketing module using a variety of marketing Rich Medias and marketing mode, essenceIt is quasi- that the marketing program made to measure for user is quickly provided, and Real-time Feedback result application YLZ streaming computing engine will be anti-Feedback data, which are included in again in calculation process, is modified optimization, further increases the accuracy of predictive behavior.
Original complicated service procedure is split by user experience map precision, service procedure is broken up, realizes service flowCheng Jie is revealed, and entire user user access scene is become more intuitive more controllable, can intervene and seek in different user's contact nodesPin improves the flexibility ratio of marketing, and can contact scene in each user and obtain user currently possible intention and demand, to battalionIt is more masterly to sell control, realizes to marketing and the more efficient purpose of user's control.User experience case map is detailed in Fig. 2And Fig. 3；
Wherein, Fig. 2 is that user experience map A is used for easy insured product and pays arrears that process is broken and user accesses scene analysis；
Fig. 3 is that user experience map B is broken for HR data product task module process of contracting for fixed output quotas, and user accesses scene analysis
Preferably, entire analysis system is established in user data operation, the fundamental coordinate system of user, the behavior of user andUser can enter in streaming computing engine directly as parameter in the feedback of different marketing nodes and calculate, point of userAnalysis is that the operation based on user is changed in real time, in the Internet era that demand shift is getting faster, give policymaker andThe faster proficiency information of marketing performers, and during being linked up with user the three-dimensional user demand of more deep building andUser's portrait.
By the above-mentioned huge data accumulation for declaring platform enterprise user, enterprise customer's label body is completed based on user informationSystem's construction, enterprise handler portrait by burying a monitoring, the action trail of record and analysis user, and rely on above-mentioned data excellentChange marketing process, completes user using user experience map and access Scenario Design, Behavior-based control analysis and requirement forecasting are in userIt accesses in scene and completes automation marketing access, form automatic, intelligence, the marketing system in thousand people, thousand face is reducing entire manpowerIn the case where cost, marketing efficiency and client's transformation in planta rate are improved.
To examine the application of this method in practice, is tested as follows: having carried out two-wheeled respectively before and after application deploymentTelephone outbound call marketing test, two-wheeled telephone outbound call marketing test result are as shown in table 1:
Table 1
| Total outgoing call number/example | Success connection/example | Percent of call completed | Intention user/example | Outgoing call intention rate | Connection intention rate |
| New version | 475 | 295 | 62.1% | 49 | 10.3% | 16.6% |
| Legacy version | 1100 | 601 | 54.6% | 16 | 1.5% | 2.7% |
It is shown according to the test testing result of table 1, uses method provided by the invention, compared to not using this present inventionThe legacy version of the method for offer, user's outgoing call connect intention conversion ratio and improve 6 times.It can be seen that using provided by the inventionMethod can significantly improve the precision of marketing.
After marketing method provided by the invention, in the case where runing the not increased basic condition of team's manpower, operational departmentMen Xin assumes responsibility for the popularization of five product lines, the course of work in a season, and service number increases ten times, community userAmount is from less than 2,000, to 30,000, efficiency of operation is significantly improved.
The accurate method for formulating marketing strategy provided in an embodiment of the present invention, by the basic label of user and predictsThe behavior of user and demand tendency combination whole process break up processing, and the battalion of quickly response and personalization can be provided for userPin scheme, precise positioning user demand improve marketing accuracy, while the cost of operation and outgoing call personnel are greatly reduced.
Another equipment for precisely formulating marketing strategy of the embodiment of the present invention, the equipment includes processor, memory,The processor is connected with the memory by bus, and machine readable code, the processor are stored in the memoryThe machine readable code in memory is executed to execute as above any method.
The embodiment of the present invention also provides a kind of computer readable storage medium, and computer journey is stored on the storage mediumSequence code, when the computer program code is computer-executed to execute as above any method.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extentPipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according toSo be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features intoRow equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solutionThe range of scheme.