A kind of brand recommended method, electronic equipment, storage medium and systemTechnical field
The present invention relates to data processing field more particularly to a kind of brand recommended method, electronic equipment, storage medium and it isSystem.
Background technique
Since current shopping at network platform has a large amount of brand articles on sale daily, and user has only seen substantially every timeThe brand of quantity is limited, how brand interested for user appears in the brand of limited quantity, just at the emphasis of research.
Be currently to be using following two mode: 1, the sensitivity based on business people to commodity and user determines product by handBoard sequence；2, classify to user, extract the feature of different user types, then use CTR (ad click rate prediction), predictionThe clicking rate of different brands, to determine sequence.But when using above two mode, with the increase of listener clustering, arrange by handThe workload of sequence increases severely, and determines feature for each user group, it is also desirable to expend great effort, the institute inside another user groupThe brand that has user to see or the same can not be entirely their independently customized interested list of brands of each user.
Summary of the invention
For overcome the deficiencies in the prior art, one of the objects of the present invention is to provide a kind of brand recommended method, energySolve the problems, such as that current brand recommended method can not be entirely their independently customized interested list of brands of each user.
The second object of the present invention is to provide a kind of electronic equipment, can solve current brand recommended method and had no wayThe problem of being all each user independently customized their interested list of brands.
The third object of the present invention is to provide a kind of storage medium, can solve current brand recommended method and had no wayThe problem of being all each user independently customized their interested list of brands.
The fourth object of the present invention is to provide a kind of brand recommender system, can solve current brand recommended method and do not doThe problem of method is entirely each user independently customized their interested list of brands.
An object of the present invention is implemented with the following technical solutions:
A kind of brand recommended method, characterized by comprising:
Order data obtains, and several order datas are obtained from the data storage device on shopping at network platform, described to orderForms data includes Brand information and user's name；
Data cleansing carries out taxonomic revision to several order datas according to the different user's names, by same useThe order data that name in an account book claims merges to obtain user's order data, by user's order data and the user's nameAs training data；
The training data is input in default recommended models by training pattern, using logistic regression algorithm and negative at randomSampling algorithm is trained the training data in the default recommended models and has been trained recommended models；
Information collection obtains any active ues list on shopping at network platform, obtains the product on sale on shopping at network platformBoard data；
Brand is recommended, and any active ues list and the branding data on sale are input to and described trained recommended modelsIt is matched and obtains recommended brands list.
It further, further include that the recommended brands list is recommended into corresponding any active ues in the user list.
Further, the brand is recommended specifically: inputs any active ues list and the branding data on saleIt has been trained in recommended models to described, it is described that recommended models has been trained to match corresponding user according to any active ues listOrder data, and associated brand name nonoculture on sale is matched in the branding data on sale according to user's order data and isRecommended brands list.
It further, further include more new data before the training pattern, in daily timing acquiring data storage deviceUpdated new order data carry out taxonomic revision to the new order data and obtain new user's order data, by the new useFamily order data incorporates in user's order data.
The second object of the present invention is implemented with the following technical solutions:
A kind of electronic equipment, characterized by comprising: processor；
Memory；And program, wherein described program is stored in the memory, and is configured to by processorIt executes, described program includes for executing a kind of brand recommended method of the invention.
The third object of the present invention is implemented with the following technical solutions:
A kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer programIt is executed by processor a kind of brand recommended method of the invention.
The fourth object of the present invention is implemented with the following technical solutions:
A kind of brand recommender system, characterized by comprising:
Order data obtains module, and the order data obtains module and is used to store dress from the data on shopping at network platformMiddle several order datas of acquisition are set, the order data includes Brand information and user's name；
Data cleansing module, the data cleansing module are used for according to the different user's names to several order numbersAccording to taxonomic revision is carried out, merge the order data of same user's name to obtain user's order data, by the useFamily order data and the user's name are as training data；
Training pattern module, the training pattern module are used to for the training data being input in default recommended models,The training data in the default recommended models is trained and is obtained using logistic regression algorithm and random negative sampling algorithmRecommended models are trained；
Information acquisition module, the information acquisition module are used to obtain any active ues list on shopping at network platform, obtainTake the branding data on sale on shopping at network platform；
Brand recommending module, the brand recommending module are used for any active ues list and the branding data on saleIt has trained recommended models to be matched described in being input to and has obtained recommended brands list.
It further, further include sending module, the sending module is used to recommend the recommended brands list describedCorresponding any active ues in user list.
It further, further include updating data module, the update data module is stored for daily timing acquiring dataUpdated new order data in device carry out taxonomic revision to the new order data and obtain new user's order data, by instituteNew user's order data is stated to incorporate in user's order data.
Further, the data cleansing module includes taxonomic revision unit and combining unit, the taxonomic revision unitFor carrying out taxonomic revision to several order datas according to the different user's names, the combining unit is used for will be sameThe order data of user's name merges to obtain user's order data, by user's order data and the user nameReferred to as training data.
Compared with prior art, the beneficial effects of the present invention are a kind of brand recommended method of the invention, by from networkObtain several order datas in data storage device on shopping platform, and according to different user title to several order datas intoThe order data of same user's name is merged to obtain user's order data, user's order data is made by row taxonomic revisionFor training data, training data is input in preset recommended models, and is adopted using using logistic regression algorithm and random bearSample algorithm is trained to the training data in default recommended models and has been trained recommended models, by any active ues list andBranding data on sale, which is input to, has trained recommended models to be matched and has obtained recommended brands list, this recommended brands list and everyOne any active ues corresponds, i.e., each corresponding one group of recommended brands list of user, it can not be entirely every for solving in the pastThe problem of a user independently customized their interested list of brands, while whole carrying out that matching is recommended to increase using training patternThe precision and efficiency recommended, improve the experience sense of user.
The above description is only an overview of the technical scheme of the present invention, in order to better understand the technical means of the present invention,And can be implemented in accordance with the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention and the accompanying drawings.A specific embodiment of the invention is shown in detail by following embodiment and its attached drawing.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes part of this application, this hairBright illustrative embodiments and their description are used to explain the present invention, and are not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is a kind of flow chart of brand recommended method of the invention；
Fig. 2 is a kind of logical architecture figure of brand recommended method of the invention:
Fig. 3 is a kind of module frame chart of brand recommender system of the invention.
Specific embodiment
In the following, being described further in conjunction with attached drawing and specific embodiment to the present invention, it should be noted that notUnder the premise of conflicting, new implementation can be formed between various embodiments described below or between each technical characteristic in any combinationExample.
As shown in Figs. 1-2, a kind of brand recommended method of the invention, including
Order data obtains, and several order datas, order numbers are obtained from the data storage device on shopping at network platformAccording to including Brand information and user's name；Many order informations can be all generated daily on shopping at network platform at present, are not hadThere is a large amount of user buy the commodity of various brands.
Data cleansing carries out taxonomic revision to several order datas according to different user title, by same user's nameOrder data merges to obtain user's order data, using user's order data and user's name as training data；In this realityApplying has a large amount of single order in the order data obtained in example by order data, this step i.e. will be same in order dataThe order of user's name, which is summarized in, is formed together a classification, such as the user's name in existing order data well are as follows:Three, Li Ming etc. will be summarized as one kind according to the order information that user's names all in order data are Zhang San, by all user namesReferred to as the order information of Li Ming is summarized as one kind, that is, completes the classification of system, by order data Brand information withUser's name corresponds.It in the present embodiment further include more new data, timing acquires from data storage device dailyUpdated new order data obtain before being dissolved into newest order data because there is new order data to generate dailyIn the user's order data got, make training data latest data, further includes to new order data before data involvementAbove-mentioned data cleansing is carried out, Brand information in new order data and user's name are subjected to summarizing.
Training data is input in default recommended models by training pattern, using logistic regression algorithm and random negative samplingAlgorithm is trained the training data in default recommended models and has been trained recommended models；Logic is returned in the present embodimentReduction method and random negative sampling algorithm are trained training data using default recommended models.Specific training algorithm is as follows: 1,Training algorithm output: for all brand b ∈ Ball, enabling b includes two vector XbAnd θb, XbFor the value vector of brand b, θbFor productThe auxiliary vector of board b；Specific derivation process are as follows: related for two brands b1, b2, then it is L (b1, b2) that certain value, which is arranged, enables LWhen (b1, b2)=1, then b1 is related to b2, and when L (b1, b2)=0, then b1, b2 are uncorrelated, then the correlation of the correlation of b1, b2Probability value is p (b1 | b2), specific such as formula (1) are as follows:
P (b1 | b2)=f (b1, b2)L(b1,b2)(1-f(b1,b2))1-L(b1,b2) (1)
Wherein, p (b1 | b2) is correlation probabilities value, and L (b1, b2) is definite value, and L (b1, b2) is that b1 and b2 is relevantSigmoid function.For user's purchaser record U={ b1, b2, b3 ... bn }, U ∈ Ball；If brand b belongs to purchaseU is recorded, NEG (b, U) is all brands in set U in addition to b；NEG(b,Ball) it is set BallIn all brands in addition to b.b∈ U, brand v ∈ NEG (b, U), brand w ∈ NEG (b, Ball)；For each brand v, v is related to b, brand v and brand w not phaseIt closes.
Information collection obtains any active ues list on shopping at network platform, obtains the product on sale on shopping at network platformBoard data；Nearly three user lists often logged on shopping at network platform are obtained, are located any active ues list, and obtainTake the branding data on sale on shopping at network platform, i.e. brand on sale on shopping at network platform.
Brand is recommended, and any active ues list and branding data on sale is input to, recommended models has been trained to be matched and obtainedTo recommended brands list.Any active ues list and branding data on sale are input to and have been trained in recommended models, recommendation has been trainedModel matches corresponding user's order data according to any active ues list, and according to user's order data in branding data on saleIn match associated brand name nonoculture on sale be recommended brands list.It usually chooses and arranges according to correlation in the present embodimentThe brand list of name previous hundred, and previous hundred brands list is recommended into corresponding user.In the present embodiment as shown in Figure 2A kind of logical architecture figure of brand recommended method, i.e., first timing acquisition, timing acquisition includes obtaining order data to obtain, active to useName in an account book list obtains, and brand message on sale obtains, and above-mentioned data are stored in the storage of Hdfs data and are stored, by order data, workJump user data, branding data on sale store respectively, are input in recommended models and are trained using order data, trained and pushed awayThe comprehensive any active ues data of model, branding data on sale progress brand recommendation are recommended, match recommended brands for each any active uesAs a result, recommended brands is read as a result, and recommended brands result is sent to user's browsing in backstage.
The embodiment of the present invention provides a kind of electronic equipment, comprising: processor；
Memory；And program, wherein program is stored in memory, and is configured to be executed by processor, journeySequence includes for executing a kind of brand recommended method of the invention.
The embodiment of the present invention provides a kind of computer readable storage medium, is stored thereon with computer program, and feature existsIn: computer program is executed by processor a kind of brand recommended method of the invention.
The embodiment of the present invention also provides a kind of brand recommender system, as shown in figure 3, specifically including: order data obtains mouldBlock, order data obtain module for obtaining several order datas, order from the data storage device on shopping at network platformData include Brand information and user's name；Data cleansing module, data cleansing module are used for according to different user titleTaxonomic revision is carried out to several order datas, the order data of same user's name is merged to obtain user's order data,Using user's order data and user's name as training data；Training pattern module, training pattern module are used for training dataIt is input in default recommended models, using logistic regression algorithm and random negative sampling algorithm to the training number in default recommended modelsAccording to being trained and trained recommended models；Information acquisition module, information acquisition module is for obtaining shopping at network platformOn any active ues list, obtain shopping at network platform on branding data on sale；Brand recommending module, brand recommending module are usedIt has trained recommended models to be matched in any active ues list and branding data on sale to be input to and obtains recommended brands list.It further, further include sending module, sending module is corresponding active in user list for recommending recommended brands listUser.Further include updating data module, updates data module for updated new in daily timing acquiring data storage deviceNew order data are carried out taxonomic revision and obtain new user's order data by order data, and new user's order data is incorporated userIn order data.Data cleansing module includes taxonomic revision unit and combining unit, and taxonomic revision unit is used for according to different useName in an account book claims to carry out several order datas taxonomic revision, and combining unit is for merging the order data of same user's nameUser's order data is obtained, using user's order data and user's name as training data.
A kind of brand recommended method of the invention, it is several by being obtained from the data storage device on shopping at network platformOrder data, and taxonomic revision is carried out to several order datas according to different user title, by the order numbers of same user's nameAccording to merging to obtain user's order data, using user's order data as training data, training data is input to presetIn recommended models, and using using logistic regression algorithm and random negative sampling algorithm to the training data in default recommended models intoRow training simultaneously has been trained recommended models, any active ues list and branding data on sale are input to trained recommended models intoRow matches and obtains recommended brands list, this recommended brands list and each any active ues correspond, i.e., each user coupleOne group of recommended brands list is answered, solving can not be entirely their independently customized interested list of brands of each user in the pastThe problem of, while the whole precision and efficiency for carrying out that matching is recommended to increase recommendation using training pattern, improve user'sExperience sense.
More than, only presently preferred embodiments of the present invention is not intended to limit the present invention in any form；All current rowsThe those of ordinary skill of industry can be shown in by specification attached drawing and above and swimmingly implement the present invention；But all to be familiar with sheet specialThe technical staff of industry without departing from the scope of the present invention, is made a little using disclosed above technology contentsThe equivalent variations of variation, modification and evolution is equivalent embodiment of the invention；Meanwhile all substantial technologicals according to the present inventionThe variation, modification and evolution etc. of any equivalent variations to the above embodiments, still fall within technical solution of the present inventionWithin protection scope.