CN109711887A - Generation method and device of mall recommendation list, electronic equipment and computer medium - Google Patents
Generation method and device of mall recommendation list, electronic equipment and computer medium Download PDFInfo
- Publication number
- CN109711887A CN109711887A CN201811621615.0A CN201811621615A CN109711887A CN 109711887 A CN109711887 A CN 109711887A CN 201811621615 A CN201811621615 A CN 201811621615A CN 109711887 A CN109711887 A CN 109711887A
- Authority
- CN
- China
- Prior art keywords
- user
- order
- rate
- history
- user data
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 101
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 230000015654 memory Effects 0.000 claims description 23
- 238000004891 communication Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000006872 improvement Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 235000013305 food Nutrition 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 235000012054 meals Nutrition 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000002073 mitogenetic effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention relates to the field of data processing, and discloses a method and a device for generating a merchant recommendation list, electronic equipment and a computer-readable storage medium. The generation method of the merchant recommendation list comprises the following steps: acquiring user data of a historical ordering user; performing model training by using user data of historical ordering users to obtain an ordering rate prediction model and an ordering price estimation model; acquiring user data of a current ordering user, inputting the user data of the current ordering user into an ordering rate prediction model to obtain predicted ordering rate, and inputting the user data of the current ordering user into an ordering price estimation model to obtain estimated ordering price; and obtaining the recommendation scores of the merchants according to the predicted ordering rate and the predicted ordering price, and generating a merchant recommendation list according to the recommendation scores. The method and the device for generating the merchant recommendation list, the electronic equipment and the computer-readable storage medium provided by the invention can be used for accurately recommending merchants aiming at different users, and the user experience and the merchant profits are improved.
Description
Technical field
The present embodiments relate to data processing field, in particular to a kind of generation method of trade company's recommendation list, device,
Electronic equipment and computer readable storage medium.
Background technique
Modern search engines can provide inquiry recommendation function for user, so that user be helped to obtain its expected search knot
Fruit.Semantic relevant inquiry is recommended in pervious inquiry recommended work primarily directed to the input inquiry of user, however, above-mentioned look into
Inquiry mode science and technology rapid development the information requirement for being unable to satisfy user today, therefore, personalized inquiry mode meet the tendency of and
It is raw.Personalization inquiry is recommended to be intended to be better described personal information requirement, and up to the present, many personalized inquiries are recommended
Method is all based on the click information of the search history record either user of user.Such as by clicking to specific user
Document is excavated, and corresponding recommendation list is generated, but this method depends on the click behavior of user, if user does not have
Click behavior or click behavior are less, then the effect that this method is just recommended without preferable personalized inquiry.Or according to
The inquiry of user's input and relevant historical record return to a recommendation list to user, possible input next to user
Inquiry is predicted.In electric business field, the generation of recommendation list is generally basede on user to place an order rate or the clicking rate of commodity.Such as
Logistic regression (Logistic Regression, LR), deep neural network (Deep Neural Network, DNN), Factor minute
The technologies such as solution machine (Factorization Machine, FM) have obtained extensive use in the rate of placing an order is estimated.
At least there are the following problems in the prior art for inventor's discovery: being based only upon the rate of placing an order or clicking rate generates pushing away for trade company
List is recommended, so that the generation of the recommendation list, according to single, the specifying information that can not accurately reflect trade company and user are to businessman
Fancy grade, it is not accurate so as to cause the recommendation list provided a user, and then affect the experience and trade company's income of user.
Summary of the invention
A kind of generation method for being designed to provide trade company's recommendation list of embodiment of the present invention, device, electronic equipment
And computer readable storage medium, accurate trade company's recommendation can be carried out for different user, promote user experience and trade company receives
Benefit.
In order to solve the above technical problems, embodiments of the present invention provide a kind of generation method of trade company's recommendation list,
Include:
Obtain the user data of single user under history;Model training is carried out using the user data of single user under the history
To obtain place an order rate prediction model and lower single price estimation model;The user data for obtaining current lower single user, will be described current
Place an order described in the user data input of lower single user obtain in rate prediction model prediction place an order rate, by the current lower single user
It obtains estimating lower single price in lower list price estimation model described in user data input;It is placed an order rate and described pre- according to the prediction
Estimate lower single price and obtain the recommended hour of trade company, and trade company's recommendation list is generated according to the recommended hour.
Embodiments of the present invention additionally provide a kind of generating means of recommendation list, comprising: first obtains module, is used for
Obtain the user data of single user under history;First model training module, for the number of users using single user under the history
According to progress model training to obtain the rate prediction model that places an order;Second model training module, for utilizing single user under the history
User data carry out model training to obtain lower single price estimation model;Second obtains module, for being applied alone under obtaining currently
The user data at family;First processing module, for predicting the rate that places an order described in the user data input of the current lower single user
Obtained in model prediction place an order rate, will be in lower single price estimation model described in the user data input of the current lower single user
To estimating lower single price;Second processing module rate and described estimate lower single price and obtains trade company for being placed an order according to the prediction
Recommended hour, and according to the recommended hour generate trade company's recommendation list.
Embodiments of the present invention additionally provide a kind of electronic equipment, including at least one processor;And with it is described extremely
The memory of few processor communication connection;Wherein, the memory, which is stored with, to be executed by least one described processor
Instruction, described instruction by least one described processor execute to realize: obtain history under single user user data;It utilizes
The user data of single user carries out model training to obtain place an order rate prediction model and lower single price estimation model under the history;
The user data for obtaining current lower single user, by the rate prediction model that places an order described in the user data input of the current lower single user
In obtain prediction place an order rate, will obtain in lower single price estimation model described in the user data input of the current lower single user it is pre-
Estimate lower single price;Place an order rate and the recommended hour for estimating lower single price acquisition trade company are predicted according to described, and are pushed away according to described
It recommends mitogenetic at trade company's recommendation list.
Embodiments of the present invention additionally provide a kind of computer readable storage medium, are stored with computer program, described
The generation method of above-mentioned recommendation list is realized when computer program is executed by processor.
Embodiments of the present invention in terms of existing technologies, by obtaining the user data of single user under history, make
The source for obtaining generation recommendation list in subsequent step is no longer single, and the user data of single user under the history is recycled to carry out mould
Type training is to obtain place an order rate prediction model and lower single price estimation model, so that utilizing different information (i.e. single user under history
User data in different information) carry out obtained different models (rate that places an order prediction model and lower unit price after model training
Lattice prediction model) more targetedly, so as to after the user data for obtaining current lower single user, will it is current under be applied alone
Obtained in the rate that places an order described in the user data input at family prediction model prediction place an order rate, will current time single user user data it is defeated
Enter to obtain estimating lower single price in lower single price estimation model more accurate, and then makes place an order according to the prediction rate and institute
It states and estimates specifying information and user that the recommended hour of the trade company that lower single price obtains can really reflect the trade company to this
The fancy grade of businessman, therefore, the trade company's recommendation list generated according to the recommended hour can carry out accurate for different user
Trade company recommend, user be based on the recommendation list can quickly find oneself favorite trade company and consume, to mention
User experience and trade company's income have been risen, has been avoided and " is based only upon the rate of placing an order or clicking rate generates the recommendation list of trade company, so that this is pushed away
The generation of list is recommended according to single, the specifying information that can not accurately reflect trade company and user to the fancy grade of businessman, thus
Cause the recommendation list provided a user not accurate, and then affect the experience and trade company's income of user " the case where generation.
Specifically, optionally, the user data of single user specifically includes under the history: the use of single user under the history
User data specifically includes: the essential attribute of single user and historical viewings data, trade company's attribute and historical trading letter under the history
The cross attribute of single user and the trade company under breath, the history, wherein the historical transactional information includes user Xia Dan trade company
Record and trade company expose user and do not obtain lower unirecord.A large amount of training number is provided in this manner for the foundation of model
According to, and these data are all authentic and valid and the data that are mutually related, and are provided for the interested dining room of accurate recommended user
Effectively reference.
As a further improvement, it is optional, under the acquisition history after the user data of single user, further includes: will
The Xia Dan trade company recording mark of single user is positive sample under the history, and trade company exposes user and does not obtain lower single recording mark
For negative sample;The negative sample is sampled so that the quantity of the negative sample is equal with the quantity of the positive sample, by institute
The negative sample after stating positive sample and sampling is as the historical transactional information.It is remote due to not obtaining the dining room that user places an order
It needs to be sampled negative sample to keep positive and negative sample proportion roughly the same much larger than the dining room that user places an order, to make to obtain
Historical sample characteristic more for reference value.
As a further improvement, optional, described placed an order according to the prediction and described estimate lower single price and obtains quotient rate
The recommended hour at family, specifically includes: placing an order rate according to the prediction and following formula obtains the rate that really places an order of the trade company: q
=p/ (p+ (1-p)/w);Wherein, q be it is described really place an order rate, p is that the prediction places an order rate, w be negative specimen sample ratio because
Son;According to rate and the recommended hour estimated lower single price and obtain dining room of really placing an order.In this manner, negative sample is utilized
This oversampling ratio factor calibrates the prediction rate of placing an order after model training, and the rate that really places an order for making to obtain is more accurate.
It is described rate and described to estimate lower single price according to described really place an order and obtain meal as a further improvement, optional
The recommended hour in the Room, specifically includes: obtaining the recommended hour: Score=q*price according to the following formula;Wherein, Score is institute
Recommended hour is stated, q is the rate that really places an order, and price estimates lower single price to be described.
In addition, it is optional, model training is carried out using the user data of single user under the history to obtain lower single price
Prediction model specifically includes: model training is carried out using the user data of single user under the history based on Xgboost algorithm,
Obtain lower single price estimation model.Make single price under the estimating obtained more accurate in this manner.
In addition, optional, the user data using single user under the history carries out model training and is placed an order with obtaining
Rate prediction model, specifically includes: carrying out model training using the user data of single user under the history based on FM algorithm, obtains
It is described to predict the rate that places an order.Make the prediction rate that places an order obtained more accurate in this manner.
In addition, optional, the user data for obtaining single user under history is specifically included: to being applied alone under the history
The frequency divisions buckets such as the user data use at family carry out sliding-model control, obtain discretization data;Using the discretization data as institute
State the user data of single user under history.
In addition, optional, it is described to the history under single user user data using etc. frequency divisions bucket carry out discretization
It handles, after acquisition discretization data, further includes: the user data of single user under the history is utilized based on Xgboost algorithm
Model training is carried out, the corresponding leaf node number of user data of single user under the history is obtained, by the leaf node
Number is used as feature growth data;The user data using single user under the history carries out model training and is placed an order with obtaining
Rate prediction model, specifically includes: carrying out model using the user data of single user under the history and the feature growth data
Training obtains the rate prediction model that places an order;Model training is carried out using the user data of single user under the history to place an order to obtain
Price estimation model, specifically includes: carrying out mould using the user data of single user under the history and the feature growth data
Type training obtains lower single price estimation model.
Detailed description of the invention
One or more embodiments are illustrated by the picture in corresponding attached drawing, these exemplary theorys
The bright restriction not constituted to embodiment, the element in attached drawing with same reference numbers label are expressed as similar element, remove
Non- to have special statement, composition does not limit the figure in attached drawing.
Fig. 1 is the flow chart of the generation method for trade company's recommendation list that first embodiment provides according to the present invention;
Fig. 2 is the flow chart of the generation method for trade company's recommendation list that second embodiment provides according to the present invention;
Fig. 3 is the structural schematic diagram of the generating means for trade company's recommendation list that third embodiment provides according to the present invention;
Fig. 4 is the structural schematic diagram for the electronic equipment that the 4th embodiment provides according to the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with attached drawing to the present invention
Each embodiment be explained in detail.However, it will be understood by those skilled in the art that in each embodiment party of the present invention
In formula, many technical details are proposed in order to make reader more fully understand the present invention.But even if without these technical details
And various changes and modifications based on the following respective embodiments, claimed technical solution of the invention also may be implemented.
The first embodiment of the present invention is related to a kind of generation method of trade company's recommendation list, detailed process as shown in Figure 1,
Include:
S101: the user data of single user under history is obtained.
About step S101, specifically, the user data of single user is specifically included under the history: the base of the user
The cross attribute of this attribute and historical viewings data, trade company's attribute and historical transactional information, the user and the trade company,
Wherein, the historical transactional information includes that user Xia Dan trade company record and trade company expose user and do not obtain lower unirecord.User
Essential attribute be age of user, gender, personality, dining preference etc.;Historical viewings data are trade company's browsing of user
Which trade company record, that is, user browsed;Trade company's attribute refers to that the working group of the trade company (for example is Sichuan cuisine shop or fire
Pot shop etc.), the scale of trade company, the geographical location of trade company etc.;Real-time deal information is the exchange hand of trade company, concluded price etc.;
The cross attribute of user and trade company is the place of user property Yu trade company's overlapped attributes, such as some user likes eating Sichuan cuisine,
And some trade company is Sichuan cuisine shop, then the user and the trade company directly there is cross attribute (Sichuan cuisine).By obtaining history sample
Eigen data can provide a large amount of training data for the foundation of model, and these data are all authentic and valid and mutual passes
The data of connection provide effective reference for the interested dining room of accurate recommended user.
It should be noted that in the present embodiment, under above-mentioned history the source of the user data of single user can be from
Different channels acquire (such as internet is manually entered) to come, and the data for being also possible to count from off-line data are (as used
The data at family itself, the data in dining room itself).In addition, the user data of single user is also not limited to the example above under history
User essential attribute and historical viewings data, trade company's attribute and historical transactional information and user and trade company cross attribute,
All data relevant to user, that is, dining room can be used as historical sample characteristic.
It is noted that obtaining in present embodiment and being applied alone under history to keep the recommendation list generated more accurate
The user data at family, can be with are as follows: to the user data of single user under the history using etc. frequency divisions bucket carry out sliding-model control, obtain
Take discretization data;It is placed an order single price under rate and the history according to history described in the discretization data acquisition.It was modeling
Cheng Zhong is needed to continuous variable discretization, and after feature discretization, model can be more stable, reduces the risk of model over-fitting, because
This, carries out sliding-model control to frequency divisions buckets such as historical sample characteristic uses, so that the data after sliding-model control only have 0 He
1 two kinds of values, in order to which subsequent computer goes on smoothly model training.The basic thought of equal frequency divisions bucket are as follows: for accurate discrete
Change, opposite quefrency should be completely the same in a section.Therefore, if two adjacent sections have very similar class
Distribution, then the two sections can merge;Otherwise, they should be held apart at, and low chi-square value shows them with similar
Class distribution.
More preferably, in present embodiment under history single user user data using etc. frequency divisions bucket carry out discretization at
It manages, after acquisition discretization data, further includes: utilize the historical sample characteristic to carry out model based on Xgboost algorithm
Training obtains the corresponding leaf node number of user data of single user under history, regard leaf node number as feature
Growth data;The user data using single user under the history carries out model training, obtains the rate prediction model that places an order, tool
Body includes: to carry out model training using the user data of single user under the history and the feature growth data, and acquisition places an order
Rate prediction model;The user data using single user under the history carries out model training, obtains lower single price estimation mould
Type specifically includes: carrying out model training using the user data of single user under the history and the feature growth data, obtains
Lower list price estimation model.Xgboost algorithm is the improvement carried out on the basis of GBDT to boosting algorithm, internal decision making
Tree uses regression tree, and for the division node of regression tree for quadratic loss function, fitting is exactly residual error;For General Loss
Function (gradient decline), fitting be exactly residual error approximation, division node enumerates the values of all features when dividing, chooses and divide
Point, it is finally predicting the result is that the prediction result of each tree is added.Xgboost algorithm has the advantages that speed is fast, effect
Large-scale data well, can be handled, multilingual is supported, support customized loss function etc..It is understood that passing through utilization
Discretization data after equal frequency divisions bucket and the feature growth data obtained based on Xgboost algorithm carry out model training jointly, energy
Prediction is set obtained in subsequent step to place an order rate and to estimate lower single price more accurate, enough so as to described in further precision
Recommendation list.
S102: model training is carried out to obtain the rate prediction model that places an order using the user data of single user under history.
About step S102, specifically, the user data of single user under the history is utilized described in present embodiment
Model training is carried out, acquisition places an order rate prediction model can be with are as follows: the number of users of single user under the history is utilized based on FM algorithm
According to model training is carried out, acquisition is described to predict the rate that places an order.It is understood that being wrapped in the user data of single user under the history
The history for including the historical user places an order rate.In traditional linear model such as LR, each feature be it is independent, if necessary
Consider feature and the direct reciprocation of feature, it may be necessary to combined crosswise manually be carried out to feature;Non-linear SVM can be to spy
Sign carries out kernel mapping, but in the case where feature height is sparse, can not be learnt well;There are also very much
Decomposition model Factorization model such as matrix decomposition MF, SVD++ etc., these models may learn between feature
The hiding relationship of interaction, but substantially each model is only applicable to specifically input and scene.For this purpose, the data sparse in height
Such as recommender system under scene, FM (Factorization Machine) algorithm can be used.FM algorithm is intersected by vector to be learnt
Mode excavate the correlation between feature, have following two points benefit: 1. can preferably dig under conditions of height is sparse
The correlation between data characteristics is dug, especially for the intersection data not occurred in training sample;2.FM is calculating target letter
It counts and can be completed in linear session when stochastic gradient descent does Optimization Learning.It can make subsequent step by such method
The prediction obtained in the rapid rate that places an order is more accurate.
S103: model training is carried out using the user data of single user under history to obtain lower single price estimation model.
About step S103, specifically, model is carried out using the user data of single user under history in present embodiment
Training, can be with to obtain lower single price estimation model are as follows: utilizes the number of users of single user under the history based on Xgboost algorithm
According to model training is carried out, lower single price estimation model is obtained.It is understood that single user under history in present embodiment
User data in include the user history under single price.It can make what is obtained in subsequent step to estimate in this manner
Lower list price is more accurate.
It will be understood by those skilled in the art that the algorithm that model training uses in above-mentioned steps is only to realize present embodiment
A kind of citing of middle model training can also train place an order rate prediction model and lower single price estimation model by other means,
It can reach technical effect identical with present embodiment.In addition, the execution sequence of step S102 and step S103 can be regardless of
Successively, it first carries out execution step S102 after step S102 executes step S103 again, first carries out step S103 or is performed simultaneously step
Rapid S102 and step S103 is feasible.
S104: obtaining the user data of current lower single user, will it is current descend single user the user data input rate that places an order it is pre-
Survey obtained in model prediction place an order rate, will be estimated in list price estimation model under the user data input of current lower single user
Lower list price.
It, specifically, can other be wireless by APP etc. when user needs the recommendation list of trade company about step S104
Communication tool sends request, and server (is such as liked eating what taste food, likes eating in the characteristic information for receiving user's request
What kind of food etc.) after, obtained in rate prediction model that this feature information input can place an order prediction place an order rate, by this feature
It obtains estimating lower single price in single price estimation model under information input.
S105: it is placed an order according to prediction and rate and estimates the recommended hour that lower single price obtains trade company, and quotient is generated according to recommended hour
Family recommendation list.
About step S105, specifically, obtaining prediction and placing an order rate and to estimate lower single price, it can be according to specific meter
It calculates formula or other modes obtains the recommended hour of trade company, server can generate trade company's recommendation list according to the height of recommended hour, such as
It is placed on recommended hour is high before recommendation list, is placed on recommended hour is low behind recommendation list.
Embodiments of the present invention in terms of existing technologies, by obtaining the user data of single user under history, make
The source for obtaining generation recommendation list in subsequent step is no longer single, and the user data of single user under the history is recycled to carry out mould
Type training is to obtain place an order rate prediction model and lower single price estimation model, so that utilizing different information (i.e. single user under history
User data in different information) carry out obtained different models (rate that places an order prediction model and lower unit price after model training
Lattice prediction model) more targetedly, so as to after the user data for obtaining current lower single user, will it is current under be applied alone
Obtained in the rate that places an order described in the user data input at family prediction model prediction place an order rate, will current time single user user data it is defeated
Enter to obtain estimating lower single price in lower single price estimation model more accurate, and then makes place an order according to the prediction rate and institute
It states and estimates specifying information and user that the recommended hour of the trade company that lower single price obtains can really reflect the trade company to this
The fancy grade of businessman, therefore, the trade company's recommendation list generated according to the recommended hour can carry out accurate for different user
Trade company recommend, user be based on the recommendation list can quickly find oneself favorite trade company and consume, to mention
User experience and trade company's income have been risen, has been avoided and " is based only upon the rate of placing an order or clicking rate generates the recommendation list of trade company, so that this is pushed away
The generation of list is recommended according to single, the specifying information that can not accurately reflect trade company and user to the fancy grade of businessman, thus
Cause the recommendation list provided a user not accurate, and then affect the experience and trade company's income of user " the case where generation.
Second embodiment of the present invention is related to a kind of generation method of recommendation list, and second embodiment is real first
It applies and has done further improvement on the basis of mode, specifically the improvement is that: in this second embodiment, being obtained described
It takes under history after the user data of single user, further includes: the Xia Dan trade company recording mark of single user under the history is positive
Sample, exposing to user and do not obtain lower single recording mark the trade company is negative sample;The negative sample is sampled so that
The quantity of the negative sample is equal with the quantity of the positive sample, hands over using the positive sample and the negative sample as the history
Easy information.To be far longer than the dining room that user places an order due to not obtaining the dining room that user places an order, to make positive and negative sample proportion substantially
It is identical, it needs to be sampled negative sample, so that the historical sample characteristic obtained is more accurate.
The detailed process of present embodiment is as shown in Figure 2, comprising:
S201: the user data of single user under history is obtained.
S202: it is positive sample by user Xia Dan trade company's recording mark, trade company exposes user and does not obtain lower unirecord mark
It is denoted as negative sample, negative sample is sampled.
About step S202, specifically, to be far longer than the meal that user places an order due to not obtaining the dining room that user places an order
The Room, to keep positive and negative sample proportion roughly the same, it is preferred that need to be sampled negative sample.If the quantity of positive sample is 10, bear
The quantity of sample is 100, then 10 are just extracted out in this 100 negative samples, so that positive and negative sample size is identical.
S203: history is obtained according to the user data of single user under history and is placed an order single price under rate and history.
S204: rate progress model training is placed an order to obtain the rate prediction model that places an order using history.
S205: model training is carried out using price single under history to obtain lower single price estimation model.
S206: obtaining the user data of current lower single user, will it is current descend single user the user data input rate that places an order it is pre-
Survey obtained in model prediction place an order rate, will be estimated in list price estimation model under the user data input of current lower single user
Lower list price.
The step S101 in step S201, step S203 to step S206 and first embodiment in present embodiment is extremely
Step S104 is similar, and in order to avoid repeating, details are not described herein again.The execution sequence of S204 and S205 can also in no particular order, first
It executes and executes step S204 after step S204 executes step S205 again, first carries out step S205 or be performed simultaneously step S204
It is all feasible with step S205.
S207: it is placed an order according to prediction and rate and estimates the recommended hour that lower single price obtains trade company, and quotient is generated according to recommended hour
Family recommendation list.
About step S207, specifically, is placed an order in present embodiment according to prediction and rate and estimate lower single price and obtain quotient
The recommended hour at family, can be with are as follows: places an order rate according to the prediction and following formula obtains the rate that really places an order of the trade company: q=
p/(p+(1-p)/w);Wherein, q be it is described really place an order rate, p is that the prediction places an order rate, and w is negative specimen sample scale factor;
According to rate and the recommended hour estimated lower single price and obtain dining room of really placing an order.It is understood that being instructed via model
The prediction got the rate that places an order is bigger than normal, and the multiple amplified be it is known, the prediction can be placed an order rate by above-mentioned formula
Really placed an order rate for inverse operation (reducing multiple identical with amplification factor), to keep marking structure more accurate.It needs
Illustrate, negative sample oversampling ratio factor w can be obtained with S202 through the above steps, such as be extracted out in 100 negative samples
10 negative samples, then negative sample oversampling ratio factor w=10/100=0.1.
In addition, according to really place an order rate and the recommendation estimated lower single price and obtain dining room in present embodiment
Point, it can be with are as follows: obtain the recommended hour: Score=q*price according to the following formula;Wherein, Score is the recommended hour, q
For the rate that really places an order, price estimates lower single price to be described.In order to make it easy to understand, below to recommending in present embodiment
A kind of feasible generating process of list is illustrated, and can be divided into the following steps:
1, data sample prepares
It is received based on history log information and Real time request log acquisition sample characteristics data, such as by channels such as internets
It is current to collect history dining record, the current dining information of user, the history exchange hand in dining room and the concluded price of user, dining room
Exchange hand and concluded price etc. are sample characteristics data, do not repeat one by one herein.
2, data prediction
For the sample characteristics data got in step 1, data prediction is carried out to it, such as: (1) for convenient for subsequent calculation
Method training, to all dense pattern eigen data using etc. frequency divisions bucket carry out sliding-model control (2) be based on Xgboost algorithm into
Row model training, the corresponding leaf node number of output sample characteristics data is as extension feature data etc., those skilled in the art
Member is it is appreciated that enumerated processing mode is only the two of them in sample characteristics data prediction mode, and there are also other
Pretreatment mode can also reach technical effect same as aforementioned two kinds of processing modes, no longer repeat one by one herein.
3, the rate that places an order model training
Pretreated sample characteristics data training pattern in utilization (1), (2), to obtain the rate prediction model that places an order, thus
Overcome static models can not follow up in real time user interest variation defect, it is to be understood that can be based in present embodiment
FM algorithm training pattern, to reach better effect.It should be noted that can be by (1), (2) in the training process of model
In model training is carried out in the pretreated equal input model of sample characteristics data to obtain the rate prediction model that places an order, can also will
(1), model training is carried out in (2) in the pretreated independent input model of sample characteristics data to obtain the rate prediction mould that places an order
Type will not influence the technical effect of present embodiment.After model training is good (having produced the rate prediction model that places an order), clothes
Characteristic information will be input in the rate prediction model that places an order by business device when receiving the characteristic information of user's request, pre- to obtain
Survey places an order rate.
4, marking is calibrated
The purpose of this step is to make obtained in step 3 the prediction rate that places an order more accurate, it is to be understood that under
The prediction rate of placing an order that single rate prediction model obtains can be greater than the rate that really places an order, and the multiple is known, it is therefore desirable to prediction
The rate of placing an order is calibrated, to obtain the rate that really places an order.In present embodiment the rate that really places an order: q can be obtained according to following formula
=p/ (p+ (1-p)/w);Wherein, q be it is described really place an order rate, p is that the prediction places an order rate, w be negative specimen sample ratio because
Son.
5, single price estimation under user
Pretreated sample characteristics data training pattern in utilization (1), (2) can to obtain lower single price estimation model
With understanding, Xgboost algorithm training pattern can be based in present embodiment, to reach better effect.It needs to illustrate
, can will be carried out in the equal input model of sample characteristics data pretreated in (1), (2) in the training process of model
Model training, can also be individually defeated by sample characteristics data pretreated in (1), (2) to obtain lower single price estimation model
Enter to carry out model training in model to obtain lower single price estimation model, will not influence the technical effect of present embodiment.?
After model training is good (produced lower single price estimation model), server when receiving the characteristic information of user's request just
Characteristic information can be input in lower single price estimation model, estimate lower single price to obtain.
6, shop recommended hour exports
For user's single request (i.e. the characteristic information of user's request), single probability is under the user of model prediction in note 4
Q, single price price under the user of model pre-estimating, obtains shop recommended hour Score:Score=q* according to following formula in 5
price.Dining room list is ranked up according to recommended hour, user is returned and completes this request.That is, user disappears having
When taking demand, it can be sent by wireless communication tool (being only exemplified by a kind of feasible transmission request method herein) to server special
Reference breath (such as want to eat the dish of what taste, where want to go to Fu Jin dining room etc.), server after receiving characteristic information,
This feature information can be separately input to place an order in rate prediction model and lower single price estimation model, with obtain prediction place an order rate and
Lower single price is estimated, places an order further according to prediction and rate and estimates the recommended hour that lower single price obtains dining room, according to the height of recommended hour
Dining room is ranked up, recommendation list is generated and returns to user, user, can be according to the recommendation after receiving recommendation list
List quickly finds the dining room for meeting oneself demand and is consumed.
Embodiments of the present invention in terms of existing technologies, are placed an order rate by obtaining user in the history that trade company places an order
And single price under history recycles the history to place an order rate so that the source for generating recommendation list in subsequent step is no longer single
Model training is carried out to obtain the rate prediction model that places an order, carries out model training using price single under the history to obtain lower unit price
Lattice prediction model, so that being obtained not after carrying out model training using different information (i.e. history place an order single price under rate and history)
Same model (and place an order rate prediction model and lower single price estimation model) is more targeted, so as to obtain user's request
Characteristic information after, will place an order described in characteristic information input obtain in rate prediction model prediction place an order rate, by the spy
It is more accurate to descend in single price estimation model to obtain estimating lower single price described in sign information input, and then makes according under the prediction
Single rate and the recommended hour for estimating the trade company that lower single price obtains can really reflect the trade company specifying information and
Therefore user can be directed to different user according to trade company's recommendation list that the recommended hour generates to the fancy grade of the businessman
It carries out accurately trade company to recommend, user, which is based on the recommendation list, can quickly find oneself favorite trade company and disappear
Take, so that the user experience is improved and trade company's income, avoids and " be based only upon the rate of placing an order or clicking rate generates the recommendation column of trade company
Table, so that the generation of the recommendation list according to single, can not accurately reflect the happiness of the specifying information and user of trade company to businessman
It is good degree, not accurate so as to cause the recommendation list provided a user, and then affect the experience and trade company's income of user " feelings
The generation of condition.
Third embodiment of the present invention is related to a kind of generating means 300 of recommendation list, as shown in figure 3, the device packet
It includes:
First obtains module 301, for obtaining user's single price in the case where the history that trade company places an order places an order rate and history;
First model training module 302, for using the history place an order rate carry out model training it is pre- to obtain the rate of placing an order
Survey model;
Second model training module 303, for carrying out model training using price single under the history to obtain lower unit price
Lattice prediction model;
Second obtains module 304, for obtaining the characteristic information of user's request;
First processing module 305, for will be obtained under prediction in the rate prediction model that places an order described in characteristic information input
Single rate will obtain estimating lower single price in characteristic information input lower single price estimation model;
Second processing module 306 rate and described estimate lower single price and obtains pushing away for trade company for being placed an order according to the prediction
It recommends point, and trade company's recommendation list is generated according to the recommended hour.
Embodiments of the present invention in terms of existing technologies, by obtaining the user data of single user under history, make
The source for obtaining generation recommendation list in subsequent step is no longer single, and the user data of single user under the history is recycled to carry out mould
Type training is to obtain place an order rate prediction model and lower single price estimation model, so that utilizing different information (i.e. single user under history
User data in different information) carry out obtained different models (rate that places an order prediction model and lower unit price after model training
Lattice prediction model) more targetedly, so as to after the user data for obtaining current lower single user, will it is current under be applied alone
Obtained in the rate that places an order described in the user data input at family prediction model prediction place an order rate, will current time single user user data it is defeated
Enter to obtain estimating lower single price in lower single price estimation model more accurate, and then makes place an order according to the prediction rate and institute
It states and estimates specifying information and user that the recommended hour of the trade company that lower single price obtains can really reflect the trade company to this
The fancy grade of businessman, therefore, the trade company's recommendation list generated according to the recommended hour can carry out accurate for different user
Trade company recommend, user be based on the recommendation list can quickly find oneself favorite trade company and consume, to mention
User experience and trade company's income have been risen, has been avoided and " is based only upon the rate of placing an order or clicking rate generates the recommendation list of trade company, so that this is pushed away
The generation of list is recommended according to single, the specifying information that can not accurately reflect trade company and user to the fancy grade of businessman, thus
Cause the recommendation list provided a user not accurate, and then affect the experience and trade company's income of user " the case where generation.
4th embodiment of the invention is related to a kind of electronic equipment, and the electronic equipment of present embodiment is it may be said that terminal side
Equipment, such as mobile phone, the terminal devices such as tablet computer, are also possible to the server of network side.
As shown in figure 4, the electronic equipment: including at least a processor 1001;And at least one processor 1001
The memory 1002 of communication connection;And the communication component 1003 with scanning means communication connection, communication component 1003 are being handled
Data are sended and received under the control of device 1001;Wherein, memory 1002, which is stored with, to be executed by least one processor 1001
Instruction, instruction by least one processor 1001 execute to realize:
Obtain the user data of single user under history;
Model training is carried out using the user data of single user under the history to obtain the rate prediction model that places an order;
Model training is carried out using the user data of single user under the history to obtain lower single price estimation model;
The user data for obtaining current lower single user, by the rate that places an order described in the user data input of the current lower single user
Obtained in prediction model prediction place an order rate, by lower single price estimation model described in the user data input of the current lower single user
In obtain estimating lower single price;
Place an order rate and the recommended hour estimated lower single price and obtain trade company are predicted according to described, and according to the recommended hour
Generate trade company's recommendation list.
Specifically, which includes: one or more processors 1001 and memory 1002, with one in Fig. 4
For processor 1001.Processor 1001, memory 1002 can be connected by bus or other modes, to pass through in Fig. 4
For bus connection.Memory 1002 is used as a kind of non-volatile computer readable storage medium storing program for executing, can be used for storing non-volatile
Software program, non-volatile computer executable program and module.Processor 1001 is stored in memory 1002 by operation
In non-volatile software program, instruction and module, thereby executing the various function application and data processing of equipment, i.e., in fact
The generation method of existing above-mentioned recommendation list.
Memory 1002 may include storing program area and storage data area, wherein storing program area can store operation system
Application program required for system, at least one function;It storage data area can the Save option list etc..In addition, memory 1002 can
It can also include nonvolatile memory to include high-speed random access memory, a for example, at least disk memory is dodged
Memory device or other non-volatile solid state memory parts.In some embodiments, it includes relative to place that memory 1002 is optional
The remotely located memory of device 1001 is managed, these remote memories can pass through network connection to external equipment.Above-mentioned network
Example includes but is not limited to internet, intranet, local area network, mobile radio communication and combinations thereof.
One or more module is stored in memory 1002, when being executed by one or more processor 1001,
Execute the generation method of the recommendation list in above-mentioned any means embodiment.
The said goods can be performed the application embodiment provided by method, have the corresponding functional module of execution method and
Beneficial effect, the not technical detail of detailed description in the present embodiment, reference can be made to method provided by the application embodiment.
Fifth embodiment of the invention is related to a kind of computer readable storage medium, is stored with computer program.Computer
Above method embodiment is realized when program is executed by processor.
That is, it will be understood by those skilled in the art that implement the method for the above embodiments be can be with
Relevant hardware is instructed to complete by program, which is stored in a storage medium, including some instructions are to make
It obtains an equipment (can be single-chip microcontroller, chip etc.) or processor (processor) executes side described in each embodiment of the application
The all or part of the steps of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey
The medium of sequence code.
It will be understood by those skilled in the art that the respective embodiments described above are to realize specific embodiments of the present invention,
And in practical applications, can to it, various changes can be made in the form and details, without departing from the spirit and scope of the present invention.
The embodiment of the present application discloses A1, a kind of generation method of trade company's recommendation list, comprising:
Obtain user's single price in the case where the history that trade company places an order places an order rate and history;
Obtain the user data of single user under history;
Model training is carried out using the user data of single user under the history rate prediction model and to place an order to obtain to place an order
Price estimation model;
The user data for obtaining current lower single user, by the rate that places an order described in the user data input of the current lower single user
Obtained in prediction model prediction place an order rate, by lower single price estimation model described in the user data input of the current lower single user
In obtain estimating lower single price;
Place an order rate and the recommended hour estimated lower single price and obtain trade company are predicted according to described, and according to the recommended hour
Generate trade company's recommendation list.
The generation method of A2, recommendation list as described in a1, the user data of single user specifically includes under the history: institute
It states and is applied alone under the essential attribute of single user and historical viewings data, trade company's attribute and historical transactional information, the history under history
The cross attribute at family and the trade company, wherein the historical transactional information includes user Xia Dan trade company record and trade company to user
It exposes and does not obtain lower unirecord.
The generation method of A3, as described in A2 recommendation list, under the acquisition history after the user data of single user,
Further include:
It is positive sample by the Xia Dan trade company recording mark of single user under the history, trade company exposes user and does not obtain down
Unirecord is labeled as negative sample;
The negative sample is sampled so that the quantity of the negative sample is equal with the quantity of the positive sample, it will be described
The negative sample after positive sample and sampling is as the historical transactional information.
The generation method of A4, recommendation list as described in A3, it is described rate to be placed an order according to the prediction and described estimate places an order
Price obtains the recommended hour of trade company, specifically includes:
Rate is placed an order according to the prediction and following formula obtains the rate that really places an order of the trade company:
Q=p/ (p+ (1-p)/w);
Wherein, q be it is described really place an order rate, p is that the prediction places an order rate, and w is negative specimen sample scale factor;
According to rate and the recommended hour estimated lower single price and obtain dining room of really placing an order.
The generation method of A5, recommendation list as described in A4, it is described really to place an order rate and described estimate places an order according to described
Price obtains the recommended hour in dining room, specifically includes:
The recommended hour is obtained according to the following formula:
Score=q*price;
Wherein, Score is the recommended hour, and q is the rate that really places an order, and price estimates lower single price to be described.
The generation method of A6, recommendation list as described in a1 carry out mould using the user data of single user under the history
Type training is specifically included with obtaining lower single price estimation model:
Model training is carried out using the user data of single user under the history based on Xgboost algorithm, under acquisition is described
Monovalent lattice prediction model.
The generation method of A7, recommendation list as described in a1, the user data using single user under the history into
Row model training is specifically included with obtaining the rate prediction model that places an order:
Model training is carried out using the user data of single user under the history based on FM algorithm, the prediction is obtained and places an order
Rate.
The generation method of A8, recommendation list as described in a1, the user data for obtaining single user under history are specific to wrap
It includes:
The frequency divisions buckets such as the user data use to single user under the history carry out sliding-model control, obtain discretization number
According to;
Using the discretization data as the user data of single user under the history.
The generation method of A9, recommendation list as described in A8, it is described to the history under the user data of single user adopt
Carries out sliding-model control with equal frequency divisions bucket, after acquisition discretization data, further includes:
Model training is carried out using the user data of single user under the history based on Xgboost algorithm, is gone through described in acquisition
The corresponding leaf node number of the user data of single user under history, regard leaf node number as feature growth data;
The user data using single user under the history carries out model training to obtain the rate prediction model that places an order, and has
Body includes:
Model training is carried out using the user data of single user under the history and the feature growth data, acquisition places an order
Rate prediction model;
Model training is carried out using the user data of single user under the history to obtain lower single price estimation model, specifically
Include:
Model training is carried out using the user data of single user under the history and the feature growth data, acquisition places an order
Price estimation model.
The embodiment of the present application discloses B1, a kind of generating means of recommendation list, comprising:
First obtains module, for obtaining the user data of single user under history;
First model training module carries out model training for the user data using single user under the history to obtain
The rate that places an order prediction model;
Second model training module carries out model training for the user data using single user under the history to obtain
Lower list price estimation model;
Second obtains module, for obtaining the user data of current lower single user;
First processing module, in the rate prediction model that will place an order described in the user data input of the current lower single user
Obtain prediction place an order rate, will be estimated in lower single price estimation model described in the user data input of the current lower single user
Lower list price;
Second processing module, for predicting place an order rate and the recommendation estimated lower single price and obtain trade company according to described
Point, and trade company's recommendation list is generated according to the recommended hour.
The embodiment of the present application discloses C1, a kind of electronic equipment, including at least one processor;And
The memory being connect at least one described processor communication;
And the communication component with scanning means communication connection, the communication component connect under the control of the processor
Receive and send data;
Wherein, the memory is stored with the instruction that can be executed by least one described processor, and described instruction is described
At least one processor is executed to realize:
Obtain the user data of single user under history;
Model training is carried out using the user data of single user under the history rate prediction model and to place an order to obtain to place an order
Price estimation model;
The user data for obtaining current lower single user, by the rate that places an order described in the user data input of the current lower single user
Obtained in prediction model prediction place an order rate, by lower single price estimation model described in the user data input of the current lower single user
In obtain estimating lower single price;
Place an order rate and the recommended hour estimated lower single price and obtain trade company are predicted according to described, and according to the recommended hour
Generate trade company's recommendation list.
C2, the electronic equipment as described in C1, the user data of single user specifically includes under the history: the history places an order
Single user and the quotient under the essential attribute and historical viewings data of user, trade company's attribute and historical transactional information, the history
The cross attribute at family, wherein the historical transactional information includes that user Xia Dan trade company record and trade company expose user and do not obtain
Lower unirecord.
C3, the electronic equipment as described in C2, the processor execute obtain for characterize the user order habit and
After the historical sample characteristic of trade company's sales situation, it is also used to:
It is positive sample by the Xia Dan trade company recording mark of single user under the history, trade company exposes user and does not obtain down
Unirecord is labeled as negative sample;
The negative sample is sampled so that the quantity of the negative sample is equal with the quantity of the positive sample, it will be described
Positive sample and the negative sample are as the historical transactional information.
C4, the electronic equipment as described in C3, the processor, which is executed, places an order rate according to the prediction and described estimate places an order
Price obtains the recommended hour of trade company, specifically:
Rate is placed an order according to the prediction and following formula obtains the rate that really places an order of the trade company:
Q=p/ (p+ (1-p)/w);
Wherein, q be it is described really place an order rate, p is that the prediction places an order rate, and w is negative specimen sample scale factor;
According to rate and the recommended hour estimated lower single price and obtain dining room of really placing an order.
The generation method of C5, recommendation list as described in C4, the processor are executed according to rate and the institute of really placing an order
It states and estimates the recommended hour that lower single price obtains dining room, specifically:
The recommended hour is obtained according to the following formula:
Score=q*price;
Wherein, Score is the recommended hour, and q is the rate that really places an order, and price estimates lower single price to be described.
C6, the electronic equipment as described in C1, the processor execute the user data using single user under the history into
Row model training to obtain lower single price estimation model, specifically:
Model training is carried out using the user data of single user under the history based on Xgboost algorithm, under acquisition is described
Monovalent lattice prediction model.
C7, the electronic equipment as described in C1, the processor execute the user data using single user under the history into
Row model training to obtain the rate prediction model that places an order, specifically:
Model training is carried out using the user data of single user under the history based on FM algorithm, the prediction is obtained and places an order
Rate.
C8, the electronic equipment as described in C1, the processor execute the user data for obtaining single user under history, tool
Body are as follows:
The frequency divisions buckets such as the user data use to single user under the history carry out sliding-model control, obtain discretization number
According to;
Using the discretization data as the user data of single user under the history.
C9, the electronic equipment as described in C8, processor user data of single user in the case where executing to the history are adopted
Sliding-model control is carried out with equal frequency divisions bucket, after obtaining discretization data, is also used to:
Model training is carried out using the user data of single user under the history based on Xgboost algorithm, is gone through described in acquisition
The corresponding leaf node number of the user data of single user under history, regard leaf node number as feature growth data;
The processor, which is executed, carries out model training using the user data of single user under the history to obtain the rate of placing an order
Prediction model, specifically:
Model training is carried out using the user data of single user under the history and the feature growth data, acquisition places an order
Rate prediction model;
The processor, which is executed, carries out model training using the user data of single user under the history to obtain lower unit price
Lattice prediction model, specifically:
Model training is carried out using the user data of single user under the history and the feature growth data, acquisition places an order
Price estimation model
The embodiment of the present application discloses D1, a kind of computer readable storage medium, is stored with computer program, the calculating
The generation method of recommendation list described in any one of A1 to A9 is realized when machine program is executed by processor.
Claims (10)
1. a kind of generation method of trade company's recommendation list characterized by comprising
Obtain the user data of single user under history;
Model training is carried out using the user data of single user under the history to obtain place an order rate prediction model and lower single price
Prediction model;
The user data for obtaining current lower single user, by the rate prediction that places an order described in the user data input of the current lower single user
Obtained in model prediction place an order rate, will be in lower single price estimation model described in the user data input of the current lower single user
To estimating lower single price;
Place an order rate and the recommended hour estimated lower single price and obtain trade company are predicted according to described, and according to recommended hour generation
Trade company's recommendation list.
2. the generation method of recommendation list according to claim 1, which is characterized in that the user of single user under the history
Data specifically include: the essential attribute of single user and historical viewings data, trade company's attribute and historical transactional information under the history,
The cross attribute of single user and the trade company under the history, wherein the historical transactional information includes user Xia Dan trade company note
Record and trade company expose user and do not obtain lower unirecord.
3. the generation method of recommendation list according to claim 2, which is characterized in that the single user under the acquisition history
User data after, further includes:
It is positive sample by the Xia Dan trade company recording mark of single user under the history, trade company exposes user and does not obtain the note that places an order
Record mark is negative sample;
The negative sample is sampled so that the quantity of the negative sample is equal with the quantity of the positive sample, by the positive sample
Originally and the negative sample after sampling is as the historical transactional information.
4. the generation method of recommendation list according to claim 3, which is characterized in that described to predict the rate that places an order according to described
And the recommended hour estimated lower single price and obtain trade company, it specifically includes:
Rate is placed an order according to the prediction and following formula obtains the rate that really places an order of the trade company:
Q=p/ (p+ (1-p)/w);
Wherein, q be it is described really place an order rate, p is that the prediction places an order rate, and w is negative specimen sample scale factor;
According to rate and the recommended hour estimated lower single price and obtain dining room of really placing an order.
5. the generation method of recommendation list according to claim 4, which is characterized in that described according to the rate that really places an order
And the recommended hour estimated lower single price and obtain dining room, it specifically includes:
The recommended hour is obtained according to the following formula:
Score=q*price;
Wherein, Score is the recommended hour, and q is the rate that really places an order, and price estimates lower single price to be described.
6. the generation method of recommendation list according to claim 1, which is characterized in that utilize single user under the history
User data carries out model training to obtain lower single price estimation model, specifically includes:
Model training is carried out using the user data of single user under the history based on Xgboost algorithm, obtains the lower unit price
Lattice prediction model.
7. the generation method of recommendation list according to claim 1, which is characterized in that described utilize is applied alone under the history
The user data at family carries out model training to obtain the rate prediction model that places an order, and specifically includes:
Model training is carried out using the user data of single user under the history based on FM algorithm, acquisition is described to predict the rate mould that places an order
Type.
8. a kind of generating means of recommendation list characterized by comprising
First obtains module, for obtaining the user data of single user under history;
First model training module carries out model training for the user data using single user under the history and is placed an order with obtaining
Rate prediction model;
Second model training module carries out model training for the user data using single user under the history and is placed an order with obtaining
Price estimation model;
Second obtains module, for obtaining the user data of current lower single user;
First processing module, for will be obtained in the rate prediction model that places an order described in the user data input of the current lower single user
Prediction place an order rate, will obtain estimating in lower single price estimation model described in the user data input of the current lower single user and place an order
Price;
Second processing module, for being placed an order rate and the recommended hour estimated lower single price and obtain trade company according to the prediction, and
Trade company's recommendation list is generated according to the recommended hour.
9. a kind of electronic equipment, including at least one processor;And
The memory being connect at least one described processor communication;
And with scanning means communication connection communication component, the communication component is received under the control of the processor with
Send data;
Wherein, the memory be stored with can by least one described processor execute instruction, described instruction by it is described at least
One processor is executed to realize:
Obtain the user data of single user under history;
Model training is carried out using the user data of single user under the history to obtain place an order rate prediction model and lower single price
Prediction model;
The user data for obtaining current lower single user, by the rate prediction that places an order described in the user data input of the current lower single user
Obtained in model prediction place an order rate, will be in lower single price estimation model described in the user data input of the current lower single user
To estimating lower single price;
Place an order rate and the recommended hour estimated lower single price and obtain trade company are predicted according to described, and according to recommended hour generation
Trade company's recommendation list.
10. a kind of computer readable storage medium, is stored with computer program, which is characterized in that the computer program is located
Reason device realizes the generation method of recommendation list described in any one of claims 1 to 7 when executing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811621615.0A CN109711887B (en) | 2018-12-28 | 2018-12-28 | Generation method and device of mall recommendation list, electronic equipment and computer medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811621615.0A CN109711887B (en) | 2018-12-28 | 2018-12-28 | Generation method and device of mall recommendation list, electronic equipment and computer medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109711887A true CN109711887A (en) | 2019-05-03 |
CN109711887B CN109711887B (en) | 2021-02-09 |
Family
ID=66258904
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811621615.0A Active CN109711887B (en) | 2018-12-28 | 2018-12-28 | Generation method and device of mall recommendation list, electronic equipment and computer medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109711887B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110400167A (en) * | 2019-06-28 | 2019-11-01 | 阿里巴巴集团控股有限公司 | User's screening technique and device |
CN110992097A (en) * | 2019-12-03 | 2020-04-10 | 上海钧正网络科技有限公司 | Processing method and device for revenue product price, computer equipment and storage medium |
CN111598677A (en) * | 2020-07-24 | 2020-08-28 | 北京淇瑀信息科技有限公司 | Resource quota determining method and device and electronic equipment |
CN111815415A (en) * | 2020-07-14 | 2020-10-23 | 北京邮电大学 | Commodity recommendation method, system and equipment |
CN112163155A (en) * | 2020-09-30 | 2021-01-01 | 深圳前海微众银行股份有限公司 | Information processing method, device, equipment and storage medium |
CN112801744A (en) * | 2020-12-01 | 2021-05-14 | 珠海必要工业科技股份有限公司 | Activity recommendation method and device, electronic equipment and storage medium |
WO2021159640A1 (en) * | 2020-02-13 | 2021-08-19 | 平安科技(深圳)有限公司 | Drug recommendation method based on artificial intelligence, and related device |
US11282052B2 (en) * | 2019-05-06 | 2022-03-22 | Advanced New Technologies Co., Ltd. | Payment channel recommendation |
WO2023000491A1 (en) * | 2021-07-19 | 2023-01-26 | 广东艾檬电子科技有限公司 | Application recommendation method, apparatus and device, and computer-readable storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105677831A (en) * | 2016-01-04 | 2016-06-15 | 拉扎斯网络科技(上海)有限公司 | Method and device for determining recommended merchants |
CN106886837A (en) * | 2017-02-21 | 2017-06-23 | 携程旅游网络技术(上海)有限公司 | Free walker stroke based on time planning recommends method |
CN107578294A (en) * | 2017-09-28 | 2018-01-12 | 北京小度信息科技有限公司 | User's behavior prediction method, apparatus and electronic equipment |
CN107657506A (en) * | 2017-09-06 | 2018-02-02 | 北京五八到家信息技术有限公司 | The intelligent recommendation method and commending system of a kind of service commodity |
CN107818478A (en) * | 2017-10-31 | 2018-03-20 | 携程计算机技术(上海)有限公司 | Overseas reward voucher distribution method and system based on provisional profit |
CN107944593A (en) * | 2017-10-11 | 2018-04-20 | 北京三快在线科技有限公司 | A kind of resource allocation methods and device, electronic equipment |
-
2018
- 2018-12-28 CN CN201811621615.0A patent/CN109711887B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105677831A (en) * | 2016-01-04 | 2016-06-15 | 拉扎斯网络科技(上海)有限公司 | Method and device for determining recommended merchants |
CN106886837A (en) * | 2017-02-21 | 2017-06-23 | 携程旅游网络技术(上海)有限公司 | Free walker stroke based on time planning recommends method |
CN107657506A (en) * | 2017-09-06 | 2018-02-02 | 北京五八到家信息技术有限公司 | The intelligent recommendation method and commending system of a kind of service commodity |
CN107578294A (en) * | 2017-09-28 | 2018-01-12 | 北京小度信息科技有限公司 | User's behavior prediction method, apparatus and electronic equipment |
CN107944593A (en) * | 2017-10-11 | 2018-04-20 | 北京三快在线科技有限公司 | A kind of resource allocation methods and device, electronic equipment |
CN107818478A (en) * | 2017-10-31 | 2018-03-20 | 携程计算机技术(上海)有限公司 | Overseas reward voucher distribution method and system based on provisional profit |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11282052B2 (en) * | 2019-05-06 | 2022-03-22 | Advanced New Technologies Co., Ltd. | Payment channel recommendation |
CN110400167A (en) * | 2019-06-28 | 2019-11-01 | 阿里巴巴集团控股有限公司 | User's screening technique and device |
CN110992097A (en) * | 2019-12-03 | 2020-04-10 | 上海钧正网络科技有限公司 | Processing method and device for revenue product price, computer equipment and storage medium |
CN110992097B (en) * | 2019-12-03 | 2024-02-13 | 上海钧正网络科技有限公司 | Processing method and device for price of revenue product, computer equipment and storage medium |
WO2021159640A1 (en) * | 2020-02-13 | 2021-08-19 | 平安科技(深圳)有限公司 | Drug recommendation method based on artificial intelligence, and related device |
CN111815415A (en) * | 2020-07-14 | 2020-10-23 | 北京邮电大学 | Commodity recommendation method, system and equipment |
CN111815415B (en) * | 2020-07-14 | 2023-10-27 | 北京邮电大学 | Commodity recommendation method, system and equipment |
CN111598677A (en) * | 2020-07-24 | 2020-08-28 | 北京淇瑀信息科技有限公司 | Resource quota determining method and device and electronic equipment |
CN112163155A (en) * | 2020-09-30 | 2021-01-01 | 深圳前海微众银行股份有限公司 | Information processing method, device, equipment and storage medium |
CN112801744A (en) * | 2020-12-01 | 2021-05-14 | 珠海必要工业科技股份有限公司 | Activity recommendation method and device, electronic equipment and storage medium |
WO2023000491A1 (en) * | 2021-07-19 | 2023-01-26 | 广东艾檬电子科技有限公司 | Application recommendation method, apparatus and device, and computer-readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109711887B (en) | 2021-02-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109711887A (en) | Generation method and device of mall recommendation list, electronic equipment and computer medium | |
TWI690880B (en) | Recommended method and device for marketing products | |
US11222273B2 (en) | Service recommendation method, apparatus, and device | |
CN104462593B (en) | A kind of method and apparatus that the push of user individual message related to resources is provided | |
CN109299994B (en) | Recommendation method, device, equipment and readable storage medium | |
CN103064945B (en) | Based on the Situational searching method of body | |
CN103295145B (en) | Mobile phone advertising method based on user consumption feature vector | |
CN108256907A (en) | A kind of construction method and computing device of customer grouping model | |
Sun et al. | Contextual intent tracking for personal assistants | |
CN109345302A (en) | Machine learning model training method, device, storage medium and computer equipment | |
CN105320724A (en) | New heuristic for optimizing non-convex function for learning to rank | |
CN108510303A (en) | Advertisement placement method and device | |
CN106600302A (en) | Hadoop-based commodity recommendation system | |
CN110020094A (en) | A kind of methods of exhibiting and relevant apparatus of search result | |
CN105677780A (en) | Scalable user intent mining method and system thereof | |
JP2021166109A (en) | Fusion sorting model training method and device, search sorting method and device, electronic device, storage medium, and program | |
CN102402625A (en) | Method and system for recommending music | |
WO2018232331A1 (en) | Systems and methods for optimizing and simulating webpage ranking and traffic | |
CN108205775A (en) | The recommendation method, apparatus and client of a kind of business object | |
CN110175895A (en) | A kind of item recommendation method and device | |
CN106557480A (en) | Implementation method and device that inquiry is rewritten | |
CN108804133A (en) | Acquisition methods, system, computer equipment and the storage medium of virtual resource | |
CN110233879A (en) | Intelligently pushing interfacial process, device, computer equipment and storage medium | |
CN112347147A (en) | Information pushing method and device based on user association relationship and electronic equipment | |
CN111680213B (en) | Information recommendation method, data processing method and 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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |