CN107194722A - A kind of Dynamic Pricing algorithm based on data mining under shared economy - Google Patents
A kind of Dynamic Pricing algorithm based on data mining under shared economy Download PDFInfo
- Publication number
- CN107194722A CN107194722A CN201710342702.1A CN201710342702A CN107194722A CN 107194722 A CN107194722 A CN 107194722A CN 201710342702 A CN201710342702 A CN 201710342702A CN 107194722 A CN107194722 A CN 107194722A
- Authority
- CN
- China
- Prior art keywords
- mrow
- feature
- price
- data
- msubsup
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0283—Price estimation or determination
- G06Q30/0284—Time or distance, e.g. usage of parking meters or taximeters
Landscapes
- Business, Economics & Management (AREA)
- Development Economics (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the Dynamic Pricing algorithm based on data mining under a kind of shared economy, include the weight optimization of preliminary pricing strategy, feature based screening, the weight optimization based on machine learning.The Dynamic Pricing algorithm based on data mining under a kind of shared economy disclosed by the invention, used by the auxiliary of multiple pricing strategy, realize the optimization and application of Dynamic Pricing algorithm, solve algorithm using it is single, be difficult to factor intervention, ageing low problem.
Description
Technical field
The present invention relates to the base under a kind of Dynamic Pricing technology based on mobile Internet, more particularly to a kind of shared economy
In the Dynamic Pricing algorithm of data mining.
Background technology
With popularizing for intelligent movable equipment and mobile payment so that the resource of Supply and Demand side, which is exchanged, to be possessed
Higher real-time.Based on this, Dynamic Pricing strategy obtains very big development:Uber Dynamic Pricing strategy obtains larger
Success, causes the highest attention of market and industry;Airbnb implemented Dynamic Pricing strategy in 2015 to its people Su Pingtai,
For pointing out supplying party one relatively reasonable rent price, realize the objective market based on data and instruct and refer to, and open
Its machine learning engineering (Aerosolve) code of source.
The current of economic resources pattern rise is shared in the whole people, Dynamic Pricing strategy is solved well to a certain extent
The arm's length pricing problem of service, but it still has many problem of need to improve, and such as price changes too fast, red-letter day or during emergency event
Price is too high, lack flexibility etc..
The rise starting of domestic Dynamic Pricing soon, existing use case at present, such as drop drop is called a taxi, Divine Land special train.But
The strategy that it is fixed a price is more single, is provided simultaneously with stronger limitation, Dynamic Pricing is not occupied mainly on its pricing practice
Status.This paper feature baseds screening technique and machine learning method, the multifactor and its weight to influence price are explored, and
Based on the dynamic pricing models of this structure automation, used by the auxiliary of multiple pricing strategy, realize Dynamic Pricing algorithm
Optimization with application, solve algorithm using it is single, be difficult to factor intervention, ageing low problem.
The content of the invention
It is an object of the invention to provide the Dynamic Pricing algorithm based on data mining under a kind of shared economy, and based on real
More flexible Dynamic Pricing strategy is realized in border effect, optimization.Pricing method proposed by the invention is divided into three levels, respectively
For preliminary pricing strategy, feature based screening carry out weight optimization pricing strategy and based on machine learning modeling test into
The pricing strategy of row weight optimization.
The technical solution adopted in the present invention is:
A kind of Dynamic Pricing algorithm based on data mining under shared economy, comprised the following steps:
Step 1, using preliminary pricing strategy, comprise the following steps:
Preliminary pricing strategy can have polytype, and relatively more flexible, such as supplying party's price, personalized subdivision price, bat
Sell the polytypes such as price.
Most service providers possess the data such as service end, transaction, customer information, are provided simultaneously with the history same period and current production
The data such as pricing information, the price information data of service end, transaction, customer information, the history same period and current production is received
Collection;Based on data mining technology, it is possible to achieve carry out client's layering based on features such as access module, purchasing model, Habit Preferences;
Reference pricing is carried out based on the linked character between price and demand, between merchandising;Based on commodity concern number, sale
The time series data of amount, stock etc. are predicted price.
There is the equity problem of personalized point of group's price, the burst period of time series forecasting price and determine in preliminary pricing strategy
Valency problem not in time.Accordingly, it would be desirable to carry out mixing auxiliary based on further pricing strategy.
Step 2, the weight optimization of feature based screening
In the characteristic extraction step of data mining, it would be desirable to by information integration as much as possible into tag file, because
This initial characteristicses collection includes all selected features.But simultaneously the feature in not all feature set all has representativeness well, and
Excessive feature can make information redundancy, make model training over-fitting, the judgment of model can be reduced on the contrary, therefore extract suitable
Feature then turns into the important step for improving modelling effect.
Have numerous Feature Selection algorithms can for evaluate different scenes under feature quality, such as principal component analysis
(principal component analysis, PCA), minimal redundancy maximal correlation algorithm (minimal-redundancy-
Maximal-relevance, mRMR), variance analysis (analysis of variance, ANOVA), bi-distribution
(binomial distribution, BD) etc., the present invention is main to realize Feature Selection using F-score methods.
Training feature vector X, k=1,2 are given, then the F-score of ith feature is represented by
HereWithRespectively feature average value of the ith feature in whole feature set and k-th of data set.Generation
The ith feature value of j-th strip sequence, N in k-th of data set of tablekIt is the protein sequence bar number in k-th of data set.
It will be seen that in F-score definition, molecule represents judgement index of the current signature between different pieces of information collection,
Denominator then represents its judgement index between each collection internal sequence.It can be seen that, molecule is bigger, and it is more accurate that classification differentiates;Denominator is smaller,
Classification differentiates that error rate is smaller.That is, F-score values are bigger, current signature more has judgement index, therefore this score value is
Can as Feature Selection Appreciation gist.
All properties based on product, and market instant demand, sample characteristics is converted into, in the feelings of a large amount of feature sets
Under condition, based in real data, whether it is purchased or selected, to carry out sample classification, the training set modeled as the later stage.
Feature Selection is carried out by F-score methods, the high feature of score is found out, and is arranged in descending order,
Influenceed as on whether ordering according to weight to light feature.
By the analysis of historical data, all characteristic informations higher than threshold value are all considered in model, with Dynamic Pricing plan
Slightly produce related.
Step 3, the weight optimization based on machine learning
SVMs is a kind of machine learning method based on statistical theory, and it is integrated with theories of learning VC dimensions
(Vapnik-Chervonenkis Dimension) and Structural risk minization principle, because being shown in terms for the treatment of classification recurrence
Color, is widely used.Its general principle is that input vector is mapped to the space of a higher-dimension by introducing kernel function, in height
Construct linear decision function in dimension space to realize recurrence, so as to solve Nonlinear Classification and regression problem well.
It is assumed here that sample set isHere l is sample number, and y is sample mesh
Mark, d is input data dimension.Then in higher dimensional space, optimal function is represented by:
Wherein αiFor the Lagrange factors, b is the threshold value of classification, and K (xi, xj) be kernel function, that is, realize nonlinear transformation
Interior Product function.
Generally there is Product function in four kinds to can select in nonlinear transformation in SVMs --- Product function in linear is more
Product function in Product function and radial direction base in Product function in formula, sigmoid.Wherein with Product function in multinomial, sigmoid inner products
Product function is studied in function and radial direction base and using at most, formula is as follows:
2) the interior Product function of polynomial form:
K (x, xi)=[(xxi)+1]q (3)
Parameter q is preassigned by user, and what is produced here is the multinomial grader of a q rank.
2) Product function in sigmoid:
K (x, xi)=tanh [v (xxi)+c] (4)
Wherein v > 0, c < 0.Using this interior Product function, then what is obtained is exactly double-deck Perceptrons Network,
Its weights and concealed nodes number are automatically determined by algorithm.
3) Product function in radial direction base:
Using above-mentioned interior Product function, obtained SVMs is that one kind is different from conventional radial basic function (RBF) method
Grader, center one supporting vector of correspondence of each of which interior Product function, wherein parameter σ is bigger, Product function in radial direction base
Extrapolability it is weaker.
Using Support vector regression (Support Vector Regression, SVR), pass through the feature to feature set
It is adjusted, is trained based on nearest and the same period real trade data, so as to builds the function of each factor and in good time price
Relation, realizes price expectation.
To keep the ageing of model, price is prepared by dynamic feature extraction and data every time, carries out dynamic modeling.
And based on optimal models is built, carry out the application and price expectation of model.
Meanwhile, by as one of feature, carrying out the classification prediction based on SVMs, realizing and strike a bargain pricing information
Whether successfully predict and probability output.System provides flexible price, provides enough frees degree for supplying party and removes conclusion valency
Lattice, and the probability that be able to can be sold based on calculation of price product, are reminded as feedback.
On the basis of the pricing data succeeded, system carries out different productions by way of the attributive character of volatility one
The weight adjustment of product attribute, so as to realize that the Automatic parameter based on attribute is set, realizes Dynamic Pricing.
It is noted that during whole modeling, demand percentage is as one of important references value, with other features
The influence that factor is produced to price is non-equivalence, is embodied as the supply of demand and has absolute influence status.Therefore, it is special
Levy in screening process, demand ratio should not be used as eliminating one of feature.It is opposite, it is necessary to carry out the weight amplification of correspondence.
I.e. in feature set normalization step, it need to give demand bigger feature spacing.
The beneficial effects of the invention are as follows the Dynamic Pricing based on data mining under a kind of shared economy disclosed by the invention
Algorithm, is used by the auxiliary of multiple pricing strategy, realizes the optimization and application of Dynamic Pricing algorithm, solves algorithm application single
First, factor intervention, ageing low problem are difficult to.
The present invention is described in further detail below in conjunction with the accompanying drawings.
Brief description of the drawings
Fig. 1 is the data mining implementation process schematic diagram of the specific embodiment of the invention;
Fig. 2 is SVMs schematic diagram.
Fig. 3 shares the Dynamic Pricing policy framework of the Dynamic Pricing algorithm based on data mining under economy for the present invention
Figure.
Embodiment
In order to deepen the understanding of the present invention, make further details of theory to the present invention with reference to the accompanying drawings and examples
It is bright.The following examples are only intended to illustrate the technical solution of the present invention more clearly, and the guarantor of the present invention can not be limited with this
Protect scope.
A kind of Dynamic Pricing algorithm based on data mining under shared economy, comprised the following steps:
Step 1, using preliminary pricing strategy, comprise the following steps:
Preliminary pricing strategy can have polytype, and relatively more flexible, such as supplying party's price, personalized subdivision price, bat
Sell the polytypes such as price.
Most service providers possess the data such as service end, transaction, customer information, are provided simultaneously with the history same period and current production
The data such as pricing information.Based on data mining technology, it is possible to achieve based on features such as access module, purchasing model, Habit Preferences
Carry out client's layering;Reference pricing is carried out based on the linked character between price and demand, between merchandising;Closed based on commodity
The time series data of note number, sales volume, stock etc. are predicted price etc..
But preliminary pricing strategy is while the problem of having certain, the equity problem of such as personalized point group's price, time sequence
Row predict the burst period price fixed a price problem etc. not in time.Accordingly, it would be desirable to which it is auxiliary to carry out mixing based on further pricing strategy
Help.
Step 2, the weight optimization of feature based screening
In the characteristic extraction step of data mining, it would be desirable to by information integration as much as possible into tag file, because
This initial characteristicses collection includes all selected features.But simultaneously the feature in not all feature set all has representativeness well, and
Excessive feature can make information redundancy, make model training over-fitting, the judgment of model can be reduced on the contrary, therefore extract suitable
Feature then turns into the important step for improving modelling effect.
Have numerous Feature Selection algorithms can for evaluate different scenes under feature quality, such as principal component analysis
(principal component analysis, PCA), minimal redundancy maximal correlation algorithm (minimal-redundancy-
Maximal-relevance, mRMR), variance analysis (analysis of variance, ANOVA), bi-distribution
(binomial distribution, BD) etc., the present invention is main to realize Feature Selection using F-score methods.
Training feature vector X, k=1,2 are given, then the F-score of ith feature is represented by
HereWithRespectively feature average value of the ith feature in whole feature set and k-th of data set.Generation
The ith feature value of j-th strip sequence, N in k-th of data set of tablekIt is the protein sequence bar number in k-th of data set.
It will be seen that in F-score definition, molecule represents judgement index of the current signature between different pieces of information collection,
Denominator then represents its judgement index between each collection internal sequence.It can be seen that, molecule is bigger, and it is more accurate that classification differentiates;Denominator is smaller,
Classification differentiates that error rate is smaller.That is, F-score values are bigger, current signature more has judgement index, therefore this score value is
Can as Feature Selection Appreciation gist.
All properties based on product, and market instant demand, sample characteristics is converted into, in the feelings of a large amount of feature sets
Under condition, based in real data, whether it is purchased or selected, to carry out sample classification, the training set modeled as the later stage.
Feature Selection is carried out by F-score methods, the high feature of score is found out, and is arranged in descending order,
Influenceed as on whether ordering according to weight to light feature.
By the analysis of historical data, all characteristic informations higher than threshold value are all considered in model, with Dynamic Pricing plan
Slightly produce related.
Step 3, the weight optimization based on machine learning
SVMs is a kind of machine learning method based on statistical theory, and it is integrated with theories of learning VC dimensions
(Vapnik-Chervonenkis Dimension) and Structural risk minization principle, because being shown in terms for the treatment of classification recurrence
Color, is widely used.Its general principle is that input vector is mapped to the space of a higher-dimension by introducing kernel function, in height
Construct linear decision function in dimension space to realize recurrence, so as to solve Nonlinear Classification and regression problem well.
It is assumed here that sample set isHere l is sample number, and y is sample mesh
Mark, d is input data dimension.Then in higher dimensional space, optimal function is represented by:
Wherein αiFor the Lagrange factors, b is the threshold value of classification, and K (xi, xj) be kernel function, that is, realize nonlinear transformation
Interior Product function.
Generally there is Product function in four kinds to can select in nonlinear transformation in SVMs --- Product function in linear is more
Product function in Product function and radial direction base in Product function in formula, sigmoid.Wherein with Product function in multinomial, sigmoid inner products
Product function is studied in function and radial direction base and using at most, formula is as follows:
3) the interior Product function of polynomial form:
K (x, xi)=[(xxi)+1]q (3)
Parameter q is preassigned by user, and what is produced here is the multinomial grader of a q rank.
2) Product function in sigmoid:
K (x, xi)=tanh [v (xxi)+c]
(4)
Wherein v > 0, c < 0.Using this interior Product function, then what is obtained is exactly double-deck Perceptrons Network,
Its weights and concealed nodes number are automatically determined by algorithm.
3) Product function in radial direction base:
Using this interior Product function, obtained SVMs is that one kind is different from conventional radial basic function (RBF) method
Grader, center one supporting vector of correspondence of each of which interior Product function, wherein parameter σ is bigger, Product function in radial direction base
Extrapolability it is weaker.
Using Support vector regression (Support Vector Regression, SVR), pass through the feature to feature set
It is adjusted, is trained based on nearest and the same period real trade data, so as to builds the function of each factor and in good time price
Relation, realizes price expectation.
To keep the ageing of model, price is prepared by dynamic feature extraction and data every time, carries out dynamic modeling.
And based on optimal models is built, carry out the application and price expectation of model.
Meanwhile, by as one of feature, carrying out the classification prediction based on SVMs, realizing and strike a bargain pricing information
Whether successfully predict and probability output.System provides flexible price, provides enough frees degree for supplying party and removes conclusion valency
Lattice, and the probability that be able to can be sold based on calculation of price product, are reminded as feedback.
On the basis of the pricing data succeeded, system carries out different productions by way of the attributive character of volatility one
The weight adjustment of product attribute, so as to realize that the Automatic parameter based on attribute is set, realizes Dynamic Pricing.
It is noted that during whole modeling, demand percentage is as one of important references value, with other features
The influence that factor is produced to price is non-equivalence, is embodied as the supply of demand and has absolute influence status.Therefore, it is special
Levy in screening process, demand ratio should not be used as eliminating one of feature.It is opposite, it is necessary to carry out the weight amplification of correspondence.
I.e. in feature set normalization step, it need to give demand bigger feature spacing.
The specific embodiment of the present invention,
So that Economic Application platform is shared in the tourism of a mobile terminal as an example, the platform provides money for personal travel entrepreneur
Gold, technology, marketing and transaction are brought together and helped, and reduce foundation threshold.Personal shared slack resources, technology, ability are advocated, volume is created
Outer income, provides the user social platform and differentiation based on tourism, personalized travelling products service.The platform using social activity as
Entrance, holds the following important traffic ingress of China Mobile Internet, the current project is on Zhangjiajie, Zhenjiang, Sanya and other places
Landing.
The Dynamic Pricing of the platform is the mode of the species auction to move App as social communication media, deployed
Carry out.Auction mechanism is:The starting price that seller is set, low price of starting auction is 10 yuan.Seller can be with self-defined simultaneously
Price markup amplitude, can also use system active agency to raise the price;The former be it is changeless, and the price markup amplitude of the latter be with work as
The increase of preceding bid amount and increased, but correspondence fixed " present price ", its price markup amplitude is also fixed.Buyer can
So that in the case of the price markup amplitude for surmounting previous personal bid record and being set not less than seller, oneself is bid or adopted
With the mode of agency's bid.Agency's bid refers to the high price that system is inputted according to buyer, when there is other buyers bid, from
It is dynamic to be bid upwards with the small price markup amount of money, to maintain the position of the high bidder of buyer, until the high bid of buyer is surpassed by other buyers
Untill crossing.In the case where system active agency is raised the price, price range is indirectly to determine there is purchase by consumer demand
The people for buying wish is more, and number of bids is more, so as to have led to higher " present price ", thus improves and adds therewith
Valency amplitude.Determined that is, the initial prices auctioned on App are sellers according to market information and the income of cost one
, price markup amplitude is changeless or only determined indirectly by demand, and is all general to all commodity.This
Plant pricing mechanism very simple, but defect is also a lot.Under existing auction mechanism, seller is difficult to collect comprehensive client and competition
Opponent's information, it is impossible to carry out depth excavation to it, also cannot carry out differential price according to client characteristics, also not accomplish to not
Same commodity make the adjustment of timely, appropriate price markup amplitude according to time change, demand, inventories, so as to can not reach
The purpose that the big satisfaction of client and seller's income are changed greatly.And the introducing of the ecommerce dynamic pricing models based on data mining then may be used
To solve this problem.System can build the task that data mining platform completes data Layer and analysis layer in model, that is, pass through
The data of both parties and exterior market are collected, data warehouse is set up, carries out data mining and form knowing for auxiliary pricing decision
Know storehouse.Then seller (including enterprises and individuals) relies on the platform, and combines self information grasp situation and technical strength progress
Dynamic Pricing decision-making.
Under C2C patterns, seller is personal, it is impossible to complete huge data collection until the task of data mining, then
Data mining platform based on dynamic bid can just help through these information buildings, then with unified price mould
Type and algorithm make corresponding pricing decision to different types of commodity according to various Dynamic Pricing strategies, including appropriate initial
The determination of price and the adjustment of price markup amplitude, the corresponding pricing strategy of different choice that each seller can be according to institute's merchandising.
Help seller to be finely divided client in addition, the network platform can set up unified standard, and different is implemented to different clients
Pricing strategy, especially auction the high loyalty client of saliency override mechanism, reach preservation client, increase customer satisfaction degree
Purpose.Certainly, to strive for new buyer, certain preferential measure can also be taken, increase client source is extended volume growth.
It is noted that embodiment described above is to the illustrative and not limiting of technical solution of the present invention, affiliated technology neck
The equivalent substitution of domain those of ordinary skill or other modifications made according to prior art, as long as not exceeding the technology of the present invention side
The thinking and scope of case, should be included within interest field of the presently claimed invention.
Claims (4)
1. the Dynamic Pricing algorithm based on data mining under a kind of shared economy, it is characterised in that comprised the following steps:
Step 1, using preliminary pricing strategy, comprise the following steps:
S1, progress Data Collection;
S2, based on data mining technology, realize client be layered;
S3, progress Reference pricing;
S4, based on data it is predicted price;
Step 2, the weight optimization of feature based screening
Feature Selection is realized using F-score methods, training feature vector X, k=1,2 is given, then the F-score of ith feature
It is expressed as
<mrow>
<mi>F</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mn>2</mn>
</msubsup>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mover>
<mi>f</mi>
<mo>&OverBar;</mo>
</mover>
<mi>i</mi>
<mi>k</mi>
</msubsup>
<mo>-</mo>
<msub>
<mover>
<mi>f</mi>
<mo>&OverBar;</mo>
</mover>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mn>2</mn>
</msubsup>
<mfrac>
<mn>1</mn>
<mrow>
<msub>
<mi>N</mi>
<mi>k</mi>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
</mfrac>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>N</mi>
<mi>k</mi>
</msub>
</msubsup>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>f</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mi>k</mi>
</msubsup>
<mo>-</mo>
<msubsup>
<mover>
<mi>f</mi>
<mo>&OverBar;</mo>
</mover>
<mi>i</mi>
<mi>k</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,WithRespectively feature average value of the ith feature in whole feature set and k-th of data set,Represent
The ith feature value of j-th strip sequence, N in k-th of data setkIt is the protein sequence bar number in k-th of data set;
In F-score definition, molecule represents judgement index of the current signature between different pieces of information collection, and denominator then represents it in each collection
Judgement index between internal sequence;It can be seen that, molecule is bigger, and it is more accurate that classification differentiates;Denominator is smaller, and classification differentiates that error rate is got over
It is small.That is, F-score values are bigger, current signature more has judgement index, therefore this score value can be used as Feature Selection
Appreciation gist;
Step 3, the weight optimization based on machine learning
Weight optimization is carried out using SVMs, kernel function is introduced, input vector is mapped to the space of a higher-dimension, in height
Linear decision function is constructed in dimension space to realize recurrence;
Sample set is (xi, yi), i=1 ..., n,{ l ,-l }, l is sample number, and y is sample object, and d is input number
According to dimension;Then in higher dimensional space, optimal function is expressed as:
Wherein αiFor the Lagrange factors, b is the threshold value of classification, and K (xi, xj) be kernel function, that is, realize the interior of nonlinear transformation
Product function.
2. the Dynamic Pricing algorithm based on data mining under a kind of shared economy according to claim 1, its feature exists
In:In step 2, comprise the following steps:
S1, all properties based on product, and market instant demand, be converted into sample characteristics, feature set formed, a large amount of
In the case of feature set, sample classification, the training set modeled as the later stage are carried out based on real data;
S2, Feature Selection carried out by F-score methods, find out the high feature of score, and arranged in descending order, work
To influence according to weight to light feature;
S3, the analysis by historical data, all characteristic informations higher than threshold value are all considered in model, with Dynamic Pricing strategy
Produce related.
3. the Dynamic Pricing algorithm based on data mining under a kind of shared economy according to claim 1, its feature exists
In:In step 3, nonlinear transformation is used for using interior Product function in SVMs, includes Product function in multinomial,
Product function in Product function and radial direction base in sigmoid, formula is as follows:
1) the interior Product function of polynomial form:
K (x, xi)=[(xxi)+1]q
(3)
Parameter q is preassigned by user, and what is produced here is the multinomial grader of a q rank;
2) Product function in sigmoid:
K (x, xi)=tanh [v (xxi)+c]
(4)
Wherein v > 0, c < 0.Using this interior Product function, then what is obtained is exactly double-deck Perceptrons Network, and it is weighed
Value and concealed nodes number are automatically determined by algorithm;
3) Product function in radial direction base:
<mrow>
<mi>K</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>exp</mi>
<mo>{</mo>
<mo>-</mo>
<mfrac>
<mrow>
<mo>|</mo>
<mo>|</mo>
<mi>x</mi>
<mo>-</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<mn>2</mn>
<msup>
<mi>s</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mo>}</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
Using above-mentioned interior Product function, obtained SVMs is point that one kind is different from conventional radial basic function (RBF) method
Class device, center one supporting vector of correspondence of each of which interior Product function, wherein parameter σ is bigger, and Product function is outer in radial direction base
Push away ability weaker.
4. the Dynamic Pricing algorithm based on data mining under a kind of shared economy according to claim 3, its feature exists
In:In step 3, comprise the following steps:
S1, using Support vector regression, be adjusted by the feature to feature set, based on nearest and the same period real trade
Data are trained, so as to build the functional relation of each factor and in good time price, realize price expectation;
S2, to keep the ageing of model, price is prepared by dynamic feature extraction and data every time, progress dynamic modeling,
And based on optimal models is built, carry out the application and price expectation of model;
S3, by the way that pricing information as one of feature, to be carried out to the classification prediction based on SVMs, realize prediction and probability
Output;System provides flexible price, and providing enough frees degree for supplying party goes to conclude price, and based on calculation of price product
The probability that can be sold, is reminded as feedback;
S4, on the basis of the pricing data succeeded, system by way of the attributive character of volatility one, carry out different product
The weight adjustment of attribute, so as to realize that the Automatic parameter based on attribute is set, realizes Dynamic Pricing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710342702.1A CN107194722A (en) | 2017-05-15 | 2017-05-15 | A kind of Dynamic Pricing algorithm based on data mining under shared economy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710342702.1A CN107194722A (en) | 2017-05-15 | 2017-05-15 | A kind of Dynamic Pricing algorithm based on data mining under shared economy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107194722A true CN107194722A (en) | 2017-09-22 |
Family
ID=59872756
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710342702.1A Pending CN107194722A (en) | 2017-05-15 | 2017-05-15 | A kind of Dynamic Pricing algorithm based on data mining under shared economy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107194722A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107832864A (en) * | 2017-12-19 | 2018-03-23 | 云南大学 | The net bidded under environment about special train distribution and pricing method |
CN109377260A (en) * | 2018-09-14 | 2019-02-22 | 江阴逐日信息科技有限公司 | User behavior analysis system towards apparel industry |
WO2019105235A1 (en) * | 2017-11-30 | 2019-06-06 | 北京京东尚科信息技术有限公司 | Pricing method and device, and computer-readable storage medium |
CN109978619A (en) * | 2019-03-25 | 2019-07-05 | 携程旅游网络技术(上海)有限公司 | Method, system, equipment and the medium of air ticket pricing Policy Filtering |
CN110570288A (en) * | 2018-06-06 | 2019-12-13 | 阿里巴巴集团控股有限公司 | Auction system, price quotation device, order generation method and order processing method |
CN110852768A (en) * | 2018-08-21 | 2020-02-28 | 北京京东尚科信息技术有限公司 | Dynamic pricing method and system, equipment and storage medium |
CN111832600A (en) * | 2019-12-31 | 2020-10-27 | 北京骑胜科技有限公司 | Data processing method and device, electronic equipment and computer readable storage medium |
CN112950257A (en) * | 2021-02-03 | 2021-06-11 | 北京金山云网络技术有限公司 | Data use pricing method and device, computer equipment and storage medium |
CN114549142A (en) * | 2022-03-17 | 2022-05-27 | 太平洋国际拍卖有限公司 | Online bidding auction information processing method and system |
-
2017
- 2017-05-15 CN CN201710342702.1A patent/CN107194722A/en active Pending
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019105235A1 (en) * | 2017-11-30 | 2019-06-06 | 北京京东尚科信息技术有限公司 | Pricing method and device, and computer-readable storage medium |
US11669875B2 (en) | 2017-11-30 | 2023-06-06 | Beijing Jingdong Shangke Information Technology Co., Ltd. | Pricing method and device, and non-transient computer-readable storage medium |
CN107832864A (en) * | 2017-12-19 | 2018-03-23 | 云南大学 | The net bidded under environment about special train distribution and pricing method |
CN107832864B (en) * | 2017-12-19 | 2021-05-14 | 云南大学 | Network contract special car distribution and pricing method under bidding environment |
CN110570288B (en) * | 2018-06-06 | 2023-07-18 | 阿里巴巴华北技术有限公司 | Auction system, bidder, order generation method, and order processing method |
CN110570288A (en) * | 2018-06-06 | 2019-12-13 | 阿里巴巴集团控股有限公司 | Auction system, price quotation device, order generation method and order processing method |
CN110852768A (en) * | 2018-08-21 | 2020-02-28 | 北京京东尚科信息技术有限公司 | Dynamic pricing method and system, equipment and storage medium |
CN109377260A (en) * | 2018-09-14 | 2019-02-22 | 江阴逐日信息科技有限公司 | User behavior analysis system towards apparel industry |
CN109978619B (en) * | 2019-03-25 | 2021-07-06 | 携程旅游网络技术(上海)有限公司 | Method, system, equipment and medium for screening air ticket pricing strategy |
CN109978619A (en) * | 2019-03-25 | 2019-07-05 | 携程旅游网络技术(上海)有限公司 | Method, system, equipment and the medium of air ticket pricing Policy Filtering |
CN111832600A (en) * | 2019-12-31 | 2020-10-27 | 北京骑胜科技有限公司 | Data processing method and device, electronic equipment and computer readable storage medium |
CN112950257A (en) * | 2021-02-03 | 2021-06-11 | 北京金山云网络技术有限公司 | Data use pricing method and device, computer equipment and storage medium |
CN112950257B (en) * | 2021-02-03 | 2024-02-09 | 北京金山云网络技术有限公司 | Data use price calculating method, device, computer equipment and storage medium |
CN114549142A (en) * | 2022-03-17 | 2022-05-27 | 太平洋国际拍卖有限公司 | Online bidding auction information processing method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107194722A (en) | A kind of Dynamic Pricing algorithm based on data mining under shared economy | |
Ahmadi et al. | New efficient hybrid candlestick technical analysis model for stock market timing on the basis of the Support Vector Machine and Heuristic Algorithms of Imperialist Competition and Genetic | |
CN109472646A (en) | A kind of financial product recommended method and device | |
CN109034915A (en) | A kind of artificial intelligent type e-commerce system using digital asset or integral as the media of exchange | |
Madaan et al. | Price forecasting & anomaly detection for agricultural commodities in India | |
CN110675029A (en) | Dynamic management and control method and device for commercial tenant, server and readable storage medium | |
CN108921602A (en) | A kind of user's buying behavior prediction technique based on integrated neural network | |
CN107851283A (en) | The segmentation and layering of the comprehensive method of investment combination of investment securities | |
CN115860880A (en) | Personalized commodity recommendation method and system based on multilayer heterogeneous graph convolution model | |
Brooks et al. | Price leadership and information transmission in Australian water allocation markets | |
Goh | How agritech is transforming traditional agriculture in emerging markets | |
Venkateswararao et al. | LT-SMF: long term stock market price trend prediction using optimal hybrid machine learning technique | |
Liao et al. | Multi-behavior RFM model based on improved SOM neural network algorithm for customer segmentation | |
Yuping et al. | New methods of customer segmentation and individual credit evaluation based on machine learning | |
Estmann et al. | Merchandise export diversification strategy for Tanzania: Promoting inclusive growth, economic complexity and structural change | |
Gupta et al. | Scope of E-commerce in agri-business in India: an overview | |
CN108830680A (en) | Personalized recommendation method, system and storage medium based on discrete disassembler | |
Huang et al. | A genetic-based stock selection model using investor sentiment indicators | |
Nassibi et al. | Demand forecasting models for food industry by utilizing machine learning approaches | |
CN107194748A (en) | Product Sale Strategy optimizes system | |
Breeden | How heterogeneous wine lots result in a price discount relative to homogeneous lots | |
Rusmee et al. | Predicting System for the Behavior of Consumer Buying Personal Car Decision by Using SMO | |
De la Torre et al. | Electricity price forecast in wholesale markets using conformal prediction: Case study in Mexico | |
Pavlíková et al. | The values and attitudes of young people | |
Zhao et al. | Application of fractal dimension of fractional Brownian motion to supply chain financing and operational comprehensive decision-making |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170922 |