CN104504578A - Active children shoe recommending method based on preference correlation in online shopping environment - Google Patents

Abstract

The invention provides an active children shoe recommending method based on preference correlation in an online shopping environment. The method consists of a model training stage and an online real-time recommending stage. The method comprises the following steps: at the model training stage, selecting a plurality of present popular children shoe products for extracting children shoe appearance elements, and performing clustering analysis on the appearance elements through a structural entropy clustering method to obtain a mainstream appearance type group; establishing a correlation model of customer individual characteristics and children shoe appearance types through multiple logistic methods, and performing optimization solution on correlation model parameters through network survey sample data; and at the online real-time recommending stage, automatically finding a children shoe appearance type which is most matched with customer characteristics according to the correlation model at the last stage by acquiring individual characteristic information of a customer in order to realize an active recommending function. Through adoption of the active children shoe recommending method, the effectiveness of the active recommending function in the online shopping environment can be enhanced, and technical support is provided for the enhancement of customer attention.

Description

Based on the children's footwear of preference association initiatively recommend method under a kind of net purchase environment

Technical field

The invention belongs to children's footwear field, the children's footwear initiatively recommend method associated based on preference under being specifically related to a kind of net purchase environment.

Background technology

China has the maximum country of children's number at present in the world, 6th national census data display, China children of less than 14 years old now about 300,011,300, make it to constitute a consumer goods market had a high potential, abnormal huge children's footwear product consumption colony also result in showing great attention to of manufacturer.Along with socioeconomic development, the consumption habit of the people is bought by solid shop/brick and mortar store gradually and is changed Online Shopping into, and how under net purchase environment, to catch consumer, be numerous manufacturers problem demanding prompt solutions.At present, main brand manufacturer has been difficult to distinguish relative superiority or inferiority in children's footwear product pattern, quality etc., buys the stage (Online Shopping environment) just carry out a kind of good method that personalized initiatively recommendation is attraction client eyeball from client.But the research at present for children's footwear product active recommend method does not almost have.

Summary of the invention

The present invention is directed to above-mentioned the deficiencies in the prior art, the children's footwear initiatively recommend method associated based on preference under providing a kind of net purchase environment, by setting up the correlation model between children's footwear outward appearance type and client's personal feature, the children's footwear outward appearance type just its preference being recommended higher to it when client carries out net purchase, thus the probability improving client's concern.

The technical solution used in the present invention is as follows:

Based on the children's footwear of a preference association initiatively recommend method under net purchase environment, comprise the following steps:

(1) collect children's footwear sample, determine the interval of L outward appearance inscape and each outward appearance inscape; Connection attribute key element and semantic attribute key element is comprised in outward appearance inscape;

(2) adopt structure entropy clustering method to carry out cluster analysis to each outward appearance inscape, obtain M main flow outward appearance type of children's footwear product;

Specifically comprise the steps:

(2.1) using the children's footwear sample in step (1) as sample data cluster sample, be X={X by sample data cluster sample labeling ₁, X ₂..., X _n, X _i={ x _i1, x _i2..., x _iL, N is sample size, and L is form factor number; Initialization cluster number K=0, cluster set R=Φ (i.e. empty set), metric threshold B=0.5;

(2.2) according to following formula estimation attribute weight parameter ω,

$\mathrm{\ω}=\frac{E_{\mathrm{ij}}^{c-\mathrm{mean}}-E_{\mathrm{it}}^{y-\mathrm{mean}}}{(E_{\mathrm{ij}}^{c-\min }-E_{\mathrm{it}}^{y-\min })+(E_{\mathrm{ij}}^{c-\mathrm{mean}}-E_{\mathrm{it}}^{y-\mathrm{mean}})}$

Wherein:

A, b represent the number of connection attribute key element and semantic attribute key element respectively, a+b=L;

J, t represent the subscript sequence number of connection attribute key element and semantic attribute key element in outward appearance inscape respectively;

with be respectively sample X _iconnection attribute key element and the minimum value of entropy of semantic attribute key element; Namely $E_{\mathrm{ij}}^{c-\min }=\underset{1\≤j\≤a}{\min }{E}_{\mathrm{ij}}^c$ With $E_{\mathrm{it}}^{y-\min }=\underset{1\≤t\≤b}{\min }{E}_{\mathrm{it}}^y;$

with be respectively sample X _iconnection attribute key element and the mean value of entropy of semantic attribute key element, namely $E_{\mathrm{ij}}^{c-\mathrm{mean}}=\frac{1}{a}\underset{j=1}{\overset{a}{\mathrm{\Σ}}}{E}_{\mathrm{ij}}^c,E_{\mathrm{it}}^{y-\mathrm{mean}}=\frac{1}{b}\underset{t=1}{\overset{b}{\mathrm{\Σ}}}{E}_{\mathrm{it}}^y;$

(2.3) the entropy E of sample is calculated according to following formula _i;

$E_i={\mathrm{\ωE}}_{\mathrm{ij}}^c+(1-\mathrm{\ω})E_{\mathrm{it}}^y,i=\mathrm{1,2},...,N,j=1,...,a,t=1,...,b;$

(2.4) k=k+1 is made;

(2.5) look in X ' and make v _k={ X _i| E (X _i)=E _min, it can be used as kth class cluster centre;

(2.6) satisfied tolerance threshold condition D (X _i, v _k) all samples of < B are included in a nearest class, are about to the D (X that satisfies condition _i, v _k) x of < B _ibe included into V _kin class, V _k={ X _i| D (X _i, v _k) < B};

(2.7) R=R ∪ V is made _k, X '=X – R;

(2.8) judge whether X ' is empty, if so, then makes M=k, obtains main flow outward appearance type V={V ₁, V ₂..., V _m, M is outward appearance number of types; Otherwise, go back to step (2.4);

(3) based on the main flow outward appearance type of the children's footwear product determined in step (2), gather client's personal feature sample data collection, and it can be used as training sample data collection Z={ (Y ₁, C ₁), (Y ₂, C ₂) ..., (Y _n, C _n); Wherein, Y _i=(y _i1, y _i2..., y _iQ) be Q the feature value of i-th client; C _i={ c _i1, c _i2..., c _iMbe the predilection grade vector of i-th client in each main flow outward appearance type;

(4) for training sample data collection, carry out client's personal feature and other the multiclass logistic association analysis of children's footwear outer appearnce, obtain the mapping weights W that each outward appearance inscape corresponds to client's personal feature;

Specifically comprise the steps:

(4.1) the mapping weight making main flow outward appearance type correspond to customer characteristic is be m class main flow outward appearance type mapping weight (m=1,2 ..., M), represent when customer characteristic integrates as Z _i=(z _i1, z _i2..., z _iQ) time, the outward appearance type of output is the probability of t class; According to logistic Function feature, be constructed as follows mathematic optimal model:

(4.2) introduce Lagrange multiplier to above formula, structure (Q+1) * M rank system of equations, solves the mapping weight that this system of equations can obtain main flow outward appearance type it is the mapping weight of m class main flow outward appearance type;

(5) the mapping weights W utilizing step (4) to obtain, calculates the preference of client to each main flow outward appearance type, and sample preference being greater than the main flow outward appearance type of 5 shows recommends client;

Wherein, calculate the method for client to the preference of each main flow outward appearance type and be the probability calculating this main flow outward appearance type of customers' preferences z is the personal feature data set of client to be measured;

P (c _t=1|Z, W) what obtain is the probability of customers' preferences t main flow outward appearance type, and probable value is converted into client's marking: c _t=P (c _t=1|Z, W) s _max, s _maxfor the maximal value of outward appearance classification preference scoring.

The present invention has following beneficial effect:

1, at present recommend the recommend method adopted based on historical viewings record for the active of children's footwear product, the method browses only according to client the recommendation that record carries out similar products, and recommendation effect is not good more.The present invention proposes the children's footwear initiatively recommend method based on customers' preferences degree, to the higher children's footwear outward appearance type of its preference of customer recommendation, consumer awareness can be improve.

2, current, the association study between children's footwear product appearance type and customers' preferences degree is not almost had.The present invention establishes the related degree model between children's footwear product appearance type and customer characteristic, first by cluster analysis condensed go out the main flow outward appearance type of children's footwear product, then the logistic correlation model of children's footwear outward appearance type and client's personal feature is set up, the outward appearance type of customers' preferences can not only be identified, the preference probability of client to various outward appearance type can also be provided, thus realize effective initiatively recommendation.

Accompanying drawing illustrates:

Fig. 1 is main flow outward appearance type and the representative sample figure thereof of canvas rubber shoe.

Embodiment

The children's footwear initiatively recommend method associated based on preference under the invention provides a kind of net purchase environment, comprises the following steps:

(1) collect children's footwear sample, determine the interval of L outward appearance inscape and each outward appearance inscape;

As shown in table 1, it is outward appearance inscape and the span thereof of canvas rubber shoe in children's footwear product.The outward appearance inscape determined in the present embodiment has 9, i.e. L=9, comprises upper area ratio shared by toe lasting, tie-down mode and with or without footwear ear etc.Comprise connection attribute and semantic attribute in outward appearance inscape, in the present embodiment, key element 1-5 belongs to connection attribute, and 6-9 belongs to semantic attribute.Dissimilar children's footwear product according to the difference of children's footwear sample, and can determine the outward appearance inscape of different number and type.(can select by experience when determining, also can be determined by group decision-making methods such as expert adjudicate methods.)

The main composition key element of table 1 canvas rubber shoe outward appearance and span thereof

(2) adopt structure entropy clustering method to carry out cluster analysis to each outward appearance inscape, obtain several main flow outward appearance types of children's footwear product;

Specifically comprise the steps:

(2.1) using the children's footwear sample in step (1) as sample data cluster sample, be X={X by sample data cluster sample labeling ₁, X ₂..., X _n, X _i={ x _i1, x _i2..., x _iL, N is sample size, and L is form factor number.Initialization cluster number K=0, cluster set R=Φ (i.e. empty set), metric threshold B=0.5;

(2.2) according to following formula estimation attribute weight parameter ω:

$\mathrm{\&omega;}=\frac{E_{\mathrm{ij}}^{c-\mathrm{mean}}-E_{\mathrm{it}}^{y-\mathrm{mean}}}{(E_{\mathrm{ij}}^{c-\min }-E_{\mathrm{it}}^{y-\min })+(E_{\mathrm{ij}}^{c-\mathrm{mean}}-E_{\mathrm{it}}^{y-\mathrm{mean}})}$

Wherein:

A, b represent the number of connection attribute key element and semantic attribute key element respectively, a+b=L; As shown in table 1, in the present embodiment, connection attribute key element and semantic attribute key element number are respectively 5 and 4, i.e. a=5, b=4;

J, t represent the subscript sequence number (sequence number layout is particular about without any, can arrange according to the natural order of key element) of connection attribute key element and semantic attribute key element in outward appearance inscape respectively;

with be respectively sample X _iconnection attribute and the minimum value of entropy of semantic attribute; Namely $E_{\mathrm{ij}}^{c-\min }=\underset{1\&le;j\&le;a}{\min }{E}_{\mathrm{ij}}^c$ With $E_{\mathrm{it}}^{y-\min }=\underset{1\&le;t\&le;b}{\min }{E}_{\mathrm{it}}^y.$

with be respectively sample X _iconnection attribute and the mean value of entropy of semantic attribute, namely $E_{\mathrm{ij}}^{c-\mathrm{mean}}=\frac{1}{a}\underset{j=1}{\overset{a}{\mathrm{\&Sigma;}}}{E}_{\mathrm{ij}}^c,E_{\mathrm{it}}^{y-\mathrm{mean}}=\frac{1}{b}\underset{t=1}{\overset{b}{\mathrm{\&Sigma;}}}{E}_{\mathrm{it}}^y.$

(2.3) the entropy E of sample is calculated according to following formula _i;

$E_i={\mathrm{\&omega;E}}_{\mathrm{ij}}^c+(1-\mathrm{\&omega;})E_{\mathrm{it}}^y,i=\mathrm{1,2},...,N,j=1,...,a,t=1,...,b,$ N is sample size;

(2.4) k=k+1 is made;

(2.5) look in X ' and make v _k={ X _i| E (X _i)=E _min, it can be used as kth class cluster centre;

(2.6) satisfied tolerance threshold condition D (X _i, v _k) all samples of < B are included in a nearest class, are about to the D (X that satisfies condition _i, v _k) x of < B _ibe included into V _kin class, V _k={ X _i| D (X _i, v _k) < B};

(2.7) R=R ∪ V is made _k, X '=X – R;

(2.8) judge whether X ' is empty, and if so, then make M=k, algorithm terminates, and obtains final cluster set (being also called main flow outward appearance type) V={V thus ₁, V ₂..., V _m, M is outward appearance number of types; Otherwise, go back to step (2.4).

According to the outward appearance inscape in table 1 in the present embodiment, after collecting 200 increments bases, adopt structure entropy clustering method to carry out cluster analysis to each outward appearance inscape, obtain 6 kinds of main flow outward appearance types (i.e. M=6) of children's footwear product, as shown in Figure 1.

(3) based on the main flow outward appearance type of the children's footwear product determined in step (2), gather client's personal feature sample data collection, and it can be used as training sample data collection Z={ (Y ₁, C ₁) ..., (Y _n, C _n);

N is sample size, and M is the quantity of main flow outward appearance type;

Y _i=(y _i1, y _i2..., y _iQ) be Q the feature value of i-th client;

C _i={ c _i1, c _i2..., c _iMbe the predilection grade vector of i-th client in each main flow outward appearance type;

Gather client's personal feature sample data collection by the mode of network surveying questionnaire, several personal characteristics to be supplied to testee and to answer for it, and allow testee carry out preference marking to each main flow outward appearance type.

Such as can gather following 12 (Q=12) personal characteristics of client: 1. the sex of consumer (head of a family); 2. the age of consumer (head of a family); 3. the education degree of consumer (head of a family); 4. the degree of understanding of pair current children's footwear information; 5. when buying, pay attention to what part of children's footwear most; 6. the most often buy the place of children's footwear; 7. consumer (head of a family) life area; 8. the average time span buying new footwear; 9. so far through how many pairs of children's footwears; 10. through how many double cloth cover rubber overshoes; 11. buy reason during children's footwear usually; The preferred way of 12. current maintenance children's footwears;

When carrying out preference marking to each main flow outward appearance type, need the implication that predefined score value represents, such as 1 point of representative is not liked very much, and 5 points represent that sensation is general, 9 points of representatives enjoy a lot, thus obtain different client's personal characteristics and the sample data in outward appearance hobby thereof.Afterwards each sample point is expressed as probability vector to the score value that each main flow outward appearance type carries out preference marking, if that is: the predilection grade of i-th client to t kind main flow outward appearance type is s, then c _it=s/s _max, s _maxfor the maximal value of outward appearance classification preference scoring.Such as, if the predilection grade of i-th client to main flow outward appearance type each in Fig. 1 is 1,5,7,3,3,1, then its predilection grade vector can be expressed as: C _i={ 1/9,5/9,7/9,3/9,3/9,1/9}.

(4) for training sample data collection, carry out client's personal feature and other the multiclass logistic association analysis of children's footwear outer appearnce, obtain the mapping weights W that each outward appearance inscape corresponds to client's personal feature;

Specifically comprise the steps:

(4.1) the mapping weight making main flow outward appearance type correspond to customer characteristic is be m class main flow outward appearance type mapping weight (m=1,2 ..., M), represent when customer characteristic integrates as Z _i=(z _i1, z _i2..., z _iQ) time, the outward appearance type of output is t class (i.e. c _it=1) probability.According to logistic Function feature, be constructed as follows mathematic optimal model:

(4.2) introduce Lagrange multiplier to above formula, structure (Q+1) * M rank system of equations, solves the mapping weight that this system of equations can obtain main flow outward appearance type it is the mapping weight of m class main flow outward appearance type.

Such as, the mapping weight that the main flow outward appearance type that the basis of Fig. 1 is calculated by 257 parts of client's personal feature sample datas (i.e. N=257) corresponds to customer characteristic is as shown in table 3, V _ibe i-th kind of main flow outward appearance type, y _ibe i-th customer characteristic.

Other customer characteristic of table 3 outer appearnce maps weight

(5) the mapping weights W utilizing step (4) to obtain, calculates the preference of client to each main flow outward appearance type, and sample preference being greater than the main flow outward appearance type of 5 shows recommends client.

Wherein, calculate the method for client to the preference of each main flow outward appearance type and be the probability calculating this main flow outward appearance type of customers' preferences z is the personal feature data set of client to be measured.

P (c _t=1|Z, W) what obtain is the probability of customers' preferences t main flow outward appearance type, and probable value is converted into client's marking: c _t=P (c _t=1|Z, W) s _max, s _maxfor the maximal value of outward appearance classification preference scoring.Table 4 is the result of calculation (s in this example of customers' preferences degree in the present embodiment _max=9).

Table 4 customers' preferences degree probability calculation result example

Claims (1)

1. under net purchase environment based on a children's footwear initiatively recommend method for preference association, it is characterized in that, comprise the following steps:

(1) collect children's footwear sample, determine the interval of L outward appearance inscape and each outward appearance inscape; Connection attribute key element and semantic attribute key element is comprised in outward appearance inscape;

(2) adopt structure entropy clustering method to carry out cluster analysis to each outward appearance inscape, obtain M main flow outward appearance type of children's footwear product;

Specifically comprise the steps:

(2.1) using the children's footwear sample in step (1) as sample data cluster sample, be X={X by sample data cluster sample labeling ₁, X ₂..., X _n, X _i={ x _i1, x _i2..., x _iL, N is sample size, and L is form factor number; Initialization cluster number K=0, cluster set R=Φ (i.e. empty set), metric threshold B=0.5;

(2.2) according to following formula estimation attribute weight parameter ω,

$\mathrm{\&omega;}=\frac{E_{\mathrm{ij}}^{c-\mathrm{mean}}-E_{\mathrm{it}}^{y-\mathrm{mean}}}{(E_{\mathrm{ij}}^{c-\min }-E_{\mathrm{it}}^{y-\min })+(E_{\mathrm{ij}}^{c-\mathrm{mean}}-E_{\mathrm{it}}^{y-\mathrm{mean}})}$

Wherein:

A, b represent the number of connection attribute key element and semantic attribute key element respectively, a+b=L;

J, t represent the subscript sequence number of connection attribute key element and semantic attribute key element in outward appearance inscape respectively;

with be respectively the minimum value of the connection attribute key element of sample Xi and the entropy of semantic attribute key element; Namely $E_{\mathrm{ij}}^{c-\min }=\underset{1\&le;j\&le;a}{\min }{E}_{\mathrm{ij}}^c$ With $E_{\mathrm{ij}}^{y-\min }=\underset{1\&le;j\&le;b}{\min }{E}_{\mathrm{ij}}^y;$

with be respectively the mean value of the connection attribute key element of sample Xi and the entropy of semantic attribute key element, namely $E_{\mathrm{ij}}^{c-\mathrm{mean}}=\frac{1}{a}\underset{j=1}{\overset{a}{\mathrm{\&Sigma;}}}{E}_{\mathrm{ij}}^c,E_{\mathrm{it}}^{y-\mathrm{mean}}=\frac{1}{b}\underset{t=1}{\overset{b}{\mathrm{\&Sigma;}}}{E}_{\mathrm{it}}^y;$

(2.3) the entropy E of sample is calculated according to following formula _i;

$E_i=\mathrm{\&omega;}{E}_{\mathrm{ij}}^c+(1-\mathrm{\&omega;})E_{\mathrm{it}}^y,i=\mathrm{1,2},...,N,j=1,...,a,t=1,...,b;$

(2.4) k=k+1 is made;

(2.5) look in X ' and make v _k={ X _i| E (X _i)=E _min, it can be used as kth class cluster centre;

(2.6) satisfied tolerance threshold condition D (X _i, v _k) all samples of < B are included in a nearest class, are about to the D (X that satisfies condition _i, v _k) x of < B _ibe included into V _kin class, V _k={ X _i| D (X _i, v _k) < B};

(2.7) R=R ∪ V is made _k, X '=X – R;

(2.8) judge whether X ' is empty, if so, then makes M=k, obtains main flow outward appearance type V={V ₁, V ₂..., V _m, M is outward appearance number of types; Otherwise, go back to step (2.4);

(3) based on the main flow outward appearance type of the children's footwear product determined in step (2), gather client's personal feature sample data collection, and it can be used as training sample data collection Z={ (Y ₁, C ₁), (Y ₂, C ₂) ..., (Y _n, C _n); Wherein, Y _i=(y _i1, y _i2..., y _iQ) be Q the feature value of i-th client; C _i={ c _i1, c _i2..., c _iMbe the predilection grade vector of i-th client in each main flow outward appearance type;

(4) for training sample data collection, carry out client's personal feature and other the multiclass logistic association analysis of children's footwear outer appearnce, obtain the mapping weights W that each outward appearance inscape corresponds to client's personal feature;

Specifically comprise the steps:

(4.1) the mapping weight making main flow outward appearance type correspond to customer characteristic is be m class main flow outward appearance type mapping weight (m=1,2 ..., M), represent when customer characteristic integrates as Z _i=(z _i1, z _i2..., z _iQ) time, the outward appearance type of output is the probability of t class; According to logistic Function feature, be constructed as follows mathematic optimal model:

(4.2) introduce Lagrange multiplier to above formula, structure (Q+1) * M rank system of equations, solves the mapping weight that this system of equations can obtain main flow outward appearance type it is the mapping weight of m class main flow outward appearance type;

(5) the mapping weights W utilizing step (4) to obtain, calculates the preference of client to each main flow outward appearance type, and sample preference being greater than the main flow outward appearance type of 5 shows recommends client;

Wherein, calculate the method for client to the preference of each main flow outward appearance type and be the probability calculating this main flow outward appearance type of customers' preferences z is the personal feature data set of client to be measured;

P (c _t=1|Z, W) what obtain is the probability of customers' preferences t main flow outward appearance type, and probable value is converted into client's marking: c _t=P (c _t=1|Z, W) s _max, s _maxfor the maximal value of outward appearance classification preference scoring.