CN101957973A - System for recommending commodities - Google Patents

Abstract

The invention discloses a system for recommending commodities. The system for recommending the commodities comprises a commodity inquiry input module, a commodity display module and a merchant display module, wherein the commodity inquiry input module is used for inputting the relevant conditions of the commodities; the commodity display module is used for displaying commodity information matched with the input conditions; the merchant display module is used for displaying merchant information corresponding to the selected commodities; and the commodity display module optimizes the display sequence of the commodities through a tree decilog model. The system for recommending the commodities has the advantages of guiding the merchants to optimize products and marketing strategies and obtain premium and effectively contributing to transaction.

Description

A kind of commercial product recommending system

Technical field

The invention belongs to the e-commerce technology field, relate to a kind of commercial product recommending system.

Background technology

Commodity comprise products ﹠ services, and commercial product recommending is exactly products ﹠ services and the preferential plan of sales promotion thereof and featured function to be showed in suitable time, place and mode to the user of most possible purchase present, to facilitate the maximization of trading volume.

The advertisement of traditional retailing enterprise, preferential promotion plan, marketing catalogue official documents and correspondence (catalog copy) are difficult to accomplish personalized commercial product recommending.After ecommerce was risen, businessman can according to user's search purpose, recommend the user with target product by the search in Website engine, has improved the efficient of commercial product recommending.But as GOOGLE, BAIDU and TAOBAO, the user searches for a keyword, and thousands of information have been returned in the website, and the real browsing information of user one all be no more than 5.The surplus of information may make the user at a loss as to what to do, and the information of returning has high similarity, ignored potential other purchase intention and the possibility of user, also ignored the characteristics of businessman's demand and commodity itself simultaneously, thereby deviated from the original object that commercial product recommending is facilitated transaction.Though some intelligent commercial product recommending system has been considered businessman's demand and commodity characteristics at present, but it is not enough to the actual diversity understanding of user intent, such as the user of search diamond ring, may be also can be interested in wedding gauze kerchief, necklace even tourism, air ticket, thereby lost the chance of commercial product recommending.

In a word, present air ticket commending system is just simply enumerated Search Results mostly, lacks effective recommendation mechanisms or only recommends with single index such as price, and subjectivity is too strong, lacks the understanding to the user.The value of recommendation results collection is not high, and the actual hit rate of user is on the low side.

Summary of the invention

Goal of the invention: the objective of the invention is at the deficiencies in the prior art, a kind of ecommerce air ticket commending system is provided, this system has fully also considered the actual diversity of user's request rightly, increase the commission merchant and selected reference, thus can be at the individual commercial product recommending of realizing personalization of particular user.

Technical scheme: a kind of commercial product recommending system, comprise the merchandise query load module, be used to import the correlated condition of commodity; The commodity displaying module is used to show the merchandise news of mating with initial conditions; Described commodity displaying module is optimized the displaying order of commodity by tree-shaped decilog model (Nested Logit Model); Businessman's display module is used for showing and the corresponding Business Information of selected commodity.

Preferable, move under the model parameter estimation online design of system with tree-shaped decilog model.

Preferable, described businessman display module is optimized agential displaying order by multinomial decilog model (Multinomial Logit Model).

Preferable, move under the model parameter estimation online design of system with multinomial decilog model.

Preferable, system also comprises the advertisement display module.

Beneficial effect: (1) native system is according to the commodity starting condition of user's input, more pay close attention to potential purchase intention of user and possibility, also pay close attention to simultaneously the characteristics of businessman's demand and commodity itself, therefore improved the recommendation efficient of commodity, more help promoting Transaction Success;

(2) native system is also paid close attention to the actual diversity of user intent, has further improved the recommendation chance of commodity;

(3) native system is by determining influence the correlated variables parameter value of commodity transaction, and then the displaying of definite commodity and businessman order, and therefore bootable businessman optimizes product and sales tactics, thus the promotion Transaction Success;

(4) native system can make businessman obtain certain premium by the design of model parameter.

Description of drawings

Fig. 1 is the structural representation of embodiments of the invention 1;

Fig. 2 is the synoptic diagram of tree-shaped decilog model;

Fig. 3 is the proposed algorithm schematic diagram of multinomial decilog model;

Fig. 4 is the synoptic diagram that model explanation of the present invention, predictive user are selected behavior;

Fig. 5 is that the proposed algorithm of multinomial decilog model realizes synoptic diagram;

Fig. 6 moves synoptic diagram down for the model parameter line.

Embodiment

Below in conjunction with accompanying drawing, by embodiment, technical solution of the present invention is elaborated, but protection scope of the present invention is not limited to described embodiment.

Embodiment 1: as shown in Figure 1, a kind of air ticket commending system comprises air ticket inquiry load module, is used for input and sets out city, arrival city and sailing date; The air ticket display module is used to show the flight that mates with initial conditions; Commission merchant's display module is used for showing and the corresponding commission merchant's information of selected air ticket; The advertisement display module is used to recommend air ticket other commodity in addition.

The air ticket display module is by the displaying order of tree-shaped decilog model optimization air ticket.As shown in Figure 2, the option i among the nest m to the effectiveness of user n is:

U _ni＝V _ni+ε _nk+ε _nki

Wherein, V _Ni=β ₁x _I1+ β ₂x _I2+ Λ+β _mx _Im

ε _NkBe the part of effectiveness at random among the nest m;

ε _NkiBe the at random effectiveness part of option i except that nest k influence.

Estimate β _mMethod be that maximum likelihood is estimated.

Option i comprises following variable: the commission merchant always offers, whether flying method makes a connection, whether be non-stop flight, aircraft type, whether be large aircraft, the departure time, airline or the like.Wherein commission merchant's variable of always offering is LOG (air ticket price+insurance price), and the parameter value of departure time variable is by a continuous utility function F (departure time) expression, and airline's variable is by set D (sequence number, title, parameter value) expression.Because user's preferences non-stop flight, large aircraft, so the flight of non-stop flight, big aircraft can obtain certain premium.

In other embodiments, because commodity of recommending or service are different, so the variable difference that relates to.

The selection probability is:

$\mathrm{Pr}\left(i\right)=\frac{e^{v_i}}{{\mathrm{\Σ}}_{j\∈N_n}{e}^{v_j}}\·\frac{{\left({\mathrm{\Σ}}_{j\∈N_n}{e}^{v_j}\right)}^u}{{\mathrm{\Σ}}_{k=1}^N{\left({\mathrm{\Σ}}_{j\∈N_k}{e}^{v_j}\right)}^u}$

0＜μ＜1

μ=1 o'clock deteriorates to the MNL model.

Model dimension parameter μ has portrayed the correlativity between the option in each nest, and μ is more little, and correlativity is big more.

Model dimension parameter μ describes close demand of the departure time.

According to the determining of variable parameter value and model dimension parameter μ, thereby determine that the user selects the probability of flight, select the size of probability to determine the order of flight displaying according to the user.

After the air ticket result of page searching was clicked certain concrete flight, the user entered " commission merchant's list page ", carries out commission merchant's selection.Commission merchant's display module is by the agential displaying order of multinomial decilog model optimization.

As Fig. 3, Fig. 4 and shown in Figure 5, option i to user's effectiveness is

U _ni＝V _ni+ε _ni

Wherein, V _Ni=β ₁x _I1+ β ₂x _I2+ Λ+β _mx _Im

ε _NiBe IID, η=0, μ=1, the stochastic variable that Gumbel distributes.

Estimate β _mMethod be that maximum likelihood is estimated.

Option i comprises following variable: the commission merchant variables L OG (air ticket price+insurance price) that always offers; Tie-in sale insurance variable, commission merchant mark variable, commission merchant's value-added service variable, the recent problem variable of commission merchant, commission merchant's own variable.Because of user preference is not paired unsalable goods up with goods that sell well insurance, scoring is higher, the scoring number is more, have value-added service, no problem commission merchant in the recent period, therefore do not pair unsalable goods up with goods that sell well insurance, scoring is higher, the scoring number is more, value-added service preferably, recent no problem commission merchant can obtain certain premium.Commission merchant's own variable is described the preference of user to commission merchant itself, and parameter value is by set D (sequence number, title, parameter value) expression.

$Q_{R^{,}}\left(R\right)=\frac{\underset{c\∈R}{\mathrm{\Σ}}{\mathrm{Pr}}_{\mathrm{\Λ}}\left(C\right)}{\underset{c\∈R^{,}}{\mathrm{\Σ}}{\mathrm{Pr}}_{\mathrm{\Λ}}\left(C\right)}$

Given selection set C={c _i, | C|=N

Given 0＜K≤N

Find the subclass R of C ^*, make

|R ^*|＝K

To the random subset R ' of C, | R ' |=K has

Q _R’(R)≤1

For the MNL model,, have any R '

$Q_{R^{,}}=\underset{c\∈R}{\mathrm{\Σ}}{\mathrm{Pr}}_A\left(c\right)=\frac{{\mathrm{\Σ}}_{c_i\∈R}{e}^{v_i}}{{\mathrm{\Σ}}_{c_j\∈A}{e}^{v_j}}$

Obviously, R ^*For selecting V among the set C _iK maximum options.

According to R ^*Middle V _iSize determine the displaying order of businessman.

Agential putting in order will influence user experience and the conversion ratio of user in commission merchant's list page, selects behavior model by the user agents of system, can optimize commission merchant's ordering; Can change commission merchant's amount of drawing a bill and prediction in advance by this model modification commission merchant ordering.

As shown in Figure 6, move under the model parameter estimation online design of system with the model parameter estimation of tree-shaped decilog model and multinomial decilog model.

As mentioned above, although represented and explained the present invention that with reference to specific preferred embodiment it shall not be construed as the restriction to the present invention self.Under the spirit and scope of the present invention prerequisite that does not break away from the claims definition, can make various variations in the form and details to it.

Claims (5)

1. a commercial product recommending system is characterized in that: comprise the merchandise query load module, be used to import the correlated condition of commodity;

The commodity displaying module is used to show the merchandise news of mating with initial conditions; Described commodity displaying module is by the displaying order of tree-shaped decilog model optimization commodity;

Businessman's display module is used for showing and the corresponding Business Information of selected commodity.

2. commercial product recommending system as claimed in claim 1 is characterized in that: move under the model parameter estimation online design of system with tree-shaped decilog model.

3. commercial product recommending system as claimed in claim 1 is characterized in that: described businessman display module is by the agential displaying order of multinomial decilog model optimization.

4. commercial product recommending system as claimed in claim 3 is characterized in that: move under the model parameter estimation online design of system with multinomial decilog model.

5. as the arbitrary described commercial product recommending system of claim 1 to 4, it is characterized in that: system also comprises the advertisement display module.

Images