CN103870604A - Travel recommendation method and device - Google Patents

Abstract

The embodiment of the invention provides a travel recommendation method and device. The method comprises the following steps of setting a utility function model of the interest point according to a set interest point type model and an interest point cost model; generating an objective function according to a history interest point score of a user and the utility function model; determining an optimal utility function model with a user reference parameter in the objective function as an optimization objective; calculating a utility value of an interest point to be calculated according to the optimal utility function model; recommending at least one interest point to selected with the smallest utility value to the user. According to the travel recommendation method and device, personalized travel reference and travel cost of the user are considered, a travel interest point with higher precision is recommended for the user.

Description

Tourism recommend method and device

Technical field

The present invention relates to travel application field, relate in particular to a kind of tourism recommend method and device.

Background technology

Tourist industry develop rapidly, becomes one of industry the biggest in the world.According to the prediction of world tourism and travelling council, will bring up to 9.6% to tourist industry in 2021 from 2011 9.1% to the contribution rate of global GDP.For providing online tourism service, visitor becomes the trend of numerous tour sites (as Expedia, take journey travelling net) development.But burgeoning online tourism information selects the sight spot that meets its individual demand to bring great difficulty to visitor.On the other hand, for obtaining more business and profit, visitor's these individual demands and preference must be understood by tourist enterprise, and improve better attractive service.Therefore, no matter to visitor still concerning tourist enterprise, intelligent tourist service all urgently development with improve.

Existing tourism commending system only utilizes user's essential information and each large website to calculate the similarity between user to the scoring at sight spot, recommend sight spot according to the similarity between user for user, this tourism commending system is difficult to recommend customer satisfaction system sight spot for user.

Summary of the invention

The invention provides a kind of tourism recommend method and device.Consider tourism preference and the tourism expense of user individual, can recommend the higher tour interest point of accuracy for user.

The invention provides a kind of tourism recommend method, comprising:

According to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest;

According to this user's historical point of interest scoring and described utility function model generation objective function;

Take the user preference parameters in described objective function as optimization aim, determine optimum utility function model;

Calculate the utility value of point of interest to be selected according to optimum utility function model;

At least one point of interest to be selected of described utility value maximum is recommended to described user.

The present invention also provides a kind of tourism recommendation apparatus, comprising:

Set up module, for according to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest;

Generation module, marks and described utility function model generation objective function for the historical point of interest according to this user;

Determination module, is optimization aim for the user preference parameters take described objective function, determines optimum utility function model;

Computing module, for calculating the utility value of point of interest to be selected according to optimum utility function model;

Recommending module, for recommending described user by least one point of interest to be selected of described utility value maximum.

The present invention's a kind of tourism recommend method and device, by according to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest; According to this user's historical point of interest scoring and above-mentioned utility function model generation objective function; Take the user preference parameters in above-mentioned objective function as optimization aim, determine optimum utility function model; Calculate the utility value of point of interest to be selected according to above-mentioned optimum utility function model; At least one point of interest to be selected of utility value minimum is recommended to described user, in the time recommending point of interest to be selected for user, considered tourism preference and the tourism expense of user individual, can recommend the higher tour interest point of accuracy for user.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is some embodiments of the present invention, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

Fig. 1 is the travel process flow diagram of an embodiment of recommend method of the present invention;

Fig. 2 is the present invention's experimental result picture of recommend method and other tourism recommend method aspect the sight spot accuracy rate of recommending of travelling;

Fig. 3 is the present invention's experimental result picture of recommend method and other tourism recommend method aspect the sight spot error rate of recommending of travelling;

Fig. 4 is the present invention's recommend method and other tourism recommend method experimental result picture aspect correlativity between the sight spot sequence of recommending and the optimal sequence at sight spot of travelling;

Fig. 5 is the travel structural representation of an embodiment of recommendation apparatus of the present invention.

Embodiment

For making object, technical scheme and the advantage of the embodiment of the present invention clearer, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment in the present invention, those of ordinary skills, not making the every other embodiment obtaining under creative work prerequisite, belong to the scope of protection of the invention.

Fig. 1 is the travel process flow diagram of an embodiment of recommend method of the present invention, and as shown in Figure 1, the executive agent of the present embodiment, for tourism recommendation apparatus, specifically can pass through software, hardware, or the mode that software and hardware combines realizes, and the method comprises:

Step 101, according to point of interest Type model and the point of interest cost model set, sets up the utility function model of point of interest.

In the present embodiment, point of interest can represent the sight spot resource in tourism, as Great Wall, and pyramid, museum etc.Also can represent the cuisines resource in tourism, as local characteristic dining room, chafing dish restaurant, western-style restaurant, Japanese cuisine shop etc., point of interest also can represent other resources in tourism, the present embodiment is not restricted.

Particularly, point of interest resource has different types and different expenses, in the time that point of interest represents the sight spot resource in tourism, expense represents the admission fee use at sight spot, in the time that point of interest represents the cuisines resource in tourism, expense represents the pre-capita consumption of cuisines, the difference of the resource in the tourism representing according to point of interest, and the implication that its expense represents also can be different.Take point of interest as tourist attractions resource is as example, sight spot resource can be divided into geological and geomorphic landscape resources, water landscape resource, biological tourist resources, humane historic site resource, religion culture resource, historic gardens resource etc. according to the difference of type, and each sight spot resource has different admission fees to use.

In the present embodiment, the point of interest Type model of setting is a function about point of interest type and this user's user preference parameters, user's preference parameter represents the value degree of this user to every kind of point of interest type, user interest point cost model is a function about the expense of this user's point of interest cost, in the utility function model of setting up point of interest, in the time that sight spot resource type is more close with user preference, utility function model value is larger, when the expense of sight spot resource cost more hour, utility function model value is larger, if utility function model value is larger, this sight spot resource is more worthwhile for user to recommend.

Step 102, according to this user's historical point of interest scoring and utility function model generation objective function.

In the present embodiment, the point of interest that each user went can be different, for different users, gather the historical point of interest that this user went, the historical point of interest of this user gathering comprises the sight spot type of each historical point of interest that this user went, admission fee is used, the scoring of this user to each historical point of interest, and according to this user's historical point of interest scoring and utility function model generation objective function, when differing, this user's historical point of interest scoring and utility function model value get over hour, objective function value is less, show that this utility function model more meets this user and select the preference of point of interest.

Step 103, take the user preference parameters in objective function as optimization aim, determines optimum utility function model.

In the present embodiment, ask for the minimum point of objective function, in the time of objective function minimalization, show that this utility function model more meets this user and select the preference of point of interest, user preference parameters is now the preference parameter of this user's optimum, the preference parameter of this user's optimum is brought in utility function model, determines this user's optimum utility function model.

Step 104, calculates the utility value of point of interest to be selected according to optimum utility function model.

In the present embodiment, point of interest to be selected is the point of interest that this user did not go, and sight spot type, the expense of point of interest to be selected are brought in optimum utility function model, calculates the utility value of each point of interest to be selected.

Step 105, recommends described user by least one point of interest to be selected of utility value maximum

Particularly, one of a utility value maximum point of interest to be selected can be recommended to user, also the utility value of the point of interest to be selected calculating can be carried out to descending sort, select front K large point of interest to be selected corresponding to utility value and recommend user, by user, a front K point of interest to be selected of recommending is selected, with user interaction process in obtain customer satisfaction system point of interest to be selected.

In the present embodiment, by according to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest; According to this user's historical point of interest scoring and utility function model generation objective function; Take the user preference parameters in objective function as optimization aim, determine optimum utility function model; Calculate the utility value of point of interest to be selected according to above-mentioned optimum utility function model; At least one point of interest to be selected of utility value minimum is recommended to described user, in the time recommending point of interest to be selected for user, considered tourism preference and the tourism expense of user individual, can recommend the higher tour interest point of accuracy for user.

Further, set up the utility function model of point of interest according to the point of interest cost model of the point of interest Type model of setting and setting, be specially:

Employing formula (1) is set up the utility function model of point of interest,

U(X,C)=α*U ₁(X)+β*U ₂(C) (1)

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent point of interest Type model, U ₂(C) represent point of interest cost model, α, β represents respectively point of interest Type model and the shared weight of point of interest cost model, for different users, its type and expense to point of interest has different degrees of concern, so for different users, α, the value of β can change according to user's request.

In the present embodiment, the utility function model of the point of interest of setting up is about point of interest Type model and point of interest cost model linear separability, can be for different users the type to point of interest and expense degree of concern different weights is set, can meet more flexibly the demand of different user.

In economics, Cobb-Douglas function can be weighed the favorable rating of consumer to commodity, and form is as follows:

u(y ₁,y ₂)=y ₁ ^a*y ₂ ^b (2)

Wherein, y ₁and y ₂represented respectively the quantity of two class commodity, a and b have described the preference 0≤a of consumer to two class commodity, b≤1, and a+b=1.

So in the present embodiment, in order to weigh the preference of user to point of interest type, the point of interest class models of setting is:

$U_1\left(X\right)={\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1---\left(1\right)$

Wherein,

x represents point of interest type vector, X=(x ₁, x ₂... x _i..., x _n), x _ifor the i dimension component of point of interest type vector X, represent whether point of interest belongs to i type, when point of interest belongs to i type, x _ivalue is 1, when point of interest does not belong to i type, x _ivalue is 0, x _i{ 0,1}, in the time that point of interest belongs to one or more different point of interest type to ∈, in the point of interest type vector being represented by X, having one or more components is 1, in the time that point of interest does not belong to any one point of interest type, in the point of interest type vector being represented by X, component value is all 0.α _irepresent the preference parameter of user to i kind point of interest type, 0≤α _i≤ 1, n represents the number of types of point of interest, forms the vector that a n who represents user interest type preference parameter ties up take the preference parameter of n point of interest type as element, and this vector representation is: Α=(α ₁, α ₂... α _i..., α _n).For different users, represent that the vectorial value of interest pattern preference parameter can be different.

Describe take point of interest as sight spot resource, as sight spot resource is divided into geological and geomorphic landscape resources, water landscape resource, biological tourist resources, humane historic site resource, religion culture resource, this six classes resource of historic gardens resource according to the difference of type, this six classes sight spot resource is used vectorial X=(x successively ₁, x ₂... x _i..., x _n) in one-component represent, if when sight spot resource is Great Wall, the sight spot resource type on Great Wall is humane historic site resource, X=(0,0,0 so, 1,0,0), if this user only has preference to this kind of sight spot resource type of humane historic site resource,, A=(0 so, 0,0,1,0,0).

In the present embodiment, U ₁(X) codomain is [0,1].When a certain point of interest does not belong to any one point of interest type, when X=(0,0,0,0,0,0), point of interest Type model is obtained minimum value, i.e. U ₁(X)=0, when type under point of interest and this user preference distribute while mating completely, point of interest Type model is obtained maximal value, i.e. U ₁(X)=1.As above-mentioned for example in, type is mated completely with this user preference distribution described in Great Wall, now point of interest Type model is obtained maximal value, U ₁(X)=1.

Further, point of interest cost model can be expressed as shown in formula (4)

$U_2\left(C\right)=\left\{\begin{array}{cc}\sqrt{1-C^2/C_{\max }^2}& C\&le;C_{\max }\\ 0& C>C_{\max }\end{array}\right.---\left(4\right)$

Wherein, C represents the expense of point of interest, C _maxthe most expensive expense in the historical point of interest that expression user went, when the expense of point of interest is higher, the value of point of interest cost model is just less, show that user selects the probability of this point of interest can be less, in the time that expense exceedes a certain special value, the value of point of interest cost model is zero, U ₂(C) codomain is [0,1], in the time that the expense of point of interest is zero, point of interest cost model value maximum, in the time that the expense of user effort is the expense of historical cost, illustrate that this point of interest has reached the receptible maximum expense of user, now point of interest cost model value minimum.

In the present embodiment, according to this user's historical point of interest scoring and described utility function model generation objective function, be specially:

Generate objective function according to formula (5);

$G\left(M\right)={\mathrm{\&Sigma;}}_{j=1}^{\left|M\right|}{(r_j-K*(\mathrm{\&alpha;}*U\left(X\right))+\mathrm{\&beta;}*U\left(C\right))}^2---\left(5\right)$

Wherein,

m represents this user's historical point of interest number, r _jrepresent the scoring of this user to j historical point of interest, K represents to shine upon constant.Due to the scoring r of user to historical point of interest _jfor any one integer in 0-5, the value of utility function model is [0,1], so mapping constant K value is 5.In the represented objective function of formula (5), objective function value is less so, shows that this utility function model more meets this user and select the demand of point of interest.

Preferably, in the time determining the optimal value of user preference parameters in objective function, adopt gradient descent method.

The thought of gradient descent method is exactly to utilize negative gradient direction to decide the new direction of search of each iteration, make each iteration can make the objective function of this optimization progressively reduce, when the target function value of the optimization obtaining after the target function value of the optimization that last iteration obtains and this iteration differs while being less than a certain threshold value, iteration finishes, and now in this objective function, corresponding A is the optimal value of this user preference parameters.

Employing formula (6) is determined optimum utility function model.

$U(X,C)=\mathrm{\&alpha;}*U_1\left(X\right)+\mathrm{\&beta;}*U_2\left(C\right)=\mathrm{\&alpha;}*({\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1)+\mathrm{\&beta;}*U_2\left(C\right)---\left(6\right)$

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent point of interest Type model, U ₂(C) represent point of interest cost model, α, β represents respectively point of interest Type model and the shared weight of point of interest cost model, X represents point of interest type vector, x _irepresent the i dimension component of point of interest type vector X, α _irepresent the preference parameter of user to i kind point of interest type, n represents the number of types of point of interest.

This user preference parameters of optimum is brought into the optimum utility function model that can obtain this user in formula (6), the correlation parameter of point of interest to be selected this user is brought in optimum utility function, can obtain the utility value of point of interest to be selected, at least one point of interest to be selected of utility value maximum is recommended to this user.

Effective result of the present embodiment can further illustrate by following emulation:

1 emulation content: the sight spot that certain user's history was gone is divided into training set and test set, travel in recommend method (GD) and employing particle cluster algorithm compute user preferences parametric technique (PSO) in the present invention, training set is for determining this user's optimum utility function model, in traditional collaborative filtering tourism recommend method (CF), training set is for determining and the highest other users of this user's similarity, to be this user recommends and other the highest user's history of its similarity were gone sight spot.Test set is used for testing travel recommend method (GD) and adopt particle cluster algorithm compute user preferences parametric technique (PSO), traditional collaborative filtering tourism recommend method to recommend the performance quality at sight spot of the present invention.Between the sequence at the sight spot that the recommendation error rate at the sight spot that the recommendation accuracy rate at the sight spot of recommending from each method, each method are recommended, each method are recommended and sight spot optimal sequencing, correlativity three aspects: is evaluated the performance separately of these methods.

2 the simulation experiment result

The experimental result of the sight spot that the each method of A is recommended aspect recommendation accuracy rate

The user preference parameters method (PSO) of travelling recommend method (GD) and adopting particle cluster algorithm to calculate with the present invention, traditional collaborative filtering tourism recommend method (CF) is respectively user and recommends front K point of interest the most attractive, in the sight spot that P represents to recommend out, this user scoring is not less than the set at 4 sight spot, T represents the set at sight spot in this user's test set, and the recommendation accuracy rate Precision@K at sight spot is expressed as shown in formula (6):

$\mathrm{Precision}@K=\frac{|P\&cap;T|}{K}\&times;100\%---\left(6\right)$

Fig. 2 is the present invention's experimental result picture of recommend method and other tourism recommend method aspect the sight spot accuracy rate of recommending of travelling, in Fig. 2, X-axis has represented the value of the point of interest number K recommending, accuracy rate is recommended in ordinate representative, as can be seen from Figure 2, the present invention travel recommend method (GD) recommend accuracy rate the highest, adopt particle cluster algorithm compute user preferences parametric technique (PSO) to recommend accuracy rate a little less than the present invention's recommend method (GD) of travelling, and traditional collaborative filtering tourism recommend method (CF) recommends accuracy rate minimum, so, the present invention's recommend method of travelling is recommending to be better than adopting aspect accuracy rate particle cluster algorithm compute user preferences parametric technique (PSO) and the traditional collaborative filtering recommend method (CF) of travelling.

The experimental result of the sight spot that the each method of B is recommended aspect recommendation error rate

With the present invention travel recommend method (GD) and adopt particle cluster algorithm compute user preferences parametric technique (PSO), traditional collaborative filtering tourism recommend method (CF) is respectively user and recommends front K point of interest the most attractive, in the time that user's scoring is less than 4, think that this user does not like this sight spot, if there are these sight spots in recommendation results, illustrate that recommendation results is inaccurate.So recommend error rate to be defined as follows: Q is the sight spot set that in the sight spot of recommending, this user's scoring is less than 4, and T is sight spot set in test set, and the recommendation error rate ErrorRate@K at sight spot is expressed as shown in formula (7):

$\mathrm{ErrorRate}@K=\frac{|Q\&cap;T|}{K}\&times;100\%---\left(7\right)$

Fig. 3 is that the present invention travels recommend method and other tourism recommend method at the experimental result picture of recommending aspect error rate, in Fig. 3, X-axis has represented the value of the point of interest number K recommending, error rate is recommended in ordinate representative, recommend the lower recommendation sight spot that represents the method for error rate more to meet this user's request, as can be seen from Figure 3, the present invention travel recommend method (GD) recommend error rate minimum, adopt particle cluster algorithm compute user preferences parametric technique (PSO) to recommend error rate a little more than the inventive method (GD), and traditional collaborative filtering tourism recommend method (CF) recommends error rate the highest, so, the present invention's recommend method of travelling is recommending to be better than adopting aspect error rate particle cluster algorithm compute user preferences parametric technique (PSO) and the traditional collaborative filtering recommend method (CF) of travelling.

The experimental result of correlativity aspect between the sequence at the sight spot that the each method of C is recommended and sight spot optimal sequence

With the present invention travel recommend method (GD) and adopt particle cluster algorithm compute user preferences parametric technique (PSO), traditional collaborative filtering tourism recommend method (CF) is respectively user and recommends front K point of interest the most attractive, the optimal sequence at the sight spot that the sight spot sequence of then each method being recommended and this user select compares, and the correlativity NDCG K of the sight spot sequence that each method is recommended and the optimal sequence at sight spot is expressed as shown in formula (8):

$\mathrm{NDCG}@K=\frac{\mathrm{DCG}@K}{\mathrm{iDCG}@K}---\left(8\right)$

Wherein,

$\mathrm{DCG}@K={\mathrm{\&Sigma;}}_{i=1}^K{\mathrm{kog}}_2\frac{2^{r_i}-1}{1+i}---\left(9\right)$

I is the subscript at sight spot in the sequence of sight spot, r _ifor this scoring of user to i sight spot, the DCG@K value under the optimal sequence that iDCG@K is sight spot.

The standard of NDCG@K is based on following two hypothesis:

(1) sight spot that correlativity is higher comes better above.

(2) sight spot low compared with correlativity, the sight spot that correlativity is high more can meet this user's demand.

Fig. 4 is the present invention's recommend method and other tourism recommend method experimental result picture aspect correlativity between the sight spot sequence of recommending and the optimal sequence at sight spot of travelling, in Fig. 4, the post figure of black represents correlativity between optimal sequence that sight spot sequence that the sight spot number that each side method is recommended is 5 and sight spot number are 5, white post figure represents correlativity between optimal sequence that sight spot sequence that sight spot number that each side's method is recommended is 10 and sight spot number are 10, the correlativity of the sight spot sequence of recommending and the optimal sequence at sight spot is larger, show that the sight spot of the method recommendation more meets this user's demand, as can be seen from Figure 4, the present invention travels recommend method (GD) in the time of K=5 and K=10, the value of NDCG@K has reached respectively 0.6502 and 0.6649, all higher than adopting the travel value of the NDCG@K of recommend method (CF) when K=5 and the K=10 of particle cluster algorithm compute user preferences parametric technique (PSO) and traditional collaborative filtering, so, the inventive method (GD) is better than adopting particle cluster algorithm compute user preferences parametric technique (PSO) and the traditional collaborative filtering recommend method (CF) of travelling between sequence and the sight spot optimal sequence at the sight spot of recommending aspect correlativity.

So, the present invention travel recommend method (GD) recommending the recommendation accuracy rate at sight spot, recommend between sequence and the sight spot optimal sequence at sight spot of recommendation error rate, the recommendation at sight spot correlativity three aspects: to be all better than adopting particle cluster algorithm compute user preferences parametric technique (PSO) and the traditional collaborative filtering recommend method (CF) of travelling.

Fig. 5 is the travel structural representation of an embodiment of recommendation apparatus of the present invention, and as shown in Figure 5, this device can comprise: set up module 501, generation module 502, determination module 503, computing module 504 and recommending module 505.

Wherein, set up module 501, for according to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest.

Generation module 502, marks and utility function model generation objective function for the historical point of interest according to this user.

Determination module 503, is optimization aim for the user preference parameters take objective function, determines optimum utility function model.

Computing module 504, for calculating the utility value of point of interest to be selected according to optimum utility function model.

Recommending module 505, for recommending this user by least one point of interest to be selected of utility value maximum.

The device of the present embodiment can execution graph 1 shown in the technical scheme of embodiment of the method, it realizes principle and technique effect is similar, repeats no more herein.

Further, set up module 501, for set up the utility function model of point of interest according to the point of interest cost model of the point of interest Type model of setting and setting, be specially:

Adopt U (X, C)=α * U ₁(X)+β * U ₂(C) set up the utility function model of described point of interest,

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent point of interest Type model, U ₂(C) represent point of interest cost model, α, β represents respectively point of interest Type model and the shared weight of point of interest cost model.

Concrete, point of interest class models is:

$U_1\left(X\right)={\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1,$ Wherein, ${\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&alpha;}}_i=\mathrm{1,0}\&le;{\mathrm{\&alpha;}}_i\&le;1,$ X represents point of interest type vector, x _irepresent the i dimension component of point of interest type vector X, α _irepresent the preference parameter of user to i kind point of interest type, n represents the number of types of point of interest.

Point of interest cost model is:

$U_2\left(C\right)=\left\{\begin{array}{cc}\sqrt{1-C^2/C_{\max }^2}& C\&le;C_{\max }\\ 0& C>C_{\max }\end{array}\right.,$ Wherein, C represents the expense of point of interest, C _maxrepresent the most expensive expense in historical point of interest that this user got on.

Generation module 502, marks and utility function model generation objective function for the historical point of interest according to this user, is specially:

According to $G\left(M\right)={\mathrm{\&Sigma;}}_{j=1}^{\left|M\right|}{(r_j-K*(\mathrm{\&alpha;}*U\left(X\right))+\mathrm{\&beta;}*U\left(C\right))}^2$ Generate described objective function;

Wherein, m represents described user's historical point of interest number, r _jrepresent j historical point of interest scoring, K represents to shine upon constant, K=5.

Determination module 503, is optimization aim for the user preference parameters take objective function, determines optimum utility function model, specifically comprises:

Adopt gradient descent method to determine the optimal value of user preference parameters in described objective function;

According to the optimal value of described user preference parameters, adopt $U(X,C)=\mathrm{\&alpha;}*U_1\left(X\right)+\mathrm{\&beta;}*U_2\left(C\right)=\mathrm{\&alpha;}*({\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1)+\mathrm{\&beta;}*U_2\left(C\right)$ Determine optimum utility function model.

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent point of interest Type model, U ₂(C) represent point of interest cost model, α, β represents respectively point of interest Type model and the shared weight of point of interest cost model, X represents point of interest type vector, x _irepresent the i dimension component of point of interest type vector X, α _irepresent the preference parameter of user to i kind point of interest type, n represents the number of types of point of interest.

Finally it should be noted that: above each embodiment, only in order to technical scheme of the present invention to be described, is not intended to limit; Although the present invention is had been described in detail with reference to aforementioned each embodiment, those of ordinary skill in the art is to be understood that: its technical scheme that still can record aforementioned each embodiment is modified, or some or all of technical characterictic is wherein equal to replacement; And these modifications or replacement do not make the essence of appropriate technical solution depart from the scope of various embodiments of the present invention technical scheme.

Claims (10)

1. a tourism recommend method, is characterized in that, comprising:

According to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest;

According to this user's historical point of interest scoring and described utility function model generation objective function;

Take the user preference parameters in described objective function as optimization aim, determine optimum utility function model;

Calculate the utility value of point of interest to be selected according to optimum utility function model;

At least one point of interest to be selected of described utility value maximum is recommended to described user.

2. method according to claim 1, is characterized in that, describedly sets up the utility function model of point of interest according to the point of interest Type model of setting and point of interest cost model, is specially:

Adopt U (X, C)=α * U ₁(X)+β * U ₂(C) set up the utility function model of described point of interest;

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent described point of interest Type model, U ₂(C) represent described point of interest cost model, α, β represents respectively described point of interest Type model and the shared weight of described point of interest cost model.

3. method according to claim 2, is characterized in that, described point of interest Type model is:

$U_1\left(X\right)={\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1$ Wherein, ${\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&alpha;}}_i=\mathrm{1,0}\&le;{\mathrm{\&alpha;}}_i\&le;1$

X represents point of interest type vector, x _irepresent the i dimension component of described point of interest type vector X, α _irepresent the preference parameter of described user to i kind point of interest type, n represents the number of types of point of interest;

Described point of interest cost model is:

$U_2\left(C\right)=\left\{\begin{array}{cc}\sqrt{1-C^2/C_{\max }^2}& C\&le;C_{\max }\\ 0& C>C_{\max }\end{array}\right.$

Wherein, C represents the expense of point of interest, C _maxrepresent the most expensive expense in historical point of interest that described user went.

4. according to the method in claim 2 or 3, it is characterized in that, described according to this user's historical point of interest scoring and described utility function model generation objective function, be specially:

According to $G\left(M\right)={\mathrm{\&Sigma;}}_{j=1}^{\left|M\right|}{(r_j-K*(\mathrm{\&alpha;}*U_1\left(X\right))+\mathrm{\&beta;}*U_2\left(C\right))}^2$ Generate described objective function;

Wherein,

m represents described user's historical point of interest number, r _jrepresent j historical point of interest scoring, K represents to shine upon constant.

5. method according to claim 4, is characterized in that, described user preference parameters in described objective function is optimization aim, determines optimum utility function model, specifically comprises:

Adopt gradient descent method to determine the optimal value of user preference parameters in described objective function;

According to the optimal value of described user preference parameters, adopt $U(X,C)=\mathrm{\&alpha;}*U_1\left(X\right)+\mathrm{\&beta;}*U_2\left(C\right)=\mathrm{\&alpha;}*({\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1)+\mathrm{\&beta;}*U_2\left(C\right)$ Determine optimum utility function model;

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent described point of interest Type model, U ₂(C) represent described point of interest cost model, α, β represents respectively described point of interest Type model and the shared weight of described point of interest cost model, X represents described point of interest type vector, x _irepresent the i dimension component of described point of interest type vector X, α _irepresent the preference parameter of described user to i kind point of interest type, n represents the number of types of described point of interest.

6. a tourism recommendation apparatus, is characterized in that, comprising:

Set up module, for according to point of interest Type model and the point of interest cost model set, set up the utility function model of point of interest;

Generation module, marks and described utility function model generation objective function for the historical point of interest according to this user;

Determination module, is optimization aim for the user preference parameters take described objective function, determines optimum utility function model;

Computing module, for calculating the utility value of point of interest to be selected according to optimum utility function model;

Recommending module, for recommending described user by least one point of interest to be selected of described utility value maximum.

7. device according to claim 6, is characterized in that,

The described module of setting up, for setting up the utility function model of point of interest according to the point of interest Type model of setting and point of interest cost model, is specially:

Adopt U (X, C)=α * U ₁(X)+β * U ₂(C) set up the utility function model of described point of interest;

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent described point of interest Type model, U ₂(C) represent described point of interest cost model, α, β represents respectively described point of interest Type model and the shared weight of described point of interest cost model.

8. device according to claim 7, is characterized in that, described point of interest class models is:

$U_1\left(X\right)={\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1$ Wherein, ${\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&alpha;}}_i=\mathrm{1,0}\&le;{\mathrm{\&alpha;}}_i\&le;1$

X represents point of interest type vector, x _irepresent the i dimension component of described point of interest type vector X, α _irepresent the preference parameter of user to i kind point of interest type, n represents the number of types of point of interest;

Described point of interest cost model is:

$U_2\left(C\right)=\left\{\begin{array}{cc}\sqrt{1-C^2/C_{\max }^2}& C\&le;C_{\max }\\ 0& C>C_{\max }\end{array}\right.$

Wherein, C represents the expense of point of interest, C _maxthe most expensive expense in the point of interest that expression user history got on.

9. according to the device described in claim 7 or 8, it is characterized in that,

Described generation module, marks and described utility function model generation objective function for the historical point of interest according to this user, is specially:

According to $G\left(M\right)={\mathrm{\&Sigma;}}_{j=1}^{\left|M\right|}{(r_j-K*(\mathrm{\&alpha;}*U\left(X\right))+\mathrm{\&beta;}*U\left(C\right))}^2$ Generate described objective function;

Wherein,

m represents described user's historical point of interest number, r _jrepresent j historical point of interest scoring, K represents to shine upon constant.

10. device according to claim 9, is characterized in that,

Described determination module, is optimization aim for the user preference parameters take described objective function, determines optimum utility function model, specifically comprises:

Adopt gradient descent method to determine the optimal value of user preference parameters in described objective function;

According to the optimal value of described user preference parameters, adopt $U(X,C)=\mathrm{\&alpha;}*U_1\left(X\right)+\mathrm{\&beta;}*U_2\left(C\right)=\mathrm{\&alpha;}*({\mathrm{\&Pi;}}_{i=1}^n{(x_i+1)}^{{\mathrm{\&alpha;}}_i}-1)+\mathrm{\&beta;}*U_2\left(C\right)$ Determine optimum utility function model;

Wherein, alpha+beta=1,0< α <1,0< β <1, U ₁(X) represent described point of interest Type model, U ₂(C) represent described point of interest cost model, α, β represents respectively described point of interest Type model and the shared weight of described point of interest cost model, X represents described point of interest type vector, x _irepresent the i dimension component of described point of interest type vector X, α _irepresent the preference parameter of described user to i kind point of interest type, n represents the number of types of described point of interest.

Images