CN103345699A - Personalized food recommendation method based on commodity forest system - Google Patents

Abstract

The invention relates to a personalized food recommendation method based on a commodity forest system. The method includes the following steps of firstly, conducting the following concept definition on the system; secondly, calculating user scores through a UserCF; thirdly, calculating the user cores through an ItemCF. With the current position of a user, the preference of the user and the behavioral habits of the user and the like taken into consideration, personalized restaurants and dishes which meet requirements of the user are recommended to the user, and the food recommendation efficiency and accuracy are improved. Meanwhile, accurate recommendation results can be converted into consuming behaviors, and the user satisfaction degree and merchant benefits are improved.

Description

A kind of personalized cuisines recommend method based on the commodity forest system

Technical field

The invention belongs to the e-commerce technology field, relate to a kind of personalized cuisines recommend method based on the commodity forest system.

Background technology

In recent years because portable terminal universal day by day, information resources obtain and propelling movement occurs in " at any time, everywhere, with oneself ", location-based mobile commending system has become commending system research field one of active research field the most.

In the existing recommendation that mainly concentrates on film, music, books, recommend as mobile blog, the m-CCS system that people such as Chiu PH, Kao GYM propose in the personalized Blog content commending system paper based on the cellphone subscriber, blog article is carried out cluster, browse record by the blog article of analyzing the mobile subscriber, obtain the mobile subscriber to the preference of different blog article types, and consider the Internet user to the clicking rate of blog article, blog article clicking rate is high and that satisfy user preference is recommended the mobile subscriber.Mobile music recommend, the behavior of browsing by the mobile subscriber (ignore, click, audition, purchase etc.) is implicitly obtained the mobile subscriber to the relative preference of music, the preference of expressing as buying behavior is better than the preference that the audition behavior is expressed, use collaborative filtering to predict into the mobile subscriber to the preference relation of other music according to the preference obtained, thereby carry out music recommend.

Comparatively speaking, at present less in the cuisines sector application more do not separate commodity with businessman, handles separately separately and modeling.

Summary of the invention

In order to solve the problems of the technologies described above, the invention provides a kind of personalized cuisines recommend method based on the commodity forest system, this method proposes the concept of commodity forest system, adopt new goods model will adopt the related businessmans of commodity (not being with position attribution) (band position attribution) structure, realize location-based personalized recommendation.

Its technical scheme is as follows:

A kind of personalized cuisines recommend method based on the commodity forest system may further comprise the steps:

(1) this system is carried out the definition of following concept:

Definition 1: classification C={C ₁, C ₂, C ₃... C _n}

Definition 2: forest F={T ₁, T ₂, T ₃... T _n}

Definition 3: tree T=(R, W, TC)

Definition 4: leaf (commodity) I=(Rs, TC ', S, U)

Wherein C is classification set, C _i∈ C, C _iDifference presentation class project, as food materials, type etc., and C _iBe defined as the extensibility generalized list, i.e. C _iComprise more detailed classification below again, as C _i(C _I1(C _I11(C _I111, C _I112...), C _I12...), C _I2(C _I21, C _I22...), C _I3... C _In);

T _iClassification tree in the expression forest; R is the root of this tree, expression the superiors class categories; W represents this tree shared weights in all classification trees of forest; TC represents child's information of this tree, and TC ∈ C; The set of paths of TC ' expression commodity (leaf) in forest, as: TC '={ C _iC _I1C _I11C _I111∪ C _iC _I2C _I21C _I212∪ ...), and every paths belongs to different trees (T) in the set; Rs represents the set of the root of the affiliated T of this leaf; U represents collection or comments on these commodity user set; S represents to have this commodity businessman set, defines S=(L, A) simultaneously, and L represents this businessman's geographical location information, and A represents other attributes of businessman,

(2) calculate user's scoring based on user collaborative filter algorithm (UserCF), its computing formula is:

$p(u,i)=\underset{v\∈S(u,K)\∩N\left(i\right)}{\mathrm{\Σ}}{w}_{\mathrm{uv}}*r_{\mathrm{vi}}$

((u K) comprises K the user the most similar with user u to S to p, and N (i) had behavior (scoring etc.) user set, w to commodity i for u, the i) scoring of the commodity i of expression user u _UvBe based on the user u of user's scoring and the similarity of user v, r _ViThe scoring of the commodity i of representative of consumer v,

And $w_{\mathrm{uv}}=\mathrm{sim}(u,v)=\frac{{\mathrm{\Σ}}_{i\∈I_{\mathrm{uv}}}(R_{u,i}-{\stackrel{\‾}{R}}_u)(R_{v,i}-{\stackrel{\‾}{R}}_v)}{\sqrt{{\mathrm{\Σ}}_{i\∈I_u}{(R_{u,i}-{\stackrel{\‾}{R}}_u)}^2}\sqrt{{\mathrm{\Σ}}_{i\∈I_v}{(R_{v,i}-{\stackrel{\‾}{R}}_v)}^2}}$

Expression user u is to the mean value of all commodity scorings, R _{U, i}The scoring of the commodity i of expression user u; In like manner Expression user v is to the mean value of all commodity scorings, R _{V, i}The scoring of the commodity i of expression user v

(3) calculate user's scoring based on article collaborative filtering (ItemCF), its computing formula is:

$p(u,j)=\underset{i\∈N\left(u\right)\∩S(i,K)}{\mathrm{\Σ}}{w}_{\mathrm{ij}}*r_{\mathrm{ui}}$

Wherein, (N (u) is the set of the commodity liked of user to p for u, the j) scoring of expression user u commodity i, and (i K) is K the set the most similar with commodity i, w to S _JiBe based on the commodity j of user's scoring and the similarity of commodity i, r _UiBe the scoring of user u commodity i, w _Ij=sim (i, j)=λ sim (i, j) _r+ (1-λ) sim (i, j) _cλ is variable coefficient, be used for adjusting article similarity sim based on user's scoring (i, j) _rWith based on the taxonomy of goods similarity sim of commodity forest system (i, j) _c

① $\mathrm{sim}{(i,j)}_r=\frac{{\mathrm{\Σ}}_{u\∈U}(R_{u,i}-{\stackrel{\‾}{R}}_u)(R_{u,j}-{\stackrel{\‾}{R}}_u)}{\sqrt{{\mathrm{\Σ}}_{u\∈U}{(R_{u,i}-{\stackrel{\‾}{R}}_u)}^2}\sqrt{{\mathrm{\Σ}}_{u\∈U}{(R_{u,j}-{\stackrel{\‾}{R}}_u)}^2}}$

②sim(i，j) _c＝∑w*sim(i，j) _c

In the formula 1, R _{U, i}The scoring of the article i of expression user u,

The mean value of the belongings product scoring of expression user u; R in like manner _{U, j}The scoring of the article j of expression user u.

In the formula 2, w represents the weights of different classification trees all classification shared ratio in gathering in whole commodity forest system.

(4) preceding n the recommendation items (Top-N) of calculated recommendation system

Calculate user in predicting scoring rank according to UserCF and ItemCF, and integrate the current geographic position of user and the distance of businessman, provide last result.

Compared with prior art, beneficial effect of the present invention: the present invention is in conjunction with the current present position of user, user preference and behavioural habits etc., recommends dining room and the vegetable of the personalization of meeting consumers' demand for it, improves efficient and accuracy that cuisines are recommended.Simultaneously, accurate recommendation results can be converted into consumer behavior, improves user satisfaction and businessman's benefit, and its beneficial effect mainly shows following two aspects:

(1) realizes stratification and the diversification of commodity classification, made and calculate the similar precision of commodity higher and computing time still less, had more personalization; (2) based on commodity classification was following in advance in the past, incorporate position (location-based) element, form user-commodity-businessman's multi-dimensional spatial structure, and then be the location-based mobile goods model of recommending to provide.

Description of drawings

Fig. 1 is the process flow diagram that the present invention is based on the personalized cuisines recommend method of commodity forest system;

Fig. 2 is commodity forest system structural drawing;

Fig. 3 is user-commodity-multi-dimensional spatial structure figure of businessman.

Embodiment

Further specify technical scheme of the present invention below in conjunction with drawings and Examples.

In Fig. 1, the cuisines recommendation LBS system among the present invention is divided into two branches to be recommended, and wherein is respectively user-commodity-businessman (User-Item-Shop) and user-businessman-commodity (User-Shop-Item).

At the User-Item-Shop branch road, be divided into the user again collection is arranged, mark, browse and collect, mark, browse Item and user that leaf node exists but at the non-existent Item of leaf node, the previous case is carried out calculated recommendation by ItemCF and UserCF.When using ItemCF, collect, mark, browse matrix and commodity forest system in conjunction with commodity-user (Item-User), calculate the higher Item of similarity, and generate commodity-commodity (Item-Item) similarity matrix; Then use the prediction scoring algorithm to calculate the interest level of user's commodity, obtain new (commodity-user) Item-User level of interest matrix.For latter event, then adopt UserCF to pass through (user-user) User-User similarity matrix and calculate the user to the interest level matrix of commodity.

At the User-Shop-Item branch road, then according to user's gps position, find the nearer Shop of distance users current location, carry out commercial product recommending by (commodity-businessman) Item-Shop matrix again.

Last two branch roads all form U (user)-I (commodity)-S (businessman) three-dimensional space model by the commodity forest system again.

A kind of personalized cuisines recommend method based on the commodity forest system may further comprise the steps:

At first based in the past achievement in research, propose commodity forest system concept, commodity are classified according to this forest system.

And

(1) this system is carried out the definition of following concept:

Definition 1: classification C={C ₁, C ₂, C ₃... C _n}

Definition 2: forest F={T ₁, T ₂, T ₃... T _n}

Definition 3: tree T=(R, W, TC)

Definition 4: leaf (commodity) I=(Rs, TC ', S, U)

Wherein C is classification set, C _i∈ C, C _iDifference presentation class project, as food materials, type etc., and C _iBe defined as the extensibility generalized list, i.e. C _iComprise more detailed classification below again.As C _i(C _I1(C _I11(C _I111, C _I112...), C _I12...), C _I2(C _I21, C _I22...), C _I3... C _In);

T _iTree in the expression forest; R is the root of this tree, expression the superiors class categories; W represents this tree shared weights in forest; TC represents child's information of this tree, and TC ∈ C; The set of paths of TC ' expression commodity (leaf) in forest, as: TC '={ C _iC _I1C _I11C _I111∪ C _iC _I2C _I21C _I212∪ ... }, and every paths belongs to different trees (T); Rs represents the set of the root of the affiliated T of this leaf; U represents collection or comments on these commodity user set; S represents to have this commodity businessman set, defines S=(L, A) simultaneously, and L represents this businessman's geographical location information, and A represents other attributes of businessman.

This system has realized stratification and the diversification of (1) commodity classification, make to calculate the similar precision of commodity higher and computing time still less; (2) based on commodity classification was following in advance in the past, incorporate position (location-based) element, form user-commodity-businessman's multi-dimensional spatial structure, improved 2 dimension structures of user-commodity in the past.

Commodity forest system structural drawing as shown in Figure 2, this figure gives two classification trees in getting out of the wood for example, and the leaf node of every tree (commodity) all is mapped on X-axis-I (commodity) axle, and be positioned at I (tem)-S (hop) plane, diagram is classified vegetable respectively by effect and period, effect such as be divided into health again under effect, improve looks, refresh oneself; Under the period, be divided into spring, summer, winter.A commodity and B commodity belong to health, C under the effect category and belong to beauty treatment under the effect category; Summer, H that G belonged under the period belong to winter; And D belongs to effect and period classification simultaneously.Can find by Z axle-S (businessman) when finding certain commodity, these commodity have existence in which businessman.

User-commodity-multi-dimensional spatial structure figure of businessman is divided into Z axle-I (businessman), X-axis-S (businessman), this 3 dimension space model of Y-axis-U (user) and is used for representing relation between user-commodity-businessman as shown in Figure 3.

The XZ face, businessman-commodity matrix plane

The information of storage businessman and commodity, even certain bar merchandise news is contained in this businessman, then has node at this commodity axle.The businessman of the attribute that has living space is combined with the commodity that do not have space attribute, and by the XZ plane, namely computable analysis goes out the affiliated relation between businessman and the commodity: comprise that an inventory records exists and a plurality of businessmans; A businessman comprises a plurality of inventory records.

The XY face, user-businessman's matrix plane

According to tuple definition before, can integrate LBS preferably directly by the position relation of this plane computations user and businessman.

The ZY face, commodity-user's matrix plane

If the user collects, browses or commented on certain bar commodity, then there are node in storing commodity and user profile at the commodity axle; Simultaneously can react certain bar inventory records intuitively and which user record be arranged.Algorithm based on collaborative filtering mainly applies to this panel data.

Above-mentioned 3 planes all can be used as the Treatment Analysis that module is independently carried out corresponding data, have well realized the particlized of commending system, and the theoretical foundation of sub-module computing is provided for the commending system backstage.

Basic idea

Algorithm is divided into UserCF, ItemCF, preceding n item is recommended (Top-N) three big modules, by UserCF, the scoring of ItemCF predictive user, recommends Top-N then respectively.

(2) UserCF calculates user's scoring, and its computing formula is:

$p(u,i)=\underset{v\∈S(u,K)\∩N\left(i\right)}{\mathrm{\Σ}}{w}_{\mathrm{uv}}*r_{\mathrm{vi}}$

((u K) comprises K the user the most similar with user u to S to p, and N (i) had behavior (scoring etc.) user set, w to commodity i for u, the i) scoring of the commodity i of expression user u _UvBe based on the user u of user's scoring and the similarity of user v, r _ViThe scoring of the commodity i of representative of consumer v,

And $w_{\mathrm{uv}}=\mathrm{sim}(u,v)=\frac{{\mathrm{\Σ}}_{i\∈I_{\mathrm{uv}}}(R_{u,i}-{\stackrel{\‾}{R}}_u)(R_{v,i}-{\stackrel{\‾}{R}}_v)}{\sqrt{{\mathrm{\Σ}}_{i\∈I_u}{(R_{u,i}-{\stackrel{\‾}{R}}_u)}^2}\sqrt{{\mathrm{\Σ}}_{i\∈I_v}{(R_{v,i}-{\stackrel{\‾}{R}}_v)}^2}}$

Expression user u is to the mean value of all commodity scorings, R _{U, i}The scoring of the commodity i of expression user u; In like manner

Expression user v is to the mean value of all commodity scorings, R _{V, i}The scoring of the commodity i of expression user v

(3) ItemCF calculates user's scoring, and its computing formula is:

$p(u,j)=\underset{i\∈N\left(u\right)\∩S(i,K)}{\mathrm{\Σ}}{w}_{\mathrm{ij}}*r_{\mathrm{ui}}$

Wherein, N (u) is the set of the commodity liked of user, and (i K) is K the set the most similar with commodity i, w to S _JiBe the similarity of commodity j and commodity i, r _UiIt is the scoring of user u commodity i.Namely with the more similar commodity of the historical interested commodity of user, more possiblely in user's recommendation list, obtain higher rank.

③w _ij＝sim(i，j)＝λsim(i，j) _r+(1-λ)sim(i，j) _c；

④ $\mathrm{sim}{(i,j)}_r=\frac{{\mathrm{\Σ}}_{u\∈U}(R_{u,i}-{\stackrel{\‾}{R}}_u)(R_{u,j}-{\stackrel{\‾}{R}}_u)}{\sqrt{{\mathrm{\Σ}}_{u\∈U}{(R_{u,i}-{\stackrel{\‾}{R}}_u)}^2}\sqrt{{\mathrm{\Σ}}_{u\∈U}{(R_{u,j}-{\stackrel{\‾}{R}}_u)}^2}}$

⑤sim(i，j) _c＝∑w*sim(i，j) _c

(4) preceding n the recommendation items (Top-N) of calculated recommendation system

This algorithm mainly applies to the cuisines industry at present, i.e. cuisines information by user collection or scoring, combining geographic location (Location-based) recommends out to have personalized Business Information, and namely Top-N is preceding n the businessman that satisfies consumer taste most that system recommendation is given the user.

Calculate user in predicting scoring rank according to UserCF and ItemCF, and integrate the current geographic position of user and the distance of businessman, provide last result.

The above only is best mode for carrying out the invention, anyly is familiar with those skilled in the art in the technical scope that the present invention discloses, and the simple change of the technical scheme that can obtain apparently or equivalence are replaced and all fallen within the scope of protection of the present invention.

Claims (1)

1. the personalized cuisines recommend method based on the commodity forest system is characterized in that, may further comprise the steps:

(1) this system is carried out the definition of following concept:

Definition 1: classification C={C ₁, C ₂, C ₃... C _n}

Definition 2: forest F={T ₁, T ₂, T ₃... T _n}

Definition 3: tree T=(R, W, TC)

Definition 4: leaf (commodity) I=(Rs, TC ', S, U)

Wherein C is classification set, C _i∈ C, C _iDifference presentation class project, as food materials, type etc., and C _iBe defined as the extensibility generalized list, i.e. C _iComprise more detailed classification below again, as C _i(C _I1(C _I11(C _I111, C _I112...), C _I12...), C _I2(C _I21, C _I22...), C _I3... C _In);

T _iTree in the expression forest; R is the root of this tree, expression the superiors class categories; W represents this tree shared weights in forest; TC represents child's information of this tree, and TC ∈ C; The set of paths of TC ' expression commodity (leaf) in forest, as: TC '={ C _iC _I1C _I11C _I111∪ C _iC _I2C _I21C _I212∪ ... }, and every paths belongs to different trees (T) in the set; Rs represents the set of the root of the affiliated T of this leaf; U represents collection or comments on these commodity user set; S represents to have this commodity businessman set, defines S=(L, A) simultaneously, and L represents this businessman's geographical location information, and A represents other attributes of businessman,

(2) UserCF calculates user's scoring, and its computing formula is:

$p(u,i)=\underset{v\∈S(u,K)\∩N\left(i\right)}{\mathrm{\Σ}}{w}_{\mathrm{uv}}*r_{\mathrm{vi}}$

Wherein, ((u K) comprises K the user the most similar with user u to S to p, and N (i) had behavior (scoring etc.) user set, w to commodity i for u, the i) scoring of the commodity i of expression user u _UvBe the similarity of user u and user v, r _ViThe scoring of the commodity i of representative of consumer v,

And $w_{\mathrm{uv}}=\mathrm{sim}(u,v)=\frac{{\mathrm{\Σ}}_{i\∈I_{\mathrm{uv}}}(R_{u,i}-{\stackrel{\‾}{R}}_u)(R_{v,i}-{\stackrel{\‾}{R}}_v)}{\sqrt{{\mathrm{\Σ}}_{i\∈I_u}{(R_{u,i}-{\stackrel{\‾}{R}}_u)}^2}\sqrt{{\mathrm{\Σ}}_{i\∈I_v}{(R_{v,i}-{\stackrel{\‾}{R}}_v)}^2}}$

(3) ItemCF calculates user's scoring, and its computing formula is:

$p(u,j)=\underset{i\∈N\left(u\right)\∩S(i,K)}{\mathrm{\Σ}}{w}_{\mathrm{ij}}*r_{\mathrm{ui}}$

Wherein, (N (u) is the set of the commodity liked of user to p for u, the j) scoring of expression user u commodity i, and (i K) is K the set the most similar with commodity i, w to S _JiBe the similarity of commodity j and commodity i, r _UiBe the scoring of user u commodity i, w _Ij=sim (i, j)=λ sim (i, j) _r+ (1-λ) sim (i, j) _c

⑥ $\mathrm{sim}{(i,j)}_r=\frac{{\mathrm{\Σ}}_{u\∈U}(R_{u,i}-{\stackrel{\‾}{R}}_u)(R_{u,j}-{\stackrel{\‾}{R}}_u)}{\sqrt{{\mathrm{\Σ}}_{u\∈U}{(R_{u,i}-{\stackrel{\‾}{R}}_u)}^2}\sqrt{{\mathrm{\Σ}}_{u\∈U}{(R_{u,j}-{\stackrel{\‾}{R}}_u)}^2}}$

⑦sim(i，j) _c＝∑w*sim(i，j) _c

(4) preceding n the recommendation items (Top-N) of calculated recommendation system

Calculate user in predicting scoring rank according to UserCF and ItemCF, and integrate the current geographic position of user and the distance of businessman, provide last result.

Images