CN103426102A - Commodity feature recommending method based on body classification - Google Patents

Abstract

The invention discloses a commodity feature recommending method based on body classification. The commodity feature recommending method comprises: (1) building a commodity feature body library; (2) building a user-commodity two-dimensional grading matrix; (3) building a user-commodity feature two-dimensional grading matrix; (4) performing recommended commodity set calculation according to a commodity feature recommending algorithm; (5) classifying the calculated recommended commodity set, and selecting K commodities with highest grade evaluated by each classification selecting user to serve as a recommended commodity set of this kind. The commodity feature recommending method based on body classification can perform evaluation calculation on features of the commodities, predication evaluation can be calculated if one commodity is registered in the system and recorded in the commodity feature body library, and the problem of new commodities can be solved. Further, the commodity feature recommending method fully utilizes level features of the body data library, and can easily achieve level inquiring and feature inquiring of commodity classification compared with a traditional data library.

Description

Product features recommend method based on Ontology

Technical field

The present invention relates to a kind of product features recommend method based on Ontology, belong to Computer Applied Technology.

Background technology

Body is a kind of knowledge base, to the description of field things, with Semantic, can be described as another kind of database.The target of body is the knowledge of catching association area, and the common understanding to this domain knowledge is provided, and determines in this field the vocabulary (term) of common approval, and provides the clearly definition of mutual relationship between these vocabulary and vocabulary from the formalization pattern of different levels.Generally speaking, the structure body can be realized knowledge sharing to a certain degree and reuse, and the ability that improves system communication, interoperability, reliability.

The implication of data is exactly semantic.Briefly, data are with regard to is-symbol.Data itself are without any meaning, and the data that only are endowed implication can be used, and at this time data just transform for information, and the implication of data is exactly semantic.

Semanteme has the territoriality feature, and the semanteme that does not belong to any field is non-existent.Semantic Heterogeneous refers to same thing existing difference on explaining, also just is presented as the difference that same thing is understood in different field.For computer science, semanteme generally refers to the explanation that the user is used for describing the computer representation (being symbol) of real world for those, and namely the user is used for contacting the approach of computer representation and real world.Semanteme is the explanation to the data symbol, and grammer is the definition for the organization regulation between these symbols and structural relation.For the information integration field, data (do not exist or implicit destructuring and semi-structured data for pattern by pattern often, often need to before integrated, define their pattern) tissue, the access of data also obtains by acting on pattern, at this moment semanteme is exactly the implication of finger print formula element (for example class, attribute, constraint etc.), and grammer is the structure of schema elements.

Semantic net is the Chinese of Semantic Web.Semantic net is exactly the network that can be judged according to semanteme.Briefly, semantic net is a kind of intelligent network that can understand human language, and it not only can understand the mankind's language, but also can make the people become light as interpersonal the interchange with the interchange between computer.

Semantic net is the imagination to future network, and in such network, information all has been endowed clear and definite implication, and machine can automatically be processed and integrated online available information.Semantic net defines the tag format of customization and carrys out expression data with the dirigibility of RDF with XML, and next step needs is exactly that the netspeak (such as OWL) of a kind of Ontology is described the clear and definite implication of the term in network documentation and the relation between them.

The architecture of semantic net is built, research to this architecture in current international coverage does not also form a gratifying tight logical description and theoretical system, Chinese scholar is also just done concise and to the point introduction on the basis of research abroad to this architecture, does not also form the elaboration of system.

The realization of semantic net needs the support of three large gordian techniquies: XML, RDF and Ontology.XML(eXtensible Marked Language, i.e. extend markup language) can allow the informant as required, self-defining mark and attribute-name, thus make the structure of XML file can complicatedly arrive any degree.It has good data memory format and extensibility, highly structural and is convenient to the advantages such as Internet Transmission, add numerous types of data and verification scheme that its distinctive NS mechanism and XML Schema support, make it become one of gordian technique of semantic net.The current discussion about the semantic net gordian technique mainly concentrates on it RDF and Ontology.

The recommended technology of main flow is collaborative filtering (Collaborative filtering) proposed algorithm at present.Its principle is very simple, is exactly the preference to article or information according to the user, finds the correlativity of article or content itself, or finds user's correlativity, and then is recommended based on these relevances.Recommendation based on collaborative filtering can be divided into three subclasses: the recommendation based on the user (User-based Recommendation), project-based recommendation (Item-based Recommendation) and the recommendation based on model (Model-based Recommendation).

(1) collaborative filtering recommending based on the user

The ultimate principle of the collaborative filtering recommending based on the user is that the preference according to all users to article or information, find " neighbours " customer group similar with preference to active user's taste, in general application, is to adopt the algorithm that calculates " K-neighbours "; Then, the historical preference information based on this K neighbour, for the active user is recommended.Fig. 1 has provided schematic diagram.Suppose that user A likes article A, article C, user B likes article B, and user C likes article A, article C and article D; From these users' historical preference information, we can find that the taste of user A and user C and preference are like comparing class, user C also likes article D simultaneously, and we can infer that user A may also like article D so, therefore article D can be recommended to user A.

Collaborative filtering recommending based on user mechanism and be all to calculate user's similarity based on demographic recommendation mechanisms, and based on " neighbours " customer group calculated recommendation, but their differences are how to calculate user's similarity, only consider user's feature itself based on demographic mechanism, and the collaborative filtering based on user mechanism is calculated user's similarity on the data of user's historical preference, its basic assumption is to like the user of similar articles that identical or similar taste and preference may be arranged.

(2) project-based collaborative filtering recommending

The ultimate principle of project-based collaborative filtering recommending is also similar, just say that it uses the preference of all users to article or information, find the similarity between article and article, then according to user's historical preference information, similar article are recommended to the user, and Fig. 2 has well annotated its ultimate principle.Suppose that user A likes article A and article C, user B likes article A, article B and article C, user C likes article A, in the time of can analyzing article A and article C from these users' history hobby like comparing class, like the people of article A all to like article C, based on these data, can infer that user C probably also likes article C, so system can be recommended article C user C.

With say above similar, project-based collaborative filtering recommending and content-based recommendation in fact all are based on the prediction of article similarity and recommend, the method that just similarity is calculated is different, and the former infers from the preference of user's history, and the latter is based on the attributive character information of article itself.

Be both collaborative filtering, based on the user with should How to choose in based on two strategies of project? project-based collaborative filtering recommending mechanism is a kind of strategy that Amazon improves on the mechanism based on the user in fact, because in most Web website, the number of article is the quantity that is far smaller than the user, and the number of article and similarity relatively stable, simultaneously project-based mechanism is more better than the real-time based on the user.But scene that neither be all is all such situation, it is contemplated that in some news commending systems, article perhaps, namely the number of news may be greater than user's number, and the renewal degree of news also have very fast, so its likeness in form degree is still unstable.So, in fact can find out, recommend tactful selection in fact with concrete application scenarios, very large relation to be arranged.

(3) collaborative filtering recommending based on model

Collaborative filtering recommending based on model just is based on the user preference information of sample, trains a recommended models, then according to the information of real-time user preferences, is predicted calculated recommendation.

The commending system major advantage of collaborative filtering has: do not need article or user are carried out to strict modeling, and do not require that the description of article is machine understandable, so this method is also field independence.The recommendation that this method is calculated is open, can share other people experience, well supports the user to find potential interest preference.

But it is also imperfect: the core of this algorithm is based on historical data, so new article and new user are had to the problem of " cold start-up ".The effect of recommending depend on the historical preference data of user the number and accuracy.In addition, owing to take historical data as basis, after crawl and modeling user's preference, be difficult to revise or develop according to user's use, thereby cause this method underaction.

Summary of the invention

Goal of the invention: in order to overcome the deficiencies in the prior art, the invention provides a kind of product features recommend method based on Ontology, be intended to the characteristics of the level by utilizing body to have, go deep into the digging user propensity to consume, thereby improve, recommend precision.

Technical scheme: for achieving the above object, the technical solution used in the present invention is:

Product features recommend method based on Ontology comprises the following steps that order is carried out:

(1) import the merchandise news in ERP, set up the product features ontology library;

(2) import user in the ERP record of doing shopping, according to implicit expression score calculation formula, build user-commodity two dimension rating matrix;

(3) access products feature noumenon storehouse, all commodity that the traversal user bought, inquire about the product features set of these commodity, generate one group of proper vector about product features, in conjunction with user in user-commodity two dimension rating matrix for the scoring of commodity, COMPREHENSIVE CALCULATING goes out the scoring of user for specific product features, thereby constructs user-product features two dimension rating matrix;

(4) set user-commodity two dimension rating matrix and user-product features two dimension rating matrix, carry out the calculating of Recommendations set according to the product features proposed algorithm;

(5) access products feature noumenon storehouse again, classified to the Recommendations set calculated, and is the Recommendations set of each classifying and selecting user K part commodity the highest to the commodity scoring as such.

Described step (2) specifically comprises the steps:

(21) import user in the ERP record of doing shopping;

(22) purchaser record for commodity j according to user i, calculate the purchase frequency F of user i for commodity j _IjFor:

F _ij＝Num/day

Wherein: day is the time interval of user i initial purchase commodity j to last purchase commodity j, and unit is sky; The total quantity of the commodity j that Num user i buys in during day;

(23) the average purchase frequency Fv of contrast commodity j _j, calculate the weighted scoring R of user i for commodity j _IjFor:

R _ij＝F _ij-Fv _j

Fv _j＝∑F _ij/N

Wherein: N is the number of users of buying commodity j; Weighted scoring R _IjMean the interest thickness of user i for commodity j; If user i is the purchase frequency F to commodity j _IjBe greater than the average purchase frequency Fv of commodity j _j, weighted scoring R _IjFor for just, otherwise for negative; Weighted scoring R _IjReacted the demand (on the other hand also be interest and concern) of user i for commodity j; Weighted scoring R _IjLarger, demand is larger, shows that user i has very large interest to the even similar commodity of commodity j;

(24) calculate the implicit expression score value IR of user i for commodity j according to implicit expression score calculation formula _IjFor:

IR _ij＝a+R _ij

Wherein: a gives the initial value of user i for the interest thickness of commodity j;

(25) obtaining user i to the user of all commodity-commodity two dimension rating matrix is:

$R1=\left[\begin{array}{cccc}{\mathrm{IR}}_{11}& {\mathrm{IR}}_{12}& ...& {\mathrm{IR}}_{1n}\\ {\mathrm{IR}}_{21}& {\mathrm{IR}}_{22}& ...& {\mathrm{IR}}_{2n}\\ ...& ...& ...& ...\\ {\mathrm{IR}}_{i1}& {\mathrm{IR}}_{i2}& ...& {\mathrm{IR}}_{\mathrm{in}}\\ ...& ...& ...& ...\\ {\mathrm{IR}}_{m1}& {\mathrm{IR}}_{m2}& ...& {\mathrm{IR}}_{\mathrm{mn}}\end{array}\right]$

Wherein: m is the total number of users amount, and n is the commodity total quantity.

Preferably, in described step (24), if user i bought commodity j, set a=3 (5 minutes systems).

Described step (3) specifically comprises the steps:

(31) access products feature noumenon storehouse, all commodity that the traversal user bought, inquire about the product features set of these commodity, generates one group of proper vector about product features;

(32) adopt the bitmap mode, proper vector is converted into to the bitmap proper vector of 0/1 form, described bitmap mode is: the feature of all commodity is made to a feature sum vector { t ₁, t ₂, t ₃, t ₄..., t _k..., t _{N '}, if commodity j has one of them or two above features, by the flag of this one or more feature, be 1, no person is labeled as 0; Suppose that commodity j has t ₁And t ₃Two features, the proper vector T of commodity j _jFor:

T _j＝{1,0,1,0,…,0,…,0}

(33) note commodity j has is characterized as t _k, the bitmap proper vector of set commodity j and user-commodity two dimension rating matrix, counting user i is for comprising feature t _kThe summation of implicit expression score value of commodity, get its mean value FR _IkAs user i to feature t _kThe implicit expression score value; Calculated amount herein is larger, can calculate in advance on backstage, then is saved in a tables of data user's of each line display proper vector;

(34) obtaining all users to the user of all product features-product features two dimension rating matrix is:

$R2=\left[\begin{array}{cccc}{\mathrm{FR}}_{11}& {\mathrm{FR}}_{12}& ...& {\mathrm{FR}}_{{1n}^{\′}}\\ {\mathrm{FR}}_{21}& {\mathrm{FR}}_{22}& ...& {\mathrm{FR}}_{{2n}^{\′}}\\ ...& ...& ...& ...\\ {\mathrm{FR}}_{i1}& {\mathrm{FR}}_{i2}& ...& {\mathrm{FR}}_{{\mathrm{in}}^{\′}}\\ ...& ...& ...& ...\\ F{R}_{m1}& {\mathrm{FR}}_{m2}& ...& {\mathrm{FR}}_{{\mathrm{mn}}^{\′}}\end{array}\right]$

Wherein: m is the total number of users amount, and n ' is the product features total quantity.

Described step (4) is specially: set user-commodity two dimension rating matrix and user-product features two dimension rating matrix, by product features vector sum user characteristics vector (a line that user-commodity rating matrix takes out according to user ID, for known quantity, what reflect is the scoring (fancy grade) of user for all product features, be not limited only to commodity) get common factor, obtain common feature, if common feature is not empty, according to the user, for the scoring of the common feature of obtaining, carry out superposition calculation (is common feature herein, and be not all product features vectors of these commodity, because according to all product features vectors, commodity are marked and are based upon all product features vectors that the user contains this commodity bundle and carried out on the basis of scoring, if this user is not marked to the part product features vector of these commodity, can't carry out superposition calculation), calculate a scoring of the prediction about these commodity, then by the prediction scoring calculated, commodity are sorted, produce the Recommendations set.

Described step (5) is specially: all Recommendations in the set of traversal Recommendations, commodity key word access products feature noumenon storehouse according to input, return to merchandise classification, according to the merchandise classification returned to Recommendations set classified, be the Recommendations set of each classifying and selecting user K part commodity the highest to commodity scorings as such.

Beneficial effect: traditional proposed algorithm, such as collaborative filtering CF proposed algorithm, all with field independence, although the versatility of recommending is very strong, but precision is not very high, it recommends precision to improve along with the reduction of user-sparse property of commodity two dimension rating matrix, therefore cause the new commodity problem, for the commodity of new warehouse-in or the commodity of the purchased mistake of user never, can't calculate the prediction scoring, thereby produce the unexpected winner commodity; Product features recommend method based on Ontology provided by the invention, directly for the feature of commodity, predicted score calculation, as long as commodity sign in in system, in the product features ontology library, had record, can calculate its prediction scoring, solve the new commodity problem; The inventive method has fully been used the level characteristic of ontology database in addition, and traditional database, more easily realized that the level of merchandise classification is inquired about and characteristic query relatively.

The accompanying drawing explanation

The basic principle schematic that Fig. 1 is the collaborative filtering recommending mechanism based on the user;

The basic principle schematic that Fig. 2 is project-based collaborative filtering recommending mechanism;

Fig. 3 is process flow diagram of the present invention;

The product features body that Fig. 4 is the retail trade simulated in embodiment;

Fig. 5 is the process of the proper vector of instant noodles of setting up in embodiment;

The bitmap proper vector that Fig. 6 is instant noodles in embodiment;

The search procedure that Fig. 7 is merchandise classification;

Fig. 8 is the classified commodity recommendation process.

Embodiment

Below in conjunction with accompanying drawing, the present invention is further described.

A kind of product features recommend method based on Ontology as shown in Figure 3, comprises the following steps that order is carried out:

(1) import the merchandise news in ERP, set up the product features ontology library:

Body is to be based upon on the model of class-based tlv triple, i.e. subject, object and predicate, characterize two objects and between certain contact; Be illustrated in figure 4 the product features body of the retail trade of simulating in this example, here member the classification of instant noodles as showing.

(2) import user in the ERP record of doing shopping, according to implicit expression score calculation formula, build user-commodity two dimension rating matrix:

(21) import user in the ERP record of doing shopping;

(22) purchaser record for commodity j according to user i, calculate the purchase frequency F of user i for commodity j _IjFor:

F _ij＝Num/day

Wherein: day is the time interval of user i initial purchase commodity j to last purchase commodity j, and unit is sky; The total quantity of the commodity j that Num user i buys in during day;

(23) the average purchase frequency Fv of contrast commodity j _j, calculate the weighted scoring R of user i for commodity j _IjFor:

R _ij＝F _ij-Fv _j

Fv _j＝∑F _ij/N

Wherein: N is the number of users of buying commodity j; Weighted scoring R _IjMean the interest thickness of user i for commodity j; If user i is the purchase frequency F to commodity j _IjBe greater than the average purchase frequency Fv of commodity j _j, weighted scoring R _IjFor for just, otherwise for negative; Weighted scoring R _IjReacted the demand (on the other hand also be interest and concern) of user i for commodity j; Weighted scoring R _IjLarger, demand is larger, shows that user i has very large interest to the even similar commodity of commodity j;

(24) calculate the implicit expression score value IR of user i for commodity j according to implicit expression score calculation formula _IjFor:

IR _ij＝a+R _ij

Wherein: a gives the initial value of user i for the interest thickness of commodity j, if user i bought commodity j, means that user i is interesting to commodity j, sets a=3 (5 minutes systems);

(25) obtaining user i to the user of all commodity-commodity two dimension rating matrix is:

$R1=\left[\begin{array}{cccc}{\mathrm{IR}}_{11}& {\mathrm{IR}}_{12}& ...& {\mathrm{IR}}_{1n}\\ {\mathrm{IR}}_{21}& {\mathrm{IR}}_{22}& ...& {\mathrm{IR}}_{2n}\\ ...& ...& ...& ...\\ {\mathrm{IR}}_{i1}& {\mathrm{IR}}_{i2}& ...& {\mathrm{IR}}_{\mathrm{in}}\\ ...& ...& ...& ...\\ {\mathrm{IR}}_{m1}& {\mathrm{IR}}_{m2}& ...& {\mathrm{IR}}_{\mathrm{mn}}\end{array}\right]$

Wherein: m is the total number of users amount, and n is the commodity total quantity.

(3) access products feature noumenon storehouse, all commodity that the traversal user bought, inquire about the product features set of these commodity, generate one group of proper vector about product features, in conjunction with user in user-commodity two dimension rating matrix for the scoring of commodity, COMPREHENSIVE CALCULATING goes out the scoring of user for specific product features, thereby constructs user-product features two dimension rating matrix:

(31) access products feature noumenon storehouse, all commodity that the traversal user bought, inquire about the product features set of these commodity, generates one group of proper vector about product features; As shown in Figure 5, be the process of the proper vector of setting up instant noodles in this example: inquire about respectively its brand, taste, taste deflection and edible way 4 category features according to the trade name of input in the product features ontology library, obtain a stack features vector of these instant noodles;

(32) adopt the bitmap mode, proper vector is converted into to the bitmap proper vector of 0/1 form, described bitmap mode is: the feature of all commodity is made to a feature sum vector { t ₁, t ₂, t ₃, t ₄..., t _k..., t _{N '}, if commodity j has one of them or two above features, by the flag of this one or more feature, be 1, no person is labeled as 0; Suppose that commodity j has t ₁And t ₃Two features, the proper vector T of commodity j _jFor:

T _j＝{1,0,1,0,…,0,…,0}

As shown in Figure 6, be the bitmap proper vector of instant noodles in this example;

(33) note commodity j has is characterized as t _k, the bitmap proper vector of set commodity j and user-commodity two dimension rating matrix, counting user i is for comprising feature t _kThe summation of implicit expression score value of commodity, get its mean value FR _IkAs user i to feature t _kThe implicit expression score value; Calculated amount herein is larger, can calculate in advance on backstage, then is saved in a tables of data user's of each line display proper vector;

(34) obtaining all users to the user of all product features-product features two dimension rating matrix is:

$R2=\left[\begin{array}{cccc}{\mathrm{FR}}_{11}& {\mathrm{FR}}_{12}& ...& {\mathrm{FR}}_{{1n}^{\′}}\\ {\mathrm{FR}}_{21}& {\mathrm{FR}}_{22}& ...& {\mathrm{FR}}_{{2n}^{\′}}\\ ...& ...& ...& ...\\ F{R}_{i1}& {\mathrm{FR}}_{i2}& ...& F{R}_{{\mathrm{in}}^{\′}}\\ ...& ...& ...& ...\\ {\mathrm{FR}}_{m1}& {\mathrm{FR}}_{m2}& ...& {\mathrm{FR}}_{{\mathrm{mn}}^{\′}}\end{array}\right]$

Wherein: m is the total number of users amount, and n ' is the product features total quantity.

(4) set user-commodity two dimension rating matrix and user-product features two dimension rating matrix, carry out the calculating of Recommendations set according to the product features proposed algorithm:

Set user-commodity two dimension rating matrix and user-product features two dimension rating matrix, by product features vector sum user characteristics to measuring common factor, obtain common feature, if common feature is not empty, according to the user, the scoring of all product features vectors of these commodity is superposeed, calculate a scoring of the prediction about these commodity, then by the prediction scoring calculated, commodity are sorted, produce the Recommendations set.

(5) access products feature noumenon storehouse again, classified to the Recommendations set calculated, and is the Recommendations set of each classifying and selecting user K part commodity the highest to the commodity scoring as such:

All Recommendations in the set of traversal Recommendations, as shown in Figure 7, according to the commodity key word access products feature noumenon storehouse of input, return to merchandise classification; As shown in Figure 8, according to the merchandise classification returned to Recommendations set classified, be the Recommendations set of each classifying and selecting user K part commodity the highest to commodity scorings as such.

The above is only the preferred embodiment of the present invention; be noted that for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (6)

1. the product features recommend method based on Ontology is characterized in that: comprise the following steps that order is carried out:

(1) import the merchandise news in ERP, set up the product features ontology library;

(2) import user in the ERP record of doing shopping, according to implicit expression score calculation formula, build user-commodity two dimension rating matrix;

(3) access products feature noumenon storehouse, all commodity that the traversal user bought, inquire about the product features set of these commodity, generate one group of proper vector about product features, in conjunction with user in user-commodity two dimension rating matrix for the scoring of commodity, COMPREHENSIVE CALCULATING goes out the scoring of user for specific product features, thereby constructs user-product features two dimension rating matrix;

(4) set user-commodity two dimension rating matrix and user-product features two dimension rating matrix, carry out the calculating of Recommendations set according to the product features proposed algorithm;

(5) access products feature noumenon storehouse again, classified to the Recommendations set calculated, and is the Recommendations set of each classifying and selecting user K part commodity the highest to the commodity scoring as such.

2. the product features recommend method based on Ontology according to claim 1, it is characterized in that: described step (2) specifically comprises the steps:

(21) import user in the ERP record of doing shopping;

(22) purchaser record for commodity j according to user i, calculate the purchase frequency F of user i for commodity j _IjFor:

F _ij＝Num/day

Wherein: day is the time interval of user i initial purchase commodity j to last purchase commodity j, and unit is sky; The total quantity of the commodity j that Num user i buys in during day;

(23) the average purchase frequency Fv of contrast commodity j _j, calculate the weighted scoring R of user i for commodity j _IjFor:

R _ij＝F _ij-Fv _j

Fv _j＝∑F _ij/N

Wherein: N is the number of users of buying commodity j; Weighted scoring R _IjMean the interest thickness of user i for commodity j;

(24) calculate the implicit expression score value IR of user i for commodity j according to implicit expression score calculation formula _IjFor:

IR _ij＝a+R _ij

Wherein: a gives the initial value of user i for the interest thickness of commodity j;

(25) obtaining user i to the user of all commodity-commodity two dimension rating matrix is:

$R1=\left[\begin{array}{cccc}{\mathrm{IR}}_{11}& {\mathrm{IR}}_{12}& ...& {\mathrm{IR}}_{1n}\\ {\mathrm{IR}}_{21}& {\mathrm{IR}}_{22}& ...& {\mathrm{IR}}_{2n}\\ ...& ...& ...& ...\\ {\mathrm{IR}}_{i1}& {\mathrm{IR}}_{i2}& ...& {\mathrm{IR}}_{\mathrm{in}}\\ ...& ...& ...& ...\\ {\mathrm{IR}}_{m1}& {\mathrm{IR}}_{m2}& ...& {\mathrm{IR}}_{\mathrm{mn}}\end{array}\right]$

Wherein: m is the total number of users amount, and n is the commodity total quantity.

3. the product features recommend method based on Ontology according to claim 2, is characterized in that: in described step (24), if user i bought commodity j, set a=3 (5 minutes systems).

4. the product features recommend method based on Ontology according to claim 1, it is characterized in that: described step (3) specifically comprises the steps:

(31) access products feature noumenon storehouse, all commodity that the traversal user bought, inquire about the product features set of these commodity, generates one group of proper vector about product features;

(32) adopt the bitmap mode, proper vector is converted into to the bitmap proper vector of 0/1 form, described bitmap mode is: the feature of all commodity is made to a feature sum vector { t ₁, t ₂, t ₃, t ₄..., t _k..., t _{N '}, if commodity j has one of them or two above features, by the flag of this one or more feature, be 1, no person is labeled as 0;

(33) note commodity j has is characterized as t _k, the bitmap proper vector of set commodity j and user-commodity two dimension rating matrix, counting user i is for comprising feature t _kThe summation of implicit expression score value of commodity, get its mean value FR _IkAs user i to feature t _kThe implicit expression score value;

(34) obtaining all users to the user of all product features-product features two dimension rating matrix is:

$R2=\left[\begin{array}{cccc}{\mathrm{FR}}_{11}& {\mathrm{FR}}_{12}& ...& {\mathrm{FR}}_{{1n}^{\′}}\\ {\mathrm{FR}}_{21}& {\mathrm{FR}}_{22}& ...& {\mathrm{FR}}_{{2n}^{\′}}\\ ...& ...& ...& ...\\ {\mathrm{FR}}_{i1}& {\mathrm{FR}}_{i2}& ...& {\mathrm{FR}}_{{\mathrm{in}}^{\′}}\\ ...& ...& ...& ...\\ F{R}_{m1}& {\mathrm{FR}}_{m2}& ...& {\mathrm{FR}}_{{\mathrm{mn}}^{\′}}\end{array}\right]$

Wherein: m is the total number of users amount, and n ' is the product features total quantity.

5. the product features recommend method based on Ontology according to claim 1, it is characterized in that: described step (4) is specially: set user-commodity two dimension rating matrix and user-product features two dimension rating matrix, by product features vector sum user characteristics to measuring common factor, obtain common feature, if common feature is not empty, according to the user, for the scoring of the common feature of obtaining, carry out superposition calculation, calculate a scoring of the prediction about these commodity, then by the prediction scoring calculated, commodity are sorted, produced the Recommendations set.

6. the product features recommend method based on Ontology according to claim 1, it is characterized in that: described step (5) is specially: all Recommendations in the set of traversal Recommendations, commodity key word access products feature noumenon storehouse according to input, return to merchandise classification, according to the merchandise classification returned to Recommendations set classified, be the Recommendations set of each classifying and selecting user K part commodity the highest to commodity scorings as such.

Images