Summary of the invention
Technical matters: the object of the invention is to provide a kind of solve property set in service recommendation in large scale with user's different recommend method of expectation to service risk when selecting service, the i.e. service recommendation method based on consumer's risk preference.
Technical scheme: the present invention will use for reference the achievement in research of up-to-date service recommendation algorithm, consider that user's request has complicacy and user's risk expectation different characteristics, a kind of service recommendation method based on consumer's risk preference is proposed, in selecting, service NOT-function attribute in conjunction with user's request and consumer's risk preference, provides service recommendation scheme.
The present invention chooses different item attributes according to user to the expection difference of unknown message and recommends, make recommendation results more meet user's demand, this service recommendation method mainly comprises attribute reduction method based on risk partiality and the service recommendation method two parts based on risk partiality.
The compositional model of the attribute reduction method based on risk partiality, comprise user's essential information, user's demand information, the NOT-function attribute information of service, user is to several parts of the score information of Service Properties, and wherein user's essential information comprises user's ID, the preference information of user to risk, and user's demand information comprises user's functional demand and non-functional requirement, the NOT-function attribute information of service comprises the price of service, the information such as QoS attribute.
Service recommendation method key based on risk partiality is user to be dissolved in commending system the risk expectation of unknown message, extract the decision attribute that meets this type of user by the method that user is classified according to the difference of risk partiality, it is combined with the preference attribute that user's request proposes, form effective decision attribute, and then carry out similarity calculating, best item is recommended to user; The processing procedure mainly comprising has user's assorting process, user's request leaching process, user's grade form processing procedure, item attribute yojan process to be recommended, attribute and user's request cohesive process, similarity computation process and optimal service selection course after yojan.
The step of the service recommendation method based on consumer's risk preference is: first by user according to different risk partiality classification of type, user sends request to service register center, service register center is extracted related service and is built scoring possibility interval table, according to the effective attribute of the different Rule Extraction of all types of user, attribute after yojan is composed to power, calculate the recommendation degree of single NOT-function attribute, then according to the recommendation degree of weight calculation service, optimal service is recommended to user, finally complete service recommendation.
Fundamental element in this recommend method, and definition and function comprise:
1) service requester, i.e. user: Service Source is filed a request, there is the entity of selecting optimal service access authorization for resource; This entity comprises fixing ID, the risk partiality classification of entity and the demand of entity to required service functional attribute and non-functional attribute;
2) ISP: the resource request proposing for service requester, the supplier of implementation information service; The attribute that this entity possesses comprises fixing ID, the functional attribute providing and the classification of non-functional attribute, functional attributes demand: one of demand details, the function of the needed service of expression user;
3) user's functional attributes demand: one of demand details, the function of the needed service of expression user;
4) user's NOT-function attribute demand: Service Properties information requirement, represents the demand of user's or a few NOT-function attributes a certain to needed service;
5) Service Properties: Service Properties information, the functional attributes that the service of listing has and NOT-function attribute, comprise price, performance, reliability, availability, security and reputation degree;
6) risk partiality: refer to for realize target, user is at risk taking kind, animus aspect big or small, and risk is exactly a kind of uncertain, and attitude, tendency that user shows in the face of this uncertainty are the imbodies of its risk partiality;
7) the user interval of marking: note a=[a
l, a
u]={ x|a
l≤ x≤a
u, a is scoring interval, a
lfor interval lower limit, a
ufor the interval upper limit;
8) scoring burst length: l
a=a
u-a
lrepresent the length of the interval a of scoring, l
b=b
u-b
lrepresent the length of interval b;
9) the user interval of marking may be spent:
for spending in the interval of marking, in order to judge the probability of two interval number sizes, a, b are scoring interval: a=[a
l, a
u], b=[b
l, b
u], l
a, l
bfor the burst length of marking described in step 8);
10) risk averse type user marks and expects expection rule: suppose a=[a
l, a
u], b=[b
l, b
u] be two scoring researching interval attribute valued, the expection rule of judging so a > b as P (> 1/2 of a>=b) or
P is that the interval of marking described in step 9) may be spent;
11) risk-neutral type user marks and expects expection rule: suppose a=[a
l, a
u], b=[b
l, b
u] be two scoring researching interval attribute valued, (a > is > 1/2 b) as P to judge so a > b expection rule;
12) risk partiality type user marks and expects expection rule: suppose a=[a
l, a
u], b=[b
l, b
u] be two scoring researching interval attribute valued, the expection rule of judging so a > b as P (> 1/2 of a>=b) or
The step of the service recommendation method based on consumer's risk preference is as follows:
The first step: service requester sends a certain services request to service register center, the list L{A of the available this service of service register center return service supplier
1, A
2... A
n, A
1, A
2... A
nrepresent to meet the service of request;
Second step: the preference attitude according to user to risk, user is classified;
The 3rd step: the possible interval table that builds the each attribute scoring of service according to user for each NOT-function attribute marking table;
The 4th step: retrain removal according to user and do not meet the service retraining;
The 5th step: build discrimination matrix according to the type selecting expection rule of consumer's risk preference;
The 6th step: according to discrimination matrix and utilize absorption method to extract effective attribute;
The 7th step: the demand NOT-function attribute proposing according to user forms and recommends attribute in conjunction with the effective attribute extracting;
The 8th step: utilize collaborative filtering to calculate the service recommendation degree of single NOT-function attribute;
The 9th step: compose and weigh for this attribute recommendation degree according to effective attribute proportion;
The tenth step: according to compose power property calculation service recommendation degree, by Top-K algorithm by the service recommendation of recommendation degree value maximum to user.
Beneficial effect: use the method under open network environment, carry out service recommendation in Service Properties collection situation in large scale and have the following advantages:
1, the NOT-function attribute of service is carried out to yojan, improve and recommend efficiency;
2, the user's who does not mark score value possibility is taken into account, building scoring may make score information more complete in interval;
3, the mode that employing classification is processed in the time of effective attribute reduction, retains the characteristic of serving as far as possible, the recommendation deviation that the loss of learning that the yojan of reduction Service Properties brings causes;
4, recommend attribute to compose power, from the advantage discrimination matrix of project scoring collection, extract effective attribute proportion and compose power for this attribute recommendation degree, multiattribute is recommended more reasonable.
Embodiment
The present invention is a kind of scheme of tactic, the existing achievement in research of Reference Services proposed algorithm, the ripe service recommendation model based on collaborative filtering of combination, users ' individualized requirement and user are introduced to recommendation process to the expectation of risk, realize a kind of user's request and expect with user the service recommendation method combining, mainly contain attribute reduction method based on consumer's risk preference and the service recommendation method two parts based on yojan attribute.
Based on a service recommendation method for consumer's risk preference, comprise user ID, consumer's risk type of preferences, demand ID, functional attributes demand, NOT-function attribute demand, quality of service attribute.The processing procedure mainly comprising has user's assorting process, user's request leaching process, user's grade form processing procedure, item attribute yojan process to be recommended, attribute and user's request cohesive process, similarity computation process and optimal service selection course after yojan.
First provide definition and the function thereof of fundamental element in the service recommendation method based on consumer's risk preference below:
1, service requester, i.e. user: Service Source is filed a request, there is the entity of selecting optimal service access authorization for resource; This entity comprises fixing ID, the risk partiality classification of entity and the demand of entity to required service functional attribute and non-functional attribute;
2, ISP: the resource request proposing for service requester, the supplier of implementation information service; The attribute that this entity possesses comprises fixing ID, the level of security of entity and technorati authority, the functional attribute providing and the classification of non-functional attribute;
3, risk partiality: refer to for realize target, user is at the animus of the aspects such as risk taking kind, size.Risk is exactly a kind of uncertain, and attitude, tendency that user shows in the face of this uncertainty are the imbodies of its risk partiality;
4, the user interval of marking: note a=[a
l, a
u]={ x|a
l≤ x≤a
u, a is scoring interval, a
lfor interval lower limit, a
ufor the interval upper limit;
5, scoring burst length: l
a=a
u-a
lrepresent scoring length of an interval degree, l
b=b
u-b
lrepresent the length of interval b;
6, the user interval of marking may be spent:
for spending in the interval of marking, in order to judge the probability of two interval number sizes, a, b are scoring interval: a=[a
l, a
u], b=[b
l, b
u], l
a, l
bit is the burst length of marking described in 5;
7, risk averse type user marks and expects expection rule: suppose a=[a
l, a
u], b=[b
l, b
u] be two scoring researching interval attribute valued, the expection rule of judging so a > b as P (> 1/2 of a>=b) or
P is that the interval of middle scoring described in 6 may be spent;
8, risk-neutral type user marks and expects expection rule: suppose a=[a
l, a
u], b=[b
l, b
u] be two scoring researching interval attribute valued, (a > is > 1/2 b) as P to judge so a > b expection rule;
9, risk partiality type user marks and expects expection rule: suppose a=[a
l, a
u], b=[b
l, b
u] be two scoring researching interval attribute valued, the expection rule of judging so a > b as P (> 1/2 of a>=b) or
10, user mark may interval building process: after user's proposition demand, the list of the available this service of service register center return service supplier.According to user, to the item attribute situation of marking, the user that will not mark takes into account, and user's score value is built to the scoring of item attribute collection with interval form may interval table;
11, item attribute leaching process to be recommended: according to user's risk partiality classification, determine the expectation of user to unknown message, adopt the regular structure advantage discrimination matrix of corresponding expection, extract and effectively recommend accordingly attribute by absorption method;
12, recommendation degree computation process: the recommendation degree of determining single attribute according to collaborative filtering, compose and weigh for this attribute according to valid genus sex ratio in advantage discrimination matrix, determine service recommendation degree, by Top-K algorithm by the service recommendation of recommendation degree value maximum to user.
One, the architecture of the service recommendation method based on consumer's risk preference
1, score information arranges process: obtain user ID, and the type of risk partiality, user provides required function attribute, for example demand film or books.User can provide preference attribute, for example, credit worthiness or certain aspect performance are explicitly called for;
2, user marks and may interval build: according to user, the scoring collection of each attribute is integrated into possible the interval table of each project NOT-function attribute scoring, with favorable comment, in comment, difference is commented to add up, with the form of hundred parts of ratios show (Fig. 1);
3 multiattribute yojan processes: mark and may interval matrix carry out NOT-function attribute reduction according to the user who builds, according to the difference of user's risk partiality, adopt respectively corresponding expection rule, judge each user group's advantage attribute by advantage discrimination, utilize absorption method to realize effective attributes extraction, if user has the demand properties of restriction, be incorporated to the property set after yojan, form final recommendation property set;
4, multiattribute recommendation process: to recommending property set to carry out similarity calculating, calculate single attribute recommendations degree, adopt attribute to compose to weigh formula the recommendation degree weight of calculating each attribute, finally form service recommendation degree, by service recommendation the highest recommendation degree to user.
Two, the workflow of the service recommendation method based on consumer's risk preference
The first step: service requester sends a certain services request to service register center, the list L{A of the available this service of service register center return service supplier
1, A
2... A
n, A
1, A
2... A
nrepresent to meet the service of request;
Second step: the preference attitude according to user to risk, user is classified;
The 3rd step: the possible interval table that builds the each attribute scoring of service according to user for each NOT-function attribute marking table;
The 4th step: retrain removal according to user and do not meet the service retraining;
The 5th step: build discrimination matrix according to the type selecting expection rule of consumer's risk preference;
The 6th step: according to discrimination matrix and utilize absorption method to extract effective attribute;
The 7th step: the demand NOT-function attribute that user is proposed forms and recommends attribute in conjunction with the effective attribute extracting;
The 8th step: utilize collaborative filtering to calculate the service recommendation degree of single NOT-function attribute;
The 9th step: compose and weigh for this attribute recommendation degree according to effective attribute proportion;
The tenth step: according to compose power property calculation service recommendation degree, by Top-K algorithm by the service recommendation of recommendation degree value maximum to user.
Three, the service recommendation method performance evaluation based on consumer's risk preference
This programme proposes a kind of service recommendation method based on consumer's risk preference to the different problem of the expectation of service risk when selection service with user for Service Properties collection is in large scale, mainly comprising that user marks may interval building process, multiattribute yojan process, multiattribute recommendation process, according to user's demand with to the difference of risk expectation, the NOT-function property set of service is effectively extracted, obtain the information that more meets user's request and recommend for user.Make a concrete analysis of as follows:
1, when rating matrix builds, the crowd of not scoring is taken into account, the unknown scoring is adopted to interval form, form the possible interval table of each project NOT-function attribute scoring, the possible situation of the disappearance part of score information is taken into account, make score information more complete.
2, aspect attribute reduction, considered the difference that user expects for unknown message, adopt corresponding algorithm and the restriction to attribute demand in conjunction with user self according to the difference of consumer's risk preference, reduce and used part attribute to recommend brought loss of learning, promoted the efficiency of recommending.
3, aspect recommendation, to composing power for the attribute of recommending after yojan, extract effective attribute according to the advantage discrimination matrix building by all types of user respective rule in 2, the ratio that each attribute is occurred in advantage discrimination matrix is as attribute weights, the thought that has adopted classification is the demand that recommendation results meets user more.
For the service recommendation method based on consumer's risk preference of the present invention is described, we provide following preferred example, the service recommendation method embodiment based on consumer's risk preference under the more detailed huge environment of description Service Properties collection.
Suppose the accommodation service of user to network request Beijing the best, and require prestige favorable comment degree to be greater than 60%, risk type selecting is evaded, and concrete steps are expressed as follows:
The first step: service requester sends a certain services request to service register center, the list L{A of the available this service of service register center return service supplier
1, A
2... A
n, A
1, A
2... A
nrepresent Beijing accommodation service;
Second step: the preference attitude according to user to risk, determine that user is risk averse type;
The 3rd step: the possible interval table (as shown in Figure 1) that builds the each attribute scoring of service according to user for each NOT-function attribute marking table;
The 4th step: delete scoring interval table according to user's NOT-function attribute demand, user proposes credit worthiness positive rating must be greater than 60%, according to user's request, will serve A
5remove;
The 5th step: build discrimination matrix according to the type selecting expection rule of consumer's risk preference;
The 6th step: according to discrimination matrix and utilize absorption method to extract effective attribute, extract effective information by absorption method from discrimination matrix, can obtain { C
3, C
7, C
1effectively to recommend property set, and hotel is comprehensive, cost performance, and security is effective attribute;
The 7th step: the demand proposing according to user, NOT-function attribute (credit worthiness) is incorporated to recommendation property set as user's request attribute, finally recommending property set is { C
1, C
2, C
3, C
7;
The 8th step: utilize collaborative filtering to calculate the service recommendation degree of single NOT-function attribute, obtain respectively C
1, C
2, C
3, C
7recommendation degree;
The 9th step: compose and weigh for this attribute recommendation degree according to effective attribute proportion in discrimination matrix;
The tenth step: according to compose power property calculation service recommendation degree, by Top-K algorithm by the service recommendation of recommendation degree value maximum to user.