CN102650997A - Element recommending method and device - Google Patents

Abstract

The invention embodiment discloses an element recommending method and an element recommending device. The method includes the steps as follows: clustering multiple users and determining the category to which each user belongs, wherein the users include multiple elements; obtaining category frequency reciprocals of all elements belonging to the current user according to the clustering result, wherein, when the distribution of the elements in the overall category is more centralized, category frequency reciprocal values of the elements are larger; and further, computing recommending values of all element of the current user according to the category frequency reciprocals, and recommending to the user according to the recommending values, wherein, when the category frequency reciprocal values of the elements are larger, the recommending values of the elements of the current user are higher. Through adopting the method and the device both disclosed by the invention, the recommendation of popular songs can be efficiently suppressed, and on the premise of keeping the accuracy unchanged fundamentally, the phenomenon that a small part of users are disgusted with the popular songs can be eliminated on one hand, and on the other hand, more non-popular songs are recommended to the users.

Description

A kind of element recommend method and device

Technical field

The present invention relates to mutual field, internet, relate in particular to a kind of element recommend method and device.

Background technology

Along with the continuous development of Internet service, it provides the more and more users interactive service.Such as, user search is professional, popular song recommendations, popular video recommendation, hot news recommendation etc.In existing commending system, like the social network structure of the CF commending system of Last.fm, there is the trend that strengthens popular singer/song through discovering it, this can cause the singer who is in the long-tail distribution to can not get due recommendation.And people find also that under study for action the user does not have and must concern with degree of accuracy the satisfaction of commending system, and degree of accuracy can not be as the sole criterion of assessment commending system performance quality.

The music recommend system of collaborative filtering of the prior art; Collaborative filtering recommending algorithm as based on cluster at first gathers into several types according to the similarity between the user; Each type is made up of several relevant users, and each user has own scoring to some clauses and subclauses (like elements such as songs).When generating recommendation list for user u, the class c that at first finds this user to belong to, the clauses and subclauses that comprise inside type of the finding c then; If the similarity of user u and type c be sim (u, c), and the average score of clauses and subclauses t in class c is ave (t; C); (u, c) (t c) calculates a mark to * ave so each clauses and subclauses all to be pressed sim.If a user belongs to a plurality of types, just mark is added up by class.Based on mark clauses and subclauses are sorted at last, generate user's recommendation list.

But, when adopting said method to recommend, can cause heat song (meaning i.e. popular especially song on network, also claims the network song) to be strengthened, and the most of user of heat song listened.Though from the degree of accuracy aspect,, recommends right ratio to recommend the ratio of mistake big, and the assessment of degree of accuracy also is on the visible clauses and subclauses of user, to carry out, so overall degree of accuracy can be higher relatively because heat song is welcome by much human.But from the effect of reality, most people had listened these heat songs, recommended the heat song not had much meanings for the user again; And for those users that dislike the heat song, can dislike very much these heat songs.

Summary of the invention

Embodiment of the invention technical matters to be solved is, a kind of element recommend method and device are provided.The effectively recommendation of suppressing heat song is keeping under the constant basically situation of degree of accuracy, can eliminate the phenomenon of few users to heat song dislike on the one hand, can recommend the song of more how non-hot topic on the other hand for the user.

In order to solve the problems of the technologies described above, the embodiment of the invention provides a kind of element recommend method, comprising:

A plurality of users are carried out cluster, confirm the affiliated classification of each user, comprise a plurality of elements among the said user;

The classification frequency of obtaining each element that belongs to the active user according to cluster result is reciprocal, and wherein, element is concentrated more in the distribution of total classification, and then the classification frequency reciprocal value of this element is big more;

According to the said classification frequency recommendation of calculating each element of said active user reciprocal, recommend to said user according to said recommendation, wherein, the recommendation of this element of the bigger then active user of classification frequency reciprocal value of element is high more.

Wherein, said classification frequency available following formula reciprocal calculates:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, ICF _iThe classification frequency of expression element i is reciprocal, | C| representes total classification number, NC _iThe classification number that expression element i occurred, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).

Accordingly, recommendation can use following formula to calculate:

$s_i=\underset{c\∈C\left(u\right)}{\mathrm{\Σ}}\mathrm{sim}(u,c)*\mathrm{COUNT}\left(\mathrm{ci}\right)*\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, S _iExpression element i recommends the recommendation of user u, the set of the classification under C (u) the expression user u, sim (u, the c) similarity of expression user u and classification c, COUNT (c _i) number of times that in current classification, occurs of expression element i.

Saidly a plurality of users are carried out cluster comprise: it is reciprocal to obtain user's frequency, according to said user's frequency inverse the user is carried out cluster, and wherein, element is concentrated more in total user's distribution, and then user's frequency reciprocal value of this element is big more.

Then user's frequency available following formula reciprocal calculates:

${\mathrm{IUF}}_i=\log \frac{n}{n_i}$

Wherein, IUF _iUser's frequency of element i among the expression user is reciprocal, and n representes total number of users, n _iTotal number of users under the expression element i.

Accordingly, then said user can be expressed as:

$u=\left\{w_i{c}_i\right\}=\{c_i{f}_i*\log \frac{n}{n_i}\}$

Wherein, u representes user u, w _iThe weight of element i among the expression user u, c _iExpression element i, f _iThe frequency that expression element i occurs in set u.

Accordingly, the embodiment of the invention also provides a kind of element recommendation apparatus, and said device comprises:

The cluster module is used for a plurality of users are carried out cluster, confirms the affiliated classification of each user, comprises a plurality of elements among the said user;

Classification frequency module reciprocal, the classification frequency that is used for obtaining according to cluster result each element that belongs to the active user is reciprocal, and wherein, element is concentrated more in the distribution of total classification, and then the classification frequency reciprocal value of this element is big more;

The recommendation computing module; Be used for recommendation according to each element of the said classification frequency said active user of calculating reciprocal; Recommend to said user according to said recommendation, wherein, the recommendation of this element of the bigger then active user of classification frequency reciprocal value of element is high more.

Wherein, classification frequency module reciprocal can be used for calculating classification frequency inverse according to following formula:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, ICF _iThe classification frequency of expression element i is reciprocal, | C| representes total classification number, NC _iThe classification number that expression element i occurred, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).

The recommendation computing module can be used for according to following formula calculated recommendation value:

$s_i=\underset{c\∈C\left(u\right)}{\mathrm{\Σ}}\mathrm{sim}(u,c)*\mathrm{COUNT}\left(\mathrm{ci}\right)*\×\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, S _iExpression element i recommends the recommendation of user u, the set of the classification under C (u) the expression user u, sim (u, the c) similarity of expression user u and classification c, COUNT (c _i) number of times that in current classification, occurs of expression element i.

It is reciprocal that said device can comprise that also user's frequency module reciprocal is used to obtain user's frequency; Said cluster module also is used for according to said user's frequency inverse the user being carried out cluster; Wherein, element is concentrated more in total user's distribution, and then user's frequency reciprocal value of this element is big more.

It is reciprocal that this user's frequency module reciprocal can be used for calculating user's frequency according to following formula:

${\mathrm{IUF}}_i=\log \frac{n}{n_i}$

Wherein, IUF _iUser's frequency of element i among the expression user is reciprocal, and n representes total number of users, n _iTotal number of users under the expression element i.

Accordingly, then the user can be expressed as:

$u=\left\{w_i{c}_i\right\}=\{c_i{f}_i*\log \frac{n}{n_i}\}$

Wherein, u representes user u, w _iThe weight of element i among the expression user u, c _iExpression element i, f _iThe frequency that expression element i occurs in set u.

Embodiment of the present invention embodiment; Has following beneficial effect: when the user carries out the recommendation of element, adopt classification frequency (Inverse Cluster Frequency) reciprocal to reduce the weight of heat song, make when the user carries out the recommendation of element; Keeping under the constant basically situation of degree of accuracy; Highlight the user and belonged to the peculiar element of classification, in fact highlighted user's personalization, can effectively improve user's satisfaction.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art; To do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below; Obviously, the accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work property, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is an idiographic flow synoptic diagram of the element recommend method in the embodiment of the invention;

Fig. 2 is a concrete synoptic diagram of forming of the element recommendation apparatus in the embodiment of the invention;

Fig. 3 is another concrete synoptic diagram of forming of the element recommendation apparatus in the embodiment of the invention;

Fig. 4 is the collaborative filtering music recommend system module figure with suppressing heat song effect in the embodiment of the invention.

Embodiment

To combine the accompanying drawing in the embodiment of the invention below, the technical scheme in the embodiment of the invention is carried out clear, intactly description, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.

In the network commending system in embodiments of the present invention; Subscriber's meter is shown the set that comprises a plurality of elements (this element is the object that will recommend to the user); Simultaneously; Characteristic according to the user is a plurality of classifications with user grouping again, in order when recommending, to highlight user's characteristic, when generating user's recommendation list; Can adopt the overheated recommendation object of method compacting of classification frequency (Inverse Cluster Frequency) reciprocal, avoid the embodiment of these overheated recommendation object compactings user personality.

With the song recommendations system is its principle of example explanation: in the song recommendations system, the user is carried out cluster; When statement different classes and user; Classification is expressed as the set that comprises a plurality of users, and subscriber's meter is shown the set that comprises a plurality of songs, and then song belongs to different users and then belongs to different clusters; And, can exist overlapping between the above-mentioned set content; One first song then possibly belong to a plurality of different users; And then also possibly belong to a plurality of different classes; So; Then carry out suitable suppressing for a lot of song of affiliated classification (promptly a first song belongs to a plurality of users simultaneously, and these users belong to a plurality of classifications again), so that highlight the tangible song of those classifications.Because the categorical measure that the heat song belongs to is generally many,, make that the song of recommending all is personalized apparent in view song, can well improve user's satisfaction so this kind measure can realize rationally suppressing what heat was sung.

As shown in Figure 1, be the element recommend method in the embodiment of the invention, this method comprises the steps.

101, a plurality of users are carried out cluster, confirm the affiliated classification of each user, comprise a plurality of elements among the said user.

As previously mentioned, in commending system, subscriber's meter is shown as the set that a plurality of elements are formed, the element in the set is the object that will recommend to the user, as, in the song recommendations system, be the first song of a monic or singer one by one or the like; Its method of confirming also can be varied, such as the song/singer who is song/singer of once listening of user, defines in the interest group that the user participates in, user-defined song/singer or the like; According to actual conditions, same song/singer can repeat to appear in different user's set.

According to above-mentioned explanation, can subscriber's meter be shown the set that a plurality of elements are formed:

u＝{w _ic _i}

Wherein, u representes user u, w _iThe weight of element i among the expression user u, c _iExpression element i works as w _iWhen being 1, then show the influence of not considering weight.

Simultaneously, this programme adopts the collaborative filtering recommending principle based on cluster, at first need gather into several classifications according to the similarity between the user, and each classification is made up of several relevant users.In commending system, be reflected as, of all categories to a plurality of user clusterings acquisitions, each classification is expressed as the set that a plurality of users form, and the user in the set distributes according to cluster.

In an embodiment of the present invention, the similarity according to the user in the time of user clustering is carried out cluster, such as the user participated in or the interest group of current participation, user-defined label, user's set in element or the like.The method of concrete cluster can be various, such as K-means or based on the MinHash of weight.

When carrying out cluster, can adopt the method for user's frequency (Inverse User Frequency) reciprocal to carry out user clustering according to the elemental distribution in user's set, can realize the power of falling the high element of frequency through this method.Because the high frequency element is not very strong to the differentiation property between classification, when the similarity of calculating between the user, the weight of high frequency element to be suppressed, experimental result shows that this behave can improve the degree of accuracy of recommendation significantly.

On the other hand, in the music recommend system, also can play the effect of suppressing the heat song well.The heat song is when calculating similarity; Often the reduction of the effect of other non-popular songs; The class of gathering is out come leading to a great extent by the heat song, adopt after user's frequency (Inverse User Frequency) reciprocal, and the class of gathering out is leading by the real interest of user; Heat song only type spreads unchecked at some heat song very much, and the influence of other types has been obtained effectively extenuating.

Carry out user clustering if adopt user's frequency method reciprocal; It is reciprocal then can to obtain user's frequency in the step 101 earlier, according to said user's frequency inverse the user is carried out cluster again, wherein; Element is concentrated more in total user's distribution, and then user's frequency reciprocal value of this element is big more.This user's frequency inverse can adopt following formula to calculate:

${\mathrm{IUF}}_i=\log \frac{n}{n_i}$

Wherein, IUF _iUser's frequency of element i among the expression user is reciprocal, and n representes total number of users, n _iThen represent the total number of users under the element i, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).

For instance, user's set of forming by following each element: A={a1, a2, a3}, B={a1, b1, b2}, C={a1, b1, c1, c2}, n=3 at this moment then, n _A1=3 (they being that the affiliated total number of users of element a1 is 3), n _B1=2, n _A2=n _A3=n _B2=n _C1=n _C2=1.

Certainly, in some other embodiment of the present invention, also can use similar aforesaid way cluster, promptly not adopt above-mentioned logarithm form reciprocal, and adopt other modes, such as directly being n/n _iOr a coefficient, like q, multiply by n/n _iOr n/n _iOther functional forms f (n/n _i) or the like.

If adopt user's frequency of logarithmic form reciprocal, then this moment, user u can be expressed as:

$u=\left\{w_i{c}_i\right\}=\{c_i{f}_i*\log \frac{n}{n_i}\}$

Wherein, f _iExpression element i gathers the frequency that occurs among the u the user.This frequency is an accumulated weight, and such as in music recommend, frequency can be expressed as the number of times that user u listens song i; And in news was recommended, the user was to the number of times or the duration of certain news page browsing, or the like.

Like this; According to above-mentioned user, when it is carried out cluster, make for when calculating similarity, obtaining compacting such as the so overheated element of heat song with the set expression; Avoid reducing the effect of other non-popular songs; The class of gathering out is leading by the real interest of user, and the heat song only type spreads unchecked at some heat song very much, and the influence of other types has been obtained effectively extenuating.

Certainly; Above-mentioned to the user clustering process in, if adopt when carrying out cluster according to the element in user's set, this element not necessarily be meant recommendation object in the foregoing description (as; Song in the music recommend system); Also can be other object, as, user's interest, age of user layer, user's sex, user job type etc.Certainly, the element of description in embodiments of the present invention is except that specifying, generally all be meant recommendation object in the commending system (as, the song in the music recommend system, special edition, singer etc.).

102, obtain the classification frequency inverse of each element that belongs to the active user according to cluster result, wherein, element is concentrated more in the distribution of total classification, and then the classification frequency reciprocal value of this element is big more.Know that through aforementioned element belongs to user's set, user's set belongs to the classification set, and identity element can belong to a plurality of user's set simultaneously, and same user's set can belong to a plurality of classification set, and identity element then possibly belong to a plurality of classification set so; Accordingly, " element is concentrated more in the distribution of total classification " is meant, the classification set that identity element belongs to is few more, in previous example, and user's set: A={a1, a2; A3}, B={a1, b1, b2}, C={a1; B1, c1, c2}, classification set: CC1={A, B}; CC2={B, C}, then element a1 belongs to two classification set simultaneously, and a2 only belongs to classification set CC1, and the relative element a1 of element a2 is comparatively concentrated in the distribution of total classification so; Accordingly, the classification frequency inverse of element a2 is greater than element a1.

In specific embodiment of the present invention, classification frequency available following formula reciprocal calculates:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, ICF _iThe classification frequency of expression element i is reciprocal, | C| representes total classification number, NC _iThe classification number that expression element i occurred, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).Certainly, in some other embodiment of the present invention, classification frequency inverse also can not be above-mentioned logarithm form reciprocal, such as directly doing | and C|/NC _iOr a coefficient, like q, multiply by | C|/NC _iOr | C|/NC _iOther functional forms f (| C|/NC _i) or the like.

Still with the concrete implication of each object in the above-mentioned example description formula, that is, the user gathers: A={a1, a2, a3}, B={a1, b1, b2}, C={a1, b1, c1, c2}, classification set: CC1={A, B}, CC2={B, C}; Then in this example | C|=2, NC _A1=2, NC _A2=1.

103, according to the said classification frequency recommendation of calculating each element of said active user reciprocal, recommend to said user according to said recommendation, wherein, the recommendation of this element of the bigger then active user of classification frequency reciprocal value of element is high more.If the classification frequency with logarithmic form is reciprocal, then recommendation can use following formula to calculate:

$s_i=\underset{c\∈C\left(u\right)}{\mathrm{\Σ}}\mathrm{sim}(u,c)*\mathrm{COUNT}\left(\mathrm{ci}\right)*\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, s _iExpression element i recommends the recommendation of user u, the set of the classification under C (u) the expression user u, sim (u, the c) similarity of expression user u and classification c (calculation of similarity degree has a lot of different formula, as, the Cosine Similarity that uses always), COUNT (c _i) number of times that in current classification, occurs of expression element i.As, calculating COUNT (c _i) time, element i occurs 1 time in the set of user a, in the set of user b, occurs 1 time, and comprises user a, b among the classification c, and then the number of times statistics that in classification c, occurs of element i is 2; Calculating COUNT (c _i) time, also can consider the frequency that element i occurs in user set, occurs 1 time in the set of user a such as element i, in the set of user b, occur 2 times, then to add up be 3 to the number of times that in classification c, occurs of element i.Certainly this also can change, and also can not consider the number of times that element occurs in certain user's set in certain embodiments of the present invention.In this formula, if adopt other classification frequency account form reciprocal (with f (| C|/NC _i) expression), then replace can obtaining corresponding recommendation computing formula to fractional part in the above-mentioned formula with this f function.

Still with the concrete implication of each object in the above-mentioned example description formula, that is, the user gathers: A={a1, a2, a3}, B={a1, b1, b2}, C={a1, b1, c1, c2}, classification set: CC1={A, B}, CC2={B, C}; Then in this example | C|=2, NC _A1=2, NC _A2=1.Then, when being CC1 for current type, COUNT (c _A1)=2; When being CC2 for current type, COUNT (c _A1)=2; When being CC1 for current type, COUNT (c _B1)=1; When being CC2 for current type, COUNT (c _B1)=2.

In some specific embodiments, if in the system of non-scoring, use be user's latent feedback (implicit feedback), use be cumulative frequency COUNT (), such as in class c, the number of times that clauses and subclauses i is clicked is COUNT (c _i); Latent feedback can comprise the click of user to news, the song that the user listened, and the film that the user has seen, or the like.If at points-scoring system, can comment 0 to 5 fen such as user in web film to film, and the average score of clauses and subclauses i in class c is that (i, c), the COUNT function in the then above-mentioned formula can use corresponding ave function to substitute to ave.

Can find out from above-mentioned formula, work as NC _iWhen smaller, explain that i relatively belongs to some type in the concentrated area, the mark that obtains will be bigger, the corresponding s that calculates _iJust bigger, when the user was recommended, the recommended possibility of element i was just bigger.

In specific embodiment, all elements in user's set can calculate such recommendation, and the size based on recommendation sorts then, and finally recommends the user by descending order.

Like this; In embodiments of the present invention when the user carries out the recommendation of element; Adopt classification frequency (Inverse Cluster Frequency) reciprocal to reduce the weight of heat song, can highlight the peculiar element that the user belongs to classification when the user carries out the recommendation of element; In fact highlight user's personalization, can effectively improve user's satisfaction.Simultaneously, when the similarity of calculating between the user, adopt user's frequency reciprocal, can suppress, also help heat song phenomenon is suppressed the weight of high frequency clauses and subclauses.

As shown in Figure 2, be the concrete composition synoptic diagram of the element recommendation apparatus in the embodiment of the invention, this device comprises:

Cluster module 20 is used for a plurality of users are carried out cluster, confirms the affiliated classification of each user, comprises a plurality of elements among the said user;

Classification frequency module 22 reciprocal, the classification frequency that is used for obtaining according to cluster result each element that belongs to the active user is reciprocal, and wherein, element is concentrated more in the distribution of total classification, and then the classification frequency reciprocal value of this element is big more.It is reciprocal that this classification frequency module 22 reciprocal can be calculated the classification frequency according to following formula:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, ICF _iThe classification frequency of expression element i is reciprocal, | C| representes total classification number, NC _iThe classification number that expression element i occurred, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).

Recommendation computing module 24; Be used for recommendation according to each element of the said classification frequency said active user of calculating reciprocal; Recommend to said user according to said recommendation, wherein, the recommendation of this element of the bigger then active user of classification frequency reciprocal value of element is high more.This recommendation computing module 24 can be according to following formula calculated recommendation value:

$s_i=\underset{c\∈C\left(u\right)}{\mathrm{\Σ}}\mathrm{sim}(u,c)*\mathrm{COUNT}\left(\mathrm{ci}\right)*\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, S _iExpression element i recommends the recommendation of user u, the set of the classification under C (u) the expression user u, sim (u, the c) similarity of expression user u and classification c, COUNT (c _i) number of times that in current classification, occurs of expression element i.

As shown in Figure 3; It is reciprocal that this device can comprise that also user's frequency module 21 reciprocal is used to obtain user's frequency, and said cluster module 20 also is used for according to said user's frequency inverse the user being carried out cluster, wherein; Element is concentrated more in total user's distribution, and then user's frequency reciprocal value of this element is big more.It is reciprocal that this user's frequency module 21 reciprocal can be calculated user's frequency according to following formula:

${\mathrm{IUF}}_i=\log \frac{n}{n_i}$

Wherein, IUF _iUser's frequency of element i among the expression user is reciprocal, and n representes total number of users, n _iTable is then represented the total number of users under the element i.

Accordingly, the user can be expressed as:

$u=\left\{w_i{c}_i\right\}=\{c_i{f}_i*\log \frac{n}{n_i}\}$

Wherein, u representes user u, w _iThe weight of element i among the expression user u, c _iExpression element i, f _iThe frequency that expression element i occurs in set u.

The concrete implication of the alternative scheme among the said apparatus embodiment and each noun is all consistent with preceding method embodiment, does not do here and gives unnecessary details.

Combine user's frequency (Inverse User Frequency) reciprocal and classification frequency (Inverse Cluster Frequency) reciprocal in the embodiment of the invention; Under the situation that the accurate rate of recommending remains unchanged basically; Slow down existing recommendation and strengthened the trend of popular element, can improve user's satisfaction.

As shown in Figure 4, be compressed to example with the heat song, the collaborative filtering music recommend system module figure with suppressing heat song effect in the embodiment of the invention is described.

Wherein, Customer group is the user who is expressed as song collection; Situation according to the song of forming the user; Utilize IUF module (being user's frequency module reciprocal) to calculate and respectively form the weight of song in this user's set among the user, utilize the cluster module that the user who forms with the song with weight is carried out cluster again.

And in calculating during the recommendation of a certain user's song to customer group, then as the flow process on the limit of keeping right among the figure, according to aforesaid cluster calculation recommendation and according to the recommendation ordering with final acquisition recommendation list.

In the example like Fig. 4, " recommendation results generation module " is the situation of not considering ICF, the recommendation results that obtains by traditional method; And " final recommendation list " is to have considered ICF, the recommendation results after the heat song is suppressed, and " ICF module " is the equal of that traditional recommendation results is reset.Certainly also can settle at one go in other embodiments of the invention, promptly directly calculate last recommendation results based on formula.

One of ordinary skill in the art will appreciate that all or part of flow process that realizes in the foregoing description method; Be to instruct relevant hardware to accomplish through computer program; Described program can be stored in the computer read/write memory medium; This program can comprise the flow process like the embodiment of above-mentioned each side method when carrying out.Wherein, described storage medium can be magnetic disc, CD, read-only storage memory body (Read-Only Memory, ROM) or at random store memory body (Random Access Memory, RAM) etc.

Above disclosedly be merely a kind of preferred embodiment of the present invention, can not limit the present invention's interest field certainly with this, the equivalent variations of therefore doing according to claim of the present invention still belongs to the scope that the present invention is contained.

Claims (12)

1. an element recommend method is characterized in that, said method comprises:

A plurality of users are carried out cluster, confirm the affiliated classification of each user, comprise a plurality of elements among the said user;

The classification frequency of obtaining each element that belongs to the active user according to cluster result is reciprocal, and wherein, element is concentrated more in the distribution of total classification, and then the classification frequency reciprocal value of this element is big more;

According to the said classification frequency recommendation of calculating each element of said active user reciprocal, recommend to said user according to said recommendation, wherein, the recommendation of this element of the bigger then active user of classification frequency reciprocal value of element is high more.

2. the method for claim 1 is characterized in that, said classification frequency is reciprocal to be calculated with following formula:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, ICF _iThe classification frequency of expression element i is reciprocal, | C| representes total classification number, NC _iThe classification number that expression element i occurred, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).

3. method as claimed in claim 2 is characterized in that, said recommendation is calculated with following formula:

$s_i=\underset{c\∈C\left(u\right)}{\mathrm{\Σ}}\mathrm{sim}(u,c)*\mathrm{COUNT}\left(\mathrm{ci}\right)*\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, S _iExpression element i recommends the recommendation of user u, the set of the classification under C (u) the expression user u, sim (u, the c) similarity of expression user u and classification c, COUNT (c _i) number of times that in current classification, occurs of expression element i.

4. like each described method in the claim 1 to 3, it is characterized in that, saidly a plurality of users are carried out cluster comprise:

It is reciprocal to obtain user's frequency, according to said user's frequency inverse the user is carried out cluster, and wherein, element is concentrated more in total user's distribution, and then user's frequency reciprocal value of this element is big more.

5. method as claimed in claim 4 is characterized in that, said user's frequency is reciprocal to be calculated with following formula:

${\mathrm{IUF}}_i=\log \frac{n}{n_i}$

Wherein, IUF _iUser's frequency of element i among the expression user is reciprocal, and n representes total number of users, n _iThen represent the total number of users that element i is affiliated.

6. method as claimed in claim 5 is characterized in that, said user can be expressed as:

$u=\left\{w_i{c}_i\right\}=\{c_i{f}_i*\log \frac{n}{n_i}\}$

Wherein, u representes user u, w _iThe weight of element i among the expression user u, c _iExpression element i, f _iThe frequency that expression element i occurs in set u.

7. an element recommendation apparatus is characterized in that, said device comprises:

The cluster module is used for a plurality of users are carried out cluster, confirms the affiliated classification of each user, comprises a plurality of elements among the said user;

Classification frequency module reciprocal, the classification frequency that is used for obtaining according to cluster result each element that belongs to the active user is reciprocal, and wherein, element is concentrated more in the distribution of total classification, and then the classification frequency reciprocal value of this element is big more;

The recommendation computing module; Be used for recommendation according to each element of the said classification frequency said active user of calculating reciprocal; Recommend to said user according to said recommendation, wherein, the recommendation of this element of the bigger then active user of classification frequency reciprocal value of element is high more.

8. device as claimed in claim 7 is characterized in that, it is reciprocal that said classification frequency module reciprocal is used for calculating the classification frequency according to following formula:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, ICF _iThe classification frequency of expression element i is reciprocal, | C| representes total classification number, NC _iThe classification number that expression element i occurred, log representes with x to be the logarithm at the end, the x span be (0,1) ∪ (1 ,+∞).

9. device as claimed in claim 8 is characterized in that, said recommendation computing module is used for according to following formula calculated recommendation value:

$s_i=\underset{c\∈C\left(u\right)}{\mathrm{\Σ}}\mathrm{sim}(u,c)*\mathrm{COUNT}\left(\mathrm{ci}\right)*\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, S _iExpression element i recommends the recommendation of user u, the set of the classification under C (u) the expression user u, sim (u, the c) similarity of expression user u and classification c, COUNT (c _i) number of times that in current classification, occurs of expression element i.

10. like each described device in the claim 7 to 9; It is characterized in that; It is reciprocal that said device comprises that also user's frequency module reciprocal is used to obtain user's frequency, and said cluster module also is used for according to said user's frequency inverse the user being carried out cluster, wherein; Element is concentrated more in total user's distribution, and then user's frequency reciprocal value of this element is big more.

11. device as claimed in claim 10 is characterized in that, it is reciprocal that said user's frequency module reciprocal is used for calculating user's frequency according to following formula:

${\mathrm{ICF}}_i=\log \frac{\left|C\right|}{{\mathrm{NC}}_i}$

Wherein, IUF _iUser's frequency of element i among the expression user is reciprocal, and n representes total number of users, n _iThen represent the total number of users that element i is affiliated.

12. device as claimed in claim 11 is characterized in that, said user can be expressed as:

$u=\left\{w_i{c}_i\right\}=\{c_i{f}_i*\log \frac{n}{n_i}\}$

Wherein, u representes user u, w _iThe weight of element i among the expression user u, c _iExpression element i, f _iThe frequency that expression element i occurs in set u.

Images