Specific implementation mode
The invention will be further described with the following Examples.
Referring to Fig. 1, a kind of Technologies of Recommendation System in E-Commerce based on user interest variation, the system includes modeling module 1, pushes awayRecommend generation module 2 and user terminal 3.
Modeling module 1 from the server of e-commerce website for obtaining user data, according to the user data of acquisitionBuild user-commodity rating matrix U and user-scoring time matrix R.
Generation module 2 is recommended to be used to generate commodity according to user-commodity rating matrix U and user-scoring time matrix R and push awayList is recommended, and commercial product recommending list is pushed into user terminal 3.
User terminal 3 is for receiving commercial product recommending list.
Preferably, user terminal 3 is smart mobile phone or tablet computer.
Preferably, user-commodity rating matrix U=(uij)m×n, the user-rating matrix R=(rij)m×n, whereinuijIt is score values of the user i to commodity j, rijBeing user i scores time of origin to commodity j, i=1,2 ..., m, j=1, and 2 ...,n。
Preferably, referring to Fig. 2, it includes information processing submodule 21 and recommendation submodule 22 to recommend generation module 2.
Information processing submodule 21 is used for user-commodity rating matrix U and user-scoring time matrix R processing,It obtains that user can be described over time to the interest attenuation degree of integration value of commercial productainterests attenuation degree.
Submodule 22 is recommended to be used to, according to obtained interest attenuation degree of integration value and user-commodity rating matrix U, calculateUser scores to the prediction of commodity, carries out descending arrangement to commodity according to obtained prediction scoring, is selected from the commodity after sequenceIt selects top n commodity and generates commercial product recommending list, and the commercial product recommending list is transmitted to user terminal 3, wherein N is self-definedRecommendations number.
Preferably, obtaining that user can be described over time to the interest attenuation of commercial productainterests attenuation degree synthesis journeyAngle value, specifically：
(1) according to user-commodity rating matrix U and user-scoring time matrix R, user couple is calculated using attenuation functionThe interest attenuation value of commodity, wherein the attenuation function is：
In formula, Ra(t) be user a interest attenuation value, a ∈ { 1,2 ..., m }, t are current time, tajIt is user a to quotientProduct j scoring time of origins, αajIt is personalized weight factors of the user a to commodity j, uajIt is score values of the user a to commodity j, j∈{1,2,…,n}；
(2) according to user-scoring time matrix R, different user is described to same part commodity using time correlation degree functionScore the degree of correlation of time of origin, wherein time correlation degree function is：
In formula, Sab(t) it is time correlation degree functional value between user a and user b, tajIt is that user a scores to commodity jTime of origin, tbjBeing user b scores time of origin to commodity j；
(3) it according to the interest attenuation value and time correlation degree functional value of step (1) and step (2), is calculated and is used using following formulaFamily a is over time to the interest attenuation degree of integration value of commodity, wherein the calculating formula of interest attenuation degree of integration value is：
In formula, Za(t) be user a interest attenuation degree of integration value, ε is weight factor, and 0 ＜ ε ＜ 1, Ra(t) it is to useThe interest attenuation value of family a, Sab(t) it is time correlation degree functional value between user a and user b,It is b couples of user a and userThe related coefficient that same part commodity j scores.
Advantageous effect：It may change for the interest of user, respectively from same user to the interest of different commodityDecaying behavior and different user describe the scoring degree of correlation of same commodity the variation feelings of user interest as time goes byCondition, which, which can care for, objectively reflects the variation tendency of user over time to commodity scoring, and then is conducive to electronics quotientBusiness platform is more accurately target user's Recommendations.
Preferably, it includes similarity calculated 220 and prediction scoring unit 221 to recommend submodule 22.
Similarity calculated 220, for calculating the similarity Sim (c, k) between target user c and other users,In, k ∈ { 1,2 ..., m }, if Sim (c, k) ＞ λth(λthIt is the threshold value of setting), then the user k is added to target user cNearest-neighbor Ω, wherein the calculating formula for calculating the similarity Sim (c, k) between target user c and other usersFor：
In formula, Simj(c, k) is the similarity value of target user c and user k to commodity j, and n is the commodity for participating in scoringNumber, and j={ 1,2 ..., n }, rcjIt is score values of the target user c to commodity j,It is that target user c puts down all commodityEqual score value, rkjIt is score values of the user k to commodity j,It is average score values of the user k to all commodity, tcjIt is that target is usedTime when family c scores to commodity j, tkjIt is time when user k scores to commodity j, Zc(t) it is the emerging of user cInterest decaying degree of integration value, Zk(t) be user k interest attenuation degree of integration value, tmaxBe in user-scoring time matrix R mostBig time, tminIt is minimum time in user-scoring time matrix R.
Advantageous effect：User-commodity rating matrix and user-scoring time matrix are combined to user c and user kThe similarity value of commodity j is calculated, which has fully considered the situation of change of user interest as time goes by, this doesMethod more meets objective law, the similarity value accuracy higher between obtained user, accurately recommends quotient to user to be follow-upProduct lay the foundation.
Preferably, prediction scoring unit 221, for according to obtained similarity, calculating target user c to not commentingThe prediction scoring of the commodity divided carries out descending arrangement to commodity according to obtained prediction scoring, is selected from the commodity after sequenceTop n commodity generate commercial product recommending list, and the commercial product recommending list is transmitted to user terminal 3, and N is customized recommendation quotientProduct number.Wherein, it calculates target user c and prediction scoring is carried out to unrated commodity, specifically utilize the prediction of lower section to score publicFormula is calculated：
In formula, Score (c, s) is prediction score values of the target user c to the commodity s that do not score before,It is all commentedDivide user to the average score of commodity s, rksIt is score values of the user k to commodity s, tcurIt is the current time for generating recommendation behavior,tcsIt is scoring times of the target user c to commodity s, tksIt is scoring times of the user k to commodity s, Sims(c, k) is target userFor c and user k to the similarity value of commodity s, Ω is the nearest neighbor set of target user c, and Ω=1,2 ..., k ...,M}。
Advantageous effect：When using prediction scoring is carried out to the commodity not scored to the similarity value of commodity between user,Pass through introducingCarry out the interests change of analog subscriber, which solves user interest variation commodity and tested and assessed in advanceThe influence of timesharing improves the accuracy of prediction scoring.In view of currently recommending time and reality to carry out the scoring time to commodityInfluence to commercial product recommending, the algorithm consider the influence that interest changes over time so that recommend output result closer to realitySituation improves the timeliness of recommendation the considerations of adding to generating the current time information recommended.
Finally it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than the present invention is protectedThe limitation of range is protected, although being explained in detail to the present invention with reference to preferred embodiment, those skilled in the art answerWork as understanding, technical scheme of the present invention can be modified or replaced equivalently, without departing from the reality of technical solution of the present inventionMatter and range.