Vol.141 (GST 2016), pp.199-203 http://dx.do.org/10.14257/astl.2016.141.43 Collaboratve Flterng Recoendaton Algorth Dong Lang Qongta Teachers College, Haou 570100, Chna, 18689851015@163.co Abstract. Ths paper proposed two dfferent odels of foraton countes, and studes whch recoended ethod s sutable under the condtons of dfferent county foraton. It proposed two sutable slarty calculaton odels n the county, and then copared the wth the tradtonal slarty odel and test several slarty calculaton odels under the condtons of dfferent county foratons. Fnally, t copares tow odels of forng countes and fnds that for non-strct dvson of county odel has a hgher accuracy and dversty of recoendaton, copared wth the strct dvson of county odel Keywords: Personalzed Recoendaton, Collaboratve flterng, Clusterng algorth 1 Introducton Indvdual recoendaton technology s the nd of technque to ae recoendatons to relatve users after analyss of contents and relatonshp based on extracted features of obects [1-2]. At present, content-based recoendaton technology and that based on collaboratve flterng are facng soe probles although they can ntally eet requreents of recoendaton [3-4]. The content-based recoendaton technology s often restrcted to the condton of acqurng recoended obect s characterstcs. To be specfc, f a ove s recoended, t needs to have nforaton such as fl ttle, fl type, drector, casts, even contents or eywords. Coparatvely, collaboratve flterng technology doesn t requre any descrptve nforaton regardng obect s feature nor s affected by such nforaton whether t s correct or not; nstead, t totally depends on user s scorngs of obects [5-6]. Even though there s coplete descrptve nforaton regardng the obect, soe studes have deonstrated that the collaboratve flterng approach reaches better recoendaton effects than content-based soluton. However, wth growng data scale, user data and the rapd enlargeent of obect data, collaboratve flterng technology eets challenges. Due to huge atrx sze, and user s partcpaton nforaton lted to a certan perod, there would be the case that atrx becoes ore and ore sparse. No atter what nd of slarty odel s adopted, t s not possble to solve the proble of data sparsty; thus the recoendaton effect degrades largely [7]. In ths case, for enorous networ data, especally the recoendaton requreents based on Internet, a ore effcent ISSN: 2287-1233 ASTL Copyrght 2016 SERSC
Vol.141(GST 2016) recoendaton ethod s used rather than the content-based or collaboratve flterng recoendaton. Here we ntroduced the ethod based on county recoendaton [8-9]. 2 Recoendaton Algorth based on County Relatonshp n Networ 2.1 Tradtonal Slarty Calculaton (TSC) Generally bpartte networ ncludes user 1 2 1 2 U { u, u,..., u..., u }, obect O { o, o,..., o..., o } and edges E { e,: uu, o O} onng the up. p n So n a bpartte networ, the slarty between two users s calculated: S ( u, u ) TS C( u ) C( u ) C( u ) C( u ) (1) In the networ, the nuber of oves chosen by each user s lted, because t s related wth ts te, energy, nterest etc. If t s calculated wth tradtonal equaton of slarty, usng denonator to dvde the nuber of each selected oves, slarty becoes lower between users who watched ore oves, but hgher between users who watched fewer fls. That s not logc. Consderng shortcong of tradtonal expresson, we proved slarty calculaton forula. 2.2 Iproved Slarty Calculaton Forula (ISC) User s ratng of obects s apped nto a 2-pont syste. It s often found n a classcal recoendaton odel. If obect (e.g. ove) ratng s fve ponts, and scorng s reduced fro a 5-pont syste to a 2-pont syste where there s only 0 and 1, t eans only need to consder whether user loves the obect. The pont not less than 3 suggests that user les the obect; otherwse, user dsles t. Although the ethod can reduce coputer processng speed and ncrease runnng effcency of the recoendaton syste, n order to enhance the accuracy of calculatng slarty, coprehensve ratng nforaton should be used nstead of condensed nforaton whch cannot be coplete. Hence, we present an proved forula to calculate the slarty. It aes full use of user s all ratng nforaton. To copute slarty between two users, we can estate dfferentaton between the. The dfference between user and u s defned as follows: D C( u) C( u ) R, R, ( u, u ) C( u ) C( u ).( R R ) IS u M ax Mn (2) 200 Copyrght 2016 SERSC
Vol.141 (GST 2016) 2.3 Slarty Calculaton Forula wth Fault-tolerant Ratng (IST) To the proved slarty calculaton ethod, an approach wth fault-tolerant echans s ntroduced. Consderng that each user s evaluaton ay be arbtrary and faulty, for nstance, when a user loves a ove but not very uch, the ratng s usually 3 or 4 ponts. A ove rated by 4 ponts ay be not better than one by 3 ponts; or a ove rated by 3 ponts ay not be worse than one by 4 ponts. In ths case, we brng as fault-tolerant ratng. The dfference degree between user u and D u R 1 s defned as follows: C( u) C( u ) R, R, R ( u, u ) C( u ) C( u ).( R R ) IST M ax Mn To classfy user accurately to the belonged county, t s necessary to consder ts connecton wth the county and defne the slarty degree between user and county and that aong countes. (1) Slarty between user and county Wth exstng forula for calculatng slarty between users, we can get the slarty degree between any user and one county, by the expresson: S UC S( u, u) ( u, C8 ) (4) C uc8 g By calculatng the average value of the slarty between the users and the county, to deterne the degree of assocaton. (2) Slarty aong countes S( u, u ) SCC ( Cg, Ch) (5) C. C uc g, u Ch g h The forula calculates the slarty between two groups of users, to deterne the degree of correlaton between the two groups. (3) 3 Experent Desgn and Dscusson 3.1 Precson The Precson rate of the recoendaton syste s n Forula6: 1 d. r P L r (6) Copyrght 2016 SERSC 201
Vol.141(GST 2016) 3.2 Dversty of Recoendaton In vew of features based on county recoendaton, we use average ntra-user dversty as the easurng ethod of syste recoendaton result. The slarty forula between two obects s defned as follows: S ( o, o ) Dversty p q a. a u, p u, q (7) ( o ). ( o ) u1 p q In the algorth based on county recoendaton, snce t s not possble to ensure that enough long recoendaton lst s provded to each user, so the length of such lst s ' L u, referrng to the length of recoendaton lst to user u. And each user n the syste acqures dfferent long recoendaton lst, whch depends on whether the nuber of selected obect to whch each user belongs s bgger than L. Now we can get the dversty easurng of recoendaton syste result. ' 1 D arg{ D ( u)}. D ( u) (8) Dversty Dversty ' Dversty u 1 4 Concluson In ths paper, two nds of dfferent county foraton odels are proposed, and the applcaton and recoendaton of the three odels are copared wth the two odels. By usng the data of the MOVIELENS data set, t s verfed that the odel based on the county foraton s not only n the recoendaton accuracy. References 1. Lu, Q.: Research on the recoendaton algorth based on collaboratve flterng. Unversty of Scence ≈ Technology Chna, 2013 2. Sun, G., Wu, Y., Lu, Q., Zhu, C., Chen, E.: A collaboratve flterng recoendaton algorth based on sequental behavor. Journal of software, 2013,11:2721-2733. 3. Su, G.: Research on E-coerce Recoendaton Algorth Based on collaboratve flterng. Shandong Noral Unversty, 2014 4. Huang, Y.: Research on Collaboratve Flterng Recoendaton Algorth Based on te clusterng and preference categores. Zheang Sc-Tech Unversty, 2014 5. Herlocer, J., Konstan, J., Terveen, L.: Evaluatng Collaboratve Flterng Recoender Syste. ACM Trans on Inforaton Syste (TOIS),2004,22(1):5-53 6. Bure, R.: Hybrd Syste for Personalzed Recoendatons. Intellgent Technques for Web Personalzaton. Sprnger Berln Hedelberg,2005:133-152 7. L, Y.: Research on Personalzed Recoendaton Algorth Based on clusterng of collaboratve flterng, Huazhong Noral Unversty, 2014 8. Zhang, L.: Research on the recoendaton algorth based on collaboratve flterng and clusterng. Jln Unversty, 2014 202 Copyrght 2016 SERSC
Vol.141 (GST 2016) 9. Zhou, J.: Research on Recoendaton Algorth of collaboratve flterng based on trust, Yanshan Unversty, 2013 10. L. D.: Method of huan body posture detecton and oton recognton based on vdeo, Central South Unversty, 2012 11. Peng, L.: Target detecton and tracng n soccer vdeo gae, Nanng Unversty of Scence and Technology, 2006 Copyrght 2016 SERSC 203