Method for contents recommendation

摘要

A method for music recommendation is provided using collaborative filtering methods while still managing to produce novel yet relevant items and by utilizing the long-tailed distribution of listening behavior of users, in which their playlists are biased towards a few songs while the rest of the songs, those in the long tail, have relatively low play counts. Also a link analysis method is applied to users with links between them to create an increasingly fine-grained approach in calculating weights for the recommended items. Results show that the method manages to include novel recommendations that are still relevant, and shows the potential for a new way of generating novel recommendations.

主权项

1. A method for contents recommendation in an online contents store comprising steps for:
providing a plurality of playlist's distributions associated with a plurality of users of the online contents store on an Internet respectively; selecting a novice user from the plurality of users, wherein the novice user has one or more novice-loved-contents in a long tail of an associated playlist's distribution as a novice-playlist's distribution; finding a first expert user with an associated playlist's distribution as a first-expert-playlist's distribution, wherein a short head of the first-expert-playlist's distribution includes at least one of the novice-loved-contents; finding a second expert user with an associated playlist's distribution as a second-expert-playlist's distribution, wherein a short head of the second-expert-playlist's distribution includes at least one of the novice-loved-contents; finding a third expert user with an associated playlist's distribution as a third-expert-playlist's distribution, wherein a short head of the third-expert-playlist's distribution includes at least one of the novice-loved-contents; assigning a weight for each of contents in the short head of each of the experts, wherein the weight depends on a sum of importances of the experts for the specific contents, wherein an importance of an expert is determined by a number of identical contents between the expert's playlist and each of the other expert's playlist and a number of identical loved-contents between the expert's playlist and each of the other expert's playlist; and recommending a first N highly-weighted contents to the novice user through the online contents store on the Internet, wherein the contents are songs, and wherein the importance of an exert, Ei, is given byImp⁡(Ej)=Ind⁡(Ej)+∑i=1Nexp⁢RegSongs⁢⁢(Ej,Ei)*Imp⁡(Ei)Nloved+1-LovedSongs⁢⁢(Ej,Ei),i≠j,where Ind (Ej) is the independent weight of the expert, Nexp is the number of total experts, Imp (Ei) is the importance of Expert Ei, and, Imp (Ei), i=1 to Nexp, are calculated recursively, RegSongs (Ej, Ei) denotes the number of same song occurrences in the playlists of experts Ei and Ej, Nloved is the total number of ‘loved’ songs in the novice's long tail, and LovedSongs (Ej, Ei) denotes the number of ‘loved’ songs that both experts share.