BRScS Approach for Resolving Heterogeneity of Data from Multiple Resources at Semantic Level

Algorithm 2

BRS_cS Model.

 Input: score dataset Score, social relation dataset Relation, review dataset Review, vocabulary V;

 Output: user representation U, item representation M, recommendation list L;

(1): Initialize , embedding size = 300, batch size = 64, negative sample = 5;

(2): for epoch = 1, 2, …, n do

(3): split the dataset score, relation, and review into training dataset (70%) and testing dataset (30%);

(4): construct positive and negative sample triplets (u, i, j) based on BPR;

(5): learn the frequency of word-review pair fw, dum and the expected value EWN-PV

(6): get review representation dum;

(7): learn U₁, U₂, c₁, c₂ from score data;

(8): get score representation ru, rm;

(9): get distance lab between users;

(10): calculate Σ(u,i,j) + λ₁L₁ − λ₂L₂;

(11): update  = {₁, ₂} with back propagation;

(12): get corresponding user and item representations U, M

(13): end for

(14): compute s;

(15): return recommendation list L