【摘要】：針對(duì)現(xiàn)有基于矩陣分解的協(xié)同過(guò)濾推薦系統(tǒng)預(yù)測(cè)精度與推薦精度較低的問(wèn)題,提出一種改進(jìn)的矩陣分解方法與協(xié)同過(guò)濾推薦系統(tǒng)。首先,將評(píng)分矩陣分解為兩個(gè)非負(fù)矩陣,并對(duì)評(píng)分做歸一化處理,使其具有概率語(yǔ)義;然后,采用變分推理法計(jì)算貝葉斯概率模型實(shí)部后驗(yàn)的分布;最后,搜索相同偏好的用戶分組并預(yù)測(cè)用戶的偏好。此外,基于用戶向量的稀疏性設(shè)計(jì)一種低計(jì)算復(fù)雜度、低存儲(chǔ)成本的推薦結(jié)果決策算法�；�3組公開(kāi)數(shù)據(jù)集的實(shí)驗(yàn)結(jié)果表明,本算法的預(yù)測(cè)性能以及推薦系統(tǒng)的效果均優(yōu)于其他預(yù)測(cè)算法與推薦算法。
[Abstract]:An improved matrix decomposition method and collaborative filtering recommendation system are proposed to solve the problem of low prediction accuracy and recommendation accuracy of collaborative filtering recommendation system based on matrix decomposition. First, the scoring matrix is decomposed into two non-negative matrices, and the score is normalized to make it have probabilistic semantics. Then, the variational reasoning method is used to calculate the posterior distribution of the real part of Bayesian probabilistic model. Search for groups of users with the same preferences and predict preferences. In addition, a recommendation result decision algorithm with low computational complexity and low storage cost is designed based on user vector sparsity. The experimental results based on three groups of open data sets show that the prediction performance of this algorithm and the effect of recommendation system are superior to those of other prediction algorithms and recommendation algorithms.
【作者單位】： 塔里木大學(xué)信息工程學(xué)院;
【基金】：國(guó)家科技支撐計(jì)劃(2013BAH27F00) 塔里木大學(xué)校長(zhǎng)基金項(xiàng)目(TDZKQN201616) 新疆南疆農(nóng)業(yè)信息化研究中心項(xiàng)目(TSAI201402)資助
【分類號(hào)】：TP391.3

【相似文獻(xiàn)】

相關(guān)期刊論文 前2條

1 曾廣平;;貝葉斯概率LSA模型權(quán)重更新算法[J];計(jì)算機(jī)工程與應(yīng)用;2009年21期

2 ;[J];;年期

，

本文編號(hào)：2212235