CAJ | 학술논문

基于信任的推荐算法将社交网络中的信任关系融入到推荐中，但数据的稀疏性迫使基于信任的方法要去考虑间接邻居，有限相似的邻居带来的长尾噪音问题降低了推荐准确度；目前已有的算法都假设用户的评分数据完全客观真实，而忽略了异常评分的存在。为了解决上述问题，文章提出新的用户相似度量方法筛选用户的信任邻居，并通过一次预测结果反馈检测并修正评分数据中的异常评分，然后进行二次预测。在真实的大规模数据集Epinions上进行实验，结果表明相比于传统的基于物品的协同过滤算法，该算法在RMSE上提高了6．0％，在M A E上提高了12.4％，说明该算法能有效缓解数据稀疏带来的上述问题，并提高预测精度。
기우신임적추천산법장사교망락중적신임관계융입도추천중，단수거적희소성박사기우신임적방법요거고필간접린거，유한상사적린거대래적장미조음문제강저료추천준학도；목전이유적산법도가설용호적평분수거완전객관진실，이홀략료이상평분적존재。위료해결상술문제，문장제출신적용호상사도량방법사선용호적신임린거，병통과일차예측결과반궤검측병수정평분수거중적이상평분，연후진행이차예측。재진실적대규모수거집Epinions상진행실험，결과표명상비우전통적기우물품적협동과려산법，해산법재RMSE상제고료6．0％，재M A E상제고료12.4％，설명해산법능유효완해수거희소대래적상술문제，병제고예측정도。
T he trust‐based approach to making recommendations introduces social relationship into the recommendation process . However ,the sparsity of the user‐item ratings forces the trust‐based ap‐proach to consider ratings of indirect neighbors ,affecting its precision of recommendation due to the result of long‐tail noise disruption brought about by dissimilar users .On the other hand ,existing ap‐proaches all assume that the ratings are completely true and objective ,and ignore that anomalous rat‐ings are also existent .To tackle the above problems ,a user similarity metric is explored that selects out trustworthy neighbors ,then detects and adjusts anomalous ratings against the results from the in‐itial prediction .In the end the prediction is re‐done .The proposed algorithm is experimented with the real Epinions datasets and a 6 .0% increase in RMSE and 12.4% in MAE are achieved compared to the popular item‐based collaboration filtering .It is concluded that the proposed approach both tackles the problems arising from data sparsity ,and enhances recommendation precision .