CAJ | 학술논문

基于热传导或物质扩散理论的推荐算法首先利用网络结构得到对象间推荐关系，然后根据对象间关系预测用户喜欢的对象，而忽略了用户偏好。为了弥补这个缺陷，根据用户已选择对象的标签，利用 TF-IDF 方法构建用户偏好模型，以用户在预测对象标签上的平均偏好作为对该对象的偏好程度，采用加权方法与现有基于网络推荐算法混合运算。经在基准数据集 MovieLens 上测试表明，通过与目前效果最好的几种基于网络推荐算法进行加权混合运算，推荐结果在推荐精度、个性化、多样化等多种评价指标方面均比原有算法有明显提高。
기우열전도혹물질확산이론적추천산법수선이용망락결구득도대상간추천관계，연후근거대상간관계예측용호희환적대상，이홀략료용호편호。위료미보저개결함，근거용호이선택대상적표첨，이용 TF-IDF 방법구건용호편호모형，이용호재예측대상표첨상적평균편호작위대해대상적편호정도，채용가권방법여현유기우망락추천산법혼합운산。경재기준수거집 MovieLens 상측시표명，통과여목전효과최호적궤충기우망락추천산법진행가권혼합운산，추천결과재추천정도、개성화、다양화등다충평개지표방면균비원유산법유명현제고。
Recommendation algorithms based on heat conduction or mass diffusion first obtain the relationship between objects according to network structure,then predict the user’s favorite objects based on these relationships,but these al-gorithms ignore user’s preference.In order to overcome this defect,the TF-IDF approach was used to construct user’s preference according to the tags contained in the objects selected by user,and the mean of preference of object’s tags was taken as the preference of the object,then a hybrid recommendation model was proposed by combining network-based algorithm and the user preference model.The benchmark datasets,MovieLens,was used to evaluate our algo-rithm,and the experimental results demonstrate that hybrid algorithm can significantly improve accuracy,diversification and personalization of recommendations.