CAJ | 학술논문

面向个性化电影推荐领域，提出一种基于多维度权重动态更新的用户兴趣模型。将电影分成演员、导演、类别、地区和时间5个维度，分别计算电影在这些维度上的相似度。采用归一化方法将电影之间的相似度转化为用户兴趣模型中的多维度权重，并应用TF-IDF算法计算各维度中特征词的权重，从而实现电影各维度权重及其特征词权重的动态更新。利用基于内容的推荐算法，在MovieLens数据集进行实验，结果表明，该模型具有较高的推荐准确率和召回率，并且能够发现用户对电影维度的偏好，解决用户兴趣漂移问题。
면향개성화전영추천영역，제출일충기우다유도권중동태경신적용호흥취모형。장전영분성연원、도연、유별、지구화시간5개유도，분별계산전영재저사유도상적상사도。채용귀일화방법장전영지간적상사도전화위용호흥취모형중적다유도권중，병응용TF-IDF산법계산각유도중특정사적권중，종이실현전영각유도권중급기특정사권중적동태경신。이용기우내용적추천산법，재MovieLens수거집진행실험，결과표명，해모형구유교고적추천준학솔화소회솔，병차능구발현용호대전영유도적편호，해결용호흥취표이문제。
For personalized movie recommendation domain,this paper proposes a user interest model based on dynamic update for multi-dimensional weight. It divides the movie into five dimensions of actor, director, categories, area and time,respectively to calculate the similarity among these dimensions of film. It uses the normalization method to change the similarity of film into multi dimension weight of the user interest model,and calculates the weights of features of each dimension in the application of TF-IDF algorithm,in order to achieve dynamic update of the film weight and dimensions of feature weight by using content-based recommendation algorithm. In the MovieLens data set for experiment, results show that,the model has higher recommendation accuracy rate and recall rate,and can find user preferences on the film dimensions,solve the problems of user interest drift.