基于多维数据融合的影视推荐系统设计

Design of video recommendation system based on multidimensional pata fusion

为提升推荐系统对传统影视的准确率,结合传统推荐算法和电影海报数据的特点,提出通过CLR+Easy Ensemble+CNN对电影海报特征和用户特征等多维特征进行提取和融合,然后再进行电影推荐。结果表明,在MovieLens 1M数据集和Top_N准确率、Top_N召回率为指标下,提出的电影推荐算法取得较高的准确率,在Top_N=5时准确率可达64.7%,明显高于另外几种推荐算法。由此说明,构建的推荐方法可为用户提供更为准确的电影推荐。

In order to improve the recommendation accuracy of recommendation system for traditional movie and television, combined with the characteristics of traditional recommendation algorithms and movie poster data, this paper proposes to extract and fuse multidimensional features such as movie poster features and user features through CLR+Easy Ensemble+CNN, and then the movie recommendation is carried out. The results show that analyzing MovieLens 1M data set and taking Top_N accuracy and Top_N recall rate as indicators, it can be found that the proposed film recommendation algorithm achieves high accuracy. When Top_N=5, the recommendation accuracy can reach 64.7%, which is significantly higher than other recommendation algorithms, and this shows that the constructed recommendation method can provide users with more accurate movie recommendation.