In the traditional content-based recommendation method, we usually use the movies users watched before or rated to represent their profile. However, there are many movies that users have never seen or rated. For an unrated movie, there are two possibilities: maybe the user likes it or does not like it. In this paper, we first focus on how to identify users' preferences for movies by using a collaborative filtering algorithm to predict the users’ movie ratings. We can then create two movie lists for each user, where one is the movies the user likes (with higher predicting or true ratings), and the other is the movies the user does not like (with lower predicting or true ratings). Based on these two movie lists, we establish a user positive profile and a user negative profile. Therefore, our algorithm will recommend to users movies that are most similar to their positive profile and most different from their negative profile. Finally, our experiments show that our method can improve the MAE index of the traditional collaborative filtering method by 12.54%, the MAPE index by 17.68%, and the F1 index by 10.16%.

在传统的基于内容的推荐方法中，我们通常使用用户之前观看过的电影或经过评级的电影来代表其个人资料。但是，有许多电影是用户从未看过或评价过的。对于未分级的电影，有两种可能性：也许用户喜欢它或不喜欢它。在本文中，我们首先关注如何通过使用协同过滤算法来预测用户的电影评分来识别用户对电影的偏好。然后，我们可以为每个用户创建两个电影列表，其中一个是用户喜欢的电影（具有较高的预测或真实评级），另一个是用户不喜欢的电影（具有较低的预测或真实评级）。基于这两个电影列表，我们建立了一个用户肯定配置文件和一个用户否定配置文件。所以，我们的算法将向用户推荐与他们的正面形象最相似且与他们的负面形象最不同的电影。最后，我们的实验表明，该方法可以将传统协同过滤方法的MAE指数提高12.54％，MAPE指数提高17.76％，F1指数提高10.16％。