Collaborative Filtering algorithms suffer from the so-called cold-start problem. In particular, when a user has rated few items, recommendations offered by these algorithms are not too accurate. Profile Expansion techniques have been described as a way to tackle this problem without bothering the user with additional information requests by increasing automatically the size of the user profile. Up to now, only collaborative approaches had been proposed for Profile Expansion. However, content-based techniques can also be used. We perform a manual analysis of a movie dataset to analyze how content features behave. According to this analysis, we propose a content-based approach, which is also combined with collaborative information. Concretely, we expose the advantages and disadvantages of the combination with a popularity feature. Moreover, a comparison to pure collaborative approaches is performed. Our approach is evaluated in a new system situation. That is, not only the active user has few ratings, but also most of the users. The results show that content-based information is useful for rating prediction. In addition, recommendations are less personalized as popularity feature acquires more relevance for item selection.

中文翻译：

---

基于内容的配置文件扩展方法

协同过滤算法存在所谓的冷启动问题。特别是，当用户对少数项目进行评分时，这些算法提供的推荐不会太准确。配置文件扩展技术已被描述为一种解决此问题的方法，而无需通过自动增加用户配置文件的大小来使用额外的信息请求来打扰用户。到目前为止，仅针对 Profile Expansion 提出了协作方法。然而，也可以使用基于内容的技术。我们对电影数据集进行手动分析，以分析内容特征的行为方式。根据这一分析，我们提出了一种基于内容的方法，该方法也与协作信息相结合。具体来说，我们揭示了与流行特征相结合的优缺点。而且，与纯协作方法进行了比较。我们的方法在新的系统情况下进行了评估。也就是说，不仅活跃用户的评分很少，而且大部分用户。结果表明，基于内容的信息对于评分预测很有用。此外，推荐的个性化程度较低，因为流行度特征与项目选择具有更多相关性。