推荐系统的冷启动问题研究

Research on cold start problem of recommendation system

潘东

11649127

硕士

信息与通信工程

骆宗伟

2018-06-07

2018-07-06

哈尔滨工业大学

深圳

互联网的迅猛发展使得我们处在信息过载的时代，个性化推荐技术由于可以帮助用户快速准确地发现自己感兴趣的信息。但由于推荐系统的工作机制需要利用用户的行为数据，所以无法对还没有数据的新用户进行个性化推荐，这称为用户冷启动问题。传统的用户冷启动解决方案有利用用户的注册信息、选择合适的物品来启动用户的兴趣、利用社交网络数据等，但它们具有严重的操作繁冗或信息虚假的问题。本文认为冷启动问题的解决关键是要快速响应用户的口味，一旦用户和系统多次交互后冷启动问题也就解决了。本文在对主流的推荐系统技术和冷启动解决方案作了深入研究后，针对视频推荐中的用户冷启动问题设计了一整套解决模型，该模型以标签的形式做基于邻域的推荐，使用机器学习的手段来将视频分类。首先将所有电影全部标签化，用户需要自己选择感兴趣的标签，系统会使用机器学习技术生成启动物品来给用户反馈，用户做反馈之后系统生成最终的推荐列表。对于机器学习部分，本题使用了K近邻、决策树、支持向量机三种分类算法分别实现模型，并使用准确率指标来考察算法的表现。本题使用Python语言实现了模型，模型使用二级推荐，整体结构呈树状，只需用户反馈两次，操作简便，响应迅速，推荐结果准确而多样，很好的解决了用户冷启动问题。

With the rapid development of the Internet, we are in the era of information overload. Personalized recommendation technology can help users quickly and accurately find the information what they are interested in. But because of the working mechanism of recommender system, it's impossible to Personalized recommend for new users who have no data, which is called user cold start problem. The traditional solution takes advantage of the user's registration information, the appropriate items to start the user's interest, social network data, etc. But they all have serious problems with complex operation or false information. In this paper, we consider the key to solve the problem is to quickly respond to the user's taste. Once the user and the system interact with each other, the cold start problem is solved.In this paper, a set of solution model is designed for the user cold start problem in video recommendation. The model is based on the neighborhood with way of label, and the video is classified by machine learning technology. First of all, all films are labeled, and the user needs to select label which is most interest in. The system will use machine learning technology to pick the starting items out. After the user’s feedback, the system will give the final recommendation list. Three classification algorithms K neighbor, decision tree and support vector machine are used to develop the model respectively, and the performance of the algorithm is valued by accuracy indicator.The model adopts secondary recommendation, just need two times feedback. It is easy to operate, quickly response and recommendation results are accurate and diverse. The model has been proved well solved the user cold start problem in video recommendation.

推荐系统 机器学习 冷启动

Recommendation system Cold start Machine learning Labeling system

中文

联合培养

南方科技大学

潘东. 推荐系统的冷启动问题研究[D]. 深圳. 哈尔滨工业大学,2018.