Abstract:

Traditional recommender algorithms have obtained great performance and have been widely used in practice, meanwhile it faces some serious problems of data sparsity and cold start. With the growth of the data around the world, and the data structures of information are increasingly richer, how to correctly capture the feature of user preferences from these data have become the key issue. The distributed representation methods provide a new solution for the issue above. By reviewing the recommender algorithms based on distributed presentation technology in recent years, this paper analyzes the differences and advantages between these algorithms and traditional one, and summarizes the practical scene of these algorithms. Finally, the prospects for future development of recommender algorithm based on distributed representation is discussed, in order to provide references for subsequent related work.

Key words: recommender algorithm, data sparsity, cold start, distributed representation technology

摘要：

传统的推荐算法在实际应用中取得不错的效果并得到广泛的使用，同时也面临严重的数据稀疏和冷启动的问题，随着全球数据总量的不断增长，数据结构愈加丰富，如何从这些数据中正确捕获用户偏好成为关键问题，分布式表示技术为此提供了新的解决思路。通过对近些年基于分布式表示技术的推荐算法进行综述，分析算法与传统推荐算法的区别和优势，对这些算法的实际应用场景进行总结，并对基于分布式表示的推荐算法的未来发展趋势进行展望，为后续相关工作提供参考。

关键词: 推荐算法, 数据稀疏, 冷启动, 分布式表示技术