Abstract: How to integrate various factors for providing users accurately with personalized products that is always a hot issues. A new near-neighbor improved algorithm is brought forward through adopting popularization and personalization recognition degree correction, which can mine the hidden information more efficiently. After shown by the experimental results, although the recall fluctuates slightly up and down on that correction algorithm relative to the traditional near-neighbor algorithm, a number of other evaluation indexes improve dramatically. The false positive rate and the depth decrease, meanwhile, the precision, F1 value and the lift increase totally. In addition, the receiver operating characteristic curve and the lifting curve both suggest that the improved algorithm has more significant recommending effects.

Key words: recommendation algorithm, recognition degree, near-neighbor algorithm, movie recommendation, recommending effect

摘要： 如何融合多方因素准确地为用户提供个性化产品一直是关注的热点问题，由此，一个新的近相邻改进算法融入了大众化认同度和个性化认同度，利于更加高效地挖掘隐藏信息。实验结果表明，相对于传统近相邻算法，认同度修正算法虽然查全率小幅度上下波动，但其他多个评价指标都得到极大提升，假正率和深度有所减少，查准率、[F1]值和提升度得以增加，并且，受试者特征曲线和提升曲线也都说明此修正算法具有更为显著的推荐效果。

关键词: 推荐算法, 认同度, 近相邻算法, 电影推荐, 推荐效果