摘要
提出一种基于weight-pooling词向量的上下文广告推荐算法,利用用户访问记录的互信息权重,计算weight-pooling词向量的余弦相似度。该算法改进了传统基于关键词匹配的推荐算法,避免了数据稀疏性和冷启动问题。通过实验分析,基于weightpooling词向量的上下文广告推荐算法在推荐效果上明显优于传统算法。
A new algorithm of context advertising recommendation based on weight-pooling word vector is proposed, which utilizes the mutual information weight of user access records to calculate the cosine similarity of weight-pooling words. This algorithm improves the traditional one based on matching key words and avoids data sparseness and cold start. After experiments and analysis, the recommendation algorithm based on the weight-pooling word vector is obviously superior to the traditional algorithm in the recommendation effect.
出处
《计算机应用与软件》
CSCD
2016年第12期224-229,共6页
Computer Applications and Software
基金
国家自然科学基金项目(61272367)
