Query recommendation is an important tool for improving searching efficiency. Traditional recommendation methods were mainly care about the frequent queries, but cannot provide good recommendations for long tail queries due to the information sparsity. Without consideration of query intents, traditional methods generated the recommendations for long tail queries, which can be greatly influenced by noise words in queries. A novel probabilistic mixture model of term-query graph was proposed in this paper, which can clearly identify query intents of users. Otherwise, a new method of assembling query intents into recommendation was introduced in the paper, which can prevent the influence from noise words by merging the recommendations of word in newcoming query according to query intents naturally. The result of experiments show the relevance of long tail query recommendation can be greatly improved by taking account of query intent.%查询推荐是一种提升用户搜索效率的重要工具.传统的查询推荐方法关注频度较高的查询,但对于那些频度较低的长尾查询,由于其信息的稀疏性而难以产生好的推荐效果.另外,传统的方法由于没有考虑查询意图对推荐结果的影响,故对长尾查询的推荐会受到查询中噪声单词的影响.该文提出了一种新的关于词项查询图(term-query graph)概率混合模型,该模型能够准确地发掘出用户的查询意图.另外,文中还提出了一种融合查询意图的查询推荐方法,该方法可以将新查询中单词的推荐结果按查询意图自然地融合起来,从而避免了噪声单词对推荐结果的影响.实验结果表明,通过考虑查询意图,可以显著提高长尾查询推荐的相关性.
展开▼
