基于朴素贝叶斯点击预测的查询推荐方法

QUERY RECOMMENDATION BASED ON NAIVE BAYES CLICK PREDICTION

查询推荐作为一种改善用户查询体验和效率的重要方式,可以帮助用户筛选并提供更加准确的查询描述。目前很多查询推荐方法主要集中在热门推荐或是基于相似度匹配的推荐上,忽略了用户的查询意图,无法有效提供个性化推荐。为此,基于对用户查询点击日志进行分析与挖掘,训练出一个朴素贝叶斯模型,针对用户输入的查询,根据历史数据预测其与URL的点击率,再利用二分图将URL的预测点击值平均分配给相对应的每个查询项,最后结合Jaccard相似度和时间相关因子综合分析用户当前输入的查询与历史中查询的相关度,并给出推荐。实验证明了该方法的可行性并取得了较好的推荐效果。

Query recommendation, as an important way to improve user-query experience and efficiency, can help users to filter and offermore accurate query descriptions. Many of the current query recommendation methods mainly focus on popular recommendation or therecommendation based on similarity matching, but neglect user’ s query intention, thus are unable to effectively provide the personalisedrecommendation. Therefore, on the basis of analysing and mining users, query-click logs, we have trained a Naive Bayes model. Aiming atthe queries inputted by the user, the model predicts CTR （click-through rate） between these queries and URL according to historical data,then uses bipartite graph to averagely assign the predicted CTR of URL to each corresponding query, and at last it combines the Jaccardsimilarity with time correlation factor to comprehensively analyse the relevance between the query currently inputted by the user and thehistorical queries, and provides the recommendations. In subsequent experiment it is proved the feasibility of this method, as well as thebetter recommendation effect achieved.