Abstract: The historical search click data of a single user is sparse, which leads inaccurate query recommendation and cannot provide diverse query.Therefore, this paper takes the log of each user as a document, and uses the vector space model to calculate the similarity between the users’ documents.The frequency of user clicking the link in the historical data is considered as the preference score of each link, and the improved Euclidean distance is used to calculate the user’s nearest neighbors. The method is used to calculate the similar user set of the current user, and the historical behavior data of similar users is added to the data of a single user. Based on the naive Bayes model, the data is trained and the click-through rate is predicted between query and links. These rates are used as weight in the click graph and spreaded for recommendation generation. Experimental results show that this method obtains higher precision and mean average precision.
vector space model,