Research On Query Recommendation Methods Supporting Exploratory Search

People's search activities become more and more complex due to the increasing complexity of the global information environment people's search activities become more and more complex and the exploratory search is an increasingly important activity for Web searchers.However,there are still many problems and challenges during the research of exploratory search,and query recommendation for exploratory search is an urgent problem to be solved.The current query recommendation methods mainly focus on optimizing users' current query which is far away from satisfying users' information needs of the whole search session.Because of lacking the analysis of the characteristics of complex search process and the behavior characteristics for users who are engaged in complex search tasks,the current query recommendation methods can only be applied to some simple search tasks,such as lookup search.It can not provide sufficient support for more complex tasks such as exploratory search.In order to solve the above problems,we have designed a new query recommendation method to support exploratory search.For better query recommendation result,we used Bayesian Rose Tree algorithm to design a model according to the behavioral characteristics of searchers in exploratory search.Based on this model,we have successfully obtained a large number of exploratory search process from search logs.And then we had accumulated a lot of high-quality exploration experiences in analyzing the exploratory search process.We found that the user's search goals are constantly changing during the exploratory search process.In order to make the recommendation results closer to the user's real needs,we built a search states prediction model and used it to catch user's search intention.Finally,we used the machine learning method to transform a large number of search experience obtained early into an exploratory search goal shift graph and to obtain the query recommendations in the search goal shift graph with the random walk algorithm.In addition,we identified search paths from the search goal shift graph and used the paths to help users quickly complete the exploratory search task.The contribution is reflected in the following aspects:(1)An identification model for exploratory search process is proposed.Because the current search task identification method mainly deals with the search task as a flat,unstructured cluster,and lacks the analysis and recognition of the task-related subtasks,it can not privode accurate identification result for more complex exploratory search.Therefore,an identification method based on hierarchical structure model is designed in this paper.The new method uses Bayesian Rose tree algorithm to mine the subtasks of complex search task from search logs.And then it uses mechine learning method to identify exploratory search processes with structure of the Bayesian Rose tree feature and the correlation between different subtasks.(2)The user's search goal has been constantly shifting in the exploratory search process,so identification and prediction of states of the entire exploratory search process has been useful for improving the accuracy of the recommendation results.This paper thus proposes a search state identification and prediction method for exploratory Search.Firstly,according to behavioral attributes such as query transitions and result clicks,the exploratory search process was divided into different states.And then we have used the gradient boosting decision Tree to automatically identify the states;finally,the weighted Markov chain algorithm is used to predict the states of exploratory search.(3)We analyzed for exploratory search processes deeply,and found that there are a lot of search goal shift phenomena in exploratory search.Based on this fact,we have designed a new query recommendation method to support exploratory search.Firstly,according to the behavioral characteristics of searchers in the search goal shift processes,all the queries submitted in the search goal shift processes were extracted from search engine logs using machine learning.And then we have used the queries to build a search goal shift graph;finally,the random walk algorithm was used to obtain the query recommendations in the search goal shift graph.(4)Through in-depth analysis of the exploratory search process,we found that it was difficult to form a relatively clear search idea in the user's mind before performing exploratory search,often searching for unnecessary information,making the entire exploratory search process full.In order to solve this problem,we have designed a search path recommendation method for exploratory search.The method utilizes the search target migration map to directly recommend a set of query paths for the user according to the initial query of the user in the initial stage of the exploratory search,thereby helping the user to quickly plan the learning content and avoid searching for the irrelevant information,thereby shortening the search time.(5)In order to ensure the accuracy and objectivity of the evaluation results,we designed a variety of exploratory search tasks and different test method for these recommendation methods proposed in this paper.