Research On Methods Of Diversifying Query Recommendation Based On User Behavior

Information explosion,which emerges with the rapid development of Internet technology is one of the characteristics of today's society.More and more users become increasingly dependent on the search engine to obtain knowledge.Query recommendation,as an important auxiliary tool for users in employing search engines,can make up for query words limitations and assist users in searching for the right contents.Query recommendation methods that based on popularity query recommendation focus on the popularity of query words and neglect the search demands of vulnerable groups.Once users search polysemy,there is a high possibility of disappointing results.Diversifying query recommendation method can properly address this problem.As is known to all,the premise of diversified recommendation method is that classification systems exist in the collection of recommended subjects.Currently,existing classification systems,for example,wikipedia,wordnet and other ontology libraries,are adopted in most of the diversified query recommendation.Problems may exist in the recommendation process due to the fact that most of these recommendation systems adopt existing classification systems.Firstly,using the existing classification system makes it quite difficult to effectively resolve the data sparse problem.Secondly,the existing classification system cannot cover all the query words in query recommendation.Most extremely popular words will not be included in the classification system,which as a result will lead to the incomplete final recommendation.Thirdly,recommendation collection obtained through diversified recommendation method is not pertinent enough.Users of different backgrounds will hold verified understandings of the same query word.Diversified recommendation method,on the other hand,provides all users with the same recommendation collection due to its search target and recommendation strategy.To solve the above-mentioned problems,this paper proposes a method,which is building a dynamic search classification system based on users' query behavior;and provides the definition and building method of dynamic classification system;and comes up with diversifying query recommendation framework based on typical user behavior model.Moreover,this paper also comes up with 3 diversifying query recommendations methods based on dynamic classification system,and provides higher quality query recommendation services for different users.Thorough research has been conducted in the following aspects:(1)Dynamic classification system was built based on users' behavior.In the diversified query recommendation,classification system of the query words mostly employed the existing classification system,which would lead to sparse data,weak coverage of the existing system and lack of pertinence of the search results.To address the above-mentioned problem,this paper proposed a method,which is building a dynamic classification system based on typical users' behavior.Using typical users'behavior to build a query word classification system;defining and constructing typical users' behavior models;proposing diversified query recommendation framework based on users' behavior were all mentioned in this paper.(2)To solve the problem of sparse data and weak coverage of the existing system,this paper suggested a diversified query recommendation method based on epidemic behavior.This method adopted the dynamic search classification system;targeted at users who had little or no search log records;and provided users with a variety of epidemic recommendation service.(3)To solve the problem of weak pertinence of the search results,it suggested to adopt a diversified query recommendation method based on diffrent search background.This approach employed a query word classification system for users with some search log records.This could help and determine the user groups.Based on users' own query behaviors,the system offers relevant recommendations to them.(4)To solve the problem of weak pertinence in the traditional diversified recommendation method,this paper put forward a new approach that combined popularity and similarity.Employing dynamic query classification system,this approach aimed at users who had numerous search log records in the log.This could help and identify the user group and could present certain operating rules.Based on user's query behavior and the user groups they are in,the system could offer users individual recommendation service.