Research On The Application Of Query Suggestion In Map Search

With the increasing of a variety of information on the Internet nowadays,people become more and more dependent on getting information needed from the vast sea of knowledge.Whereas how to help users quickly locate and get the information of what they want has become a problem.Using query suggestion technology to recommend relevant keywords for users in the processing of searching,has been widely applied to major search engines and e-commerce merchandise retrievals.As one of the most fundamental characteristics of the network search engines,query suggestion is regarded as a very effective way of human-computer interaction in information retrieval.But in daily life,query keywords of users are not definite,or the query keywords have inherent ambiguity,which leads to that search engines are very difficult to accurately locate the query information of users.Thus,although the query suggestion techniques have been widely applied in major search engines,recommendation results of the same keyword varies a lot.So to excogitate the query suggestion algorithm that can help users find what they desire accurately and quickly is very important.Except for the full text search engines which are most general,the application of query suggestion techniques in other industries is far from meeting the broad requirements of users,such as the areas of journalism,booking and so on.With the popularity of GPS and GPS-enabled mobile devices,how to allow users to obtain accurate query recommendation is becoming increasingly important in the map search.This thesis focused on map search engine,and used the user query logs in map search as a dataset(i.e.POI data).In order to facilitate the smooth progress of the experiment,data set preprocessing was needed before the experiment.Because of the huge data sets,of which 300,000 records were collected for the experiment.Given the sparsity of map data,the thesis added new features,the weight of query terms calculated by special algorithms,and extracted the query terms with the largest weight as keywords,in which way to strengthen the relationship between the query terms.In the experiment,the thesis first transformed query suggestion problem into an un-weighted bipartite graph of query-pguid,and constructed vector space model VSM(Vector Space Model)based on this un-weighted bipartite graph of query-pguid,and then established a standardized random matrix with normalization processing followed.Next,the thesis employed a modified restarting-style random walk algorithm,ItemRank,to measure the similarity between vertices in the graph.After the identification of recent popular search records by clustering,top k query keywords were selected and recommended to users.Finally,evaluation methods for a recommendation algorithm were introduced to compare the results of this algorithm and other algorithms and to analyze the effects of the experiment.Experiments showed that the algorithm could be applied and get a high degree of accuracy in map search.