Research On Ranking And Query Expansion Based On Polysemy

Internet has become a common way for modern people to obtain relevant information with the rapid growth of Web technology.The emergence of information retrieval systems such as search engine has dramatically enhanced the speed and accuracy of information acquisition through the internet.However,the modern information retrieval systems represented by search engine can not greatly satisfy user's need for information access.In order to make the search results closer to the user's current search needs,the search ranking and the query expansion technology are introduced and greatly improve the accuracy of the search results,thereby improving the user's search experience.However,the existing search ranking methods,usually based on traditional ranking algorithms like PageRank,do not consider the user's search intent.As a result,the search results do not satisfy the search needs and desire of users which hurts the user's search experience.Similar problems exist in query extension which can be used to improve the searching recall rate to a certain extent.However,due to incomprehension of user's search intent,the accuracy of query expansion still can not satisfy the user's current search needs.In the search ranking and query expansion problems we have discussed above,thereason why the search results can not match the user's interest is that the search system usually utilize only the query input of the user,as a consequence the search ranking and query expansion can only depend on the current user query input without consideration of the user's search intent.However,by analyzing the query condition,one of the root causes for this phenomenon is the existence of polysemy in query words.Due to common existence of word polysemy,queries are usually also polysemous.Therefore,for each meaning of the query,one type of result set corresponds to it.The search system can not distinguish between these results,but only mix them in the list of result sets according to the ranking algorithm,thus causing difficulties for users to find out the results which match their search intent.Targeting at the problems caused by the word polysemy in search ranking and query expansion,this dissertation proposes UIMP,a User Interest Model for Polysemy.Based on this Model,we deal with the impact of word polysemy in search ranking and query expansion.The main content of this dissertation is aiming at the problem of inaccuracy and mismatch of search results and user search intent caused by query polysemy in search ranking and query expansion.The motivation of this dissertation is that,by establishing a model based on polysemous query and user interest to match interest-related and filter out interest unrelated meanings of queries with user's current search intent in the process of search ranking and query expansion,to achieve the goal that the search results can better fit the user's search intent.This paper presents in-depth research in the following aspects.(1)User Interest Model for Polysemy(UIMP).The key reason why the search ranking and query expansion results can not fit the user's current search intention caused by polysemy is the lack of acquisition of the user's current search intention in relevant processing methods.In this dissertation,through in-depth analysis of users'possible interest sources,we select four types of effective user interest sources to extract users' current search interest,and establish the UIMP model.Based on the UIMP model,the most effective user interest can be selected to determine a definite meaning of the query which fits the user's current search interest in order to improve the accuracy of ranking and query expansion.(2)Ranking method based on UIMP.Through observation and related research,this dissertation finds that users usually only browse the results of the top position after the search engine returns the list of search results set.Considering the user's browsing habits,search ranking becomes the key to improve the user's search experience,and ranking is precisely the core step of the search process.Due to the lack of the current search intent of users,the existing ranking methods can only fit the ranking preference of the mass search,but the ranking experience of individuals is not significantly improved.Therefore,this dissertation proposed a ranking method based on UIMP,and designs ranking algorithm based on user interest to increase the ranking value of the results which fit the user interest,so as to improve the user's search experience.(3)Ranking method of search target gradation based on UIMP.With the continuous increase of Internet information,people are not satisfied with only using the Internet to find relevant information,but regard the search engine as a tool to discover related knowledge,which is particularly evident in the search process represented by exploratory search.In this kind of search process,there is an obvious characteristic that the user's interest is constantly changing.Therefore,based on UIMP,this dissertation proposed an interest gradation model to deal with the ranking results of this kind of search process,so that each ranking result in the search process can be more accurate and to some extent guides the user to discover relevant knowledge which they interested in.(4)Query expansion method based on UIMP.Existing query expansion methods lack the consideration of users' interests,which leads to the fact that when a query has a polysemy,the set of expanded words itself represents different meanings,and may also lead to the expansion words deviating from the user's current interests.The method proposed in this dissertation is an expansion word selection method based on user interest.The purpose of this method is to analyze the relevance between the meanings of the query and the user's current interest by user's interest,and then expands the query from a certain meaning based on the user's current search interest.Because of the integration of user interest factors,the method in this dissertation concentrates the expanded words on the most possible meaning direction of the current query,thus improving the efficiency and accuracy of the expansion.(5)Popular words query expansion method based on UIMP.With the continuous development of search technology,the way people search has gradually formed a certain regularity.In this dissertation,we observed that a large proportion of searchers will search the same content within a certain period of time,which is called search hotspot.The corresponding query usually contains a large number of popular words.Existing query expansion methods usually do not consider the popularity of queries,but use a unified query expansion method for processing,which resulting in query expansion effect can not reflect the query fever and the users' interest.This dissertation proposed a popular query expansion method based on UIMP,and established a word polysemy model based on popular queries,so that query expansion can obtain higher processing efficiency with popular queries.