Research On Agent-based Result Merging In Meta-search Engine

Given the ranked lists of documents returned by multiple component search engines in response to a given query,the purpose of result merging in meta-search is to combine these lists into a single one.As one of the key technologies,Result merging technology is directly affects the users' satisfaction evaluation.The existing result merging methods often use a single ranking algorithm,but it will cause distinct performance when results overlap rate is different.So it will affect the performance of meta-search engine and lead to significant difference when user searches different query words.In order to solve this problem,we propose a dynamic result merging method based on multi-agent.This method selects multiple ranking algorithms as member algorithms,and uses intelligent agent to sense the results overlap rate under current query words.When the overlap rate changed,our method will pick out a best algorithm from the member algorithms to merge the results.By this mean,the precision of the meta-search and users' satisfaction will be increased.This paper makes some main contributions including design dynamic result merging method and result merging subsystem based on multi-agent.The specific contents are as follows:(1)By analyzing the limitations of a single algorithm,the strategy of dynamic scheduling ranking algorithm is determined.The strategy selects multiple algorithms as member ranking algorithm,and takes the results overlap rate as the algorithm scheduling condition.We use static empirical learning to determine the dynamic scheduling strategy,that is,the optimal algorithm at the current result overlap rate.(2)We design duplicated web pages detection during the process of results merging,propose a method and a process of the detection,and determine the overlap rate formula.(3)Since the results returned by the component search engine do not contain the initial correlation scores of the web pages,we use the logistic regression equation to estimate the initial correlation scores of the Comb MNZ algorithm and the SDM algorithm.(4)According to the selection criteria,the member ranking algorithm of dynamic scheduling is determined,and the algorithm idea and formula of these algorithms are analyzed.(5)We use the user click log data to analyze the user's preference for the component search engines and the user's search topic interest.We use component search engines weight and user interest weight to correct the result of the dynamic merging and realize the result merging based on user's interest.(6)In this paper,the Agent-based Result Merging Subsystem is implemented on the "iSearch" meta-search engine,the system structure diagram is analyzed and the agent modules in the subsystem are analyzed.(7)We design experiments to verify the work done by the paper,the experiment is divided into three parts.First,we compare our method with the Borda Fuse,Comb MNZ and others in the dynamic result merging experiment.It is verified that our method has a high precision.Second,we compare the result sequence of the login user and the unregistered user when they search same key words in the experiment of result merging based on the user's interest.It is verified that the system can meet the user's interest demand after adding the user interest factor.Third,we compare the agent-based result merging system and unused agent system in the performance experiment.It is verified that the agent-based result merging system has a certain superiority in time performance.