Optimization Of Information Retrieval Based On The Elevance Judgement

The application of Information Retrieval has significantly improved users’information obtaining efficiency. The essential of information retrieval is a paring process between user demands and user sets, in which users express their needs via initial inquiry and accord to some retrieval model to figure out the information they need in retrieval system. Thus, Information retrieval is the retrieval between user needs and retrieval objects. Before21st century, information retrieval has been focused on research on a systematical level. Researchers have made endeavors to improve retrieval performance for the information system via exploration of retrieval principles, organization formats of information sources and grading algorithm.Together with the technology and society improvement as well as human-computer interaction and intelligent retrieval blooming, human effect plays a much more important role in information retrievals and people-oriented relevant feedback will definitely become the main style of information retrievals. Researchers gradually make deep investigation of impacting elements in and technology improvement for the information retrieval, in which relevance feedback technology is a typical example. Various of elements will affect the information relevance. After comprehensive explorations by researchers, it is pointed out that users determined the quality of information query and are the basis of information query; hence, user is the key impact factor of relevant feedback. The core of information retrieval system is correlation evaluation and user end-member is the dominant part of correlation evaluation since the evaluation process is affected by users’self condition and surrounding environment and is altered by subjectivity. In the system retrieval research, especially for mainland china research, the query-expansion based information retrieval have been well developed while the user-based relevance feedback research is rare. However, few case and experimental studies have been conducted.In my research, information modification technologies and current study situation will be reviewed systematically; information retrieval definition, evaluation, affecting factors and enhancing technologies will be summarized with an emphasis description of user’s relevance feedback technologies. Furthermore, a new information retrieval system is built up. We take five international standard test sets which are Cranfield, Medline, CISI, NPL and CACM as corpus and create system index; meanwhile, we adopt widely used vector special model as information retrieval model and utilize TF-IDF algorithm to calculate weights. The system can optimize the query vectors via modified formulas of Ide Dec-hi. Based on the system we examined three information retrieval experimental groups. The first set goes without relevance feedback. Users are not designed to participate in the feedback. After the input of retrieval inquiries, retrieval system export retrieval document set. This part is utilized as BASELINE and retrieval results are taken as the frame of reference. The second part is created based on correlation feedback. After the system export the retrieval results, users will distinguish relevant references and irrelevant references among top surface N (referring to experiments by Salton and Buckley, N is set to15) references. Afterwards, second retrieval is made. According to users’ feedback results, the systems will calculate the similarity between reference sets and query set with the Ide Dec-hi algorithm and send back the retrieval results to the users lately. This part is the control group. The third part, which is similar to the second part, is operated based on relevance criterion set and relevance feedback. The difference is that the relevance criterion by users is not made randomly but with given criterion. This part is experimental group. Our empirical studies revealed that Relevance feedback technology and criterion set based method can greatly enhance information retrieval.Finally, flaws and perspectives of my search are promoted..