An Information Retrieval Model Based On Multilayer Markov Network

With the explosive development of internet, information retrieval technology iskeeping innovation as the internet resource having an exponential growth. Users comeinto contact with plentiful information using information retrieval system, whichbrings great convenience in common life, learning and work. The key problem is howto search contents that using request quickly and accurately in front of vast dates.The method matching terms in document with query terms simply leads toterrible query result, therefore, it has been a research focus that improving theretrieval accuracy by mining more effective information. Previous studies have shownwhen the additional information combines with the information retrieval process, itwill lead to a better retrieval effect. As for a specific query, we can take fullyadvantage of the existing queries correlation information, terms and documents forquery expansion and reconstruction. On this way, we propose an information retrievalmodel based on multilayer Markov network. The Markov network is constructed bythe correlation of query network, term network and document network. The modelcan be integrated correlation between terms, documents and queries. In order to cutdown calculation, the model of clique was proposed. The experiments on the standarddata sets have indicated that our model can not only integrate information of threeaspects more effectively but also improve the effect of retrieval.There are several innovations in this paper as follows:1. Through learning the document, Markov net can be built according to thecorrelation of words, documents and queries respectively.2. Threshold value was set to avoid too much noise information. Also a model ofclique was proposed to cut down calculation. By exacting word group, documentgroup and query clique information respectively, different weights were distributed tothese three information, which could be used in the formula of document and queryprobability, so that the final probability to document and query comes out.