Huo Hua, Feng Boqin
(School of Electronics and Information Engineering, Xi'an Jiaotong University, Xi'an 710049, China)
Abstract: In order to increase the performance of the combined retrieval
system, a multi-engine fusion method based on a mixture model was presented. The method
describes the relevant score distribution of the relevant and non-relevant documents using
Gaussian density function and exponential density function respectively. Based on the
algorithm of the mixture model the relevant scores are normalized, the scores of
non-relevant documents are estimated and combined, which consider both the difference
between relevant and non-relevant documents in the score distribution and the retrieval
performances of the member search engine estimated by users. Experimental results show
that the average search accuracy is improved by 37.8% compared with member engines,and
also higher than three often used fusion methods of Sum-CombSUM, Sum-CombMNZ, and
Standard-CombSUM.
Keywords: relevance score; mixture model; multi-engine fusion; score combining