Tolerant Information Retrieval: Neural Networks for adaptivity and flexibility in searching Export

@phdthesis{citeulike:3379783, abstract = {Information retrieval deals with vague queries and vague models of user be-havior. Neural networks are a method to process vague information and implement cognitive skills. This thesis provides an overview of the state-of-the-art of neural networks in information retrieval by analysis, clustering and evaluation of a large number of systems. The shortcomings of existing models have led to the development of the COSIMIR model, which is based on the neural backpropagation algorithm. It learns from examples to compare queries and documents, a central process in information retrieval. The cognitive approach replaces a formal model and leads to higher adaptivity and tolerance with regard to user interests. The transformation network is another system which is based on the backpropagation algorithm and which makes the retrieval of heterogeneous data possible. In several experiments, the COSIMIR model and the transformation network are tested with real world data. The COSIMIR model achieves good results for factual retrieval. The experiments with the transformation network and alternative methods lead to different results for different data sets. Which method performs best depends to a considerable extent on the data. Where the different methods are of comparable quality, the overlap of the results is relatively low. It is therefore recommended that fusion methods be employed.}, author = {Mandl, Thomas}, citeulike-article-id = {3379783}, keywords = {adaptive, ann, back-propagation, connectionist, ir, model}, month = {June}, posted-at = {2008-10-06 12:34:15}, priority = {0}, school = {University of Hildesheim}, title = {Tolerant Information Retrieval: Neural Networks for adaptivity and flexibility in searching}, year = {2001} }

TY - THES ID - citeulike:3379783 L3 - citeulike-article-id:3379783 N2 - Information retrieval deals with vague queries and vague models of user be-havior. Neural networks are a method to process vague information and implement cognitive skills. This thesis provides an overview of the state-of-the-art of neural networks in information retrieval by analysis, clustering and evaluation of a large number of systems. The shortcomings of existing models have led to the development of the COSIMIR model, which is based on the neural backpropagation algorithm. It learns from examples to compare queries and documents, a central process in information retrieval. The cognitive approach replaces a formal model and leads to higher adaptivity and tolerance with regard to user interests. The transformation network is another system which is based on the backpropagation algorithm and which makes the retrieval of heterogeneous data possible. In several experiments, the COSIMIR model and the transformation network are tested with real world data. The COSIMIR model achieves good results for factual retrieval. The experiments with the transformation network and alternative methods lead to different results for different data sets. Which method performs best depends to a considerable extent on the data. Where the different methods are of comparable quality, the overlap of the results is relatively low. It is therefore recommended that fusion methods be employed. TI - Tolerant Information Retrieval: Neural Networks for adaptivity and flexibility in searching PB - University of Hildesheim KW - adaptive KW - ann KW - back-propagation KW - connectionist KW - ir KW - model AU - Mandl, Thomas PY - 2001/06// ER -