Recurrent neural network learning for text routing
Stefan Wermter · 1999
This paper describes new recurrent plausib-ility networks with internal recurrent hys-teresis connections. These recurrent con-nections in multiple layers encode the se-quential context of word sequences. We show how these networks can support text routing of noisy newswire titles according to different given categories. We demon-strate the potential of these networks us-ing an 82 339 word corpus from the Re-uters newswire, reaching recall and preci-sion rates above 92%. In addition, we care-fully analyze the internal representation us-ing cluster analysis and output representa-tions using a new surface error technique. In general, based on the current recall and pre-cision performance, as well as the detailed analysis, we show that recurrent plausibility networks hold a lot of potential for develop-ing learning and robust newswire agents for the internet. 2