Application of the Recommendation Architecture for discovering associative similarities in text

We investigate the use of the Recommendation Architecture (RA) for discovering associative similarities in text documents. RA is a connectionist model that simulates the pattern synthesizing and pattern recognition functions of the human brain. For this purpose a set of experiments has been carried out to adjust the parameters of the system to classify newsgroup postings belonging to 10 different categories. The variation and the poor quality of such a data set poses an interesting challenge to any intelligent classification system. A suitable feature selection scheme is devised to represent the input document set. Then the input is organized by the system into a hierarchy of repeating patterns that sets up a preferred path to the output. We report on the key findings of this experiment and the features of the Recommendation Architecture model that makes it suitable for classification of noisy and complex real world data.