How to improve robustness in Kohonen maps and display additional information in Factorial Analysis: Application to text mining

Abstract

This article is an extended version of a paper presented in the WSOM'2012 conference [1]. We display a combination of factorial projections, SOM algorithm and graph techniques applied to a text mining problem. The corpus contains 8 medieval manuscripts which were used to teach arithmetic techniques to merchants.Among the techniques for Data Analysis, those used for Lexicometry (such as Factorial Analysis) highlight the discrepancies between manuscripts. The reason for this is that they focus on the deviation from the independence between words and manuscripts. Still, we also want to discover and characterize the common vocabulary among the whole corpus.Using the properties of stochastic Kohonen maps, which define neighborhood between inputs in a non-deterministic way, we highlight the words which seem to play a special role in the vocabulary. We call them fickle and use them to improve both Kohonen map robustness and significance of FCA visualization. Finally we use graph algorithmic to exploit this fickleness for classification of words.

Search for Meaning Through the Study of Co-occurrences in Texts

“…A first idea would be to use hierarchical cluster analysis or clustering methods based on graph theory as those we used in the last part of Bourgeois et al [2015]. The problem is that the data we want to deal with is actually too big for those algorithms to be efficient -those ones having running time asymptotically equivalent to n 3 or worse.…”

“…However, due to the limitations of FCA (in terms of projection quality, for example), we propose to use a variant of the SOM algorithm which deals with contingency tables (see Oja and Kaski [1999] for other applications of SOM to text mining). See Cottrell et al [1998], Bourgeois et al [2015] for a definition of this variant of SOM, called KORRESP.…”

“…These pairs are called fickle pairs. This question was addressed by de Bodt et al [2002] in a bootstrap frame and used for text mining in Bourgeois et al [2015].…”

Advances in Computational Intelligence

“…The corpus (Lamassé [2012], Bourgeois et al [2015]) includes eight treaties on arithmetical education written in the 15 th century, in vernacular language (French). Their purpose was to teach merchants about arithmetics, but also to develop theory and knowledge.…”

Web mining based on one-dimensional Kohonen’s algorithm: analysis of social media websites

“…The Kohonen algorithm is a neural network proposed by Kohonen, which maps a distribution of vectors of any dimension into a lower-dimensional space, generally one or two, while maintaining a high degree of topological ordering, or neighborhood preservation. It is widely adopted to visualize high-dimensional data sets and to mine online data [1][2][3][4].…”