### abstract ###
In data analysis new forms of complex data have to be considered like for example (symbolic data, functional data, web data, trees, SQL query and multimedia data,

)
In this context classical data analysis for knowledge discovery based on calculating the center of gravity can not be used because input are not  SYMBOL  vectors
In this paper, we present an application on real world symbolic data using the self-organizing map
To this end, %%@ we propose an extension of the self-organizing map that can handle symbolic data \\  keywords:  Classification, Self organizing map, symbolic data, dissimilarity
### introduction ###
The self-organizing map(SOM) introduced by Kohonen  CITATION  is an unsupervised neural network method which has both clustering and visualization %%@ properties
It can be considered as an algorithm that maps a high dimensional data space,  SYMBOL , to a lower dimension, generally 2, and which %%@ is called a map
This projection enables the input data to be partitioned into "similar" clusters while preserving their topology
Its most similar %%@ predecessors are the k-means algorithm  CITATION  and the dynamic clustering method  CITATION , which operate as a SOM without topology preservation %%@ and therefore without easy visualization
In data analysis, new forms of complex data have to be considered, most notably symbolic data (data with an internal structure such as interval %%@ data, distributions, functional data, etc ) and semi-structured data (trees, XML documents, SQL queries, etc )
In this context, classical data %%@ analysis based on calculating the center of gravity can not be used because input are not  SYMBOL  vectors
In order to solve this problem, %%@ several methods can be considered depending on the type of data (for example projection operators %%@ for functional data  CITATION )
However, those methods are not fully general and an adaptation of every data analysis algorithm to the resulting %%@ data is needed
The Kohonen's SOM is based on the center of gravity notion and unfortunately, this concept is not applicable to many kinds of complex data
In this paper we propose an adaptation of the SOM to dissimilarity data as an alternative solution
Our goal is to modify the SOM algorithm to allow its implementation on dissimilarity measures rather than on raw data
To this end, we take one's inspiration from the work of Kohonen  CITATION
To apply the method, only the definition of a dissimilarity for each type of data is necessary and so complex data can be processed
