How to Deal with Heterogeneous Data?

In the context of Web 2.0, text content is often heterogenous (i.e. lexical heterogeneity). For instance, some words may be shortened or lengthened with the use of specific graphics (e.g. emoticons) or hashtags. Specific processing is necessary in this context. For instance, with an opinion classification task based on the message SimBig is an aaaaattractive conference!, the results are generally improved by removing repeated characters (i.e. a). But information on the sentiment intensity identified by the character elongation is lost with this normalization. This example highlights the difficulty of dealing with heterogeneous textual data content. The following sub-section describes the heterogeneity according the document types (e.g. images and texts).