Towards improving WEBSOM with multi-word expressions

Abstract

Large quantities of free-text documents are usually rich in information and covers several topics. However, since their dimension is very large, searching and filtering data is an exhaustive task. A large text collection covers a set of topics where each topic is affiliated to a group of documents. This thesis presents a method for building a document map about the core contents covered in the collection. WEBSOM is an approach that combines document encoding methods and Self-Organising Maps (SOM) to generate a document map. However, this methodology has a weakness in the document encoding method because it uses single words to characterise documents. Single words tend to be ambiguous and semantically vague, so some documents can be incorrectly related. This thesis proposes a new document encoding method to improve the WEBSOM approach by using multi word expressions (MWEs) to describe documents. Previous research and ongoing experiments encourage us to use MWEs to characterise documents because these are semantically more accurate than single words and more descriptive.