Learning - based text mining in analysis of free - text responses to the ( 2013 ) Wales Cancer Patient Experience Survey

 Learning-based text mining has the potential to save time and resources in analysing free-text data from patients.  The possibility of using this approach, and the quality of the results that it produces, are dependent upon the size and quality of the training data sets available for sorting the free-text material.  Care must be taken when verifying the data sorted by text mining, ensuring full coverage of the free-text material and where necessary undertaking manual coding of the remainder of unsorted data. Researchers must also remain alert to the potential presence of novel data that does to map to the existing taxonomy of thematic categories used to classify responses.  In future, a rules-based approach to text mining may be preferable to a learning-based approach. The former allows for direct control over data sorting through manual construction of rules, and offers the potential for integrating expert knowledge into the sorting of data (this is not practical with the latter).  Exploration of rules-based text mining in analysis of free-text comments from patients is currently under way at the University of Southampton, Faculty of Health Sciences, in partnership with Nominet UK 1 .