I am currently working on a research project where I am building supervised classification models to assess text fragments. My general methodological approach roughly follows the design science principle as stated by Hevner and Peffers. Yet, I am wondering whether there is a special/adapted version of the general DSR framework that is specially tailored to machine learning artefacts.
Dear, as much i know machine learning, The first step is to find correct data with proper labeling for training your machine. secondly use different ML algorithms to check the validation and testing accuracy.
Marco:
Are some methodologies for data sciences research, the most popular are KDD and CRISP-DM. A general approach for text mining are:
- Collect data
- Preparing data (tokenization, steaming, stopword remove, etc.). This is one or the most important step.
- Applied classification model
- Evaluated
- Deploy
As Amin say, you need a labeling dataset to evaluate you proposed model. One popular dataset are Reuters. One aspect that need study if you text data have a single (1) or multi label (2) assignations. If you answer is 1, you can use any of the existent classification algorithm (instance, kernel, tree based method or neural network approach). If the answer is 2, your problem if a more complex, and you need to probe no imbalance label, and use one multi label classifications algorithm. For multilabel recommend MULAN or MEKA for Java development and scikit-multilearn for Python, are many implementations for Matlab too.
Best regards
It seems you are using a 'validated solution' approach. This is one of several approaches within the DRS methodology.
See: van Turnhout, Koen, et al. "Design patterns for mixed-method research in HCI." Proceedings of the 8th Nordic Conference on Human-Comp )
For more details. It should be fairly easy to link your approach to the framework.
Thank you all for your indeed helpful answers. They gave me several useful hints for my research.
Finally I would like to add one additional publication I came across during my further studies that may be helpful to others reading this QA:
Hirt, R., Niklas, K. and Satzger, G. 2017. “An End-to-End Process Model for Supervised Machine Learning Classification: From Problem to Deployment in Information Systems,” in DESRIST 2017 Research in Progress Proceedings of the 12th International Conference on Design Science Research in Information Systems and Technology, A. Maedche, J. vom Brocke and A.R. Hevner (eds.), Karlsruhe, Germany: KIT, pp. 55–63.
- Badges
- Science topic