Support Vector Mind Map of Wine Speak
Models created by blackbox machine learning techniques such as SVM can be difficult to interpret. It is because these methods do not offer a clear explanation of how classifications are derived that is easy for humans to understand. Other machine learning techniques, such as: decision trees, produce models that are intuitive for humans to interpret. However, there are often cases where an SVM model will out preform a more intuitive model, making interpretation of SVM trained models an important problem. In this paper, we propose a method of visualizing linear SVM models for text classification by analyzing the relation of features in the support vectors. An example of this method is shown in a case study into the interpretation of a model trained on wine tasting notes.
KeywordsModel visualization SVM Support vector weight
This work was supported by JSPS KAKENHI Grant Number 15J04830.
- 2.Sakai, T., Hirokawa, S.: Feature words that classify problem sentence in scientific article. In: Proceedings of the 14th International Conference on Information Integration and Web-Based Applications and Services, pp. 360–367 (2012)Google Scholar
- 4.Hirokawa, S., Flanagan, B., Suzuki, T., Yin, C.: Learning winespeak from mind map of wine blogs. In: Yamamoto, S. (ed.) HCI 2014, Part II. LNCS, vol. 8522, pp. 383–393. Springer, Heidelberg (2014)Google Scholar
- 5.Flanagan, B., Wariishi, N., Suzuki, T., Hirokawa, S.: Predicting and visualizing wine characteristics through analysis of tasting notes from viewpoints. In: Stephanidis, C., Tino, A. (eds.) HCII 2015 Posters. CCIS, vol. 528, pp. 613–619. Springer, Heidelberg (2015). doi: 10.1007/978-3-319-21380-4_104 CrossRefGoogle Scholar