Selecting Characteristic Patterns of Text Contributions to Social Networks Using Instance-Based Learning Algorithm IBL-2
| Autoři | |
|---|---|
| Rok publikování | 2017 |
| Druh | Článek ve sborníku |
| Konference | ENTERPRISE AND COMPETITIVE ENVIRONMENT |
| Fakulta / Pracoviště MU | |
| Citace | |
| www | https://ece.pefka.mendelu.cz/sites/default/files/imce/ece_2015_final.pdf |
| Klíčová slova | machine learning; instance-based algorithms; IBL-2; text mining; social networks; typical patterns; computational complexity reduction; classification |
| Popis | The presented research focuses on selecting typical patterns of textual entries written using a natural language (English) in a social network booking.com, which publishes sentiment of customers that used an accommodation service. This work deals with the possibility to find the patterns via text mining based on a machine-learning tool known as Instance-Based Learning (IBL). To reduce high computational demands of the basic algorithm IBL-1 (k-nearest neighbors), IBL-2 does not store sample candidates the function of which is successfully carried out by the already stored samples. The textual data are represented as bag-of-words with sparse vectors. Because the non-linearly increasing computational complexity depends on the number of samples as well as on their vocabulary, the potential candidates are firstly freed of common insignificant terms and then the vector sparsity is strongly decreased by removing words having a low frequency in relation to the number of samples. Then, IBL-2 rejects to store samples that duplicate the functionality of the already stored ones. As a result, the database contains only (or mainly) significant samples that represent characteristic patterns, which may be used for classification or another type of a following social network analysis. |