Focusing on the recent Big-Data boom, we develop vegetable price prediction models incorporating unstructured web-based data obtained from various online web-sites such as news, blogs, cafes, and so on. For empirical analysis, we employ Bayesian structural time series (BSTS) models with four unstructured indices using a text-mining tool, Textom; the amount of buzzwords, the amount of search keywords, the 'term frequency-inverse document frequency' (TF-IDF), and the 'degree-centrality-weighted term frequency' (DCTF). Then, the models are applied to three vegetable products of garlic, onion, and pepper in Korea. Results show that prediction performances can be remarkably improved by the introduction of unstructured indices for all products. The degree of improvement and the selection of unstructured indices can vary by vegetable products with their market and web-based environments.