| Abstract |
Currently, many product sellers like to advertise their supplement products on web. However, there are some ads showing messages to deceive consumers. This work presents a system to extraction supplement products advertisement data from web and classifies the illegal ads that show misleading properties. Therefore, we proposed a method to automatic search and extract ads text from multiple websites using defined supplements keywords. Then, the extracted ads texts were preprocessed by word segmentation, stop words eliminate methods, and classified by the misleadingness words database that be prohibited by the Food and Drug Administration of Thailand. All illegal classified ads would be computed TF-IDF vectors and stored in an illegal reference database. However, some illegal ads avoided to use the prohibited words that they can be classified as legal. Therefore, they would be re-classified by measuring the similarity with all ads in the reference database. The experimental results show that the proposed system can detect forbidden ads with an accuracy of 0.775. |
Publication
Journal Publication
ไทย