| Summary: | 碩士 === 國立中興大學 === 資訊管理學系所 === 104 === These days, people can continuously receive all kinds of information immediately because computer technology, communication product and Internet become more convenient and more universal. Everyone can use keywords to find out the information, which is not only just words but also with images, that they want to know by web search engine in the Internet. However, to offer people the information, which are documents or articles with images, what they need exactly is not a simple task. Therefore, many researchers begin to study how to give appropriate topic or annotation to those documents or images automatically. In text mining, Bag-of-Words Model (BOW) is a regular way to represent a document or an article. The words in the document were ranked by the weights calculated by TF-IDF (Term Frequency–Inverse Document Frequency). Some researchers consider the image in an article can be annotated through its surrounding text. Nevertheless, can the annotations, which is produced by any surrounding text and computation methods, tally with the cognition of human beings exactly? According our research, we found that caption is better than article’s content about the effect of annotation. Furthermore, the things depicted by article’s content do not match the image sometimes. For that reason, this research is going to analyze the relationship between the result and the cognition of human beings.
|
|---|
