Automated document metadata extraction

Abstract

Web documents are available in various forms, most of which do not carry additional semantics. This paper presents a model for general document metadata extraction. The model, which combines segmentation by keywords and pattern matching techniques, was implemented using PHP, MySQL, JavaScript and HTML. The system was tested with 40 randomly selected PDF documents (mainly theses). An evaluation of the system was done using standard criteria measures namely precision, recall, accuracy and F-measure. The results show that the model is relatively effective for the task of metadata extraction, especially for theses and dissertations. A combination of machine learning with these rule-based methods will be explored in the future for better results.