Interactive annotation process Sample Clauses

Interactive annotation process. Figure 3.2. The workflow of interactive annotation process In the online learning mode (See Figure 3.2) of information extraction, the user either manually extracts information from each document, or inspects and correct, if necessary, any prefilled values by the system in the “Output Form”. As the user moves through the document collection, the system learns features of correct and incorrect answers by comparing system extracted values and manually revised ones. Through this process, the decision model improves, and the amount of information that the system is able to correctly prefill grows over time. As one might expect, the "Output Form" for the first few documents may be empty. When perform classification, user specifies the following information to construct a classification attribute: input attributes, labels of output classes, and machine learning model of classification. While processing each instance, user specifies the class label for each classification attribute. When new instance gets loaded, the system populates labels for classification attributes automatically. User then verify and revise system generated class label to further improve the accuracy of classification. Once the decision model accrues an acceptable level of accuracy, the user has the option to turn off manual review and to allow the system to complete the extraction/classification for the remaining documents in batch mode. A demo video can be found in YouTube [8].