Split documents into batches
Goal: Divide documents into groups by layouts to facilitate labeling.
Input:
- Snapshot with 609 OCRed documents.
Output:
- Batches ready for tagging split on the basis of "secretary of state" office in which they were filed.
In order to decrease the number of tagging mistakes and increase speed, all documents should be split into batches by structure. To assign a document a particular layout name it's recommended to look the documents and try to define all unique names that will be used as unambiguous identifiers of a particular layout: for example, if we use issuer_name as an indicator for classification, it'll be necessary to define the full list of issuers in the data set - Tampa Holdings, Workfusion Inc. etc. The list of layouts will be used then in classification task that will make the process of documents' sorting more convenient. The task also has one useful option - if during classification you encounter a new layout (for example, new issuer), which is not in your list, you'll be able to add it manually right in the task.
In our example, we'll use "secretary of state" as structure indicator. During documents analysis we identified 4 layouts: New Hampshire, Massachusetts, Georgia and California.
As the result of classification task we have the following layouts' representation:
- Georgia: 211
- New Hampshire: 144
- Massachusetts: 138
- California: 118