posted Jun 9, 2015, 4:25 AM by Eric Medvet
updated May 26, 2016, 3:47 AM
Many data-related activities at organizations of all sizes are concerned with low-level string processing, such as format transformation and validation, data cleaning, substring extraction and classification, and so on. Problems of this sort occur routinely in a one-off fashion as part of specific processes or activities that cannot be integrated in long-lived workflows, such as analysis of data gathered from the web or from other enterprise sources. The input stream may range from structured or semistructured data, such as database tables or spreadsheets, to unstructured data, such as text in natural language, as well as data whose structure is not available, such as a web page or a pdf invoice. These tasks are a vital ingredient of virtually every organization but are difficult to address efficiently: they are usually too simple to justify the cost and latency of a full-blown IT project, yet they are not simple enough to be solved by non-IT specialists.