Putting Self-Supervised Token Embedding on the Tables

摘要

Information distribution by electronic messages is a privileged means oftransmission for many businesses and individuals, often under the form ofplain-text tables. As their number grows, it becomes necessary to use analgorithm to extract text and numbers instead of a human. Usual methods arefocused on regular expressions or on a strict structure in the data, but arenot efficient when we have many variations, fuzzy structure or implicit labels.In this paper we introduce SC2T, a totally self-supervised model forconstructing vector representations of tokens in semi-structured messages byusing characters and context levels that address these issues. It can then beused for an unsupervised labeling of tokens, or be the basis for asemi-supervised information extraction system.