Support cases resolution retrieval
Coveo provides search and recommendation software for customer support systems in which customers can ask for help by entering a case description and, on the other side, support agents must find the solution. In this context, queries to Coveo are in the form of long texts describing a problem and potential solutions are knowledge articles or help documents.
Support cases resolution retrieval (SCRR), at its very core, requires a mapping of casual English text, possibly with grammatical mistakes, to well-formed formal English documents. This mapping problem closely resembles a well-studied problem in natural language processing (NLP) that is neural machine translation (NMT). However, unlike most publicly available datasets of NMT, SCRR aims to map a long query sequence (the support case) to an even longer target sequence (the resolution document). Although, recement language model based approaches such as GPT, BERT, XLNET etc. have provided means of overcoming this issue, practical uses are yet to be explored.