In pickup and delivery services, transaction classification based on customerprovided free text is a challenging problem . This categorization is important for the business: it helps understand the market needs and trends, and also assist in building a personalized experience for different segments of the customers . In this paper, we focus on a specific use-case where a single category drives each transaction . We propose a cost-effectivetransaction classification approach based on semi-supervision and knowledgedistillation frameworks . This framework, with ALBERT as astudent and RoBERTa as teacher, is further referred to as R-ALBERT in thispaper . The model is in production and is used by business to understandchanging trends and take appropriate decisions . The approach identifies the category of a transaction using free text input given by the customer . We use weak labelling and noticethat the performance gains are similar to that of using human-annotatedsamples. The performance gains were similar to those of using unadapted ALBERt. This framework is further known as RoBERT, which is further called R-AlBERT

Author(s) : Rohan Sukumaran

Links : PDF - Abstract

Code :
Coursera

Keywords : albert - transaction - classification - business - semi -

Leave a Reply

Your email address will not be published. Required fields are marked *