2021/04/28 by Shalin Shah, Shah, Shalin, Ryan Siskind +1
Computer Science · Decision Sciences · #Computation and Language (cs.CL) #Data Quality and Management #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2105.03316
openalex publication_date 2021/04/28 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Named entity recognition (NER) has been studied extensively and the earlier\nalgorithms were based on sequence labeling like Hidden Markov Models (HMM) and\nconditional random fields (CRF). These were followed by neural network based\ndeep learning models. Recently, BERT has shown new state of the art accuracy in\nsequence labeling tasks like NER. In this short article, we study various\napproaches to task specific NER. Task specific NER has two components -\nidentifying the intent of a piece of text (like search queries), and then\nlabeling the query with task specific named entities. For example, we consider\nthe task of labeling Target store locations in a search query (which could be\nentered in a search box or spoken in a device like Alexa or Google Home). Store\nlocations are highly ambiguous and sometimes it is difficult to differentiate\nbetween say a location and a non-location. For example, "pickup my order at\norange store" has "orange" as the store location, while "buy orange at target"\nhas "orange" as a fruit. We explore this difficulty by doing multi-task\nlearning which we call global to local transfer of information. We jointly\nlearn the query intent (i.e. store lookup) and the named entities by using\nmultiple loss functions in our BERT based model and find interesting results.\n