Research on Equivalence Discrimination of Social Security Problems

摘要

Problem equivalence discrimination aims to compare two problems and determine whether they are semantically equivalent, it is a binary classification problem. This is a crucial task for QA. This paper studies the problem equivalence discrimination in the field of Chinese social security, proposes a question equivalence discrimination system based on Bert + Focal-Loss, and constructs the question data set in the field of social security. It solves the problem of imbalance between positive and negative samples, so as to input more negative samples for pre training, improve the accuracy of the system, construct the question data set of social security field according to a certain annotation proportion, fill in the lack of data set in professional field, and expand the data set according to the users' wishes.