US 12,147,768 B2
Natural language bias detection in conversational system environments
Gandhi Sivakumar, Bentleigh (AU); Naeem Altaf, Round Rock, TX (US); Wesley M Devine, Durham, NC (US); and Rizwan Dudekula, Bommanahalli (IN)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on May 18, 2021, as Appl. No. 17/324,043.
Prior Publication US 2022/0374604 A1, Nov. 24, 2022
Int. Cl. G06F 40/30 (2020.01); G06F 16/9032 (2019.01); G06N 20/00 (2019.01)
CPC G06F 40/30 (2020.01) [G06F 16/90332 (2019.01); G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A method for detecting natural language (NL) bias by a conversational system comprising:
determining an NL bias in one of either a set of training questions used to train a machine learning model used by the conversational system;
selecting a user intent or a user question received by the conversational system;
determining NL characteristics of the user question and NL characteristics of the set of training questions;
comparing the NL characteristics of the user question to the NL characteristics of the set of training questions;
identifying the NL bias associated with the machine learning model to preferentially associating the user queries to the user intent; and
adjusting the NL bias of the training questions or the user question by performing a corrective action, wherein performing the corrective action is based on identifying a distribution of intents, and wherein the corrective action includes retraining the machine learning model; and
correctly classifying, by the retrained machine learning model, a type of user question as exemplified by the user question.