US 12,248,640 B2
Artificial intelligence based hybrid system and method for generation of word predictions based on language modelling
Rahul Prasad, Gurugram (IN); Ankit Prasad, Gurugram (IN); and Sumegha Yadav, Gurugram (IN)
Assigned to TALENT UNLIMITED ONLINE SERVICES PRIVATE LIMITED, (IN)
Filed by TALENT UNLIMITED ONLINE SERVICES PRIVATE LIMITED, Delhi (IN)
Filed on Jun. 28, 2022, as Appl. No. 17/851,695.
Claims priority of application No. 202211012599 (IN), filed on Mar. 8, 2022.
Prior Publication US 2023/0288990 A1, Sep. 14, 2023
Int. Cl. G06F 3/023 (2006.01); G06F 40/274 (2020.01); G06F 40/30 (2020.01)
CPC G06F 3/0237 (2013.01) [G06F 3/0236 (2013.01); G06F 40/274 (2020.01); G06F 40/30 (2020.01)] 14 Claims
OG exemplary drawing
 
1. An Artificial Intelligence (AI)-based hybrid system for dynamic word prediction and suggestion, the system comprising:
an electronic device comprising an input device, at least one processor, a memory unit operably and communicatively coupled to the at least one processor, and a display unit;
a sensor unit comprising a plurality of sensors and operably coupled to the at least one processor of the electronic device; and
an AI engine operably coupled to the sensor unit, the at least one processor, and a messaging application deployed on the electronic device, and wherein the AI engine is configured to:
identify a plurality of elements of an input message in an input language on the messaging application;
replace each of the plurality of elements of the input message with a corresponding integer present in a vocabulary of the input language and transform the input message into an array of specific numbers, based on the plurality of elements;
convert the array of specific numbers into a two-dimensional matrix of embeddings, wherein an embeddings include semantic information of the input message representing a context of the input message;
convert the two-dimensional matrix into a normal distribution array and generate a feature map, and process sequential information present within said feature map;
process said feature map to generate a one-dimensional array of a predetermined shape, and apply an activation function to said one-dimensional array to convert said one-dimensional array into an output array of probabilities based on said vocabulary, and wherein a shape of said output array of probabilities is based on a size of said vocabulary; and
select one or more words from said output array of probabilities as word suggestions corresponding to the input message.