| CPC G06Q 30/0631 (2013.01) [G06Q 10/087 (2013.01); G06Q 30/0234 (2013.01)] | 17 Claims |
|
1. A computer-implemented method for identifying product in a distributor inventory system that fulfills a product request made via a natural language query, said method comprising:
receiving a natural language query as a product request, said natural language query comprising a plurality of words in a sequential order;
vectorizing each of the words and thereby generating a plurality of corresponding word-vectors through a word embedding model that is trained on product-specific vocabulary;
aggregating the plurality of word-vectors to form a query embedding by concatenating the plurality of word-vectors;
processing the query embedding utilizing a trained product category classifier ML model and thereby predicting in which of a plurality of predefined product categories the requested product belongs;
generating, based on the plurality of sequential order words of the natural language query, a forward sequence vector and a backward sequence vector;
selecting a trained ML model specific to the predicted product category from a plurality of product category-specific trained models that are trained for different product categories; and
concatenating the forward and backward sequence vectors and processing that concatenation using the trained ML model specific to the predicted product category and thereby identifying one or more product attributes embodied in the natural language query that each correspond to a predetermined key-characteristic of the category;
generating an output comprising an indication of the one or more product attributes and an indication of the predicted product category and providing the output to a search engine.
|