Method and electronic device to assign appropriate semantic categories to documents with arbitrary granularity

A method for automatically assigning at least one semantic category of at least one document using an electronic device, wherein the method comprises:

• receiving, by the electronic device, the at least one document embedding indicating a semantic representation of the at least one document;
• determining, by the electronic device, a probable set of semantic categories of a plurality of semantic categories associated with the at least one document embedding based on an execution of the at least one document embedding on a plurality of proto-models, wherein each proto-model of the plurality of proto-models is a compressed version of a semantic model associated with a respective semantic category, wherein each semantic model comprises a set of vector embeddings of the at least one document used in training at least one semantic category of the plurality of semantic categories;
• receiving, by the electronic device, the semantic model associated with each of the probable set of semantic categories;
• executing, by the electronic device, the at least one document embedding on the received semantic model, wherein executing the at least one document embedding on the received semantic model comprises:
• determining similarity between the at least one document embedding and each embedding of the received semantic model;
• determining whether the highest similarity among the similarity between the at least one document embedding and each embedding of the received semantic model, is greater than a model threshold;
• and
• converting the highest similarity for each received semantic model that has the highest similarity greater than the respective model threshold, into a confidence score, wherein the confidence score is computed based on a difference between the highest similarity and the respective model threshold, wherein the confidence score is computed as:
• Confidence score=0.5+0.5*(HS−Th)/(1−Th)
• wherein HS is the highest similarity for each received semantic model that has the highest similarity greater than the respective model threshold, and wherein Th is the respective model threshold;
• and
• automatically assigning, by the electronic device, the at least one semantic category out of the probable set of semantic categories, of the at least one document embedding, wherein the at least one semantic category is associated with the received semantic model that has the highest confidence score.