Fecha de publicación: -- Fuente: WIPO "tomato" A machine-learning architecture (MLA) can generate a modifier recommendation for a target base object. In some examples, the MLA can include an embedding layer that generates an embedded set of previously selected modifiers, an embedded set of base object features, and an embedded set of candidate modifier features. The MLA can also include an average embedding layer that generates an average embedding of the embedded set of previously selected modifiers; a concatenation layer that concatenates the average embedding with the embedded set of base object features and the embedded set of candidate modifier features, to thereby produce a concatenated embedding; and a feed-forward neural network layer that receives the concatenated embedding as input and generates dense embeddings as output. The MLA can further include an output layer that receives the dense embeddings as input and generates a final score for the candidate modifier based on the dense embeddings.