| CPC G06F 16/353 (2019.01) [G06F 16/313 (2019.01)] | 20 Claims |
|
1. A computer-implemented method comprising:
responsive to receiving a sequence of an input document, generating by a neural network, using a trained keyphrase generation model, a set of keyphrases corresponding to the input document, a keyphrase in the set of keyphrases comprising a word summarizing a portion of the input document wherein an encoder of the neural network receives the sequence of the input document and outputs a context and wherein a decoder of the neural network receives the context and outputs the keyphrase in the set of keyphrases;
calculating, for the keyphrase in the set of keyphrases, a relevance score, the relevance score measuring a similarity between an embedding of the keyphrase in the neural network and an embedding of the input document in the neural network wherein the embedding of the keyphrase and the embedding of the document are generated by a contextual model; and
adjusting, according to a diversity balancing function, the relevance score, the adjusting resulting in an adjusted relevance score wherein performing the adjusting is further based on the neural network configured with equal weight wherein the neural network receives a vector of a keyphrase-input document pair in tandem with a vector of an entity and textual description of the entity.
|