| CPC G06F 16/345 (2019.01) [G06F 40/284 (2020.01)] | 16 Claims |
|
1. A document summarizing apparatus, the document summarizing apparatus comprising:
a processor;
a memory storing one or more programs configured to be executed by the processor; and
the one or more programs include instructions for;
an encoding unit configured to receive document data including one or more sentences and convert the document data into a token defined in a predetermined unit to generate a feature vector;
an extraction summary unit configured to receive the feature vector and calculate probability values that each sentence of the one or more sentences corresponds to a summary, and generate, for each sentence, an attention vector for a token weight to be applied to each token included in a sentence, based on a probability value of the sentence; and
a decoding unit configured to receive the feature vector and the attention vector and generate abstract summary data, based on the feature vector and the attention vector.
|
|
6. A document summarizing apparatus, the document summarizing apparatus comprising:
a processor;
a memory storing one or more programs configured to be executed by the processor; and
the one or more programs include instructions for;
an encoding unit configured to receive document data including one or more sentences and convert the document data into a token defined in a predetermined unit to generate a feature vector;
an extraction summary unit configured to receive the feature vector and calculate probability values that each sentence of the one or more sentences corresponds to a summary to generate a probability vector for each sentence, and generate, for each sentence, an attention vector for a token weight to be applied to each token included in a sentence, based on a probability value of the sentence;
a candidate data generator configured to generate a plurality of summary candidate data from the one or more sentences, based on the probability vector for each sentence; and
a decoding unit configured to receive the feature vector, the attention vector, and the plurality of summary candidate data, and generate abstract summary data for each summary candidate data.
|
|
9. A document summarization method performed by a computing device comprising at least one processor and a computer-readable storage medium storing one or more programs implemented by the at least one processor, the method comprising:
an encoding operation of receiving document data including one or more sentences, converting the document data into a token defined in a predetermined unit, and generating a feature vector;
an extraction summary operation of receiving the feature vector, calculating probability values that each sentence of the one or more sentences corresponds to a summary, and generating, for each sentence, an attention vector for a token weight to be applied to each token included in a sentence, based on a probability value of the sentence; and
a decoding operation of generating abstract summary data by receiving the feature vector and the attention vector as inputs.
|
|
14. A document summarization method performed by a computing device comprising at least one processor and a computer-readable storage medium storing one or more programs implemented by the at least one processor, the method comprising:
an encoding operation of generating a feature vector by receiving document data including one or more sentences and converting the document data into a token defined in a predetermined unit;
an extraction summary operation of receiving the feature vector and calculating probability values that each sentence of the one or more sentences corresponds to a summary to generate a probability vector for each sentence, and generating, for each sentence, an attention vector for a token weight to be applied to each token included in a sentence, based on a probability value of the sentence;
a candidate data generation operation of generating a plurality of summary candidate data from the one or more sentences, based on the probability vector for each sentence; and
a decoding operation of receiving the feature vector, the attention vector, and the plurality of summary candidate data, and generating abstract summary data for each summary candidate data.
|