METHOD FOR ABSTRACTIVE DOCUMENT SUMMARIZATION BASED ON A CONTEXT BASED COVERAGE MECHANISM, RECORDING MEDIUM AND DEVICE FOR PERFORMING THE METHOD

摘要

The automatic summarization method using document context-based coverage is based on a long-short term memory network (LSTM) cell-based bidirectional encoder, using the input word, the embedding vector of the input word, and the hidden state and cell state of the previous step. Calculating a hidden state; Calculating an attention weight of the input step at each decoding step in a unidirectional decoder, based on a cell state of the decoder and a hidden state of the encoder; Calculating a context vector for the input word based on the importance of the input step and the weighted sum of the hidden states of the input step; Generating a word having the highest generation probability in a current step based on the context vector and the hidden state of the decoder; And arranging the words by calculating alignment energy based on a context coverage vector obtained by accumulating context vectors of all the calculated steps. Accordingly, a summary can be derived centering on the meaning of the document.