VOICE CONVERSATION METHOD AND SYSTEM OF ENHANCED WORD FEATURES

摘要

The speech conversation method for enhancing the word quality according to the present invention comprises: receiving a question and answer corpus consisting of a plurality of sentences and generating first meta data consisting of words for each of the plurality of sentences; After receiving the first meta data for each of the plurality of sentences, the first word vectors are generated by adding a first embedding vector to each of the words of the first meta data, and the first word vectors are bidirectional LSTM processing To generate the first sentence embedding information, add a second embedding vector to each of the words of the first meta data to generate second word vectors, and bidirectional LSTM processing of the second word vectors to embed the second sentence Generating information; Generates second meta data consisting of words for the question sentence, adds a third embedding vector to each of the words of the second meta data to generate third word vectors, and bidirectionally integrates the third word vectors. Generating a question sentence embedding information by processing LSTM; The first sentence embedding information for each of the plurality of sentences and the question sentence embedding information are received, and first information indicating a relationship between the first sentence embedding information for each of the plurality of sentences and the question embedding information Generating; Generating second information by implementing a weighted thumb of the first information for each of the plurality of sentences and the second sentence embedding information for each of the plurality of sentences; Outputting third information by adding second information and the question information to each of the plurality of sentences; And calculating a probability between the plurality of sentences and question information through the third information, and predicting and outputting a sentence having the highest probability as an answer sentence.