Generating and Measuring Similar Sentences Using Long Short-Term Memory and Generative Adversarial Networks

The two problems of measuring the semantic similarity (MSS) between sentences and generating a similar sentence (GSS) for given one are particularly challenging. Since these naturally have logical connections, this article proposes algorithms to deal with them together. main contributions in four aspects. 1) We propose new algorithm called syntactic long short-term memory (SSLSTM) computing similarity. model used by SSLSTM computes representation vector merging results separately running LSTM networks, other related that is generated based on features words sentence. score calculated distance representations vectors. 2) A GAN framework proposed generative adversarial network (SSGAN). GSS an MSS incorporated as modules generator discriminator SSGAN. unique design SSGAN that, input triple GAN, will produce three additional items, process them. Three versions proposed: classic (C-SSGAN), hybrid (H-SSGAN), black-box (B-SSGAN). 3) Two paradigms emerge from patterns SSGAN, (B-GAN) (H-GAN), respectively, which potentials be generally applied NLP problems. 4) series experiments different settings designed test effects B-SSGAN, show B-SSGAN has considerable boosting both chosen algorithms. Several executed compare some representative state-of-the-art advantages terms amount error overall performance. There experiments. performances measured using algorithm. Multiple performance measures considered describe algorithms’ holistically, including efficiency achieved relative training time, indicates CNN-based (SSCNN) most training-efficient comparison.