Systems and methods for continual updating of response generation by an artificial intelligence chatbot

Inventors

Ma, Weicheng • Cao, Kai • Pan, Bei • Chen, Lin • Li, Xiang

Assignees

Cambia Health Solutions Inc

Abstract

Methods and systems are provided for a natural language processing system comprising a chatbot adapted for dialog generation. In one example, the system may include a combination of a variational autoencoder (VAE) and a generative adversarial network (GAN) for generating natural responses to input queries. The VAE may convert queries into vector embeddings that may then be used by the GAN to continuously update and improve responses provided by the chatbot.

Core Innovation

The continual dialog generator for an artificial intelligence natural language processing chatbot generates continually updating response dialogs by using a Variational-Seq2Seq dialog generator architecture that combines a variational autoencoder (VAE) with a generative adversarial network (GAN). An input query is entered into a user interface and received by the VAE, which embeds the input query into vector representations. The vector representations are converted into response embeddings through a decoder and dense layers, and the response embeddings are converted into responses by a GAN generator (GAN-G).

A GAN discriminator (GAN-D) evaluates generated responses by comparing generated responses to seed answer or anticipated responses, and loss calculation based on that evaluation is used to update GAN components. Competition between the GAN generator and the GAN discriminator drives updating of the discriminator, and the VAE is updated by backpropagation based on updating of the discriminator of the GAN. Continual improvement of the dialog generator is described based on question-answer pairs and user responses feedback.

The architecture evaluates variations of the VAE encoder and decoder configurations using BLEU and ROUGE. The system workflow includes a testing process versus a training process, and the continual dialog generation is presented as using the combined VAE and GAN to produce natural responses and continually updating response generation for the chatbot.

Claims Coverage

The independent claims cover three aspects: a training method that takes user queries through a VAE into a GAN and then displays responses; a dialog generator system that computes and outputs responses using a VAE combined with a GAN; and a method specifically for continually updating chatbot responses by training with GAN competition and backpropagation updates.

Across the independent claims, the core coverage is the integration of a VAE to embed queries into vector representations and a GAN to convert those representations into responses, with response output in a user interface; and a continually updating training approach where the discriminator is updated through generator-discriminator competition and the VAE is updated via backpropagation based on discriminator updates.

Stated Advantages

Documented Applications