Artificial intelligence snapchat: Visual conversation agent

摘要

Visual conversation is a dialog in which parties exchange visual information. The key novelty presented in this paper is an artificial intelligence-driven visual conversation automation method. We will present a state of the art Artificial Intelligence Snapchat Visual Conversation Agent (AISVCA). AISVCA uses our proposed artificial intelligence-driven visual conversation automation method to create received image caption and generate an appropriate reasonable visual response. These functionalities are achieved by using a combination of Convolutional Neural Network (CNN), Long Short-Term Memory Neural Network (LSTM) and, Latent Semantic Indexing method (LSI). CNN and LSTM are used to create image captions and, LSI is used to assess the semantic similarity between captions generated from personalized image dataset, and captions that are extracted from the received image content. We will show that AISVCA, using the proposed method can generate a visual response that is basically indistinguishable from a human visual response. To evaluate the proposed approach, we measured the accuracy of the proposed system and, conducted a user study to test communication quality. In the user study, we analyzed source credibility and interpersonal attraction of the AISVCA. The user study results showed that there are no significant differences in communication quality between a visual conversation with AISVCA and visual conversation with the human agent.