Emotional chat type reply generation method with theme perception

Abstract

The invention discloses an emotional chat type reply generation method with theme perception. The method is characterized in that a generation model composed of an encoder, a theme perception module,an emotion perception module and a decoder is adopted; an external topic common sense introduced into a generation model is fused with two attention mechanisms, semantic and emotion information of a session is effectively captured by using hidden variables under a variational auto-encoder framework, and an emotion reply with topic perception is generated. Compared with the prior art, the method has the advantages that the generated session and the input session are better in topic correlation; according to the method and the device, the reply which better conforms to the preset emotion category can be generated, the reply with emotion can be more accurately generated, the generated reply and the input session can be ensured to be under the same session theme, two attribute requirements ofa conversation system are simultaneously met, and the method and the device have wider practical significance.

Description

technical field [0001] The invention relates to the technical field of task-based dialogue systems, in particular to a recurrent neural network-based method for generating an emotional chat reply with theme awareness for a chat robot. Background technique [0002] At present, one of the hottest researches in the dialogue system is the chat dialogue system. This kind of system does not need to complete a clear task. Its purpose is often to accompany the user to chat and let the machine simulate the chat between people to the greatest extent. . Studies have shown that chatbots capable of expressing emotions can improve user satisfaction. Giving chatbots the ability to perceive and express emotions is an important manifestation of artificial intelligence in emotional computing. In addition, by analyzing the corpus, we also found that the conversations of real people are not only rich in topics, but also have high topic relevance and consistency between sentences. [0003] Th...

AI Technical Summary

Problems solved by technology

The current methods of generating emotional responses based on the Seq2seq model, due to the addition of emotional factors, many tend to generate trivial or generally relevant responses, which have almost no practical meaning, such as "haha", "I love you", " I hate you"

In addition, current methods often tend to ignore the topic correlation between the generated reply and the input conversation, resulting in the situation where the answer is not what was asked.

Images