Research On Dialogue Model Based On Improved Codec Mechanism And Sentiment Recognition

Dialogue models can realize human-computer interaction without any limitation of topics in various social scenarios by actively recognizing input information and generating responses.The existing dialogue models can’t recognize and analyze the fine-grained emotions embedded in text utterances,and the information on topic distribution and contextual semantics of dialogue texts can’t be fully utilized in the process of emotional dialogue,resulting in the lack of emotional resonance and lessthan-ideal anthropomorphic effect of the response utterances generated by the model in specific dialogue contexts.To address the above problems,a dialogue model based on improved codec mechanism and sentiment recognition is proposed.The main research contents of this thesis are as follows.(1)A dialogue generation model based on improved codec mechanism is proposed,which can enhance the dynamic perception and learning ability of different feature information during the training process.Firstly,the multi-layer gated structure is used to sense the topic distribution,context information and sentence pattern characteristics of paragraphs,and capture the background information related to the dialogue content.Secondly,text data is trained and tested by encoder-decoder.Finally,a high-quality dialogue generation model is constructed based on MLG and codec to improve the perception and learning effect of the model on the background information contained in the text data and the behavior patterns reflected.(2)Furthermore,the sentiment analysis method is embedded into the encoderdecoder mechanism to capture the emotional information contained in the statement,which solves the problem that the response anthropomorphism generated by the model is not ideal.Firstly,a fine-grained sentiment dictionary is constructed and calculation rules are set up to identify and analyze the emotional granularity of conversational sentences.Secondly,the emotional features are embedded into the codec structure,and to enhance the attention of the codec process to the emotional information.Finally,based on the emotion contrast algorithm,the corresponding anthropomorphic responses are generated according to different emotional granularity,enhancing the ability of the model to generate emotional responses according to specific characteristic information in complex dialogue scenes.The experimental results on Chinese and English dialogue datasets show that the proposed model can generate personalized responses in line with contextual logic and dialogue background,and achieve a higher degree of anthropomorphism in emotional dialogue process,and is superior to the traditional dialogue model in both automatic and manual evaluation indexes.Figure [25] Table [18] Reference [64]...