Design And Implementation Of Multi-Modal Emotion Analysis System

With the continuous development of artificial intelligence technology,more and more intelligent technologies are applied in all aspects of people's lives.How to intelligently recognize people's emotional changes is a very difficult technology.Because there are many ways to express people's emotions,and the expression of emotions in facial expressions,speaking voices,and different languages also changes.At present,a single algorithm for identifying people's emotions through a modal data is constantly emerging,and the application of algorithms is also reflected in various aspects.However,the emotion recognition of a single model does not conform to the way people express their emotions,and is easily affected by other unavoidable external factors.The performance of the emotion analysis system is not high.Emotion analysis algorithms that combine modal information for analysis.Therefore,it is of practical value to use multi-modal sentiment analysis algorithms to transfer emotion recognition applications from a single mode to a combination of multiple modalities for recognition analysis.Therefore,this thesis uses multi-modal sentiment analysis technology to design a more realistic and accurate sentiment analysis system.This thesis designs and implements a multi-modal emotion recognition system from the aspects of system requirements analysis,architecture design,concrete implementation,and system testing.The system can use the three modalities data of text,voice and video files to call the corresponding algorithm module for single and multi-modal sentiment analysis.The system is analyzed from a functional perspective,including three major modules,a data module,an algorithm module,and a display module.1.The data module realizes the collection of data information,including user information and modal emotion data.As long as the user logs in and registers and clicks on the file upload,the data collection can be completed,simplifying the user operation process and improving the operability of the system.The preprocessing process data of text,voice,and video modalities into format data that conforms to the input data of the algorithm model,which simplifies the process of processing data in the model call and improves the flexibility of the system;use data management to upload user information and users The original data,as well as various modal characteristic data,are managed and stored in a unified manner,and the methods of data addition,deletion,and modification are defined,which facilitates the data call of each module and realizes the persistence of data.2.The algorithm module implements two emotion recognition algorithms for users to choose flexibly and intuitively feel the superiority of the multimodal emotion analysis algorithm.The two algorithms selected in this thesis have achieved a good sentiment analysis effect with an accuracy of 83%when performing model training tests,indicating that the algorithm can solve the sentiment classification problem well and improve the reliability of the system.After the algorithm is called,the sentiment analysis result file is generated for visual display by the front-end module.3.the display module uses data streaming to achieve the display of multimedia data,optimize system performance,dynamically display the results of sentiment analysis in the form of charts,and improve the user experience.In the end,this thesis conducted a functional and performance test of the system.The test results verified that the various modules of the system can work in harmony with each other,and confirmed the accuracy and reliability of the system in emotion recognition.It can stably and continuously complete the task of emotion analysis.