Design And Implementation Of Mobile Terminal Emotion Analysis System Based On Machine Learning

With the continuous development and progress of the society,people's focus has gradually changed from physiological needs to psychological needs,and more and more people are concerned about physical and mental health.In such cases,combined with the large demand of market and rapid development of computer science technology,emotion recognition has received widespread attention,and becoming a hot research topic in today's society,like in the health,online education,remote sensing and many other fields,with a very broad application prospects and the ground scene.Current emotion recognition technologies are mainly divided into four directions: based on facial or voice expression,human behavior,physiological signals,and traditional text.In these cases,emotion recognition requires users to input relevant information actively,which probably makes users inconvenient.At the same time,ordinary smart phones are also equipped with a lot of sensor devices nowadays,which can automatically collect relevant information from people and the surrounding environment without users' participation.Combining the advantages of mobile sensing technology and related machine learning algorithms of emotion recognition,this thesis designs and implements a method of emotion recognition based on android mobile terminal,which is significantly more accurate than the traditional single machine learning algorithm.First of all,the emotion recognition algorithm we proposed in this thesis is essentially a multi-classifier ensemble algorithm,which achieves better results by integrating classical machine learning classifiers.After the sensor information is preprocessed,a reasonable feature vector is obtained and input into the multi-classifier integrated emotion recognition algorithm for prediction.The main problem of traditional multi-classifier integration algorithm is how to choose the base classifier and what combination strategy should be used.In order to solve these problems,the main work of this thesis is as follows: in the construction of base classifier,the diversity strategy is adopted to increase its diversity and complementarity;In the selection of base classifier,we combined clustering algorithm with silhouette to determine the number of classifiers,and use inconsistent measures to evaluate their differences.In combination strategy,we learn the weights of base classifier based on its prior probability and conditional probability.Secondly,based on the above algorithm,this thesis designs and implements a emotion recognition system on the mobile terminal of android phones,including requirement analysis,architecture design and the final implementation details.Through this system,the training data of the algorithm can be obtained,and the user can interact with the system,which truly makes the requirement of " recognize the emotion of users which based on their smartphone data" a reality.At the end of the thesis,the emotion recognition algorithm and the system are tested and analyzed respectively,we display the test results,summarize the advantages and disadvantages,and look forward the future research direction and related work.