Design And Implementation Of Online Learning Sentiment Analysis Model Based On Multi-modal Data

As an emerging product of "Internet+Education",the development of online learning is in full swing in recent years,but the separation of teachers and students leads to the lack of emotional communication,which is a huge challenge for online learning.This thesis proposes an online learning sentiment analysis model based on multi-modal data to understand the learner's emotional tendency and learning state more deeply,which provides a new perspective for online learning sentiment analysis.Firstly,aiming at the current research status and existing problems of online learning sentiment analysis,set up an online learning sentiment analysis model based on four kinds of modal data:comment text,emoticon,facial expression and limb movement.Four key steps are mainly included in this modal:acquisition of multi-modal data,preprocessing of multi-modal data,feature extraction of multi-modal data and fusion of multi-modal data.Crawler technology and camera capture technology are used to acquire data in the data collection phase.In the preprocessing phase,for comment text and emoticon,three steps are included:comment text extraction,word segmentation and part of speech tagging;for facial expressions and limb actions,image enhancement,grayscale,geometric normalization three steps are included.In the feature extraction phase,the comment text adopts the TF-IDF method;the emoticon adopts the method of constructing the sentiment dictionary;the facial expression uses the LBP histogram to extract the facial features;the limb movement adopts the relative size of the face and the reference face to judge leaning forward or leaning back.Finally,weighted summation method based on the support vector machine posterior probability is adopted to fuse multi-modal data.Then,the online learning sentiment analysis model based on multi-modal data proposed in this thesis is tested,the experimental result shows that:Compared with the existing work,the model has stronger practicability and promotion,because it does not need to wear equipment and collect voice,reduces the learning cost and learning environment requirements.Compared with the single modal data sentiment analysis model,the model has higher accuracy and can understand the learner's emotional tendency and learning state more deeply.Finally,this thesis implements an online learning sentiment analysis prototype system based on multi-modal data,the development environment and tools of the system are given,learning behavior acquisition,learning behavior detection and multi-modal sentiment analysis functional modules are emphatically introduced.Experiments show that the system works well and can analyze the emotional tendency and understand the learning status of online learners.