Design And Implementation Of Bullet Screen Emotion Analysis System For Bilibili

At the beginning of the establishment of bilibili,it was positioned as a video website for ACG(animation,comics and games)content.Later,it gradually entered the public's attention and evolved into a fashion culture and entertainment community with rich content and good creative atmosphere.Now bilibili is not only welcomed by young people who love animation and games,but also joined special areas such as science and technology,music,ghost animals,dance,etc.following the public's preferences,and its most characteristic is its bullet screen.All kinds of bullet screens provide convenience for users to discuss animation plots and popular science while watching.The bullet screen itself also contains the real-time feelings and views of the sender when watching the video.After the up master makes the original video and uploads it,you can also know the user's views and emotional trend on the content through the bullet screen,which also provides a basis for the up master to make more high-quality videos and standardize the video content on the platform.Based on this,the main research contents and work of this paper are as follows:Compared with ordinary text,bullet screen text has its unique style.It not only contains more network words and character expressions,but also has a large number of "different meanings of the same word",which is difficult to analysis emotion.However,there are relatively few articles on Emotional Analysis of barrage in China,and most articles tend to academic research,with low degree of data visualization and weak applicability.Moreover,due to the characteristics of barrage itself,short text accounts for the majority,colloquial phenomenon is prominent,there are many network terms,non-standard terms,etc.,so there are still great challenges in accurate emotional analysis of barrage.Firstly,the construction method of bullet screen common word dictionary is proposed.On the basis of the existing emotion dictionary,the influence of facial expression,mood auxiliary words,degree adverbs and negative adverbs on the text is fully considered,and the network new words are greatly expanded through the new word discovery algorithm to add the bullet screen common word dictionary.At the same time,this paper proposes an improved text emotion value calculation method based on the bullet screen common word dictionary,which takes into account the number of negative words before emotion words,the order of degree adverbs and negative adverbs,the strengthening or weakening of clauses by turning conjunctions and progressive conjunctions,and the syntactic features of the text(such as interrogative sentences and exclamatory sentences).Secondly,a new bullet screen emotion analysis based on the bullet screen common word dictionary and a-lstm hybrid model is proposed.The main experimental process has two steps:in the first step,the bullet screen text to be analyzed is matched with the bullet screen common word dictionary,and the text emotion value based on the bullet screen common word dictionary is obtained by using the improved text emotion value calculation method.In the second step,the topic words of each "high-energy segment" are selected by topic clustering method,and then the word segmentation results are quantified.After that,the text with emotion annotation is obtained through the training data of LSTM network model based on attention mechanism.Finally,with a certain threshold hybrid model,the range of values is controlled by sigmoid function to obtain the final emotional characteristics of bullet screen text.Experiments show that this method has better effect and higher accuracy than the general bullet screen emotion analysis method.Finally,on the basis of determining the system architecture,Django+Vue front-end and back-end separation technology is used.Through asynchronous interaction with the server through Ajax language,various functional modules of the system are realized,including user management,emotion analysis Result display and task management.After the user inputs the video URL,the system completes the collection and preprocessing of the barrage data and inputs it into the emotion analysis module.After the task is completed,the barrage emotion analysis results and visual analysis data are displayed on the page.Finally,through the system performance test,it is proved that the system can better meet the needs of users and has certain application value.