Research On Topic Classification Algorithm And Automatic Outline Strategy For Bullet Notes

With the development of Internet technology and being affected by surging online learning caused by the pandemic,online education has also encountered many problems while developing rapidly.The main problem is how to better integrate offline and online courses and to improve students' ability of independent learning.When studying online,students are facing problems like lack of interaction and difficulty to grasp the key points since the video is too long.Therefore,there is an urgent need for a tool that can analyze the course content by linking the course with the textbook.To improve students' ability of independent learning,the key lies in accurately feedback students' learning status.But the current methods such as learning time and check-in are vulnerable to be forged.Hence,ways that can automatically and accurately feedback students' learning status are required.In response to the above problems,this thesis focuses on the bullet notes tool on the online education platform by analyzing topic classification and automatic outline of bullet notes.Besides,optimization models are proposed based on the benchmark models,and comparison experiments are designed to prove the feasibility.This thesis puts stress on studying the bullet notes data produced during automatic control principle teaching.Since there is no suitable subject data set of this field,it is suggested to create a data set with textbooks first.And visualize the original data of bullet notes to grasp the basic characteristics of the data.To integrate the content of videos and textbooks,an experiment is done in the benchmark model.In response to the wrong cases,corresponding improvement schemes are proposed.The self-attention convolution model is employed for solving the long-term dependency problem.The sub-word optimization is used in response to the text representation problem.And the two-stage training of focal loss is used for the hard sample.The experiment proves that the optimized model has achieved an improvement by 5.56% to 10.08% on all levels of subjects.In the end,the thesis analyzes the notes with the improved classification model.The thesis proposes the automatic outline task of bullet notes and divides the task into key-point detector and key-point summary.The key-point detector provides assessment indicators that ensure no repetition and no omission.Key point summary employs the same way of assessment as text summary.And it proposes the benchmark model for the key point detector and key point summary.Based on the result of the benchmark model experiment,the feasibility of forming the best number of key points by taking advantage of the notes features is discussed and the batch clustering scheme bytaking advantage of the notes features is formed.The scheme of using the book data set as the background corpus and the scheme of optimizing the characteristics of high-quality users are discussed and the optimized model is constructed.Experiments show that the optimized model is significantly improved in performance and can effectively solve the problem of automatic outline.Finally,the outline system is designed and implemented.The two models in the full text are deployed in the system by taking advantage of the formed annotation corpus.And the test proves that the system can work effectively.