Research And Application Of Editor Recommendation System Based On Multi-Attention Sequence Model

In the field of education,various online education platforms are constantly emerging.Due to cost advantages and application flexibility,many offline educational institutions have begun to shift to online education.But online education poses new challenges to the way teachers and students interact.Currently,the most common teaching interaction methods are video conferencing and live streaming etc.its user experience when answering questions online is poor,either the appropriate tool cannot be quickly found in the editor,or the input process is relatively cumbersome.Therefore,it is necessary to provide a convenient editor system for teachers and students.With the improvement of computer computing power,natural language processing has made remarkable achievements in many fields of application.Multi label text classification is a classic task in natural language processing.In educational resources,their content classification has a strong hierarchical correlation,and the classification results of the previous label have a strong impact on the latter label.The Seq2Seq(Sequence to Sequence)mechanism can capture the association of front and rear tags.This article will use Seq2 Seq to focus on the internal connections of tags to implement editor component recommendations.The main work is as follows:(1)The GRUSeq2Seq(Gate Recurrent Unit Seq2Seq)model is proposed.The GRU(Gate Recurrent Unit)network is used as the basic unit of the Seq2 Seq encoder and decoder to simplify the Seq2 Seq based model SGM(Sequence Generation Model for Multi label Classification).Through experimental comparison,GRUSeq2 Seq can improve training speed while ensuring performance indicators.(2)The MASM(Multi Attention Seq2 Seq Model)model is proposed.GRUSeq2 Seq is further improved by combining two attention mechanisms,SelfAttention and Luong-Attention,to enhance the ability of the model to focus on global information.Experiments have shown that MASM has improved its main performance indicators compared to GRUSeq2 Seq.(3)Deconstruct editor.Split the general online editor,conduct secondary development on the components in the editor,and build a component set of the editor.Enable components to compose lightweight editors on demand.(4)Design a editor recommendation system based on MASM model.Use the MASM model to classify multi label text,and then map the classification label with the feature information in the scene to the editor component set to obtain the recommendation results of the component,generating a personalized editor.