Research On Online Course Recommendation Algorithm Based On Course Reviews

In recent years,with the outbreak of the epidemic,online courses have become an integral part of students’ learning,with teams of educational institutions and universities launching online courses on different platforms.While online education platforms provide users with a vast amount of learning resources,they also pose the problem of "information disorientation".How to quickly find the learning resources you need and are interested in from the vast amount of learning resources has become a challenge for online education platforms to solve.Therefore,the development of personalised recommendation technology provides technical support for the above problem,and can effectively solve the problem of information disorientation caused by users facing information overload of educational resources,thus improving users’ online learning experience.However,traditional personalised recommendation models use the same type of structured data for feature selection,while the data in actual online course platforms are unstructured and heterogeneous.To address the above problems,this paper studies a Ro BERTa-Bi LSTM model to process text information,extract the hidden information in the comments,dig out the deeper meaning after the interaction between features through Deep FM based recommendation model,and finally make recommendations,the main research work carried out in this paper is as follows:A Ro BERTa-Bi LSTM aspect-level sentiment analysis model incorporating an attention mechanism is proposed,which is used to process unstructured textual information and extract the hidden meanings of textual information.The method first uses the Ro BERTa model to convert course review text and aspectual words into word vectors;then,a bidirectional LSTM is used to extract the deep semantics of course reviews and aspectual words in both directions,and then the attention mechanism is used to learn the importance of aspectual words in the review text.Finally,the updated vector representation is multiplied with the output of the Bi LSTM model and input to the fully connected layer for classification prediction.Experimental accuracy and F1 values of 84.6% and 0.834 were achieved on the Bilibili platform dataset.The Deep FM recommendation model incorporating attention mechanism and deep crossover network is proposed.Firstly,the course review information features and user information features are embedded onto a low-dimensional dense space as model inputs through One-Hot coding;then,the inputs are concatenated to AFM model and DCN model,where AFM is able to extract low-order signatures combinations and DCN is able to crossover to extract high-order features combinations;finally,the the concatenation is followed by classification prediction,setting a threshold value of 0.5,and the output value greater than the threshold value is recommended,otherwise it is not recommended.According to the experimental results,this recommendation model uses information features of courses and users to deeply mine the cross-meanings between features and features,and the experimental results show that the mean absolute error and root mean square error of this model on the Bilibili platform dataset are reduced by 0.0581 and0.1109 respectively compared to the Deep FM model.This paper shows that aspect-level sentiment analysis is used to deeply mine This paper shows that the information of course reviews can be used to assist the recommendation algorithm of online courses by aspect-level sentiment analysis,which can improve the recommendation effect.