Classification And Recommendation System Of Knowledge Points Based On Mobile Phone Photography

The ever-changing Internet education has entered the public’s vision with the rapid development of information technology.It has become a leader in the education industry with rich Internet resources and different from the traditional education model.The learning method of searching questions and obtaining answers and analysis through mobile phone photos has been welcomed by the majority of students and parents.However,at present,most photo search Apps simply provide students with the answers and analysis of the questions,and can not deeply obtain the knowledge points investigated behind the questions.Although a few Apps provide analysis and relevant knowledge points,they only roughly train students in exercises by accumulating the number of questions,and can’t get learning feedback from students.Over time,the learning mechanism of taking photos and searching questions is easy to increase students’ inertia and make them lose the process of thinking.At this time,the text classification model of knowledge points for Internet education is particularly important.It can divide the knowledge points in detail,accurately locate the deficiencies in the learning process of students,and improve the learning efficiency of students.This thesis proposes a text classification algorithm based on FastText text classification and bidirectional encoder representation from transformers model and multi-dimensional feature extraction of word frequency inverse document frequency(TF-IDF).The algorithm uses the BERT model to extract the features of the text,stitches the extracted word vector and the word vector extracted by FastText.It uses the TF-IDF algorithm to extract the keywords of the text,retain the key information of the text,strengthen the influence of a single word on the document,and obtain the word vector with weight information.At the same time,the sentence vector and word vector extracted by BERT model are fused,and the attention mechanism is applied to train the classification model to realize text classification.The accuracy of the model is tested and optimized by using the public data set and the topic data set of a wellknown Internet enterprise.After that,according to the characteristics of Internet education,this thesis designs a personalized hybrid recommendation system integrating a variety of mainstream recommendation algorithms and offline and real-time recommendation mechanisms.The system divides the recommendation results of modelbased recommendation algorithm and content based recommendation algorithm into different sections to show to students to meet their different learning needs.At the same time,TF-IDF algorithm is used to calculate the weight of students’ historical learning data,so as to obtain the knowledge points that students need to make up most at present and form a feedback mechanism.Then this thesis embeds the trained model and personalized hybrid recommendation system into an Internet Education App developed based on Django,obtains the text through OCR technology for the topic pictures uploaded by students,classifies the knowledge points,analyzes the weak knowledge points of students,and strengthens the training of students by means of after-school exercises,famous teacher explanation and so on.Finally,the students are divided into groups.Through the analysis of the data such as the answer accuracy of students in different groups,it shows that the personalized hybrid recommendation system has a certain effectiveness.