Researchon Mathematical Problem Recommendation Algorithm Based On Multidimensional Similarity Calculation

Personalized recommendation is the product of the integration and development of the Internet and people’s needs.It can help people find accurate and effective information from massive Internet resources.Nowadays,the vigorous development of online education is accompanied by the rapid expansion of online education resources.In terms of education,personalized recommendation still needs further development.An excellent recommendation algorithm can help students and teachers obtain suitable exercise resources among the massive resources.It can not only help students save the time and cost of finding exercise resources,but also It can grasp the weak points of students,carry out effective targeted exercises,improve students’ learning efficiency,and at the same time provide educational institutions with better teaching resource recommendation services,which has important research significance and application value.The research object of this paper is mathematics exercises in primary and junior high schools.It analyzes the characteristic attributes of mathematics exercises and develops an algorithm model for the similarity of exercises.The characteristics of mathematics exercises are relatively complex,including not only various classification labels,but also text information and many mathematical formulas.Therefore,starting from the characteristic attributes of the exercises themselves,this paper proposes a similar mathematical exercise recommendation algorithm that integrates mathematical formulas,text semantics,and exercise labels in exercises,and evaluates and optimizes the effect of the recommendation algorithm.Students find more accurate math problems and make recommendations.The main research work of this paper is as follows:(1)Propose and implement a mathematical formula similarity calculation method based on symbolic path.This article converts the mathematical formula in Math ML format into a binary tree structure,the operands are all leaf nodes,and the operators are all non-leaf nodes.According to this feature,all the paths from the root node to the parent node of the leaf node are extracted and the logic of the mathematical formula is calculated.Structural similarity,extracting the child nodes of each symbol node and calculating the similarity of semantic information in the mathematical formula,and the weighted calculation of the two to represent the similarity of the overall mathematical formula.(2)For other feature attributes of exercises,this paper proposes a similarity calculation method based on semantics and labels(Exercise Recommendation Algorithm of Semantic and Label Similarity,ERASLS).The main idea of this method is to calculate the similarity of the two labels through the cosine similarity after the topic expression form label and the knowledge point label are vectorized in one-hot form.At the same time,on the doc2 vec pre-training model,all the topic text information in the exercise set is extracted to train a unique mathematical exercise text word vector model.After word segmentation and cleaning of the topic text,the input model obtains a distributed word vector representation.The text of the topic,and use it to calculate the text similarity between the two exercises.The second algorithm proposed in this paper is obtained by weighting the label and text similarity.(3)On the basis of the research on these two algorithms,this paper proposes a math problem recommendation algorithm that combines mathematical formulas and text semantics with exercise labels.The model has improved the recommendation accuracy by more than 17%.Students can use this algorithm to carry out targeted intensive training,so as to continuously improve their understanding and application of such exercises.