| Collaborative problem-solving ability is one of the key skills of students who need special attention in the current education reform.However,China has not yet had systematic research on the ability of students to solve problems in cooperation,and lacks the evaluation criteria and strategies for high-level skills in cooperation and problem-solving skills.For a long time,the evaluation of the ability to solve cooperative problems has always been a difficult problem that hinders the research and promotion of cooperative problem solving capabilities.Based on the analysis of existing literature,this study firstly analyzes the characteristics of collaborative problem solving ability based on the evaluation practice of PISA2015 and ATC21 S.Based on the characteristics of this capability,this paper explores the characteristics of collaborative problem solving ability evaluation,and then based on literature research and related evaluation.In practice,an evaluation plan for the problem solving ability of middle school students based on evidence center design was proposed.The program uses the four-person QQ discussion group chat mode,sets the task situation of the energy conservation plan,collects the chat records of 67 middle school students’ cooperative problem solving ability,and extracts the chat text through the python language.By combing the evaluation framework of PISA2015 and ATC21 S,this paper proposes a framework for the evaluation of collaborative problem solving in this research,with a total of 18 indicators.Among them,9 indicators belong to cooperative,namely report their own task status,do not understand their own tasks,reject tasks,understand and guide the progress of members,request teammates to criticize,suggest,information,task assignment,irrelevant or interfere with topics,call,respond to teammates,take the initiative task.9 indicators belong to problem solving,namely not knowing how to solve the sub-problem,clarifying the status of the task,confirming the information needed to complete the task,decomposing and merging the task,providing solutions to the sub-problem,providing the path to complete the task,and adjusting the solution to the sub-problem Promote the progress of the task and restate the goal(content,form).In this paper,five different convolution kernels and five different feature planes are selected respectively,and the words vector of external training are loaded with word2 vec.The over-fitting is adjusted by means of discarding and early termination,and finally a better effect is trained.Convolutional neural network classifier,which has a correct rate of 85% on the verification set.Finally,the validity of the research collaborative problem assessment framework was tested.The bifactor model was used to test and four special factors were selected,and the bifactor model had a good fitting effect(CFI=0.92).Through the load analysis of bifactor model items and general factors and special factors,it is shown that these indicators can better measure the factors,which proves that this is an effective evaluation framework. |
