Design And Implementation Of Automated Essay Scoring System With High Concurrency

AES(Automatic Essay Scoring)is to extract a variety of features in the composition through various methods such as machine learning and deep neural network model,and then score comprehensively,give the scoring results and comments,and further give modification suggestions.Feature extraction in composition includes composition theme,parallelism sentences,figurative sentences,anthropomorphic sentences,poetry quotation,emotion analysis,language fluency and so on.Different feature extraction methods and models are not exactly the same.They all improve the accuracy of recognition in continuous iteration and improvement algorithms.Due to the main energy invested in the improvement of these feature extraction algorithms,there is a slight lack of attention in the user experience part of composition scoring.User experience includes but is not limited to the speed of feature extraction,waiting time for composition scoring,display method of scoring results,suggestions given after scoring,concurrent service volume of composition scoring system,etc.From the perspective of high concurrency,this paper analyzes the resource type and composition processing time of the composition scoring module in the current intelligent writing platform,and quantitatively analyzes the performance bottlenecks such as small concurrency and long waiting time of composition processing in the system through concurrency testing.Aiming at the bottleneck problem,a distributed automatic composition scoring system with high concurrency is built by using software such as Flask,Celery,Rabbit MQ and Zabbix.The stress test data show that the composition processing speed of the highly concurrent composition scoring system has been significantly improved,the processing time of each composition has been significantly shortened,the waiting time of composition processing tasks has been significantly reduced,and the user experience has been significantly improved.In order to control the concurrency in a better way and achieve the purpose of peak shaving and valley leveling,this paper introduces Pareto principle to improve the algorithm based on the original whale optimization algorithm(WOA).Through the test and verification of 23 standard test functions,the improved whale optimization algorithm with Pareto principle which is called WOA-PP algorithm inherits the excellent quality of jumping out of the local optimal solution,significantly improves the optimization accuracy,and can converge to the theoretical optimal solution with faster speed and fewer iterations.The actual test shows that when the composition scoring task is overstocked,WOA-PP algorithm can optimize the current optimal number of concurrent services,expand the capacity horizontally,make full use of system resources and shorten the overall processing time of composition.At the same time,when the number of composition processing tasks is small,the minimum number of concurrent services is automatically optimized,and the horizontal capacity is reduced to reduce the number of concurrent services,so as to save system resources on the premise of ensuring that the processing of composition tasks is not blocked.