Rule-based Human Knowledge Converging Under Crowdsourcing Platform

With the development of artificial intelligence,the research on data mining algorithm is more and more in-depth.At present,data mining methods can be divided into two categories.One is pure-machine algorithm.This type of algorithm is often able to make good use of the machine in storage,computing and other advantages and it can be more rapid and low-cost to complete the task.Nevertheless,this approach also shows more and more limitations on handling some complex issues,especially the mixed with human experience,such as the understanding of semantics is still a gap in natural language processing.The other is to try to combine human knowledge with the ability of the computer.This is a widely used method of using the corresponding rules from the experts to carry out the corresponding mining tasks,but this method is too expensive and the scarcity of experts in the field is difficult to expand.Rules is an effective form of knowledge.Traditional rule learning algorithm uses the data to produce some rules,however,this kind of rules are generally poor quality,even unable to be used in large-scale practical applications.Despite of the use of the rules provided by specialized experts can achieve good results,there is a serious dependence on the experts,seriously hampered the development of this approach.Crowdsourcing is a way to use crowd wisdom to complete complex tasks widely studied.However,since crowdsourcing workers tend to be people with different backgrounds and differing knowledge levels,it is a very challenging question to collect their knowledge with high quality.To address the above problems,this paper has entered into the in-depth systematic research.The main work and contribution include the following aspects:· Propose a crowd-based rule generation and quality assessment model.Firstly,a framework is proposed to collect the knowledge and experience of crowdsourcing workers to generate rules.Meanwhile,it can improve the quality of the rules provided by training for the package workers.Then,an online rule collection system is designed to motivate and train the crowds of workers to generate the initial rules.Then,for the application of emotional analysis,we propose a rule quality assessment model to measure the quality of the rules.· Propose an active learning-based training model.The ability of crowdsourcing workers to directly affect the performance of the crowdsourcing system,because the workers in the platform are often the lack of appropriate professional skills,the training of the package workers is particularly important.Aiming at the training problem of all the workers,this paper proposes a model,which can enhance the relevant professional knowledge and complete the corresponding tasks at the same time.· Design a refinement algorithm for crowd-generated rules.Firstly,through the redundancy elimination and rule refinement,the initial rules are obtained in the previous pre-processing,elimination from the crowd of workers collected by the redundant rules;and then,through a greedy rule set extraction algorithm,we can get the most effective core rule set.The Experimental results show that the core rule set can achieve similar or even better results with all rules under fewer rules after refining the core rule set.