Research On Processing And Analyzing Big-data Of Inter-city Bus Passengers Based On Deep Learning

With the network of people's life and production developing,the data produced in the daily life shows explosive growth.A variety of billions of data is generated in all cities every day.The data contains the current urban development patterns and trends,and it contains great value of the data and business opportunities.Therefore,it is very necessary to analyze these data.In this thesis,two problems are solved in the way of processing and analyzing big data of the passenger of a long-distance bus company.On the one hand,we correct the problem of Chinese address ambiguity of member data in long-distance bus company.On the other hand,high-quality members are screened out from member group.In view of the problem of Chinese address ambiguity of member data,a method of deep learning is applied to realize the feature extraction of Chinese address in this thesis and use a method of classification based on cosine similarity to achieve the purpose of disambiguating the meaningless.The main research contents include the preprocessing of Chinese address,design and training of the model of "autoencoder",classification based on cosine similarity and analysis of the error of the elimination of ambiguity.At last,the deep neural network structure with three layers of hidden layer of "96-48-24-7-24-48-96" is presented here.More than 800 thousand ambiguous addresses are corrected successfully and the accuracy rate reaches 99.8%.In view of the problem of screening out high-quality members from member group,firstly,the traditional method based on evaluation index is used to screen out high quality members.Secondly,the relevant information of these members is quantized to input the information to the deep neural network to extract the abstract features of high quality members.Lastly,all these abstract features are used to re-screen.In this study,the training process of different structure depth network is also compared,and it is concluded that the deep neural network with three layers of hidden layer is more suitable for the screening of the work of the high quality members than other neural network structures.The selection of high-quality members of the contribution of consumption is greater than the other network structure selecting members of it.In this thesis,the deep learning and big data processing is combined to put forward the research of text disambiguation and recommendation system based on data feature.The data feature extraction based on deep learning is an approximate lossless compression of raw data,which both ensures the efficiency of the calculation from the data dimension and retains the original data information in the maximum degree in order to make the processing of large amounts of data possible.The research of this article also has the promotion,which can be used in the electronic commerce,the industry production,the medical health and so on.