Research And Design Of Interactive Dataanalsis System

With the presence of big data era and the prosperity of related technologies,the industry starts to realize and benefit from the value of data.However,this trend continuously pulls up the demand on efficiency and quality of data analysis.The new generation of data analysis software require standardized data and knowledge organization,diverse forms of data interaction,flexible application scenarios and the feature of active data exploration driven by the momentum sourcing from the end uses.Motivated by these emerging demands,this thesis propose an interactive data analysis system for data analysis applications leveraging the practical experience of building an auxiliary data analysis robot for the accounting and operating workloads in a retail company.The main research contribution are summarized as follows.To address the issues of knowledge extraction,experience accumulation and reuse in data analysis scenarios,this thesis build an enterprise-wide data analysis knowledge graph.Firstly,the causal analysis events are extracted from the internal analysis report to build an analysis event-centric knowledge graph,then building an indicator-centric knowledge graph contains business and financial statements analysis indicators,database's metadata and their relationship.Finally,the relationship between indicators is inferred from the causal event relationship,which integrates analysis reports,indicators,analysis data and the association relationships of them.The knowledge graph provides a basis for data analysis and interactive recommendation of related data,as well as new ideas for enterprise data resource management.For the two key parts of natural language understanding and dialogue management in interactive data analysis,this thesis designs a joint model of intent recognition and slot filling based on BERT-Text CNN,and improves the effect of the model by integrating the features of high and low layers of BERT.Experiments show that the model effectively improves the accuracy of intent recognition and slot filling.Meanwhile,a dialogue management method based on knowledge graph and Q?A log is designed,the dialogue state is tracked through the combination of state machine and slot filling method,the interactive options recommended by the knowledge graph and Q?A logs improve the quality and quality of interactive Q?A efficiency.Finally,this thesis designs an interactive data analysis system through analyzing the application requirements of industry in data analysis scenarios,which provide an interactive data question and answer analysis services for uses,including indicators retrieval,data query,data analysis and other capabilities.The system is tested through different dimensions such as task completion,interactive guidance capability and user satisfaction.The results show that the system has the effect of guiding and auxiliary analysis in the data analysis.At the same time,the system has the advantages of convenience and efficiency,and has certain research significance and application value.