User Tools

Site Tools


projects:year8:8a.012.sbu

8a.012.SBU - The Intelligent Dashboard with Blockchain-enabled Collaboration Capabilities

Project - Summary

Dashboards play a critical role in representing the state of a system/organization & enhance its decision-making process. Building custom dashboards tailored for a specific task/role/organization can be a time-consuming and expensive process. In this project, we present an intelligent visual data exploration tool & automated end-to-end dashboard building framework. For any given tabular dataset, it generates a set of possible univariate & bivariate charts. Each chart is assigned a value based on how statistically interesting it is. Users are presented with the most interesting charts which accelerates the knowledge discovery process. The system also generates a set of fully automated dashboards. Users can choose from the set of pre-built dashboards or select charts to generate a custom dashboard.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Klaus Mueller PI mueller@cs.stonybrook.edu (631) 632-1524 Stony Brook University
Darius Coelho Student dcoelho@cs.stonybrook.edu Not available Stony Brook University
William LaBar Project Mentor william.labar@cgifederal.com (337) 281-2066 CGI
Sumit Shah Project Mentor Sumit.Shah@cgifederal.com (202) 309-8790 CGI
Funded by: CGI

Project - Novelty of Approach

  • Most chart recommendation systems rely purely on statistical analysis of datasets. Others require users to specify the tasks and attributes that they are interested and attributes that they are interested and then recommend charts based on these user preferences.
  • Our approach attempts to further improve the current systems by analyzing the metadata (datasets name and attributes) with an ML module to fine-tune the ranking generated by the statistical approach. The ML Module will attempt to generate the most appropriate data analysis tasks for a particular attribute or attribute pair.
  • We identify and filter out highly ranked charts that show similar information and then use the remaining charts to construct a multi-chart dashboard, a feature we have not seen in other systems.
  • Finally, most current blockchain based data sharing services store data items on the blockchain. Our DataChain approach stores transforms to the data via the blockchain. This allows users to either have the whole dataset or just a subset of it, while still maintaining the capability of sharing and applying the same data transforms. Such an approach is beneficial as the blockchain achieves consistency, security, immutability and provenance with a single protocol/algorithm unlike other systems that require multiple solutions to achieve the same functionality.

Project - Deliverables

Deliverables
1 Implementation of a recommender system based on the combined ranking generated by the statistical analysis module and ML model
2 Dashboard composer with implementation
3 Evaluation of the system

Project - Benefits to IAB

Dashboard.ai presents a novel approach for visual data exploration and building custom dashboards. It offers the perfect tradeoff between automation and flexibility. It employs statistical metrics to assign an ‘interestingness’ score to attributes of a dataset. It then uses NLP techniques to analyze domain specific documents to determine domain-specific ‘interestingness’ and analytical tasks related to these attributes. The system then uses this information to recommend visualizations in a dashboard which expedites the data exploration process.

Dashboard.ai is highly useful in scenarios where users do not have the knowledge or time to create effective dashboards to explore interesting features in their data. Additionally, it allows users apply different “styles” of analyses to their dashboard. In the car dataset analysis, for example, only using car reviews from the Autocar magazine would apply the Autocar style of analysis, and only using car reviews from the Car and Driver magazine would apply the Car and Driver style of analysis. It should be noted that our relation extraction model was trained over questions that are relatively simple in structure as compared to complex sentences found in review articles. In future work we will train a new model over complex sentences that we have extracted from 17 data-based news articles that we collected.

Project - Documents

FilenameFilesizeLast modified
8a.012.sbu_final_report.docx694.0 KiB2020/08/17 16:10
8a.012.sbu_mid-year_report.docx236.2 KiB2019/12/11 10:34
8a.012.sbu_executive_summary_updated_10.12.2019.docx57.7 KiB2019/10/01 08:51
8a.012.sbu_year_8_pitch.pptx63.1 KiB2019/09/18 17:28

Life Form Feedback

Year 8 Project Poster Session Feedback (Fall 2019)
Real name Great Progress On Course Needs Change Off Course Abstain
Kimmo Valtonen (kimmo.valtonen)     
Sumit Shah (sumit.shah)     
Count:02000
projects/year8/8a.012.sbu.txt · Last modified: 2021/06/02 16:30 by sally.johnson