Master's Thesis Siddhesh Kandarkar

Computergestützte Beschreibung von Tendenzen und Mustern in Zeitreihendaten in natürlicher Sprache

Computer Assisted Natural Language Description of Trends and Patterns in Time Series Data

Business Intelligence and Analytics Software provides useful visualizations and dashboards to depict the health of an organization. Dashboards typically contain multiple graphical representations of big data to summarize and communicate structured data. However, they can easily overwhelm viewers who are not data scientists or experts. Especially if the graphical representation or dashboard is novel and/or displays multiple features of and relationships and data points within the dataset, it often fails to convey key insights. Even with the availability of insightful analytical dashboards, business stakeholders often prefer intuitive and concise textual executive summaries for understanding business insights. The process of manually writing executive summaries can become tedious, repetitive and time-consuming for the data experts, especially when they have to write thousands of such reports on a regular basis. Thus, automating the process of writing executive summaries can prove beneficial for the data experts.

The objective of this master ’s thesis is - 1) survey manual and automated approaches used in writing textual executive summaries for time-series data, 2) survey state-of-the-art Natural Language Generation (NLG) techniques for Data-to-Text Generation, 3) propose a prototype which can assist the data experts in writing textual executive summaries which are relevant, intuitive, accurate and timely.

This thesis is offered at the Chair of Software Engineering for Information Systems Business in cooperation with the Celonis. Celonis is a vendor of intelligent big data technology that analyzes and visualizes every process in a company based on data traces in IT systems.

Related Resources

https://paperswithcode.com/task/data-to-text-generation