Dialogbasierte Benutzungsschnittstelle für interaktive Datenanalyse in natürlicher Sprache

The purpose of data analysis is to extract new information from existing data. Users of data analysis applications usually do not make a single, perfect request. Instead, data analysis is typically an iterative process in which a user executes a query, obtains a result, and then performs a new query based on it. In general, users in departments do not necessarily have programming skills or knowledge of a formal query language. Natural Language Interfaces to Databases serve to overcome this problem. In this context, the guided interaction paradigm guides users through the process of query formulation by presenting possible values in a concise manner and by allowing users to formulate a query step by step. Following the guided interaction paradigm, this work presents a dialog-based natural languge interface for interactive data analysis. A knowledge-based approach is chosen because pure machine-learning approaches are unsuitable in this case. The basis for query generation are analysis graphs and analysis situations as well as a machine-readable definition of the conceptual multidimensional model. The user can first formulate a multidimensional query as natural language text. The system then tries to create an analysis situation using constraint satisfaction and heuristic rules, based on lexical and semantic similarity. By using navigation operators, the analysis situation can be refined. The development of a universal NLIDB system was not a goal of this thesis.