Model-driven restricted-domain adaptation of question answering systems for business intelligence

Business Intelligence (BI) applications no longer limit their analysis to structured databases, but they also need to obtain actionable information from unstructured sources (e.g. data from the Web, etc.). Interestingly, Question Answering (QA) systems are good candidates for these purposes, since they allow users to obtain concise answers to questions stated in natural language from a collection of text documents. Traditionally, QA systems include patterns for dealing with a large spectrum of general questions, namely open-domain question answering (ODQA). However, BI users should be aware of asking questions related to a specific activity of the business (e.g. healthcare, agricultural, transportation, etc.). Therefore, adapting ODQA systems to new restricted domains is an increasingly necessity for these systems to be precisely used in BI. Unfortunately, research addressing this topic has two main drawbacks: (i) patterns are manually tuned, which requires a huge effort in time and cost, and (ii) tuning of patterns is based on analyzing potential questions to be answered, which is not a realistic situation since, in restricted domains, questions are highly complex and difficult to be acquired. To overcome these drawbacks, this paper presents a novel approach based on model-driven development in order to use knowledge resources to automatically and effortlessly adapt patterns of ODQA systems to be useful for restricted-domain BI scenarios.