Research in language technologies for the Industrial sector

Chatbi presents a research project whose objective is to design and develop a dialogue system that, by combining information extraction techniques, semantic representation, and question answering, allows access to business or industrial information, facilitating its consultation through a simpler and more natural dialogue-based interaction.

The project's main objective is broken down into the following technological objectives:
A. Design and development of an information extraction system capable of interpreting and structuring the information contained in manuals and other documents, written in natural language, into a knowledge graph.

B. Design and development of a knowledge base that allows the storage of this information once interpreted, so that it can be efficiently consulted by a dialogue system that guides the user.

C. Design and development of a system that allows the semantic interpretation of user queries and the execution of queries against the corresponding knowledge base.

D. Design and development of a dialogue system that interacts with the user, allowing them to ask questions and guiding them in the search for answers, based on the information available in the corresponding knowledge graph.

To maximize future market entry, work will focus on four areas:
+ Validation:
1. Specification of two pilot projects for validating the resulting system in a realistic environment that facilitates market entry.

2. Establishment of an evaluation methodology to objectively quantify the system's effectiveness, both in terms of average query time and the volume of queryable information.

+ Dissemination:
3. Academic dissemination of the project results.

4. Transfer of results to industry and analysis of commercial exploitation possibilities.

As a result of the ChatBI project, a system is expected that will allow access to and utilization of documents and records (e.g., user manuals, incident reports, etc.) that would otherwise be difficult to utilize.