This system, first, processes the input text with a NLP pipeline (coreNLP): sentence splitting, tokenization, part-of-speech tagging, Named Entities recognition, coreference resolution and dependency analysis. Then, departing from this  information, it generates sets of triples expressing concepts and relations contained in each of the sentences.


This baseline chatbot handles two different types of scenarios: (1) the user wants to find a recipe meeting his criteria, and (2) the user asks a question relative to the cooking domain. For the first scenario, the system accesses a database which contains 1,064 recipes from Wikipedia:Cookbook. The database contains information about the name, the details, the ingredients, the variations, the procedure and the categories for each recipe. For the second scenario, the system accesses the unstructured data consisting in sentences from 784 non-recipe documents that can be found on Wikipedia:Cookbook.


  • Demo of a baseline question answering system (internal demo)

This baseline question answering system consists in a subset of an industrial chatbot system, which can read and understand a textual documentation, and automatically generate all the questions and answers that can be asked about the product or service in question. There is no graphical user interface to use it. Instead, there is a Python notebook  that shows how to programmatically build a knowledge base, either from a text or from a set of questions and answers, and how to make queries.