The construction of a smart assistant requires the implementation of software components for data acquisition, data extraction, intent recognition, recommendation and so on. In this master’s thesis we implement a smart assistant for meal suggestion and center the focus more specifically on the allergen and nutrient recognition from semi-structured HTML data from restaurant websites.
To solve this problem we implement algorithms for menu text separation to stand-alone dishes included inside the menu, rule based algorithm for allergen detection from text and allergen detection algorithm using neural network.
The software components mentioned above are used to implement a chatbot that provides users enriched and customized previews of daily menus. It is integrated into different communication platforms (Microsoft Teams, Discord, Slack and Facebook Messenger), where the conversation is held in Slovene.
A case study with users has shown, that the assistant makes menu and restaurant selection for lunch easier for the user. The algorithm for the automatic extraction of allergens with a neural network reaches an accuracy of 68% (F1 score), which is suitable for warning users about the possible content of allergens in a dish, although it is wise for the user to check this information with restaurant staff.
|
