Chatbot for flood awareness
Research Summary
The power of Artificial Intelligence (AI) allows for extending its application in various areas of water management. One of the already widely known ability of AI techniques is building non-linear regression models which can be employed in forecasting future flood events, using numerous available sufficient past hydrological and geospatial data; which is important to support management of the surging cases of disastrous flood events. At the same time there has been a limited number of attempts to utilize quite a different AI technology, such as the AI chatbot, in creating awareness based on the forecasted flood information. Thus the viability of employing an AI chatbot mobile application for creating flood awareness and encouraging citizen participation in flood risk management, was explored in this research. Yearly flood forecast data was obtained for the study area, Nigeria; which was classified into eight hydrological units consisting of 36 states of the federation with a total of 774 local areas. These forecast reports were obtained from the Nigerian Hydrological Agency (NHIA), and were extracted to form a large parts of the model inputs. An AI technique known as the Natural language processing (NLP), was employed in the manipulation and conversion of text to computer interpretable language. Training data (known as intents) were created in the form of json with numerous conversational text with several patterns and possible responses per intents. A Python code was developed and documented with libraries such as nltk, numpy, tflearn and tensorflow, to execute the various preprocessing and processing stages where the texts were converted to machine language for further processing. To optimize the functionality of the bot, an advanced NLP platform known as the Google Dialogflow, was used for processing, training and testing a total of 180 bot intents with series of patterns and responses. Finally, a mobile application was developed for integrating and deploying the AI chatbot for user interaction. This was performed using the flutter framework, which operates using the Dart programming language for the front-end, resulting in an app created with one codebase, to be used on both the Android and the iOS platform. The end product of the research is a robotic chatbot mobile application that is user-friendly, and is capable of providing flood awareness and guidance information to the user without the need for a secondary human agent. Furthermore, a pilot testing of the app on a relatively large sample population is carried out to investigate its user-ability and efficiency for use by the targeted population. Performance chats evaluating the responses were carried out to determine these levels of satisfaction. The presented research brings the AI chatbot technology and natural language processing techniques into the practical realm of HydroInformatics, and shows their excellent potential for citizen’s involvement in water management.
Problem Description & Study Area
Problem
High level of limited literacy and technological experience of the affected population especially in the rural and remote regions in the developing countries, could impact the ability to surf the net in search of available flood related information.
Lack of easily accessible means of getting flood guidance information by the flood affected population
Study Area
Nigeria lies within the west coastal region of Africa, between latitudes 4oN and 14oN and longitudes 3oE and 15oE.
Population of about 200million people occupying a total of approximately 923,800 sq.km.
There are basically two seasons in Nigeria namely; the raining season and the dry season
Significant factors causing flood: 1.) regime of the soil moisture of the lower lying plain during peak periods of the raining season, 2.) topography, 3.) dam operations especially from outside the nation’s borders,
Approach Employed in the solution
System Design
Response Network
Steps followed in the Data Procesing with the NLP Technique
