Chatbot for Information Retrieval from Unstructured Natural Language Documents

Sammanfattning: This thesis brings forward the development of a chatbot which retrieves information from a data source consisting of unstructured natural language text. This project was made in collaboration with the company Jayway in Halmstad. Elasticsearch was used to create the search function and the service Dialogflow was used to process the natural language input from the user. A Python script was created to retrieve the information from the data source, and a request handler was written which connected the tools together to create a working chatbot. The chatbot correctly answers questions with a accuracy of 72% according to testing with a sample of n = 25. The testing consisted of asking the chatbot questions and determining if the answer is correct. Possible further research could be done to explore how chatbots might help the elderly or people with disabilities use the web with a natural dialogue instead of a traditional user interface.