Chat Bot
The goal aimed for when creating a retrieval-based bot is to have a model which ranks how appropriate all predefined responses are to a given input and then replies with the highest ranked.
Abstract
This projects implementation of a retrieval based chatbot relies on the bot being able to make two decisions to choose an appropriate response from a set. Firstly, it needs to decide which of the arguments from the set is the most similar to the users given input. The bot then has to decide whether the input and the most similar claim agrees with each other. If the input and the most similar claim do agree, the bot should return a known counterargument to the claim. If they do not, the bot should return an affirming argument to the claim. With the prepared data, a database containing all known claims and their respective pro and con arguments can be constructed. The database is then also filled with features which enables similarity comparison between different, but and also new claims. The database is constructed using pandas. Pandas is powerful and widely used python library for data manipulation and analysis. To make the bot appear more human, some answers to non-discussion related inputs where added, such as greetings. As long as the bot’s inputs are similar to what the bot "knows", its answers can be somewhat reasonable.