I produced this application for my university dissertation, as I really wanted to work with NLP and poetry. I'm a big poetry fan, so I personally got a lot out of my project! If you're interested in glancing at my dissertation, you can find it here: https://docs.google.com/document/d/e/2PACX-1vRmK5bMVz7HPTCFTvNUHdgGV417vvQMuLgxihf1YcrtP4Q2dez16jKj6hQy2SLo-SoKBS3SYvk4SxdC/pub
Recommender systems are applications that use product features to recommend similar products to users. Content-based recommender systems are a subset of recommender systems that use features gained from analysing the content itself. Content-based recommender systems require a strong degree of specific domain understanding to extract relevant features for comparison. Each and every different type of content has a different set of features that need to be extracted. This project will be focused around building a web application that recommends poetry to users, based on users previously liked poems and relevant features gathered from the text. The central question answered in this dissertation is, what features of poems are predictive of style, and can they be extracted to create a viable poetry recommender.
Study participants were used to determine if the final recommender provides better recommendations than a random poetry selection. Qualitative interviews were also carried out to assess if the chosen features were understood by the end users, and if users trusted the final poetry recommendation system.
It was found that features could be successfully extracted from poetry and used to generate a poetry recommendation system with recommendations that surpassed a random control. Participants of the qualitative interviews enjoyed the use of the recommender system, and all showed decreased skepticism in a machines ability to recommend poetry.