Better book search with AI

OBA would like to upgrade the search capabilities for their library catalogue. Currently, search is very limited in that it requires an exact book title and/or author name. Not only are typos and misspellings fairly common in practice, but users often search for books by referring to the book’s content. For example: "I'm looking for a book about a boy who takes the train to a wizard school." How do we connect this query to the library database and get information about Harry Potter books as output? Often, users do not even have a specific book in mind, but instead search for books on a particular topic, like "Olympic Games". In that case, the goal would be to output books that mention the Olympic Games rather than books whose title contains those specific words.

We proposed an incremental approach, where more complex and/or experimental features can be added gradually to the core system. As a first step, we built a knowledge graph based on the OBA database and added a named entity recognition layer on top of the existing data. This enables searching for characters, organizations, locations, and other entities mentioned in book descriptions. For example, a search for “boeken over de Tweede Wereldoorlog in Amsterdam” (books about World War II in Amsterdam) will yield all books that mention both Amsterdam and WW2. This augmented knowledge graph is a foundation for further steps that could additionally enhance the search experience. For example, large language models can be integrated with the knowledge graph in the future.