Processing complex question with Watson
Hello I would like to use Watson to parse compound questions from user. For example, "Which movie was directed by director who also directed Apocalypse Now". The functionality should be able to identify what thing was asked by the user (movie) and related conditions, possibly resulting in a hierarchal structure:
{movie directed by:{ director directed:{ Apocalypse Now } }
Does Watson have such features
There is no single Watson service that will do this, but it is possible to craft such an application using a combination of the natural language classifier, concept insights and retrieve and rank services. You will need to train the classifier and load in your documents and craft a ranker. I would probably create the retrieve capability with documents that provide you with search parameters into something like IMDb.
You could use Alchemy API to do keyword extraction. Using your question above returns this with knowledge graph switch on.
"Which movie was directed by director who also directed Apocalypse Now"
Apocalypse -> 0.922952 -> /books/apocalypse
movie -> 0.789507 -> /industries/media/movie
director -> 0.778883 -> organizations/individuals/participants/representatives/director
On the face of it, it appears to almost capture the items you need to generate your structure. You would probably need to play with it a bit more.
Another option is to use Watson Explorer Studio to create custom annotators (UIMA) and then reference them in your application. It means you could easily create a movies dictionary, or any other that has custom domain terms. There is a free unsupported old version, which should at least work with earlier UIMA versions. It will give you a feel to what it can do. I'd recommend watching the training videos. I also did a video before (again not the latest version). It shows you can use it with a Java application.