jjfiv/poetry-identification

Poetry Identification Code from my dissertation runs on zip files containing DJVUXML from the Internet Archive.

Poetry-Identification

Poetry Identification Code from my dissertation runs on zip files containing DJVUXML from the Internet Archive.

Where did this model come from?

For details about where this model came from, or what it does, refer to my dissertation for now.

@phdthesis{foley2019thesis,
  author = {John Foley},
  title = {{Poetry: Identification, Entity Recognition, and Retrieval}},
  year = {2019},
  school = {University of Massachusetts},
}

Can I get some data for this?

Data from my dissertation is available at CIIR/downloads/poetry. The training data used to build the model is there, as well as the output of this model on the 50,000 books from the INEX 2007 challenge (basically a random sample of Internet Archive books).

How do I run the code?

You'll need a bunch of DJVU-XML books available in a zip file. I have so many of these -- email me and we can work something out :)

Prepare

1. Get Rust.
2. gunzip ../models/forest-05-2019.json.gz # Extract the model; it's too big for github otherwise -- only need to do this once.

Build and run the code:

cd classification
cargo build --release
./target/release/classification --model ../models/forest-05-2019.json --books input_books.zip > input_books.poetry.jsonl

The classification binary once built is very portable because Rust does static linking -- you can build it once and copy it to a cluster of Linux machines fairly easily.

About this Code

This code is written in Rust. There are two packages: djvuxml-rs which is a pretty generic way to interact with internet-archive scanned book files, and classification which runs through using a JSONified Random Forest model and makes predictions at the page level. The files on CIIR/downloads/poetry -- Poetry50K collection were generated from de-duplicating the output of this code.

Help? Where's the code for XXX?

I'm slowly cleaning up and open-sourcing all the code. If you're looking for a piece that's not made it public yet, please don't hesitate to contact me! File an issue here or check out my personal website to find my latest academic email.