Skip to main content

Open Book Genome Project

Project description

Welcome to the Open Book Genome Project (OBGP) Sequencer™, an open-source Book Processing Pipeline of responsibly vetted community "modules" which classify, sequence, and fingerprint book fulltext to reveal public insights about Books.


Want to get started immediately?

Try the OBGP Sequencer™ Google Colab Notebook:

How it Works

Each month, the OBGP Sequencer™ gets run against the fulltext of more than 1M books, generating valuable public insights for book lovers and researchers around the globe. OBGP Sequencer™ consists of carefully vetted community-contributed modules which aim to responsibly help increase the discoverability and usefulness of books, e.g.:

  • Identifying urls, isbns, and citations within the text
  • Generating word frequency mappings
  • Guessing grade reading levels

Who we are

OBGP is an independent, community-run, not-for-profit committee of open-source and book enthusiasts who want to responsibly further the effort of making books as useful and accessible as possible.



If you want to run the OBGP Sequencer™ pipeline, run:

pip install obgp

Local Development

git clone  # get the code
virtualenv venv && source venv/bin/activate  # setup a virtual environment
cd sequencer  # change into project directory
pip install -e .  # install the library (and re-run in project root as you make changes)

Docker Usage

First, create a directory called obgp_dir if it doesn't already exist. mkdir -p obgp_dir

Next, while in the sequencer directory, run: export OBGP_S3=~/.config/ia.ini OBGP_DIR=./obgp_dir;docker-compose up

Use docker container ls to get container ID. Then docker exec -it obgp /bin/sh replacing the container ID with the actual ID

You should now be in the container shell and can execute commands!


Once you've install either the production code or build your developer code, you may proceed to start python and import the runner.pipeline with whatever modules you'd like.

Let's say you want to process the book which has identifier hpmor on First, you would define your Sequencer as follows:

from bgp import Sequencer, STOP_WORDS
from bgp.modules.terms import NGramProcessor, WordFreqModule
s = Sequencer({
    'words': NGramProcessor(modules={
        'term_freq': WordFreqModule()
    }, n=1, stop_words=STOP_WORDS)

Then, you would pass this book identifier into the Sequencer to sequence the book to get back a genome Sequence object:

genome = s.sequence('hpmor')

Saving & Uploading Results

If your internetarchive tool is configured against an account with sufficient permissions, you can then upload your genome results back to an item (we'll arbitrarily pick the identifier bgp) by running:

>>> genome.write_results_to_item('bgp')

This will upload the genome.results as json to <book_identifier>_results.json (e.g. hpmor_results.json) unless otherwise specified by overriding params.

You will then be able to see your file hpmor_results.json within the bgp item's file downloads:

If you want to run a default test to make sure everything works, try:

from bgp import DEFAULT_SEQUENCER
genome = DEFAULT_SEQUENCER.sequence('9780262517638OpenAccess')


This pipeline allows a user to sequence a list of books from a jsonl in the following format:

{"identifier": "samplebook"}
{"identifier": "9780262517638OpenAccess"}

The pipeline then automatically chooses the most probable isbn for the book and attempts to update ia metadata accordingly while keeping a filesystem based database of all these actions.

Record Filesystem Action
How do we determine which books we've successfully uploaded a genome Touches GENOME_UPDATED_{identifier}
How do we determine the ISBNs of all books we’ve sequenced so far Touches ISBN_1234567890_{identifier}
How do we determine which books were sequenced but had no ISBN Touches UPDATE_NONE_{identifier}
How do we know which books attempted updating but failed UPDATE_FAILED_{identifier}
How do we know which books succeeded at updating and succeed UPDATE_SUCCEED_{identifier}
How do we know if item already has isbn metadata and is skipped UPDATE_CONFLICT_{identifier}
How do we know how many new urls were found in a book URLS_{number_of_urls}_{identifier}

The user can can grep and pipe to wc -l which tells them how many for each status and lists those items

Example Usage

Here is a archive item with the identifier samplebook. It has an isbn in the text of the book but no isbn metadata.

Screen Shot 2021-07-15 at 6 06 08 PM

There is a jsonl file with the items to be sequenced listed on new lines.

{"identifier": "samplebook"}
{"identifier": "9780262517638OpenAccess"}

To use the pipeline, run python samplebook.jsonl

You can specify the amount of pipeline processes to run concurrently with --p {number of processes} as a parameter. For example: python --p 4 samplebook.jsonl

If we tree results/bgp_results now we get:

├── 9780262517638OpenAccess
│   ├── ISBN_9780262517638_9780262517638OpenAccess
│   ├── UPDATE_CONFLICT_9780262517638OpenAccess
│   ├── URLS_290_9780262517638OpenAccess
│   └── book_genome.json
└── samplebook
    ├── ISBN_0787959529_samplebook
    ├── UPDATE_SUCCEED_samplebook
    ├── URLS_0_samplebook
    └── book_genome.json

2 directories, 8 files

If we check the metadata for the book on again we can see that the isbn field has been updated.

Screen Shot 2021-07-15 at 6 06 08 PM

There is also a file generated named URLS_0_samplebook which indicates that 0 non '' were extracted.

In the case of another item like 9780262517638OpenAccess, the filename indicates that 290 new urls were extracted. If we look at the contents of that file we will see all the unique urls extracted separated by newlines.
... etc.

The 9780262517638OpenAccess directory also shows that a url was extracted with the ISBN_9780262517638_9780262517638OpenAccess record, but was the ia item was not updated because a isbn already existed with the UPDATE_CONFLICT_9780262517638OpenAccess

Contributing a Module

  1. Please read the whitepaper and look through our community list of proposed or requested modules
  2. Propose a "module" by creating a github issue
  3. Get the code: Fork this git repository and clone it to your workspace. Create a new branch for your module (named after its corresponding github issue number and title: e.g. git checkout -b 12/module/find-isbns). Install
  4. Create a new module to the modules/ directory
  5. Test your module locally using Internet Archive's unrestricted collection of ~800k books
  6. open a Pull Request so your contribution may be reviewed.

Technical Overview

  • Book Genome Project extends/overloads @jjjake's internetarchive tool (invisibly using bad practices) in bgp/ with functions to fetch xml / plaintext (in a smart, memoized way)
  • Programmer builds a Sequencer (which is a list) of Sequences (a Sequence essentially does 1-pass on the data). Currently, the only sequences we have are 1gram and 2gram and these could be done in a single pass.
  • Each Sequence specified a list of modules to get run as it steps
  • The result is the top-level Sequencer can print out its .results as a dict

Sequencers, Processors & Modules

A Sequencer tells the Book Genome Project what tasks should be run and what results should be derived when processing a book's genome.

When a OGBP Sequencer is defined, it is loaded with a list of Processors, and these Processors with Modules.

A Processor is an abstraction which is responsible for fetching a specific representation of a Book (e.g. plaintext, xml, abbyy), splitting it into predefined logical units (e.g. characters, words, sentences, paragraphs, pages, chapters, entire text), stepping over each of these logical units, and sending them to it's registered Modules.

One example hypothetrical Processor might be called XMLSentenceProcessor. This Processor may be responsible for fetching (i.e. downloading) a Book in XML format with word-level markup. The run() method of the Processor's interface might parse and split the structured XML data into sentences, iterate through each sentence, and forward them to each of its registered Modules for processing. This hypothetical SentenceProcessor might be loaded with several Modules, such as a TotalSentenceCountModule and an SentenceWordCountStatsModule which, respectively, keeps track of the total number of sentences within the book and calculates the average number of words per sentence, etc.

In many cases a developer may find that the package's out-of-the-box bgp.DEFAULT_SEQUENCER is a great place to start, either as a Sequencer to run or a good example for extending.

Genome Schema

This is the reference schema used in genome json files:

  "metadata": {
    "identifier": "(ia identifier)",
    "version": "(commit)",
    "timestamp": "r(Unix Epoch)",
    "sequence_time": "r(sequence process seconds)",
    "source": {
      "txt": {
        "time": "r(txt download seconds)",
        "bytes": "r(txt bytes)"
      "xml": {
        "time": "r(xml download seconds)",
        "bytes": "r(xml bytes)"
    "processors": {
      "1gram": {
        "tokenization_time": "r(1gram tokenization process seconds)",
        "total_tokens": "r(1gram count)",
        "total_time": "r(1gram processor process seconds)",
        "modules": {
          "urls": {
            "time": "r(url process seconds)"
          "1grams": {
            "time": "r(1gram frequency process seconds)"
      "2gram": {
        "tokenization_time": "r(2gram tokenization process seconds)",
        "total_tokens": "r(2gram count)",
        "total_time": "r(2gram processor process seconds)",
        "modules": {
          "2grams": {
            "time": "r(2gram frequency process seconds)"
      "fulltext": {
        "total_time": "r(fulltext processor process seconds)",
        "modules": {
          "readinglevel": {
            "time": "r(reading level process seconds)"
      "pagetypes": {
        "total_time": "r(pagetype processor process seconds)",
        "modules": {
          "copyright_page": {
            "time": "r(copyright page process seconds)"
          "backpage_isbn": {
            "time": "r(copyright page process seconds)"
  "urls": [
  "1grams": [
                "r(1gram frequency)"
  "2grams": [
                "r(2gram frequency)"
  "readinglevel": {
              "readability": {
                "flesch_kincaid_score": "r(flesch kincaid score)",
                "smog_score": "r(smog score)"
              "lexile": {
                "min_age": "(Lower age in range)",
                "max_age": "(Upper age in range)"
  "copyright_page": [
                "page": "(copyright page)",
                "isbns": [
  "backpage_isbn": [

Public Testing Data sets

Here's a corpus of ~800k item identifiers of public domain books (of varying quality/appearance/language) which may be used for testing your module: (~19mb)


Please open an issue and request a slack invite

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Built Distribution

obgp-0.0.42-py2.py3-none-any.whl (15.2 kB view hashes)

Uploaded py2 py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page