No Description

Michael Ekstrand af73d15a92
Document parser fields
fix url
1 year ago
Bump DVC version
1 year ago
commit loc-books
1 year ago
System requirements
1 year ago
Bump DVC version
1 year ago
Bump DVC version
1 year ago
Bump DVC version
1 year ago
Bump DVC version
1 year ago
Use Curl instead of wget/aria2 for downloads
1 year ago
Document parser fields
11 months ago
Add Windows command support
1 year ago
Add Windows command support
1 year ago
Re-run LOC MDS extract
1 year ago
Refactor Rust and add transcripts to ntriples
1 year ago
Refactor Rust and add transcripts to ntriples
1 year ago
Cluster statistics and exploration
1 year ago
Bump DVC version
1 year ago
Cluster statistics and exploration
1 year ago
Add license
2 years ago
cluster information
1 year ago
Add documentation
1 year ago
Add Windows command support
1 year ago
Get path-based statuses working
1 year ago
Use Curl instead of wget/aria2 for downloads
1 year ago
Bump DVC version
1 year ago
simplify + goodreads import
2 years ago
Add Rust-based ISBN-parsing logic
1 year ago
Bump DVC version
1 year ago
Data Pipeline
DVC Managed File
Git Managed File
Stage File
External File

This repository contains the code to import and integrate the book and rating data that we work with. It imports and integrates data from several sources in a single PostgreSQL database; import scripts are primarily in Python, with Rust code for high-throughput processing of raw data files.

If you use these scripts in any published research, cite our paper:

Michael D. Ekstrand, Mucun Tian, Mohammed R. Imran Kazi, Hoda Mehrpouyan, and Daniel Kluver. 2018. Exploring Author Gender in Book Rating and Recommendation. In Proceedings of the 12th ACM Conference on Recommender Systems (RecSys '18). ACM, pp. 242–250. DOI:10.1145/3240323.3240373. arXiv:1808.07586v1 [cs.IR].

Note: the limitations section of the paper contains important information about the limitations of the data these scripts compile. Do not use the gender information in this data data or tools without understanding those limitations. In particular, VIAF's gender information is incomplete and, in a number of cases, incorrect.

In addition, several of the data sets integrated by this project come from other sources with their own publications. If you use any of the rating or interaction data, cite the appropriate original source paper. For each data set below, we have provided a link to the page that describes the data and its appropriate citation.


  • PostgreSQL 10 or later with orafce and pg_prewarm (from the PostgreSQL Contrib package) installed.
  • Python 3.6 or later with the following packages:
    • psycopg2
    • numpy
    • tqdm
    • pandas
    • colorama
    • chromalog
    • natural
    • dvc
  • The Rust compiler (available from Anaconda)
  • 2TB disk space for the database
  • 100GB disk space for data files

It is best if you do not store the data files on the same disk as your PostgreSQL database.

The environment.yml file defines an Anaconda environment that contains all the required packages except for the PostgreSQL server. It can be set up with:

conda create -f environment.yml

We use Data Version Control (dvc) to script the import and wire its various parts together. A complete re-run, not including file download time, takes approximately 8 hours on our hardware (24-core 2GHz Xeon, 128GiB RAM, spinning disks).

Configurating Database Access

All scripts read database configuration from the DB_URL environment variable, or alternately a config file db.cfg. This file should look like:

host = localhost
database = bookdata

This file additionally supports branch-specfic configuration sections that will apply to work on different Git branches, e.g.:

host = localhost
database = bookdata

database = bdorig

This setup will use bookdata for most branches, but will connect to bdorig when working from the master branch in the git repository.

This file should not be committed to Git. It is ignored in .gitignore.

Initializing and Configuring the Database

After creating your database, initialize the extensions (as the database superuser):


The default PostgreSQL performance configuration settings will probably not be very effective; we recommend turning on parallelism and increasing work memory, at a minimum.

Downloading Data Files

This imports the following data sets:

Several of these files can be auto-downloaded with the DVC scripts; others will need to be manually downloaded.

Running Everything

You can run the entire import process with:

dvc repro

Individual steps can be run with their corresponding .dvc files.


The import code consists of Python, Rust, and SQL code, wired together with DVC.

Python Scripts

Python scripts live under scripts, as a Python package. They should not be launched directly, but rather via, which will make sure the environment is set up properly for them:

python sql-script [options] script.sql

SQL Scripts

Our SQL scripts are run with a custom SQL script runner (the sql-script Python script), that breaks them into chunks, handles errors, and tracks dependencies and script status. The script runner parses directives in SQL comments; for example:

--- #step ISBN ID storage

is a step called "ISBN ID storage". Each step is processed in a transaction that is committed at the end, so steps are atomic (unless marked with #notx).

These are the directives for steps:

  • #step LABEL starts a new step with the label LABEL. Additional directives before the first SQL statement will apply to this step.
  • #notx means the step will run in autocommit mode. This is needed for certain maintenance commands that do not work within transactions.
  • #allow CODE allows the PostgreSQL error 'code', such as invalid_table_definition. The script will not fail if the step fails with this error. Used for dealing with steps that do things like create indexes, so if the index already exists it is fine to still run the script.

In addition, the top of the file can have #dep directives, that indicate the dependencies of this script. The only purpose of the #dep is to record dependencies in the database stage state table, so that modifications can propagate and be detected; dependencies still need to be recorded in .dvc files to run the import steps in the correct order.

DVC Usage and Stage Files

Running the scripts here with raw dvc does not work. You need to use the wrapper script, as in:

./ repro

The wrapper script sets up DVC to recognize our special pgstat://stage URLs for tracking the status of database import stages in the live database.

Import is structured as a concept of stages map almost 1:1 to our DVC step files. They manage database-side tracking of data and status.

Each import stage includes pgstat://stage as an unached output stage, as in:

- path: pgstat://bx-import
  cache: false

From the command line, uncached outptus are created by using -O instead of -o.

Each script that requires another stage to be run first depends on pgstat://stage as a dependency.

This wires together all of the dependencies, and uses the current state in the database instead of files that might become out-of-sync with the database to track import status.

The stage name matches the name of the .dvc file.

The reason for this somewhat bizarre layoutis that if we just wrote the output files, and the database was reloaded or corrupted, the DVC status-checking logic would not be ableto keep track of it. This double-file design allows us to make subsequent steps depend on the actual results of the import, not our memory of the import in the Git repository.

In-Database Status Tracking

Import steps are tracked in the stage_status table in the database. For completed stages, this can include a key (checksum, UUID, or other identifier) to identify a 'version' of the stage. Stages can also have dependencies, which are solely used for computing the status of a stage (all actual dependency relationships are handled by DVC):

  • stage_deps tracks stage-to-stage dependencies, to say that one stage used another as input.
  • stage_file tracks stage-to-file dependencies, to say that a stage used a file as input.

The source_file table tracks input file checksums.

Projects using the book database can also use stage_status to obtain data version information, to see if they are up-to-date.

Utility Code

The bookdata package contains Python utility code, and the src directory contains a number of utility modules for use in the Rust code. To the extent reasonable, we have tried to mirror design patterns and function names.

Design for Datasets

The general import philosophy is that we import the data into a PostgreSQL table in a raw form, only doing those conversions necessary to be able to access it with PostgreSQL commands and functions. We then extract information from this raw form into other tables and materialized views to enable relational queries.

Each data set's import process follows the following steps:

  1. Initialize the database schema, with an SQL script under schemas/.
  2. Import the raw data, controlled by DVC steps under import/. This may be multiple steps; for example, OpenLibrary has a separate import step for each file. Actual import is usually handled by Rust or Python code.
  3. Index the data into relational views and tables. This is done by SQL scripts under index/.

Data integration then happens after the data sets are indexed (mostly - a few indexing steps depend on the book clustering process).

Adding a New Dataset

If you want to add a new data set, there are a few steps:

  1. Set up the initial raw database schema, with an SQL script and corresponding DVC file under schemas/. This should hold the data in a form that matches as closely as practical the raw form of the data, and should have minimal indexes and constraints. For a new schema ds-schema, you create two files:

    • ds-schema.sql, containing the CREATE SCHEMA and CREATE TABLE statements. We use PostgreSQL schemas (namespaces) to separate the data from different sources to make the whole database more manageable. Look at existing schema definitions for examples.
    • ds-schema.dvc, the DVC file running ds-schema.sql. This should contain a few things:


      Run the schema file

      cmd: python ../ sql-script ds-schema.sql

      Depend on the file and initial database setup


      • path: ds-schema.sql
      • path: pgstat://common-schema outs:

        a transcript of the script run

      • path: ds-schema.transcript

        the status of importing this schema

      • path: pgstat://ds-schema cache: false ```

      When you run ./ repro schemas/ds-schema.dvc, it will run the schema script and fill in the other values (e.g. checksums) for the dependencies and outputs.

  2. Download the raw data files into data and register them with DVC (dvc add data/file for the simplest case), and document in this file where to download them. For files that it is reasonable to auto-download, you can create a more sophisticated setup to download them, but this is often not necessary.

  3. Identify, modify, and/or create the code needed to import the raw data files into the database. We have importers for several types of files already:

    • If the data is in CSV or similar form, suitable for PostgreSQL's COPY FROM command, the pcat import tool in the Rust tools can copy the file, decompressing if necessary, directly to the database table.

    • If the data is in JSON, we have importers for two forms of JSON in the import-json Rust tool, the source for which is in src/commands/import_json/. Right now it supports OpenLibrary and GoodReads JSON files; the first is a tab-separated file containing object metadata and the JSON object, and the second is a simple object-per-line format. The accompanying file ( and define the data format and the destination tables. For many future JSON objects, will be the appropriate template to start with, and add support for it to the appropriate places in

    • If the data is in MARC-XML, the Rust parse-marc command is your starting place. It can process both multiple-record formats (e.g. from VIAF) or single-document formats (from the Library of Congress), and can decompress while importing.

    If you need to write new import code, you may need to make sure it properly records stage dependencies and status. At a minimum, it should record each file imported and its checksum as a file for the stage, along with the stage begin/end timestamps. Look at the meta-schema.sql file for the specific tables. The and support modules provide code for recording stage status in Python and Rust, respectively.

  4. Set up the import process with an appropriate .dvc step in import/. This step should depend on the schema (pgstat://ds-schema), and have as one of its uncached outputs the import process status (pgstat://ds-import, if the file is named ds-import.dvc). Some importers require you to explicitly provide the stage name as a command-line argument.

  5. Write SQL commands to transform and index the imported data in a script under index/. This script may do a number of things:

    • Map data set book ISBNs or other identifiers to ISBN IDs.
    • Extract relational tables from structured data such as JSON (e.g. the book author lists extracted from OpenLibrary).
    • Create summary tables or materialized views.

    See the existing index logic in index/ for examples.

  6. Create a .dvc stage to run your index script; this works like the one for the schema in (1), but is under index/ and depends on pgstat://ds-import (or whatever your import stage is named).

  7. Create or update data integrations to make use of this data, as needed and appropriate.

    If the new data contains ISBN/ID links that you want to include in book clustering, add support to and update the cluster.dvc file to also depend on your data set's index stage (e.g. pgstat://ds-index).

  8. If appropriate, add a dependency on the last stage of your processing to Dvcfile.

All dependencies should be through the pgstat:// URLs, so that they are computed from current database status.

Book Identifiers

Each data set comes with its own identification scheme for books:

  • LCCN
  • OpenLibrary key
  • ASIN
  • GoodReads book and work identifiers

We integrate these through two steps. First, we map ISBNs to numeric IDs with the isbn_id table. This table contains every ISBN (or ISBN-like thing, such as ASIN) we have seen and associates it with a unique identifier.

Second, we map ISBN IDs to clusters with the isbn_cluster table. A cluster is a collection of related ISBNs, such as the different editions of a work. They correspond to GoodReads or OpenLibrary 'works' (in fact, when a GoodReads or OpenLibrary work is available, it is used to generate the clusters).

This allows us to connect ratings to metadata with maximal link coverage, by pulling in metadata across the whole book cluster.