nchah/freebase-triples nchah/freebase-triples

  • Stars
    star
    196
  • Rank 198,553 (Top 4 %)
  • Language
    Shell
  • License
    BSD 2-Clause "Sim...
  • Created over 8 years ago
  • Updated over 5 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
nchah/freebase-triples
github

nchah

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

A methodology to process triples data from the Freebase data dumps.

freebase-triples

This repository contains Bash and Python code to process the Freebase RDF triples data dumps. Portions of this README document steps to clean and analyze the Freebase data dumps.

Citation

The accompanying paper is available at arXiv.

BibTeX entry for citation:

@article{chah2017freebase,
  title={Freebase-triples: A Methodology for Processing the Freebase Data Dumps},
  author={Chah, Niel},
  journal={arXiv preprint arXiv:1712.08707},
  year={2017}
}

The code is archived with a DOI on Zenodo: DOI

Niel Chah. (2017, November 10). nchah/freebase-triples v1.1.0 (Version v1.1.0). Zenodo. http://doi.org/10.5281/zenodo.1045306

Table of Contents

Background

Freebase

The Freebase Wikipedia article provides a good overview of the knowledge base's nearly decade-long lifetime from its beginnings at Metaweb Technologies, Inc. in 2007, its acquisition by Google in 2010, and its eventual shutdown and move to Wikidata towards 2015-2016. Freebase data could be accessed through the Freebase API and the online Query Editor. At the time, queries were written in the Metaweb Query Language (MQL).

Although the shutdown of freebase.com was set for as early as June 30, 2015 according to this initial Google+ post, the actual website was still accessible for quite a long time. On May 2, 2016, freebase.com was finally closed off as announced in this Google Group post. The freebase.com link now redirects to the Google Developers page for the remaining Freebase triples data dumps.

A screenshot of freebase.com on May 2, 2016 before it was shut down. freebase.com screenshot

Freebase Data Dumps

A data dump of 1.9 billion Freebase Triples in N-Triples RDF format is available on the developers page under the CC-BY license. The freebase.com URL also redirects to this page following its shutdown. The Developers page lists the file as 22 GB gzip compressed and 250 GB uncompressed, although a recent download exceeds this file size (a May 2016 download amounted to >30 GB compressed and >400 GB uncompressed).

Examining the compressed data with Z commands on the command-line:

$ # Scan through the data with zmore or zless
$ zmore freebase-rdf-latest.gz

<http://rdf.freebase.com/ns/american_football.football_player.footballdb_id>    <http://rdf.freebase.com/ns/type.object.type>   <http://rdf.freebase.com/ns/type.property>      .
<http://rdf.freebase.com/ns/american_football.football_player.footballdb_id>    <http://rdf.freebase.com/ns/type.object.name>   "footballdb ID"@en      .
<http://rdf.freebase.com/ns/american_football.football_player.footballdb_id>    <http://rdf.freebase.com/ns/type.property.unique>       "true"  .
<http://rdf.freebase.com/ns/american_football.football_player.footballdb_id>    <http://rdf.freebase.com/ns/type.property.expected_type>        <http://rdf.freebase.com/ns/type.enumeration>   .
<http://rdf.freebase.com/ns/american_football.football_player.footballdb_id>    <http://www.w3.org/2000/01/rdf-schema#label>    "footballdb ID"@en      .
<http://rdf.freebase.com/ns/american_football.football_player.footballdb_id>    <http://rdf.freebase.com/ns/type.property.schema>       <http://rdf.freebase.com/ns/american_football.football_player>  .

You can also grep it:

$ zgrep '/ns/film.film>' -m 5 freebase-rdf-latest.gz

<http://rdf.freebase.com/ns/film.film_song_relationship.film>   <http://rdf.freebase.com/ns/type.property.expected_type>    <http://rdf.freebase.com/ns/film.film>  .
<http://rdf.freebase.com/ns/film.film_song_relationship.film>   <http://www.w3.org/2000/01/rdf-schema#range>    <http://rdf.freebase.com/ns/film.film>  .
<http://rdf.freebase.com/ns/g.112ygbz6_>    <http://rdf.freebase.com/ns/type.object.type>   <http://rdf.freebase.com/ns/film.film>  .
<http://rdf.freebase.com/ns/g.112ygbz6_>    <http://www.w3.org/1999/02/22-rdf-syntax-ns#type>   <http://rdf.freebase.com/ns/film.film>  .
<http://rdf.freebase.com/ns/g.113qbnjlk>    <http://rdf.freebase.com/ns/type.object.type>   <http://rdf.freebase.com/ns/film.film>  .

Each triple is encoded in the aforementioned N-Triples format. The subject, predicate, and object values on each line are "< >" enclosed and tab separated. Each line terminates with a "." and a newline.

Viewing this with vim, using :set list to show these hidden characters:

</american_football.football_player.footballdb_id>^I</type.object.name>^I"footballdb ID"@en^I.$
</astronomy.astronomical_observatory.discoveries>^I</type.object.name>^I"Discoveries"@en^I.$
</automotive.body_style.fuel_tank_capacity>^I</type.object.name>^I"Fuel Tank Capacity"@en^I.$
</automotive.engine.engine_type>^I</type.object.name>^I"Engine Type"@en^I.$
</automotive.trim_level.max_passengers>^I</type.object.name>^I"Maximum Number of Passengers"@en^I.$

In This Repository

Directory

An overview of the directory structure.

$ tree
.
β”œβ”€β”€ README.md
β”œβ”€β”€ data
β”‚Β Β  └── schema
β”‚Β Β      β”œβ”€β”€ fb-rdf-pred-schema-domains-ids-1-byalpha-desc
β”‚Β Β      β”œβ”€β”€ fb-rdf-pred-schema-domains-ids-1-byalpha-typeinfo
β”‚Β Β      β”œβ”€β”€ fb-rdf-pred-schema-properties-ids-1-byalpha-desc
β”‚Β Β      β”œβ”€β”€ fb-rdf-pred-schema-properties-ids-1-byalpha-typeinfo
β”‚Β Β      β”œβ”€β”€ fb-rdf-pred-schema-types-ids-1-byalpha-desc
β”‚Β Β      β”œβ”€β”€ fb-rdf-pred-schema-types-ids-1-byalpha-typeinfo
β”‚Β Β      β”œβ”€β”€ unique-predicates-sorted
β”‚Β Β      └── unique-types-sorted-and-counts
β”œβ”€β”€ images
β”‚Β Β  β”œβ”€β”€ screenshot-cayley-visualization.png
β”‚Β Β  └── screenshot-freebase-com.png
└── scripts
    β”œβ”€β”€ python
    β”‚Β Β  β”œβ”€β”€ queries
    β”‚Β Β  β”‚Β Β  β”œβ”€β”€ queries-common--properties
    β”‚Β Β  β”‚Β Β  β”œβ”€β”€ queries-common--types
    β”‚Β Β  β”‚Β Β  β”œβ”€β”€ queries-schema-for-domains-types-properties
    β”‚Β Β  β”‚Β Β  β”œβ”€β”€ queries-slices-for-all-domains
    β”‚Β Β  β”‚Β Β  β”œβ”€β”€ queries-type--properties
    β”‚Β Β  β”‚Β Β  └── queries-type--types
    β”‚Β Β  β”œβ”€β”€ s2-c1-extract-triples.py
    β”‚Β Β  β”œβ”€β”€ s2-c2-extract-schema.py
    β”‚Β Β  β”œβ”€β”€ s2-c3-extract-schema-ids.py
    β”‚Β Β  β”œβ”€β”€ s2-c4-extract-schema-desc.py
    β”‚Β Β  β”œβ”€β”€ s2-c5-extract-schema-typeinfo.py
    β”‚Β Β  └── s2-c6-extract-merge-triples.py
    └─── shell
     Β Β  β”œβ”€β”€ s0-run-parse-extract-triples.sh
     Β Β  β”œβ”€β”€ s1-parse-triples.sh
     Β Β  β”œβ”€β”€ s2-extract-triples.sh
     Β Β  └── s3-query-triples.sh

Scripts

Languages

The scripts in this repo are written in Bash and Python. Each script is named and ordered to reflect the ETL stages outlined here. Bash/Shell scripts handled the initial parsing stages for the massive data dumps. Python, with its many libraries, is ideal to use after the initial processing.

  • Bash commands used: awk, cat, cut, grep, gsed*, less, more, parallel, pv, sed, sort, wc, zless, zmore, zgrep

*gsed is GNU sed. macOS's sed does not handle '\t' as tab characters so gsed is preferred in some instances.

Operating Scripts

Pausing a job on the command line can be done with CTRL+Z. All stopped and background jobs can be listed with the jobs command. To bring background jobs back into the foreground, use fg [job]. Unless specified differently, scripts were run on a MacBook Pro (Early 2015, 2.7 GHz Intel Core i5, 8 GB memory).

ETL Changes

ETL refers to the extraction, transformation, and loading of large datasets in the data science field. This section tracks the ETL changes that can be applied to the triples data.

The data dumps encode Freebase data in a few ways that are different from the usual usage on Freebase.com.

  • Notes:
    • "/" is replaced by "." for topic mids and domains/types/properties (e.g. /m.abcdef or /film.film instead of /m/abcdef or /film/film).
    • URLs to freebase.com or w3.org are used, not just the Freebase mids. All freebase.com addresses no longer work following the site shutdown but remain in the data dump as unique identifiers.
    • A mix of freebase.com and w3.org schemas are used, especially as predicates in the triples.
    • There are over 1.9 billion triples and thus the same amount of lines in the entire data dump.
    • The triples are already sorted alphabetically in some columns, but this is applied either inconsistently or according to a pattern that needs to be discovered.
    • Unique identifiers are enclosed in "<,>". Strings are written in the format: "string"@language_namespace.

Tasks

  1. Simplifying Data
    • [s1-c1] - Convert N-Triples (Wikipedia) format to a leaner format. Working with the full URIs conforms to the standard, but can be unwieldy to use for this project.
    • [s1-c2] - Removing the "< >" format enclosing each value.
    • [s1-c3] - Convert "." back to "/" in the domain, type, and property schemas to return a more Freebase-like format (e.g. "/award/award_winner" for the type).
  2. Slicing Data
    • [s2-c1] - Create predicate-based slices for each domain, type, and property.
    • [s2-c2, 3, 4, 5] - Extract data on the schema (ontology) from the slices.
    • [s2-c6] - Extract and merge a slice's data concerning a domain(s) with other slices, like name, desc, etc.
  3. Querying Data
    • [s3-c1] - Query triples by specific prediciates, by domain, or other criteria of interest.
    • [s3-c2] - Obtain analytics/statistics on the data distribution, shape of the data, etc.
    • [s3-c3] - Merge data together to understand a specific domain, object, etc.
  4. Interpreting/Visualizing Data

Sample Analysis

Cayley

Once the data is cleaned and ready, the Cayley graph database and platform can be used to analyze triples data. Cayley is an open-source graph database that draws on Freebase and the Knowledge Graph. It is maintained at cayleygraph/cayley.

A Gremlin query using the Cayley Visualize function in the browser:

$ # On the command line:
$ ./cayley http --dbpath=data/testdata.nq

> // Running Cayley in the browser locally
> // Visualizing instances of /award/award_winner
> // Set GetLimit() as getting All() creates a large cluster of thousands of nodes
> g.V("/award.award_winner").Tag("source").Out("/type.type.instance").Tag("target").GetLimit(10)

Cayley visualization screenshot

License

Freebase data is licensed under the CC-BY license. The Freebase API also has additional Terms and Conditions.

Sources

Some sources may no longer be available following the deprecation of the Freebase API on June 30, 2015.

Announcements Timeline

Many Freebase and Knowledge Graph related updates are posted on the once active freebase-discuss Google Group and the Google+ community.

Freebase

  • Bollacker, K., Evans, C., Paritosh, P., Sturge, T., & Taylor, J. (2008, June). Freebase: a collaboratively created graph database for structuring human knowledge. In Proceedings of the 2008 ACM SIGMOD international conference on Management of data (pp. 1247-1250). AcM.
  • Google, Freebase Data Dumps, https://developers.google.com/freebase/data, August 15, 2017.

Google Developers Resources

Freebase API

Knowledge Graph Search API

More Repositories

1

github-traffic-stats

Get statistics on web traffic to your GitHub repositories.
Python
121
star
2

word2vec4everything

Python CLI to apply word2vec to all sorts of text documents.
Python
49
star
3

freebase-mql

A review of the deprecated Freebase knowledge base and Metaweb Query Language (MQL). A brief comparison of MQL and GraphQL.
42
star
4

knowledge-graph-api

Simple script to query Google's Knowledge Graph API.
Python
41
star
5

movielens-recommender

Data Analytics and Machine Learning project for a University of Toronto graduate course.
Python
34
star
6

chrome-extension-starter

Starter kit for Chrome (and Firefox) Extension and Theme development.
JavaScript
30
star
7

google-vision-api-image-annotate

Annotating images with the Google Cloud Vision API.
Python
7
star
8

tech-graph

A force-directed graph of notable tech companies, their products, and services.
JavaScript
5
star
9

nlpml-project

Natural language processing project
Python
3
star
10

machine-learning-scripts

various Natural Language Processing scripts
Python
2
star
11

topsites-html-comments

A script to create a collection of the comments left in the HTML code of various high traffic sites as ranked by Alexa.com. A search for easter eggs, if there are any.
Python
2
star
12

mongo2elasticsearch

Load data from MongoDB to Elasticsearch.
Shell
1
star
13

url-alert

Chrome extension to parse URLs for internationalized domain names
HTML
1
star