linguatools / disco

compute semantic similarity between arbitrary words and phrases in many languages

DISCO API

Java API for word embeddings

This is the source code repository for the linguatools DISCO API. For more information on DISCO visit http://www.linguatools.de/disco/disco_en.html.

Quickstart

Install DISCO API

Download the source code by cloning this repository:

git clone [email protected]:linguatools/disco.git

Go into the repository folder and build the executable jar with dependencies:

cd disco/
./gradlew shadowJar

For instructions on command line usage call DISCO API without any parameters:

java -jar build/libs/disco-3.0.0-all.jar

or consult the web page.

Import a vector file from fastText

Download a fastText vector file in text format and unpack it:

wget https://s3-us-west-1.amazonaws.com/fasttext-vectors/word-vectors-v2/cc.de.300.vec.gz
gunzip cc.de.300.vec.gz

Download DISCO Builder:

wget http://www.linguatools.de/disco/DISCOBuilder-1.1.1.tar.bz2
tar jxf DISCOBuilder-1.1.1.tar.bz2

Convert the vector file into a DISCO DenseMatrix:

java -Xmx8g -cp DISCOBuilder-1.1.1/DISCOBuilder-1.1.1-all.jar de.linguatools.disco.builder.Import -in cc.de.300.vec -out cc.de.300.col.denseMatrix -wsType COL 

Query the new DISCO word space from the command line with the DISCO API:

java -Xmx4g -jar ~/repos-linguatools/disco/build/libs/disco-3.0.0-all.jar cc.de.300.col.denseMatrix/cc.de.300-COL.denseMatrix -s Haus Wohnung COSINE
0.64413786

Java API

To include DISCO in your Maven or Gradle project see below or visit the DISCO page on JitPack.

Gradle

Add this to your build.gradle file:

repositories {
    maven { url 'https://jitpack.io' }
}
dependencies {
    compile 'com.github.linguatools:disco:v3.0.0'
}

Maven

Add this to your pom.xml file:

<repositories>
	<repository>
	    <id>jitpack.io</id>
	    <url>https://jitpack.io</url>
	</repository>
</repositories>

<dependency>
    <groupId>com.github.linguatools</groupId>
    <artifactId>disco</artifactId>
    <version>v3.0.0</version>
</dependency>

Example Java code

DISCO disco = DISCO.load("cc.de.300-COL.denseMatrix");
float sim = disco.semanticSimilarity("Haus", "Häuschen", 
      	    	DISCO.getVectorSimilarity(SimilarityMeasure.COSINE));
System.out.println("similarity between 'Haus' and 'Häuschen': "+sim);
// get word vector for "Haus" as map
Map<String,Float> wordVectorHaus = disco.getWordvector("Haus");
// get word embedding for "Haus" as float array
float[] wordEmbeddingHaus = ((DenseMatrix) disco).getWordEmbedding("Haus");
// solve analogy x is to "Frau" as "König" is to "Mann"
List<ReturnDataCol> result = Compositionality.solveAnalogy("Frau", "König", "Mann", disco); 

Documentation

• javadoc
• DISCO home page

How to get word spaces for DISCO?

• import vector files (text format) from word2vec, GloVe or fastText using DISCO Builder. There are pre-computed vector files from fastText for 157 languages.
• you can download ready-to-use native DISCO word spaces (high-dimensional distributional count vectors) and DISCO word embeddings (low-dimensional predict vectors) imported from word2vec for several languages at http://www.linguatools.de/disco/disco-download_en.html.
• you can create your own high-dimensional distributional count vectors from a text corpus using DISCO Builder.

Features

• native Java API
• the API provides many useful methods for computing text similarity, solving analogies, clustering of similar words, compositional semantics, etc.
• efficient storage of high-dimensional sparse matrices (distributional count vectors) as well as low-dimensional dense matrices (word embeddings)
• higher-order word similarities can be stored and retrieved efficiently
• API is open source with Apache license.

disco's Issues

Information request : how do I create a word-space on my own??

The available wordspaces for English are from 2013 as I understand from the file names. Is there are a new wordspace for English available? Or how can I make one?