CogComp-time

CogComp's Natural Language Processing Libraries and Demos: Modules include lemmatizer, ner, pos, prep-srl, quantifier, question type, relation-extraction, similarity, temporal normalizer, tokenizer, transliteration, verb-sense, and more.

CogComp-time

CogComp-time is a temporal extraction and normalization framework, designed to detect the temporal phrase of given text, and normalize them according to TIMEX3 standard

Usage

If you want to use this in your project, you need to take two steps. First add the dependencies, and then call the functions in your program. Here is how you can add maven dependencies into your program:

    <repositories>
        <repository>
            <id>CogCompSoftware</id>
            <name>CogCompSoftware</name>
            <url>http://cogcomp.org/m2repo/</url>
        </repository>
    </repositories>

    <dependencies>
        <dependency>
            <groupId>edu.illinois.cs.cogcomp</groupId>
            <artifactId>illinois-time</artifactId>
            <version>VERSION</version>
        </dependency>
    </dependencies>

Note: Make sure to change the pom.xml parameter VERSION to the latest version of the project.

In general, the best way to use the illinois-time is through the TemporalChunkerAnnotator class. Like any other annotator, it is used by calling the addView() method on the TextAnnotation containing sentences to be tagged.

import edu.illinois.cs.cogcomp.annotation.AnnotatorException;
import edu.illinois.cs.cogcomp.annotation.AnnotatorService;
import edu.illinois.cs.cogcomp.core.datastructures.ViewNames;
import edu.illinois.cs.cogcomp.core.datastructures.textannotation.TextAnnotation;
import edu.illinois.cs.cogcomp.core.datastructures.textannotation.View;
import edu.illinois.cs.cogcomp.core.utilities.configuration.Configurator;
import edu.illinois.cs.cogcomp.core.utilities.configuration.ResourceManager;
import edu.illinois.cs.cogcomp.pipeline.common.PipelineConfigurator;
import edu.illinois.cs.cogcomp.pipeline.main.PipelineFactory;

import java.io.IOException;
import java.util.Properties;

public class demo {
    public static void main (String []args) throws IOException, AnnotatorException {
        Properties rmProps = new TemporalChunkerConfigurator().getDefaultConfig().getProperties();
        // Set to "True" if you want to use HeidelTime for normalization
        rmProps.setProperty("useHeidelTime", "False");
        TemporalChunkerAnnotator tca = new TemporalChunkerAnnotator(new ResourceManager(rmProps));

        Properties props = new Properties();
        props.setProperty( PipelineConfigurator.USE_POS.key, Configurator.TRUE );
        props.setProperty( PipelineConfigurator.USE_SENTENCE_PIPELINE.key, Configurator.TRUE );
        AnnotatorService pipeline = PipelineFactory.buildPipeline(new ResourceManager(props));
        String text = "Saturday morning";
        // If you don't specify DCT, we use today's date as default DCT
        tca.addDocumentCreationTime("2017-08-06");
        TextAnnotation ta = pipeline.createAnnotatedTextAnnotation("corpus", "id", text);
        tca.addView(ta);
        View temporalViews = ta.getView(ViewNames.TIMEX3);
        System.out.println(temporalViews);
    }
}

For developers, after you make any changes to this module, run ` mvn install -pl temporal-normalizer -am -DskipTests to recompile (make sure the current directory is under the main project -- cogcomp-nlp`).

Framework Overview

Illinois-time is divided into two parts: extractor and normalizer. Our extractor is a statistical model based on Illinois-chunker, a shallow parser. We train our chunker using TimeBank and AQUAINT newspaper dataset.

Our normalizer is a deterministic rule-based system. Another option is to use HeidelTime to normalize.

The whole module is tested on TempEval3 dataset, evaluated using an evaluation tool provided by TempEval3. You can run this benchmark script by doing sh scripts/benchmark.sh (make sure your CWD is cogcomp-nlp) to see the performance. You should see the following results (some unnecessary printed information is omitted, we only show the result using our extractor + normalizer here):

Using our extractor + our normalizer:

=== Timex Performance ===

Strict Match F1 P R
  79.35 89.91 71.01
Relaxed Match F1 P R
  83.4 94.5 74.64
Attribute F1 Value Type
  70.45 74.49

Models

When usingTemporalChunkerAnnotator, the models are loaded automatically from the directory specified in the Property TemporalChunkerConfigurator.MODEL_PATH

Thus, to use your own models, simply place them in this directory and they will be loaded; otherwise, the model version specified in this project’s pom.xml file will be loaded from the Maven repository and used.

Note : To use your own models, exclude the illinois-time-models artifact from the illinois-time dependency in your pom.xml

Training

For the training of our extraction model, please refer to Chunker, with training data substituted by temporal related text.

Performance Benchmark

Run:

cd temporal-normalizer
sh scripts/benchmark.sh

in terminal to get results printed on console.

We provide a TemporalNormalizerBenchmark for you to generated evaluated files. You can use the following options:

  1. -verbose, this is optional
  2. -useGoldChunk, optional, if not set, temporal chunker will be used
  3. -inputFolder , mandatory
  4. -outputFolder , mandatory

We use TempEval3 dataset for evaluation, you can download https://www.cs.york.ac.uk/semeval-2013/task1/index.php%3Fid=data.html.

Citation

Please cite the following paper if you used this package.

@inproceedings{ZhaoDoRo12,
    author = {Ran Zhao, Quang Do and Dan Roth},
    title = {A Robust Shallow Temporal Reasoning System},
    booktitle = {NAACL},
    month = {6},
    year = {2012},
    url = "http://cogcomp.org/papers/ZhaoDoRo12.pdf",
}