Initial commit for the Dataless Classifier (closes #556)

README.md

@@ -33,6 +33,7 @@ Each library contains detailed readme and instructions on how to use it. In addi
 | [commasrl](commasrl/README.md) | This software extracts relations that commas participate in. |
 | [similarity](similarity/README.md) | This software compare objects --especially Strings-- and return a score indicating how similar they are. |
 | [temporal-normalizer](temporal-normalizer/README.md) | A temporal extractor and normalizer.  |
+| [dataless-classifier](dataless-classifier/README.md) | Classifies text into a user-specified label hierarchy from just the textual label descriptions |
 | [external-annotators](external/README.md) | A collection useful external annotators.  |
 
 

core-utilities/src/main/java/edu/illinois/cs/cogcomp/core/datastructures/ViewNames.java

@@ -88,6 +88,9 @@ public class ViewNames {
 
     public static final String WIKIFIER = "WIKIFIER";
 
+    public static final String DATALESS_ESA = "DATALESS_ESA";
+    public static final String DATALESS_W2V = "DATALESS_W2V";
+
     /**
      * @deprecated Replaced by {@link #CLAUSES_CHARNIAK}, {@link #CLAUSES_BERKELEY},
      *             {@link #CLAUSES_STANFORD}
@@ -150,6 +153,8 @@ public static ViewTypes getViewType(String viewName) {
             case SHALLOW_PARSE:
             case QUANTITIES:
             case WIKIFIER:
+            case DATALESS_ESA:
+            case DATALESS_W2V:
             case CLAUSES_CHARNIAK:
             case CLAUSES_STANFORD:
             case CLAUSES_BERKELEY:

dataless-classifier/README.md

@@ -0,0 +1,45 @@
+# CogComp-DatalessClassifier
+Given a label ontology, and textual descriptions of those labels, Dataless-Classifier is capable of classifying arbitrary text into that ontology.
+
+It is particularly useful in those scenarios where it is difficult/expensive to gather enough training data to train a supervised text classifier. Dataless-Classifier utilizes the semantic meaning of the labels to bypass the need for explicit supervision. For more information, please visit our main project [page](http://cogcomp.org/page/project_view/6).    
+
+
+Some key points:
+- The Main classes for the Dataless Annotators are:
+  * **ESADatalessAnnotator** for the ESA-based Dataless Annotator
+  * **W2VDatalessAnnotator** for the Word2Vec-based Dataless Annotator
+- Dataless Annotators add the **DATALESS_ESA** and **DATALESS_W2V** views to the input `TextAnnotation` respectively, and it requires the presence of a **TOKENS** view with the end-user's desired Tokenization.
+- Since Labels/Topics are inferred at the Document-Level, all topic annotations span the entire document.
+- Sample invocation has been provided in the main functions of each annotator.
+- Both annotators load up embeddings in memory, and thus can easily consume upto **10GB RAM**.
+
+
+## Label Hierarchy
+Dataless Classification requires the end-user to specifcy a Label hierarchy (with label descriptions), which it classifies into. The Label hierarchy needs to be provided using a very specific format:
+* **labelNamePath**: Specify your label id to label name mapping here in the `labelID \t labelName` format 
+  (label id can be any ID specific to your system, however we use the label name itself as ID in our sample hierachy for readibility)
+* **labelHierarchyPath**: The first line of this file should contain tab-separated list of Top-Level nodes in the hierarchy (i.e. the ones directly connected to the root). Then, every following line should specify the connections in the hierachy in the `parentLabelID \t childLabelID1 \t childLabelID2 \t ...` format.
+* **labelDescPath**: Dataless' performance hinges on good label descriptions, which you specify in this file in the `labelID \t labelDescription` format.
+
+We provide a sample 20newsgroups hierarchy with label descriptions inside data/hierarchy/20newsgroups, where:
+* idToLabelNameMap.txt should be used as labelNamePath
+* parentChildIdMap.txt should be used as labelHierarchyPath
+* labelDesc\_Kws\_simple.txt should be used as labelDescPath
+
+We also provide improved 20newsgroups label descriptions in *labelDesc\_Kws\_embellished.txt* which corresponds to the label descriptions used in [2], whereas the *labelDesc\_Kws\_simple.txt* corresponds to the label descriptions used in [1].
+
+## Embeddings
+ESA and Word2Vec Embeddings are fetched from the DataStore on demand.
+
+## Config
+A sample config file with the default values has been provided in the config folder .. *config/project.properties*
+
+To check whether you are properly set to use the project or not, run:
+* `mvn -Dtest=ESADatalessTest#testPredictions test` to test the ESADatalessAnnotator.
+* `mvn -Dtest=W2VDatalessTest#testPredictions test` to test the W2VDatalessAnnotator.
+
+If you use this software for research, please cite the following papers:
+
+[1] Chang, Ming-Wei, et al. "Importance of Semantic Representation: Dataless Classification." AAAI. Vol. 2. 2008.
+
+[2] Song, Yangqiu, and Dan Roth. "On Dataless Hierarchical Text Classification." AAAI. Vol. 7. 2014.

dataless-classifier/config/project.properties

@@ -0,0 +1,25 @@
+## Use ResourceManager to read these properties
+# curatorHost = trollope.cs.illinois.edu
+# curatorPort = 9010
+
+## Target Label Hierarchy 
+labelHierarchyPath = data/hierarchies/20newsgroups/parentChildIdMap.txt
+labelNamePath = data/hierarchies/20newsgroups/idToLabelNameMap.txt
+labelDescPath = data/hierarchies/20newsgroups/labelDesc_Kws_simple.txt
+# labelDescPath = data/hierarchies/20newsgroups/labelDesc_Kws_embellished.txt
+
+## Classifier configuration    
+inferenceBottomUp = True
+classifierThreshold = 0.99
+classifierLeastK = 1
+classifierMaxK = 3
+
+## ESA Configuration
+#esaPath = data/embeddings/esaEmbedding/esa_vectors.txt
+#esaMapPath = data/embeddings/esaEmbedding/idToConceptMap.txt
+#esaDimension = 100
+
+## W2V Configuration
+#w2vPath = data/embeddings/w2vEmbedding-100/w2v_vectors.txt
+#w2vDimension = 200
+

dataless-classifier/data/electronicsTestDocument.txt

@@ -0,0 +1 @@
+yes i know it s nowhere near christmas time but i m gonna loose net access in a few days maybe a week or if i m lucky and wanted to post this for interested people to save till xmas note bell labs is a good place if you have a phd and a good boss i have neither subject xmas light set with levels of brightness another version of a variable brightness xmas light set this set starts with a blinker bulb string diagram orginal way set 0v b b 0rtn modified set for level brightness string 0v 0k w string b 0v rtn note no mods to wiring to the right of this point only one blinker is used note that the blinker would not have as much current thru it as the string bulbs because of the second string of bulbs in parallel with it that s why the use of the 0k w resistor here to add extra current thru the blinker to make up for the current shunted thru the second string while the blinker is glowing and the second string is not glowing when the blinker goes open this resistor has only a slight effect on the brightness of the strings s slightly dimmer s slightly brighter or use a w 0v bulb in place of the 0k resistor if you can get one caution do not replace with a standard c bulb as these draw too much current and burn out the blinker c approx w what you ll see when it s working powerup string will light at full brightness and b will be lit bypassing most of the current from the second string making them not light b will open placing both strings in series making the string that was out to glow at a low brightness and the other string that was on before to glow at reduced brightness be sure to wire and insulate the splices resistor leads and cut wires in a safe manner level brightness xmas light set for easter

dataless-classifier/data/graphicsTestDocument.txt

@@ -0,0 +1 @@
+i m looking for some recommendations for screen capture programs a couple of issues ago pc mag listed as editor s choices both conversion artist and hijaak for windows anyone have any experience with those or some others i m trying to get an alpha manual in the next few days and i m not making much progress with the screen shots i m currently using dodot and i m about to burn it and the disks it rode it on it s got a lot of freaky bugs and oversights that are driving me crazy tonight it decided that for any graphic it writes out as a tiff file that s under a certain arbitrary size it will swap the left and right sides of the picture usually it confines itself to not copying things to the clipboard so i have to save and load pix for editing in paintbrush or crashing every hour or so the one nice thing it has though is it s dither option you d think that this would turn colors into dots which it does if you go from say colors to colors but if you go from or colors to b w you can set a threshold level for which colors turn to black and which turn to white for me this is useful because i can turn light grays on buttons to white and the dark grays to black and thereby preserve the d effect on buttons and other parts of the window if you understood my description can you tell me if another less buggy program can do this as well much thanks for any help signature david delgreco what lies behind us and what lies technically a writer before us are tiny matters compared delgreco rahul net to what lies within us oliver wendell holmes david f delgreco delgreco rahul net recommendation for screen capture program

dataless-classifier/data/hierarchies/20newsgroups/idToLabelNameMap.txt

@@ -0,0 +1,26 @@
+politics	politics
+religion	religion
+computer	computer
+autos.sports	autos.sports
+science	science
+sales	sales
+talk.politics.guns	talk.politics.guns
+talk.politics.mideast	talk.politics.mideast
+talk.politics.misc	talk.politics.misc
+alt.atheism	alt.atheism
+soc.religion.christian	soc.religion.christian
+talk.religion.misc	talk.religion.misc
+comp.sys.ibm.pc.hardware	comp.sys.ibm.pc.hardware
+comp.sys.mac.hardware	comp.sys.mac.hardware
+comp.graphics	comp.graphics
+comp.windows.x	comp.windows.x
+comp.os.ms.windows.misc	comp.os.ms.windows.misc
+rec.autos	rec.autos
+rec.motorcycles	rec.motorcycles
+rec.sport.baseball	rec.sport.baseball
+rec.sport.hockey	rec.sport.hockey
+sci.electronics	sci.electronics
+sci.crypt	sci.crypt
+sci.med	sci.med
+sci.space	sci.space
+misc.forsale	misc.forsale

dataless-classifier/data/hierarchies/20newsgroups/labelDesc_Kws_embellished.txt

@@ -0,0 +1,26 @@
+politics	politics gun fbi guns weapon compound israel arab jews jewish muslim gay homosexual sexual
+religion	religion atheist christian atheism god islamic christian god christ church bible jesus christian morality jesus god religion horus
+computer	computer bus pc motherboard bios board computer dos mac apple powerbook graphics image gif animation tiff window motif xterm sun windows windows dos microsoft ms driver drivers card printer
+autos.sports	autos.sports car ford auto toyota honda nissan bmw bike motorcycle yamaha baseball ball hitter hockey wings espn
+science	science circuit electronics radio signal battery encryption key crypto algorithm security doctor medical disease medicine patient space orbit moon earth sky solar
+sales	sales sale offer shipping forsale sell price brand obo
+talk.politics.guns	gun fbi guns weapon compound
+talk.politics.mideast	israel arab jews jewish muslim
+talk.politics.misc	gay homosexual sexual
+alt.atheism	atheist christian atheism god islamic
+soc.religion.christian	christian god christ church bible jesus
+talk.religion.misc	christian morality jesus god religion horus
+comp.sys.ibm.pc.hardware	bus pc motherboard bios board computer dos
+comp.sys.mac.hardware	mac apple powerbook
+comp.graphics	graphics image gif animation tiff
+comp.windows.x	window motif xterm sun windows
+comp.os.ms.windows.misc	windows dos microsoft ms driver drivers card printer
+rec.autos	car ford auto toyota honda nissan bmw
+rec.motorcycles	bike motorcycle yamaha
+rec.sport.baseball	baseball ball hitter
+rec.sport.hockey	hockey wings espn
+sci.electronics	circuit electronics radio signal battery
+sci.crypt	encryption key crypto algorithm security
+sci.med	doctor medical disease medicine patient
+sci.space	space orbit moon earth sky solar
+misc.forsale	sale offer shipping forsale sell price brand obo

dataless-classifier/data/hierarchies/20newsgroups/labelDesc_Kws_simple.txt

@@ -0,0 +1,26 @@
+politics	politics politics guns politics mideast politics
+religion	religion atheism society religion christianity christian religion
+computer	computer computer systems ibm pc hardware computer systems mac macintosh apple hardware computer graphics computer windows x windowsx computer os operating system microsoft windows
+autos.sports	autos.sports cars motorcycles baseball hockey
+science	science science electronics science cryptography medicine science space
+sales	sales for sale discount
+talk.politics.guns	politics guns
+talk.politics.mideast	politics mideast
+talk.politics.misc	politics
+alt.atheism	atheism
+soc.religion.christian	society religion christianity christian
+talk.religion.misc	religion
+comp.sys.ibm.pc.hardware	computer systems ibm pc hardware
+comp.sys.mac.hardware	computer systems mac macintosh apple hardware
+comp.graphics	computer graphics
+comp.windows.x	computer windows x windowsx
+comp.os.ms.windows.misc	computer os operating system microsoft windows
+rec.autos	cars
+rec.motorcycles	motorcycles
+rec.sport.baseball	baseball
+rec.sport.hockey	hockey
+sci.electronics	science electronics
+sci.crypt	science cryptography
+sci.med	science medicine
+sci.space	science space
+misc.forsale	for sale discount

dataless-classifier/data/hierarchies/20newsgroups/parentChildIdMap.txt

@@ -0,0 +1,7 @@
+politics	religion	computer	autos.sports	science	sales
+politics	talk.politics.guns	talk.politics.mideast	talk.politics.misc
+religion	alt.atheism	soc.religion.christian	talk.religion.misc
+computer	comp.sys.ibm.pc.hardware	comp.sys.mac.hardware	comp.graphics	comp.windows.x	comp.os.ms.windows.misc
+autos.sports	rec.autos	rec.motorcycles	rec.sport.baseball	rec.sport.hockey
+science	sci.electronics	sci.crypt	sci.med	sci.space
+sales	misc.forsale

dataless-classifier/pom.xml

@@ -0,0 +1,71 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project xmlns="http://maven.apache.org/POM/4.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/maven-v4_0_0.xsd">
+    <parent>
+        <artifactId>illinois-cogcomp-nlp</artifactId>
+        <groupId>edu.illinois.cs.cogcomp</groupId>
+        <version>4.0.0</version>
+    </parent>
+
+    <modelVersion>4.0.0</modelVersion>
+
+    <artifactId>illinois-datalessclassification</artifactId>
+    <name>Illinois Dataless Classifier</name>
+    <description>Classifies Text into the given label hierarchy from just the textual label descriptions</description>
+
+    <dependencies>
+		<dependency>
+			<groupId>org.cogcomp</groupId>
+			<artifactId>cogcomp-datastore</artifactId>
+			<version>1.9.10</version>
+		</dependency>
+    	<dependency>
+      		<groupId>edu.illinois.cs.cogcomp</groupId>
+      		<artifactId>illinois-core-utilities</artifactId>
+      		<version>4.0.0</version>
+      	</dependency>
+      	<dependency>
+      		<groupId>edu.illinois.cs.cogcomp</groupId>
+      		<artifactId>illinois-tokenizer</artifactId>
+      		<version>4.0.0</version>
+      	</dependency>
+		<dependency>
+			<groupId>org.slf4j</groupId>
+			<artifactId>slf4j-log4j12</artifactId>
+			<version>1.7.12</version>
+			<optional>true</optional>
+		</dependency>
+      	<dependency>
+      		<groupId>net.sf.jung</groupId>
+      		<artifactId>jung-api</artifactId>
+      		<version>2.0.1</version>
+      	</dependency>
+      	<dependency>
+      		<groupId>net.sf.jung</groupId>
+      		<artifactId>jung-graph-impl</artifactId>
+      		<version>2.0.1</version>
+      	</dependency>
+      	<dependency>
+      		<groupId>commons-cli</groupId>
+      		<artifactId>commons-cli</artifactId>
+      		<version>1.4</version>
+      	</dependency>	
+    </dependencies>
+
+    <build>
+        <pluginManagement>
+            <plugins>
+                <plugin>
+                    <groupId>org.apache.maven.plugins</groupId>
+                    <artifactId>maven-surefire-plugin</artifactId>
+                    <version>2.20.1</version>
+                    <configuration>
+                        <!--increase the memory requirements if you need more space-->
+                        <argLine>-Xmx15g</argLine>
+                    </configuration>
+                </plugin>
+            </plugins>
+        </pluginManagement>
+    </build>
+
+
+</project>

dataless-classifier/script/testESADataless.sh

@@ -0,0 +1,3 @@
+#mvn compile
+#mvn dependency:copy-dependencies
+nice java -Xmx10g -cp ./target/*:./target/dependency/* edu.illinois.cs.cogcomp.datalessclassification.ta.ESADatalessAnnotator $@

dataless-classifier/script/testW2VDataless.sh

@@ -0,0 +1,3 @@
+#mvn compile
+#mvn dependency:copy-dependencies
+nice java -Xmx10g -cp ./target/*:./target/dependency/* edu.illinois.cs.cogcomp.datalessclassification.ta.W2VDatalessAnnotator $@