1) adapt a number of codes in several classes as a consequence of Tag…

…gingTokenClusteror implementation in AcknowledgmentParser class; 2) add new feature vectors for improving acknowledgement model; 3) automatically split data (80/20)-train-evaluate a new acknowledgment model; 4) add new n-fold evaluation results. Old results are still kept for comparison purposes.
kermitt2 · Oct 25, 2019 · c85abd9 · c85abd9
1 parent 3e4171a
commit c85abd9
Show file tree

Hide file tree

Showing 19 changed files with 259,512 additions and 180,107 deletions.
diff --git a/grobid-core/src/main/java/org/grobid/core/engines/AcknowledgmentParser.java b/grobid-core/src/main/java/org/grobid/core/engines/AcknowledgmentParser.java
@@ -1,7 +1,7 @@
 package org.grobid.core.engines;
 
 /**
- * @created by Tanti
+ * @created by Tanti, 2019
  */
 
 import org.apache.commons.collections4.CollectionUtils;
@@ -55,12 +55,12 @@ public List<AcknowledgmentItem> processing(String input) {
     }
 
     public List<AcknowledgmentItem> processing(List<LayoutToken> tokens) {
-        List<AcknowledgmentItem> acknowledgments = null;
+        List<AcknowledgmentItem> acknowledgments = new ArrayList<>();
         if (CollectionUtils.isEmpty(tokens)) {
             return null;
         }
         try {
-            String headerAcknowledgment = FeaturesVectorAcknowledgment.addFeaturesAcknowledgment(tokens);
+            String headerAcknowledgment = FeaturesVectorAcknowledgment.addFeaturesAcknowledgment(tokens, null);
             String resAcknowledgment = label(headerAcknowledgment);
             acknowledgments = resultExtractionLayoutTokens(resAcknowledgment, tokens);
             return acknowledgments;

diff --git a/grobid-core/src/main/java/org/grobid/core/engines/label/TaggingLabels.java b/grobid-core/src/main/java/org/grobid/core/engines/label/TaggingLabels.java
@@ -90,7 +90,7 @@ public class TaggingLabels {
     public final static String GLOSSARY_LABEL = "<glossary>";
     public final static String BACK_LABEL = "<back>";
 
-    // acknowledgment
+    // for acknowledgment model
     public final static String AFFILIATION = "<affiliation>";
     public final static String EDUCATIONAL_INSTITUTION = "<educationalInstitution>";
     public final static String FUNDING_AGENCY = "<fundingAgency>";
@@ -356,7 +356,7 @@ protected static void register(TaggingLabel label) {
         register(MONOGRAPH_BACK);
         register(MONOGRAPH_OTHER);
 
-        // acknowledgment
+        // for acknowledgment model
         register(ACKNOWLEDGMENT_AFFILIATION);
         register(ACKNOWLEDGMENT_EDUCATIONAL_INSTITUTION);
         register(ACKNOWLEDGMENT_FUNDING_AGENCY);

diff --git a/grobid-core/src/main/java/org/grobid/core/features/FeaturesVectorAcknowledgment.java b/grobid-core/src/main/java/org/grobid/core/features/FeaturesVectorAcknowledgment.java
@@ -1,9 +1,6 @@
 package org.grobid.core.features;
 
-import org.grobid.core.analyzers.GrobidAnalyzer;
-import org.grobid.core.exceptions.GrobidException;
 import org.grobid.core.layout.LayoutToken;
-import org.grobid.core.utilities.OffsetPosition;
 import org.grobid.core.utilities.TextUtilities;
 import org.grobid.core.utilities.UnicodeUtil;
 
@@ -12,8 +9,9 @@
 import java.util.regex.Matcher;
 
 /**
- * Class for features used for parsing acknowledgment chunk.
- * Created by Tanti, 2019
+ * Class contains feature vectors for acknowledgment parser model.
+ *
+ * @Created by Tanti, 2019
  */
 
 public class FeaturesVectorAcknowledgment {
@@ -25,6 +23,8 @@ public class FeaturesVectorAcknowledgment {
     public boolean singleChar = false;
     public boolean properName = false;
     public boolean commonName = false;
+    public boolean locationName = false;
+    public boolean countryName = false;
     public boolean affiliation = false;
     public boolean educationalInsitution = false;
     public boolean fundingAgency = false;
@@ -90,6 +90,18 @@ public String printVector() {
         else
             res.append(" 0");
 
+        // location name
+        if (locationName)
+            res.append(" 1");
+        else
+            res.append(" 0");
+
+        // country name
+        if (countryName)
+            res.append(" 1");
+        else
+            res.append(" 0");
+
         // lexical information (9)
         if (affiliation)
             res.append(" 1");
@@ -139,6 +151,8 @@ public String printVector() {
         // punctuation information (1)
         res.append(" " + punctType); // in case the token is a punctuation (NO otherwise)
 
+        res.append(" ").append(wordShape);
+
         // label - for training data (1)
         if (label != null)
             res.append(" " + label + "\n");
@@ -149,7 +163,7 @@ public String printVector() {
     }
 
     /**
-     * Add feature for acknowledgment parsing.
+     * Add feature for acknowledgment parsing with string as input.
      */
     static public String addFeaturesAcknowledgmentString(List<String> lines) throws Exception {
 
@@ -285,19 +299,21 @@ static private FeaturesVectorAcknowledgment addFeaturesAcknowledgment(String lin
     /**
      * Add feature for acknowledgment parsing with tokens as input.
      */
-    static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws Exception {
+    static public String addFeaturesAcknowledgment(List<LayoutToken> tokens, List<String> labels) throws Exception {
         FeatureFactory featureFactory = FeatureFactory.getInstance();
 
         StringBuilder acknowledgment = new StringBuilder();
 
         String previousTag = null;
         String previousText = null;
         FeaturesVectorAcknowledgment featuresVectorAcknowledgment = null;
-        int sentenceLenth = tokens.size(); // length of the current sentence
-        for (int n=0; n < tokens.size(); n++) {
+        for (int n = 0; n < tokens.size(); n++) {
             LayoutToken token = tokens.get(n);
             String tag = null;
 
+            if ((labels != null) && (labels.size() > 0) && (n < labels.size()))
+                tag = labels.get(n);
+
             boolean outputLineStatus = false;
 
             String text = token.getText();
@@ -311,7 +327,7 @@ static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws
 
             // parano normalisation
             text = UnicodeUtil.normaliseTextAndRemoveSpaces(text);
-            if (text.trim().length() == 0 ) {
+            if (text.trim().length() == 0) {
                 continue;
             }
 
@@ -334,7 +350,7 @@ static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws
                     featuresVectorAcknowledgment.lineStatus = "LINESTART";
                     outputLineStatus = true;
                 }
-            } else if (tokens.size() == n+1) {
+            } else if (tokens.size() == n + 1) {
                 if (!outputLineStatus) {
                     featuresVectorAcknowledgment.lineStatus = "LINEEND";
                     outputLineStatus = true;
@@ -345,18 +361,13 @@ static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws
                 outputLineStatus = true;
             }
 
-            // single character
-            if (text.length() == 1) {
-                featuresVectorAcknowledgment.singleChar = true;
+            // capitalisation
+            if (Character.isUpperCase(text.charAt(0))) {
+                featuresVectorAcknowledgment.capitalisation = "INITCAP";
             }
 
-            // capital
             if (featureFactory.test_all_capital(text)) {
-                featuresVectorAcknowledgment.capitalisation = "ALLCAPS";
-            }
-
-            if (Character.isUpperCase(text.charAt(0))) {
-                featuresVectorAcknowledgment.capitalisation = "INITCAP";
+                featuresVectorAcknowledgment.capitalisation = "ALLCAP";
             }
 
             if (featuresVectorAcknowledgment.capitalisation == null) {
@@ -376,17 +387,32 @@ static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws
             if (featuresVectorAcknowledgment.digit == null)
                 featuresVectorAcknowledgment.digit = "NODIGIT";
 
-            // common name
-            if (featureFactory.test_common(text)) {
-                featuresVectorAcknowledgment.commonName = true;
+            // single character
+            if (text.length() == 1) {
+                featuresVectorAcknowledgment.singleChar = true;
             }
 
             // proper name
             if (featureFactory.test_names(text)) {
                 featuresVectorAcknowledgment.properName = true;
             }
 
-            // find the punctuations
+            // common name
+            if (featureFactory.test_common(text)) {
+                featuresVectorAcknowledgment.commonName = true;
+            }
+
+            // location name
+            if (featureFactory.test_city(text)) {
+                featuresVectorAcknowledgment.locationName = true;
+            }
+
+            // country name
+            if (featureFactory.test_country(text)) {
+                featuresVectorAcknowledgment.countryName = true;
+            }
+
+            // punctuations
             Matcher m0 = featureFactory.isPunct.matcher(text);
             if (m0.find()) {
                 featuresVectorAcknowledgment.punctType = "PUNCT";
@@ -395,7 +421,6 @@ static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws
             if (featuresVectorAcknowledgment.punctType == null)
                 featuresVectorAcknowledgment.punctType = "NOPUNCT";
 
-            // token containing special character
             if ((text.equals("(")) | (text.equals("["))) {
                 featuresVectorAcknowledgment.punctType = "OPENBRACKET";
             } else if ((text.equals(")")) | (text.equals("]"))) {
@@ -414,6 +439,7 @@ static public String addFeaturesAcknowledgment(List<LayoutToken> tokens) throws
 
             previousTag = tag;
             previousText = text;
+
         }
 
         return acknowledgment.toString();

diff --git a/grobid-core/src/test/java/org/grobid/core/engines/AcknowledgmentParserTest.java b/grobid-core/src/test/java/org/grobid/core/engines/AcknowledgmentParserTest.java
@@ -1,11 +1,9 @@
 package org.grobid.core.engines;
 
 import org.grobid.core.data.AcknowledgmentItem;
-import org.grobid.core.data.Acknowledgment;
 import org.grobid.core.factory.AbstractEngineFactory;
 import org.junit.Before;
 import org.junit.BeforeClass;
-import org.junit.Ignore;
 import org.junit.Test;
 
 import java.util.List;
@@ -19,8 +17,7 @@
  */
 
 public class AcknowledgmentParserTest {
-       AcknowledgmentParserOld targetOld;
-       AcknowledgmentParser targetNew;
+       AcknowledgmentParser target;
 
     @BeforeClass
     public static void setInitialContext() throws Exception {
@@ -29,47 +26,40 @@ public static void setInitialContext() throws Exception {
 
     @Before
     public void setUp() throws Exception {
-        targetOld = new AcknowledgmentParserOld();
-        targetNew = new AcknowledgmentParser();
-    }
-
-    @Ignore ("old method")
-    @Test
-    public void processing_akwnowledgmentOld_shouldWork() throws Exception {
-        List<Acknowledgment> output = targetOld.processing("This research was supported by the Deutsche Forschungsgemeinschaft through the SFB 649 \"Economic Risk\". http://sfb649.wiwi.hu-berlin.de ISSN 1860-5664");
-
-        final Acknowledgment acknowledgment = output.get(0);
-        assertThat(acknowledgment.getFundingAgency(), is("the Deutsche Forschungsgemeinschaft"));
-        assertThat(acknowledgment.getProjectName(), is("the SFB 649 \" Economic Risk \""));
+        target = new AcknowledgmentParser();
     }
 
     @Test
     public void processing_akwnowledgmentNew1_shouldWork() throws Exception {
-        List<AcknowledgmentItem> output = targetNew.processing("This research was supported by the Deutsche Forschungsgemeinschaft through the SFB 649 \"Economic Risk\". http://sfb649.wiwi.hu-berlin.de ISSN 1860-5664");
-        AcknowledgmentItem acknowledgment = new AcknowledgmentItem();
+        List<AcknowledgmentItem> output = target.processing("This research was supported by the Deutsche Forschungsgemeinschaft through the SFB 649 \"Economic Risk\". http://sfb649.wiwi.hu-berlin.de ISSN 1860-5664");
+
 
-        acknowledgment = output.get(0);
+        assertThat(output.get(0).getText(), is("the Deutsche Forschungsgemeinschaft"));
+        assertThat(output.get(0).getLabel(), is("fundingAgency"));
 
-        assertThat(acknowledgment.getText(), is("the Deutsche Forschungsgemeinschaft"));
-        assertThat(acknowledgment.getLabel(), is("fundingAgency"));
+        assertThat(output.get(1).getText(), is("the SFB 649 \"Economic Risk\""));
+        assertThat(output.get(1).getLabel(), is("projectName"));
     }
 
     @Test
     public void processing_akwnowledgmentNew2_shouldWork() throws Exception {
-        List<AcknowledgmentItem> output = targetNew.processing("This research was funded by Computational Science grant #635.000.014 from " +
+        List<AcknowledgmentItem> output = target.processing("This research was funded by Computational Science grant #635.000.014 from " +
             "the Netherlands Organization for Scientific Research (NWO). " +
             "Mikas Vengris, Denitsa Grancharova and Rienk van Grondelle provided the data modeled in Section 5.6. Rob Koehorst, " +
             "Bart van Oort, Sergey Laptenok, Ton Visser and Herbert van Amerongen provided the data modeled in Section 6.3. " +
             "Joris Snellenburg is thanked for constructive comments on the text. Uwe Ligges and Martin Mächler collaborated " +
             "in the implementation of the nls options described in Section B. Achim Zeileis contributed helpful suggestions regarding the figures.");
 
-        assertThat(output.get(1).getText(), is("the Netherlands Organization for Scientific Research (NWO)"));
-        assertThat(output.get(1).getLabel(), is("fundingAgency"));
+        assertThat(output.get(1).getText(), is("#635.000"));
+        assertThat(output.get(1).getLabel(), is("grantNumber"));
 
-        assertThat(output.get(2).getText(), is("Mikas Vengris"));
-        assertThat(output.get(2).getLabel(), is("individual"));
+        assertThat(output.get(2).getText(), is("the Netherlands Organization for Scientific Research (NWO)"));
+        assertThat(output.get(2).getLabel(), is("fundingAgency"));
 
-        assertThat(output.get(3).getText(), is("Denitsa Grancharova"));
+        assertThat(output.get(3).getText(), is("Mikas Vengris"));
         assertThat(output.get(3).getLabel(), is("individual"));
+
+        assertThat(output.get(4).getText(), is("Denitsa Grancharova"));
+        assertThat(output.get(4).getLabel(), is("individual"));
     }
 }
diff --git a/grobid-home/models/acknowledgment/config.json b/grobid-home/models/acknowledgment/config.json
@@ -1,22 +1,19 @@
 {
-    "model_name": "acknowledgment",
-    "model_type": "BidLSTM_CRF",
-    "embeddings_name": "glove-840B",
-    "char_vocab_size": 111,
-    "case_vocab_size": 8,
-    "char_embedding_size": 25,
-    "num_char_lstm_units": 25,
-    "max_char_length": 30,
-    "max_sequence_length": null,
-    "word_embedding_size": 300,
-    "num_word_lstm_units": 100,
-    "case_embedding_size": 5,
-    "dropout": 0.5,
-    "recurrent_dropout": 0.25,
-    "use_char_feature": true,
-    "use_crf": true,
-    "fold_number": 1,
-    "batch_size": 20,
-    "use_ELMo": false,
-    "use_BERT": false
+  "case_vocab_size": 8,
+  "model_name": "acknowledgment",
+  "num_char_lstm_units": 25,
+  "num_word_lstm_units": 100,
+  "case_embedding_size": 5,
+  "use_crf": true,
+  "fold_number": 1,
+  "char_vocab_size": 206,
+  "dropout": 0.5,
+  "model_type": "BidLSTM_CRF",
+  "batch_size": 20,
+  "recurrent_dropout": 0.25,
+  "max_char_length": 30,
+  "char_embedding_size": 25,
+  "use_char_feature": true,
+  "embeddings_name": "glove-840B",
+  "word_embedding_size": 300
 }