From c5ee9af442c500ec43fc28808903cfca2417ac12 Mon Sep 17 00:00:00 2001
From: Chris Kamphuis <mail@chriskamphuis.com>
Date: Thu, 29 Aug 2019 15:31:00 +0200
Subject: [PATCH] Add date filter to background linking reranker (#786)

---
 .../index/generator/WapoGenerator.java        |  3 +-
 .../lib/NewsBackgroundLinkingReranker.java    | 30 ++++++++++++++-----
 .../java/io/anserini/search/SearchArgs.java   |  6 +++-
 3 files changed, 30 insertions(+), 9 deletions(-)

diff --git a/src/main/java/io/anserini/index/generator/WapoGenerator.java b/src/main/java/io/anserini/index/generator/WapoGenerator.java
index aea88c89ce..8ad917049f 100644
--- a/src/main/java/io/anserini/index/generator/WapoGenerator.java
+++ b/src/main/java/io/anserini/index/generator/WapoGenerator.java
@@ -45,7 +45,7 @@ public class WapoGenerator extends LuceneDocumentGenerator<WashingtonPostCollect
   public static final String FIELD_RAW = "raw";
   public static final String FIELD_BODY = "contents";
   public static final String FIELD_ID = "id";
-  
+
   private static final String PATTERN = "<.+>";
   public static final List<String> CONTENT_TYPE_TAG = Arrays.asList("sanitized_html", "tweet");
 
@@ -86,6 +86,7 @@ public Document createDocument(WashingtonPostCollection.Document wapoDoc) {
     // This is needed to break score ties by docid.
     doc.add(new SortedDocValuesField(FIELD_ID, new BytesRef(id)));
     doc.add(new LongPoint(WapoField.PUBLISHED_DATE.name, wapoDoc.getPublishDate()));
+    doc.add(new StoredField(WapoField.PUBLISHED_DATE.name, wapoDoc.getPublishDate()));
     wapoDoc.getAuthor().ifPresent(author -> {
       doc.add(new StringField(WapoField.AUTHOR.name, author, Field.Store.NO));
     });
diff --git a/src/main/java/io/anserini/rerank/lib/NewsBackgroundLinkingReranker.java b/src/main/java/io/anserini/rerank/lib/NewsBackgroundLinkingReranker.java
index a34c5d26d9..bce520aff1 100644
--- a/src/main/java/io/anserini/rerank/lib/NewsBackgroundLinkingReranker.java
+++ b/src/main/java/io/anserini/rerank/lib/NewsBackgroundLinkingReranker.java
@@ -34,6 +34,7 @@
 
 import static io.anserini.index.generator.LuceneDocumentGenerator.FIELD_BODY;
 import static io.anserini.index.generator.LuceneDocumentGenerator.FIELD_ID;
+import static io.anserini.index.generator.WapoGenerator.WapoField.PUBLISHED_DATE;
 
 /*
 * TREC News Track Background Linking task postprocessing.
@@ -54,29 +55,44 @@ public ScoredDocuments rerank(ScoredDocuments docs, RerankerContext context) {
     }
     
     // remove the duplicates: 1. the same doc with the query doc 2. duplicated docs in the results
-    Set<Integer> duplicates = new HashSet<>();
+    Set<Integer> toRemove = new HashSet<>();
     for (int i = 0; i < docs.documents.length; i++) {
-      if (duplicates.contains(i)) continue;
-      String docid = docs.documents[i].getField(FIELD_ID).stringValue();
+      if (toRemove.contains(i)) continue;
       if (computeCosineSimilarity(queryTermsMap, docsVectorsMap.get(i)) >= 0.9) {
-        duplicates.add(i);
+        toRemove.add(i);
         continue;
       }
       for (int j = i+1; j < docs.documents.length; j++) {
         if (computeCosineSimilarity(docsVectorsMap.get(i), docsVectorsMap.get(j)) >= 0.9) {
-          duplicates.add(j);
+          toRemove.add(j);
+        }
+      }
+    }
+
+    if (context.getSearchArgs().backgroundlinking_datefilter) {
+      try {
+        int luceneId = NewsBackgroundLinkingTopicReader.convertDocidToLuceneDocid(reader, queryDocId);
+        Document queryDoc = reader.document(luceneId);
+        long queryDocDate = Long.parseLong(queryDoc.getField(PUBLISHED_DATE.name).stringValue());
+        for (int i = 0; i < docs.documents.length; i++) {
+          long date = Long.parseLong(docs.documents[i].getField(PUBLISHED_DATE.name).stringValue());
+          if (date > queryDocDate) {
+            toRemove.add(i);
+          }
         }
+      } catch (Exception e) {
+        e.printStackTrace();
       }
     }
   
     ScoredDocuments scoredDocs = new ScoredDocuments();
-    int resSize = docs.documents.length - duplicates.size();
+    int resSize = docs.documents.length - toRemove.size();
     scoredDocs.documents = new Document[resSize];
     scoredDocs.ids = new int[resSize];
     scoredDocs.scores = new float[resSize];
     int idx = 0;
     for (int i = 0; i < docs.documents.length; i++) {
-      if (!duplicates.contains(i)) {
+      if (!toRemove.contains(i)) {
         scoredDocs.documents[idx] = docs.documents[i];
         scoredDocs.scores[idx] = docs.scores[i];
         scoredDocs.ids[idx] = docs.ids[i];
diff --git a/src/main/java/io/anserini/search/SearchArgs.java b/src/main/java/io/anserini/search/SearchArgs.java
index 69f87723f7..773451f060 100644
--- a/src/main/java/io/anserini/search/SearchArgs.java
+++ b/src/main/java/io/anserini/search/SearchArgs.java
@@ -66,7 +66,11 @@ public class SearchArgs {
   @Option(name = "-backgroundlinking.weighted", usage = "Boolean switch to construct boosted query for TREC News Track Background " +
       "Linking task. The terms scores are their tf-idf score from the query document")
   public boolean backgroundlinking_weighted = false;
-  
+
+  @Option(name = "-backgroundlinking.datefilter", usage = "Boolean switch to filter out articles published after topic article " +
+      "for the TREC News Track Background Linking task.")
+  public boolean backgroundlinking_datefilter = false;
+
   @Option(name = "-stemmer", usage = "Stemmer: one of the following porter,krovetz,none. Default porter")
   public String stemmer = "porter";