Negative keyword boosting (bst-mug#130)

- Adds experiment with negative keyword boosting - Updates KeywordExperimenter to test strings from a manually generated file. - As it improves NDCG from 0.7728 to 0.79, enables it by default. - This fixes bst-mug#130.
michelole · Jul 31, 2017 · df3e9ba · df3e9ba
1 parent 9ebf19e
commit df3e9ba
Show file tree

Hide file tree

Showing 5 changed files with 125 additions and 3 deletions.
diff --git a/src/main/java/at/medunigraz/imi/bst/trec/Experimenter.java b/src/main/java/at/medunigraz/imi/bst/trec/Experimenter.java
@@ -19,6 +19,8 @@ public static void main(String[] args) {
 		final File synonymTemplate = new File(RunnerDemo.class.getResource("/templates/synonym.json").getFile());
 		final File regexpDrugsTemplate = new File(
 				RunnerDemo.class.getResource("/templates/regexp-drugs.json").getFile());
+		final File negativeBoostKeywordsTemplate = new File(
+				RunnerDemo.class.getResource("/templates/negative-boost-keywords.json").getFile());
 		final Gene.Field[] expandTo = { Gene.Field.SYMBOL, Gene.Field.DESCRIPTION };
 
 		ExperimentsBuilder builder = new ExperimentsBuilder();
@@ -55,6 +57,9 @@ public static void main(String[] args) {
 
 		builder.newExperiment().withGoldStandard(Experiment.GoldStandard.FINAL).withTarget(Experiment.Task.PUBMED)
 				.withTemplate(regexpDrugsTemplate).withWordRemoval();
+
+		builder.newExperiment().withGoldStandard(Experiment.GoldStandard.FINAL).withTarget(Experiment.Task.PUBMED)
+				.withTemplate(negativeBoostKeywordsTemplate).withWordRemoval();
 
 		Set<Experiment> experiments = builder.build();
 

diff --git a/src/main/java/at/medunigraz/imi/bst/trec/KeywordExperimenter.java b/src/main/java/at/medunigraz/imi/bst/trec/KeywordExperimenter.java
@@ -14,7 +14,7 @@ public class KeywordExperimenter {
 	public static void main(String[] args) {
 		final File keywordTemplate = new File(
 				KeywordExperimenter.class.getResource("/templates/keyword.json").getFile());
-		final File keywordsSource = new File(KeywordExperimenter.class.getResource("/keywords/").getFile());
+		final File keywordsSource = new File(KeywordExperimenter.class.getResource("/negative-keywords/").getFile());
 
 		ExperimentsBuilder builder = new ExperimentsBuilder();
 

diff --git a/src/main/java/at/medunigraz/imi/bst/trec/RunnerDemo.java b/src/main/java/at/medunigraz/imi/bst/trec/RunnerDemo.java
@@ -8,7 +8,7 @@
 
 public class RunnerDemo {
 	public static void main(String[] args) {
-		final File pmTemplate = new File(RunnerDemo.class.getResource("/templates/regexp-drugs.json").getFile());
+		final File pmTemplate = new File(RunnerDemo.class.getResource("/templates/negative-boost-keywords.json").getFile());
 
 		final File ctTemplate = new File(RunnerDemo.class.getResource("/templates/improved-ct.json").getFile());
 

diff --git a/src/main/resources/templates/keyword.json b/src/main/resources/templates/keyword.json
@@ -79,7 +79,8 @@
 								"abstract"
 							],
 							"tie_breaker": 0.3,
-							"type": "best_fields"
+							"type": "best_fields",
+							"boost": -1
 						}
 					}
 				}

diff --git a/src/main/resources/templates/negative-boost-keywords.json b/src/main/resources/templates/negative-boost-keywords.json
@@ -0,0 +1,116 @@
+{
+	"bool": {
+		"must": [
+			{
+				"multi_match": {
+					"query": "{{disease}}",
+					"fields": [
+						"title^2",
+						"abstract",
+						"keyword",
+						"meshTags"
+					],
+					"tie_breaker": 0.3,
+					"type": "best_fields",
+					"boost": 1
+				}
+			},
+			{
+				"multi_match": {
+					"query": "{{gene}}",
+					"fields": [
+						"title^2",
+						"abstract",
+						"keyword",
+						"meshTags"
+					],
+					"tie_breaker": 0.3,
+					"type": "best_fields"
+				}
+			}
+		],
+		"should": [
+			{
+				"match": {
+					"_type": "extra"
+				}
+			},
+			{
+				"query_string": {
+					"query": "*mab *nib *cin *one *ate *mus *lin",
+					"fields": [
+								"title^2",
+								"abstract",
+								"keyword",
+								"meshTags"
+							]
+				}
+			},
+			{
+				"bool": {
+					"should": {
+						"multi_match": {
+							"query": "cancer carcinoma tumor",
+							"fields": [
+								"title^2",
+								"abstract",
+								"keyword",
+								"meshTags"
+							],
+							"tie_breaker": 0.3,
+							"type": "best_fields"
+						}
+					}
+				}
+			},
+			{
+				"bool": {
+					"should": {
+						"multi_match": {
+							"query": "gene genotype DNA base",
+							"fields": [
+								"title^2",
+								"abstract",
+								"keyword",
+								"meshTags"
+							],
+							"tie_breaker": 0.3,
+							"type": "best_fields"
+						}
+					}
+				}
+			},
+			{
+				"bool": {
+					"should": {
+						"multi_match": {
+							"query": "therapy treatment prognosis prognostic survival patient resistance recurrence targets malignancy study therapeutical outcome",
+							"fields": [
+								"title^2",
+								"abstract"
+							],
+							"tie_breaker": 0.3,
+							"type": "best_fields"
+						}
+					}
+				}
+			},
+			{
+				"bool": {
+					"should": {
+						"multi_match": {
+							"query": "transcript paraffin tumorigenesis embedded formalin fish tissue probes detection screening",
+							"fields": [
+								"title^2",
+								"abstract"
+							],
+							"tie_breaker": 0.3,
+							"type": "best_fields",
+							"boost": -1
+						}
+					}
+				}
+			}
+		]
+	}
+}