Load and query the Microsoft Concept Graph in Neo4j https://concept.research.microsoft.com/Home/Introduction

ms-concepts-import.sh

function import_extract_first {
  echo "name:ID(Concept)" > concepts.txt
  cat data-concept-instance-relations.txt | cut -d $'\t' -f 1 | sort | uniq >> concepts.txt
  
  echo "name:ID(Instance)" > instances.txt
  cat data-concept-instance-relations.txt | cut -d $'\t' -f 2 | sort | uniq >> instances.txt

  echo $':END_ID(Concept)\t:START_ID(Instance)	relations:int' > is_a.hdr
  
  $NEO4J_HOME/bin/neo4j-import --into concepts.db --id-type string --delimiter TAB --bad-tolerance 13000000 --skip-duplicate-nodes true --skip-bad-relationships true \
  --nodes:Concept concepts.txt \
  --nodes:Instance instances.txt \
  --relationships:IS_A is_a.hdr,data-concept-instance-relations.txt
}

function import_skip_duplicates {
  echo $'name:ID(Concept)\t:IGNORE\t:IGNORE' > concept.hdr
  echo $':IGNORE	name:ID(Instance)\t:IGNORE' > instance.hdr
  echo $':END_ID(Concept)\t:START_ID(Instance)	relations:int' > is_a.hdr
  
  $NEO4J_HOME/bin/neo4j-import --into concepts.db --id-type string --delimiter TAB --bad-tolerance 13000000 --skip-duplicate-nodes true --skip-bad-relationships true \
  --nodes:Concept concept.hdr,data-concept-instance-relations.txt \
  --nodes:Instance instance.hdr,data-concept-instance-relations.txt \
  --relationships:IS_A is_a.hdr,data-concept-instance-relations.txt
}

if [ ! -f data-concept-instance-relations.txt ]; do
  echo Download & Unzip from here: https://concept.research.microsoft.com/Home/Download
  exit 1
#  curl -IL -o data-concept.zip 'https://concept.research.microsoft.com/Home/DownloadData?key=S54IIWamuQc9YD1WQA6tzKULMMZ79xK6&h=2127477893' 
#  unzip -j data-concept.zip
fi

export NEO4J_HOME=${NEO4J_HOME-/data/versions/neo4j-enterprise-3.0.7}

import_extract_first

echo $'
CREATE CONSTRAINT ON (i:Instance) ASSERT i.name IS  UNIQUE;\n
CREATE CONSTRAINT ON (c:Concept)  ASSERT c.name IS  UNIQUE;' | $NEO4J_HOME/bin/neo4j-shell -path concepts.db

echo << EOF

IMPORT DONE in 1m 27s 888ms.
Imported:
  17878053 nodes
  33377320 relationships
  51255373 properties
Peak memory usage: 410.36 MB

cypher runtime=compiled profile
MATCH (i:Instance {name:"apple"})-[r:IS_A]->(c:Concept)
RETURN i.name, r.relations, c.name ORDER BY r.relations DESC LIMIT 10;

MATCH (i:Instance {name:"apple"})-[r:IS_A]->(c:Concept)
> RETURN i.name, r.relations, c.name ORDER BY r.relations DESC LIMIT 10;
+----------------------------------------+
| i.name  | r.relations | c.name         |
+----------------------------------------+
| "apple" | 6315        | "fruit"        |
| "apple" | 4353        | "company"      |
| "apple" | 1152        | "food"         |
| "apple" | 764         | "brand"        |
| "apple" | 750         | "fresh fruit"  |
| "apple" | 568         | "fruit tree"   |
| "apple" | 483         | "crop"         |
| "apple" | 280         | "corporation"  |
| "apple" | 279         | "manufacturer" |
| "apple" | 257         | "firm"         |
+----------------------------------------+
10 rows
20 ms

explain
MATCH (a:Instance {name:"apple"})-[:IS_A]->(c:Concept)<-[:IS_A]-(b:Instance {name:"pie"})
USING INDEX a:Instance(name)
USING INDEX b:Instance(name)
MATCH (c)<-[:IS_A]-(o) WHERE o <> a and o <> b
WITH o, count(*) as freq order by freq desc limit 10
RETURN o.name, freq;

export a_name="apple"
export b_name="pie"
export b_name="ipad"
MATCH (a:Instance {name:{a_name}})-[:IS_A]->(c:Concept)<-[:IS_A]-(b:Instance {name:{b_name}})
USING INDEX a:Instance(name)
USING INDEX b:Instance(name)
MATCH (c)<-[:IS_A]-(o)
WITH o, count(*) as freq order by freq desc SKIP 2 limit 10
RETURN o.name, freq;


export a_name="apple"
export b_name="pie"
# export b_name="ipad"
cypher runtime=compiled profile
MATCH (a:Instance {name:{a_name}})-[r1:IS_A]->(c:Concept)<-[r2:IS_A]-(b:Instance {name:{b_name}})
USING INDEX a:Instance(name)
USING INDEX b:Instance(name)
RETURN c.name, r1.relations+r2.relations as freq order by freq desc limit 10;


+--------------------------+
| c.name            | freq |
+--------------------------+
| "device"          | 1139 |
| "mobile device"   | 998  |
| "brand"           | 772  |
| "item"            | 396  |
| "product"         | 320  |
| "player"          | 201  |
| "technology"      | 191  |
| "apple product"   | 182  |
| "client"          | 147  |
| "portable device" | 139  |
+--------------------------+
10 rows
16 ms



export a_name="apple"
export b_name="pie"
# export b_name="ipad"
MATCH (a:Instance {name:{a_name}})-[r1:IS_A]->(c:Concept)<-[r2:IS_A]-(b:Instance {name:{b_name}})
USING INDEX a:Instance(name)
USING INDEX b:Instance(name)
RETURN c.name, r1.relations+r2.relations as freq order by freq desc limit 10;


+----------------------+
| c.name        | freq |
+----------------------+
| "fruit"       | 6316 |
| "food"        | 1408 |
| "item"        | 345  |
| "product"     | 268  |
| "dessert"     | 259  |
| "flavor"      | 221  |
| "baked goods" | 209  |
| "ingredient"  | 184  |
| "business"    | 144  |
| "snack"       | 144  |
+----------------------+
10 rows
15 ms


MATCH (a:Instance {name:"apple"})-[:IS_A]->(c:Concept)<-[:IS_A]-(b:Instance {name:"pie"})
USING INDEX a:Instance(name)
USING INDEX b:Instance(name)
RETURN count(*);

explain
with "the apple engineer is eating the apple" as sentence
unwind split(sentence," ") as word
match (i:Instance {name:word})
with collect(i) as instances
unwind range(0,length(instances)-2) as idx
with idx, instances[idx] as a, instances[idx+1] as b
MATCH (a)-[:IS_A]->(c:Concept)<-[:IS_A]-(b)
MATCH (c)<-[:IS_A]-(o) WHERE o <> a AND o <> b
WITH idx, a, b, o, count(*) as freq order by freq desc
RETURN idx, a.name, b.name, collect(o.name)[0..5] as meanings order by idx;


+-----------------------------------------------------------------------------------------------------------------------+
| idx | a.name     | b.name     | meanings                                                                              |
+-----------------------------------------------------------------------------------------------------------------------+
| 0   | "the"      | "apple"    | ["vehicle","light","money","tobacco","television"]                                    |
| 1   | "apple"    | "engineer" | ["bank","university","doctor","school","google"]                                      |
| 2   | "engineer" | "is"       | ["coos","fructose","armour","starseeds","centaur"]                                    |
| 3   | "is"       | "eating"   | ["compensation","ukuyigxoba","next","offside","process learning"]                     |
| 4   | "eating"   | "the"      | ["effective honest","crushed leg","drogue parachutes","marine sergeant","nano sized"] |
| 5   | "the"      | "apple"    | ["family","sandwich","vehicle","poison","door"]                                       |
+-----------------------------------------------------------------------------------------------------------------------+

with "the apple engineer is eating the apple" as sentence
unwind split(sentence," ") as word
match (i:Instance {name:word})
WHERE NOT (i)-[:IS_A]->(:Concept {name:"stop words"})
with collect(i) as instances
unwind range(0,length(instances)-2) as idx
with idx, instances[idx] as a, instances[idx+1] as b
MATCH (a)-[r1:IS_A]->(c:Concept)<-[r2:IS_A]-(b)
WITH idx, a, b, c, r1.relations+r2.relations as freq order by freq desc
RETURN idx, a.name, b.name, collect({concept:c.name, relations:freq})[0..5] as concepts order by idx;


+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| idx | a.name     | b.name     | concepts                                                                                                                                                                                                     |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| 0   | "apple"    | "engineer" | [{concept -> "company", relations -> 4363},{concept -> "firm", relations -> 260},{concept -> "product", relations -> 182},{concept -> "person", relations -> 172},{concept -> "vendor", relations -> 169}]   |
| 1   | "engineer" | "is"       | [{concept -> "word", relations -> 25}]                                                                                                                                                                       |
| 2   | "is"       | "eating"   | [{concept -> "word", relations -> 4}]                                                                                                                                                                        |
| 3   | "eating"   | "apple"    | [{concept -> "fruit", relations -> 6316},{concept -> "activity", relations -> 736},{concept -> "item", relations -> 253},{concept -> "product", relations -> 182},{concept -> "business", relations -> 144}] |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+


with "the apple engineer is eating the apple" as sentence
unwind split(sentence," ") as word
match (i:Instance {name:word})
MATCH (i)-[r1:IS_A]->(c:Concept {name:"stop words"})
return i,c limit 10;

EOF

Michael,

The "the apple engineer is eating the apple" example performs well but it's a lucky test case. Try removing "is" from the phrase and the query doesn't complete. It appears to thrash or something but never returns. You can remove the "the"s as well as the "is" because "the" is an instance of the concept "stop words" i.e. it gets removed anyway. For example, "apple engineer eating apple"

The issue is with the 3rd MATCH clause (where "o" is unbounded). I'm guessing there is some sort of relationship data explosion happening for one of the 3 instance pairs: "apple engineer", "engineer eating" or "eating apple".

Any ideas for a better formulation of this query?

Best regards,
Michael Herman (Toronto)

with "apple engineer eating apple" as sentence
unwind split(sentence," ") as word
MATCH (i:Instance {name:word})
with collect(i) as instances
unwind range(0,length(instances)-2) as idx
with idx, instances[idx] as a, instances[idx+1] as b
MATCH (a)-[:IS_A]->(c:Concept)<-[:IS_A]-(b)
MATCH (c)<-[:IS_A]-(o) WHERE o <> a AND o <> b
WITH idx, a, b, o, count(*) as freq order by freq desc
RETURN idx, a.name, b.name, collect(o.name)[0..5] as meanings order by idx;

When I set the starting phrase, one at a time, to each of the 3 instance pairs: "apple engineer", "engineer eating" and "eating apple", each one, individually runs fine. ...1 minute or less ...usually about half a minute.