The demand to access large amounts of heterogeneous structured data is emerging as a trend for many users and applications. However, the effort involved in querying heterogeneous and distributed third-party databases can create major barriers for data consumers. At the core of this problem is the semantic gap between the way users express their information needs and the representation of the data. This work aims to provide a natural language interface and an associated semantic index to support an increased level of vocabulary independency for queries over Linked Data/Semantic Web datasets, using a distributional-compositional semantics approach. Distributional semantics focuses on the automatic construction of a semantic model based on the statistical distribution of co-occurring words in large-scale texts. The proposed query model targets the following features: (i) a principled semantic approximation approach with low adaptation effort (independent from manually created resources such as ontologies, thesauri or dictionaries), (ii) comprehensive semantic matching supported by the inclusion of large volumes of distributional (unstructured) commonsense knowledge into the semantic approximation process and (iii) expressive natural language queries. The approach is evaluated using natural language queries on an open domain dataset and achieved avg. recall=0.81, mean avg. precision=0.62 and mean reciprocal rank=0.49.

1. Natural Language Queries over
Heterogeneous Linked Data Graphs:
A Distributional-Compositional Semantics Approach
André Freitas and Edward Curry
Insight Centre for Data Analytics
International Conference on Intelligent User Interfaces
Haifa, 2014

2. Talking to your (Big) Data

3. Motivation

4. Shift in the Database Landscape
 Heterogeneous, complex and large-scale databases.
 Very-large and dynamic “schemas”.
circa 2014
circa 2000
10s-100s attributes
1,000s-1,000,000s attributes

5. Databases for a Complex World
How do you query data on this scenario?

6. Vocabulary Problem for Databases
Query: Who is the daughter of Bill Clinton married to?
Semantic Gap
Possible representations
Semantic approximation
= Commonsense Knowledge

7. Semantics for a Complex World
Formal World
Real World
Distributional Semantics
Query Approach

8. Does it work?

9. Addressing the Vocabulary Problem for
Databases (with Distributional Semantics)
Gaelic: direction

10. Solution (Video)

11. More Complex Queries (Video)

12. Treo Answers Jeopardy Queries (Video)
http://bit.ly/1hWcch9

13. Evaluation
 102 natural language queries (Test Collection: QALD 2011).
 Avg. query execution time: 1.52 s (simple queries) – 8.53 s
(all queries).
Dataset (DBpedia 3.7 + YAGO): 45,767 predicates,
5,556,492 classes and 9,434,677 instances

14. Comparative Evaluation

15. Query Approach

16. Distributional Semantics
“Words occurring in similar (linguistic) contexts are
semantically related.”
 If we can equate meaning with context, we can simply
record the contexts in which a word occurs in a
collection of texts (a corpus).
 This can then be used as a surrogate of its semantic
representation.

17. Distributional Semantic Model
function (number of times that the words occur in c1)
c1
0.7
0.5
husband
spouse
cn
c2
child
Commonsense is here

18. Semantic Relatedness
c1
husband
spouse
Works as a semantic ranking function
θ
cn
c2
child

19. Approach Overview
Query
Query Analysis
Query Features
Query Planner
Query Plan
Core semantic approximation &
composition operations
Ƭ-Space
Database
Distributional
semantics
Large-scale
unstructured data
Commonsense
knowledge

20. Approach Overview
Query
Query Analysis
Query Features
Query Planner
Query Plan
Core semantic approximation &
composition operations
Ƭ-Space
RDF Data
Explicit Semantic
Analysis (ESA)
Wikipedia
Commonsense
knowledge

21. Ƭ-Space
r
p
e

22. Core Operations
Query

23. Core Operations
Query
Search &
Composition
Operations

24. Search and Composition Operations

Instance search
- Proper nouns
- String similarity + node cardinality

Class (unary predicate) search
- Nouns, adjectives and adverbs
- String similarity + Distributional semantic relatedness

Property (binary predicate) search
- Nouns, adjectives, verbs and adverbs
- Distributional semantic relatedness

Navigation

Extensional expansion
- Expands the instances associated with a class.

Operator application
- Aggregations, conditionals, ordering, position


Disjunction & Conjunction
Disambiguation dialog (instance, predicate)

25. Core Principles
 Minimize the impact of Ambiguity, Vagueness, Synonymy.
 Address the simplest matchings first (heuristics).
 Semantic Relatedness as a primitive operation.
 Distributional semantics as commonsense knowledge.

26. Question Analysis
Transform natural language queries into triple
patterns
“Who is the daughter of Bill Clinton married to?”
Bill Clinton
daughter
married to
PODS
(INSTANCE)
(PREDICATE)
(PREDICATE)
Query Features

27. Query Plan
Map query features into a query plan.
A query plan contains a sequence of core operations.
(INSTANCE)
(PREDICATE)
(PREDICATE)

(1) INSTANCE SEARCH (Bill Clinton)

(2) p1 <- SEARCH PREDICATE (Bill Clintion, daughter)

(3) e1 <- NAVIGATE (Bill Clintion, p1)

(4) p2 <- SEARCH PREDICATE (e1, married to)

(5) e2 <- NAVIGATE (e1, p2)
Query Features
Query Plan

28. Instance Search
Query:
Bill Clinton
daughter
Instance Search
Linked
Data:
:Bill_Clinton
married to

29. Predicate Search
Query:
Linked
Data:
Bill Clinton
daughter
married to
:child
:Bill_Clinton
:Chelsea_Clinton
:religion
:Baptists
:almaMater
...
(PIVOT ENTITY)
:Yale_Law_School
(ASSOCIATED
TRIPLES)

30. Predicate Search
Query:
Bill Clinton
daughter
married to
Which properties are semantically related to „daughter‟?
Linked
Data:
:child
:Bill_Clinton
:Chelsea_Clinton
:religion
...
:Baptists
sem_rel(daughter,child)=0.054
sem_rel(daughter,child)=0.004
:almaMater
:Yale_Law_School
sem_rel(daughter,alma mater)=0.001

31. Navigate
Query:
Linked
Data:
Bill Clinton
daughter
:child
:Bill_Clinton
:Chelsea_Clinton
married to

32. Navigate
Query:
Linked
Data:
Bill Clinton
daughter
:child
:Bill_Clinton
:Chelsea_Clinton
(PIVOT ENTITY)
married to

33. Predicate Search
Query:
Linked
Data:
Bill Clinton
daughter
:spouse
:child
:Bill_Clinton
married to
:Chelsea_Clinton
(PIVOT ENTITY)
:Mark_Mezvinsky

34. Results

35. Conclusions
 The compositional-distributional model supports a schemaagnostic natural language query mechanism over a large
schema (open domain) database
 Comprehensive and accurate semantic matching
- Avg. recall=0.81, map=0.62, mrr=0.49
 Medium-high expressivity
- 80% of queries answered
 Interactive query execution time
- Avg. 1.52 s (simple queries) – 8.53 s (all queries) / query
 Better recall and query coverage compared to baselines with
equivalent precision
 Low adaptation effort for new datasets