0.25
0.5
0.75
1.25
1.5
1.75
2
An Algebraic Foundation for Knowledge Graph Construction
Best Research Nominee
Related categories
Presentation
An Algebraic Foundation for Knowledge Graph Construction00:00
Mapping languages describe KG construction from heterogeneous types of data sources00:13
Diverse options in terms of syntax and expressiveness00:37
Let’s say, you choose dedicated mapping languages for your KG construction tasks00:55
Dedicated mapping languages lack solid formal foundations (i.e. , no formal semantics)01:06
Lack of formal foundations is a problem, both for developers and users01:34
Lack of formal foundations is a problem, both for developers and users02:00
Contributions of our work02:18
Contributions of our work02:50
General mapping process by contemporary mapping engines02:59
Two kinds of concepts needed to be defined03:23
Data model of our algebra: a variation of the relational model03:45
Values: RDF terms and an error value 04:03
Mapping algebra introduces five operators04:14
Mapping algebra introduces five operators04:37
Two new operators defined for the mapping algebra04:39
Source04:56
An abstraction of data objects as a key ingredient05:16
An abstraction of data objects as a key ingredient05:25
An abstraction of data objects as a key ingredient05:45
Running example with a JSON data as input06:04
JSON input data with JSONPath as the query language06:25
Extract each records in the JSON data with JSONPath query q06:40
Sequence of JSON objects generated from the “records” array07:00
ℙ maps attributes (for the mapping relation) to query expressions07:15
Extraction of JSON fields “firstname” and “age” using parameter ℙ07:28
Type cast JSON values to data-typed RDF literals07:50
Mapping relation (A,I) is generated with A = { "name", "age" }08:08
After the definition of data model and source operator08:19
Two attributes with only RDF Literals are not enough to generate RDF triples/quads08:31
We need an operator to extend mapping relation with new attributes and derive new RDF terms08:46
We need an operator to extend mapping relation with new attributes and derive new RDF terms09:01
Extend09:12
Concatenate the name with a base IRI and save it in temp attribute09:21
Generate IRI using values in the existing tempattribute09:47
Also possible to add constant RDF terms and copy existing values from the mapping relation10:05
Special attributes s, p,and oreserved for generating the RDF dataset10:33
RDF dataset is generated in a similar way to a SPARQL CONSTRUCT with triple pattern ?s?p?o.10:52
We have captured the semantics of the whole mapping process11:04
Contributions of our work11:10
Steps taken to translate arbitrary RML documents to our mapping algebra11:24
Normalize arbitrary RML document11:33
Translates normalized RML document to a mapping plan composed with our mapping algebra11:39
An example normalized RML document11:43
Mapping operations of RML are translated to algebra operators11:48
Mapping operations of RML are translated to algebra operators11:59
Project only the special attributes s, p and o at the end12:05
Prototype implementation of the translation algorithm verified with official RML v1.1.2 test cases12:11
Contributions of our work12:32
Two groups of equivalences in the paper12:39
Projection pushing for potential reduction in the size of the intermediate mapping relations12:56
Depending on the operator applied on mapping relation r, the intermediate result will be different13:11
Our formal foundation13:47
An Algebraic Foundation for Knowledge Graph Construction14:29