Knowledge Graphs

1. 00:00This is knowledge graphs lecture number five knowledge graph applications
2. 00:05and we start this first part of the lecture with the last chapter
3. 00:10on ontologies and now we are going for ontologies in action.
4. 00:16What does that mean? So usually when we are talking about ontologies
5. 00:19we can distinguish different types and categories of ontologies.
6. 00:24There are on one hand the so called top level ontologies that
7. 00:28try to explain the entire world and put the entire world into top-level categories.
8. 00:34For that reason they are also referred to as upper ontology or
9. 00:38foundation ontology. Then we have domain ontologies if it comes
10. 00:43only lets say to describing the knowledge of a specific domain,
11. 00:47we have task ontologies for task, and then we have application ontologies.
12. 00:52Let's have a closer look into each one.
13. 00:56Again so we start with the upper ontology.
14. 01:00These kind of ontologies usually are rather general. They are
15. 01:04cross domain ontologies trying to describe the entire world
16. 01:09and they represent very general concepts, for example time, space
17. 01:14or what is an event. And as you might remember and this also
18. 01:19reflects the example that you see here on the picture, this
19. 01:23by the way is the so called Porphyrian tree of knowledge which
20. 01:27is based on the ten Aristotlian categories. You remember
21. 01:31aristotle when he was dealing with metaphysics he tried to
22. 01:36create a system of categories in which each single thing in the world
23. 01:41could be sorted into. And exactly this tree of knowledge here
24. 01:46is kind of a let's say decision algorithm that was connected
25. 01:50to that kind of ontology. But if I want to describe you know
26. 01:53the world in ten categories of course these categories have to be
27. 01:57rather general. So all of the ontologies that try to model the entire world
28. 02:03they are top-level ontologies and they are rather generic, which
29. 02:08means they are independent of a specific domain or of a specific problem
30. 02:13that should be solved.
31. 02:15If I want to solve a specific problem in a specific domain
32. 02:20then I come closer to so called domain ontologies.
33. 02:24And they are usually fundamental concepts according to a generic domain
34. 02:29are explained and are modeled. So if you connect this to the
35. 02:35upper ontology what you do there is you specialize the terms
36. 02:38that have been introduced in the top level ontology. And here
37. 02:42you have also a smart example for example ice cream might be
38. 02:45an interesting domain and of course everything which is related
39. 02:48to ice cream and can be subsumed under ice cream then will
40. 02:52be modelled in your ice cream ontology.
41. 02:58Besides the domain ontology we have the so called task ontologies and there
42. 03:05fundamental concepts according to a general activity or task are modeled
43. 03:10independent from a specific domain and again like for the domain
44. 03:16ontology also for the task ontology what you do there is you
45. 03:19specialized against some terms that previously have been introduced within your upper
46. 03:25ontology or your top level ontology.
47. 03:29And now if you are going to combine both which means you are
48. 03:32looking at a specific domain and you want to solve a specific problem there
49. 03:38you combine task and type ontology and you come up with a so called application ontology.
50. 03:44This means you you have a specialized ontology focused on a
51. 03:48specific task in a specific domain and this is often a specialization
52. 03:53of both task and domain ontology and often it specifies roles
53. 03:58played by domain entities for specific activities.
54. 04:03An example could be you know how to bake a cheesecake which
55. 04:06means you have on the one cake on the one hand the cheesecake domain or
56. 04:11let's say cooking domain and a special task there baking and
57. 04:15you bake a cheesecake. This then will be an application ontology.
58. 04:20So this is only for the distinction on which level we deal with
59. 04:26if we are talking about ontologies and applying ontologies within knowledge graphs
60. 04:31for specific problems.
61. 04:35Another way to distinguish or to differentiate different ontologies
62. 04:40would be their level of generality and their level of expressivity.
63. 04:46Expressivity first means ok that can range from rather informal
64. 04:50which is more natural language related to rather formal which is then more
65. 04:55mathematical logic related. And you deal on the one hand informally
66. 05:00with rather lightweight ontologies and if you go to full fledged
67. 05:04first order logic you deal with so called heavyweight ontologies because
68. 05:09there reasoning also the more complex the ontology becomes
69. 05:14becomes also much more difficult complex time consuming.
70. 05:19That's the problem. So you start usually if you try to formalize ontologies
71. 05:23according to that schema you start with something like
72. 05:27a controlled vocabulary. This is nothing else but a list of
73. 05:31terms that have been agreed upon and you use these kind of
74. 05:34terms to describe knowledge. But it's only terms.
75. 05:38As soon as you endorse these terms let's say with a description and an explanation
76. 05:44you end up in a so called glossary. This is a list of terms
77. 05:47where each of these terms is explained but the explanation
78. 05:51again is here given in natural language.
79. 05:55If on the other hand you start to describe the relation
80. 05:59between the single words you are using then you have a so called Thesaurus.
81. 06:06And if you are going to make some of these relations more formal
82. 06:09like for example you define a hierarchy
83. 06:13then you have this a huge sorry then you have this taxonomy you see here
84. 06:19which is in our timeline here referred to as an informal is a
85. 06:23or a formal is a
86. 06:27representation. Informal in that sense means you have a hierarchy where
87. 06:31each subclass is allowed to have several super classes and
88. 06:35formal means there you have a strict separation which means each subclass only
89. 06:40is derived from a specific super class and not more. So you avoid
90. 06:45ambiguity and become more unique more formal. So this is exactly
91. 06:49the way you go. And then you introduce more concept by concept
92. 06:54which is again more formal and you end up in the end with value
93. 06:58restrictions logical constraints and then let's say things like
94. 07:03disjunctiveness, inversiveness, part-of-ness. So you have
95. 07:07then specific relations with a specific semantic that can only
96. 07:10be described in description logics of first order logic and
97. 07:14then of course we are in the field of the so called really
98. 07:17really heavyweight ontology. So most times
99. 07:20ontologies are not described in first order logic because this
100. 07:23is way too complicated and the point is also then the
101. 07:28complexity behavior is rather bad.
102. 07:31So we are in that are in the realm of description logics usually
103. 07:35for describing the stuff and you have seen that OWL or the
104. 07:39OWL dialect that we are using here that's OWL2DL is closely
105. 07:45connected to a specific description logic that we are using.
106. 07:50But beware there are dragons.
107. 07:53And to quote famous J R R Tolkien, "it does not do to leave
108. 07:58a live dragon out of your calculations if you live near him."
109. 08:02So what does that mean?
110. 08:04This means don't mistake your ontology that you have designed defined for reality.
111. 08:10And I will give you a rather prominent example, you know what should be
112. 08:15avoided what kind of pitfalls there are.
113. 08:18Probably you know Jorge Luis Borges, this is a rather famous author and he was
114. 08:23librarian of the National Library of Argentina in Buenos Aires.
115. 08:27So a rather famous guy he wrote lots of interesting philosophical
116. 08:31texts and also science fiction novels. So this guy wrote in
117. 08:35nineteen forty two in his essay the analytical language of John Wilkins
118. 08:41the following. So he introduced an ontology as
119. 08:45various categories of animals from "a certain Chinese
120. 08:50encyclopedia" to show that of course animals can be categorized
121. 08:55also in a complete different way, he said for example.
122. 08:58According to that certain Chinese encyclopedia we distinguish
123. 09:02animals in the following way -
124. 09:05first class: those that belong to the emperor, second class: embalmed ones,
125. 09:09then those that are trained, suckling pigs, then mermaids or sirens,
126. 09:15fabulous ones, stray dogs, those that are included in this classification,
127. 09:20those that tremble as if they were mad, honorable ones,
128. 09:24those drawn with a very fine camel hair brush, et cetera, and those
129. 09:29that have just broken the flower vase, and those that at a distance resemble flies,
130. 09:35which of course completely captures reality, doesn't it?
131. 09:40Yeah you see if you compare this of course to let's say taxonomies from modern biology
132. 09:46of course that looks completely different.
133. 09:51Message that you have to take home -
134. 09:55ontologies are not the reality. However if we want to make use
135. 10:00of ontologies and populate ontologies with instances.
136. 10:03Then we are dealing with knowledge graphs and so far
137. 10:07you might remember that in that lecture we haven't given you
138. 10:10any definition of knowledge graphs and this will happen in the
139. 10:13next part of the lecture.

To enable the transcript, please select a language in the video player settings menu.