BoW Ontology

We started modeling on December, 18th, 2020. After 70 minutes the first 100 base classes have been modeled with 120 triples resulting in in modeling speed of 1.7 triple / minute. There were no specific requirements for the task, except that as many concepts as possible contained on the website should be modeled. In contrast to the English website , which essentially only contains a list of water body types, the German website is very detailed and well suited as a knowledge base.

The system of German rivers, which we had already modeled in 2012, fits in with this. This also results in the attributions (data and object properties) for flows. Of course, the triple instances of the flows could be entered/modeled much faster, because in principle it is only data acquisition. After the first phase, there were 120 triples for classes and about 1,300 triples for particulars of flows. This corresponds to an intensionality in section int of 120 / 1420 = 8.5% and an extensionality in section ext of 1300 / 1420 = 91.5%.

The 2nd phase covers the period from 12/24/2020 to 12/30/2020. During this time, a number of schema optimizations were made. In addition, value sets were introduced for water body color, water body size, ecological status, etc. In total, the number of triples in the scheme grew to 796. An additional 12:23 hours were spent on this in the period, making a total of 14.4 hours. This results in a speed of 796 triples / 862 minutes = 0.92 triples / minute. This means that the modeling speed has roughly halved in the 2nd phase. This is due to the fact that reflecting on improvement opportunities and revising takes much more time.

The fact that we needed fewer triples than for modeling in RDF/OWL is partly due to the fact that our naming conventions mean that we always implicitly/semantically have a class for things with a ^ as a prefix, an object property for <> or ◊ and an object property for . a data property. Otherwise approx. 214 triples would be required additionally. Compared to OWL modeling, we would then arrive at around 1050 triples and a modeling speed of 1.22 triples per minute. Approximately 220 minutes = 3.7 hours and sub-class relationship 170 minutes = 2.8 hours were spent purely on modeling the classes, i.e. 6.5 hours for the class hierarchy. This results in: 220 / 3.7 = 59 classes/concepts per hour, 60 subclass relationships per hour and 390 / 6.5 = 60 classes + subclasses per hour.

The English and German Wikipedia pages on formal concept analysis contain simpler examples of the Body of Water term latices.

Lets first analyse the BoW-ontology in terms of number of classes, properties and featuress. A number of ca. 170 base concepts has been identified. The first 20 of them are shown in the following table where each valid feature is marked with X, e.g. (Bach, PD:ganzjährig) resp. (Creek, PeriodType:perennial) . The total number of properties for BoW (body of water) is around 65. Less than twenty of them have only a valid feature for one concept. The most frequent features are (German/English):

1. GT-See/BoWType-Lake:16
2. GT-Meer/BoWType-Sea: 24
3. GT-Bucht/BoWType-Bay: 7
4. GT-Wasserstrasse/BoWType-Waterstreet: 15
5. FS-fliessend/FloatingType-floating: 23
6. FS-fliessendFloatingType-standing: 31
7. OU-oberirdisch/LocationType-aboveground: 103
8. OU-unterirdisch/LocationType-underground: 7
9. NT-natürlich/NatureType-natural: 76
10. NT-künstlich/NatureType-artificial: 9
11. SS-Salzwasser/WaterType-Saltwater: 27
12. SS-Süßwasser/WaterType-Freshwater: 44
13. BR-schmal/WidthType-small: 9
14. BR-breit/WidthType-large: 7
15. TF-flach/ShallowType-shallow: 12
16. TF-sehr_tief/ShallowType-very_deep: 22
17. PD-jährlich/PeriodType-perennial:79

The Ontology4 search page retrieves BoW types by combinations of features. It computes the intersection of superclasses containing all values. Because the values have been stored in English and in German it works for both languages.The following results have been recherched manually, but of course one could automize it to find out which of the combination of the values deliver a substantial number of results to be candidate for PC combinations.

1 Keyword: About a dozen values yield around 10 or more basic concepts

• aboveground /oberirsch:107
• underground /unterirdisch:9
• natural /natürlich:75;
• artificial /künstlich: 12
• standing /stehend: 51
• floating /fliessend: 18
• fresh_water /Süßwasser: 35
• salt_water /Salzwasser
• open /offen: 16
• perennial /ganzjaehrig: 74
• flat /flach: 12
• very_deep /sehr_tief: 18

2 Keywords ⊓ :from (artificial/natural, aboveground/underground, floating/standing, inland/open) follow 24 combinations

1. artificial ⊓ aboveground: 11
2. artificial underground: 1
3. artificial ⊓ floating: 3
4. artificial ⊓ standing: 3
5. natural ⊓ aboveground: 60
6. natural ⊓ underground: 3
7. natural ⊓ floating
8. natural ⊓ standing: 35
9. standing ⊓ aboveground: 30
10. floating ⊓ aboveground: 24
11. standing ⊓ underground: 4
12. floating ⊓ underground: 2
13. inland ⊓ artificial: 10
14. inland ⊓ natural: 55
15. inland ⊓ aboveground: 51
16. inland ⊓ underground: 5
17. inland ⊓ floating: 19
18. inland ⊓ standing: 41
19. open ⊓ artificial: 0
20. open ⊓ natural: 12
21. open ⊓ aboveground: 12
22. open ⊓ underground: 0
23. open ⊓ floating: 1
24. open ⊓ standing: 10

3 Keywords: only 2 combinations of three keywords deliver more than 10 basic classes

• natural ⊓ aboveground ⊓ floating: 13
• natural ⊓ aboveground ⊓ standing: 25

4 Keywords: combining previous with saltwater/sweet_water create 32 more relevant partitions:

• natural ⊓ aboveground ⊓ standing ⊓ saltwater: 13
• natural ⊓ aboveground ⊓ standing ⊓ freshwater: 9
• natural ⊓ aboveground ⊓ floating ⊓ freshwater: 10

The number of triples on the class level for the combination of 4 keywords is reduced from 32 * 4 = 128 to 32 * (4 - 3) = 32 . This also holds for any particulars of one of the 32 PCs. In general the percentage of saving is 100 - 1 / NbrOfKeywords *100, e.g. 100 - 1 / 4 *100  = 75%.

The following analysis of the Body of Water (BoW) features was provided to us by Thomas Zeh and is based on the procedure presented in the Attribute Implications section. The 82 classes and 17 features from the Tables were included in the investigation. Classes that had no properties were excluded. A total of around 100 feature implications were derived, many of which are acceptable with few exceptions. Some typical ones are discussed below:

The following implications seem not to be acceptable:

• (5) inland ⇒ aboveground
• (14) underground, natural ⇒ inland, aboveground : not acceptable, underground and aboveground exclude each other
• (69) aboveground, flat ⇒ natural
• (21) floating, freshwater ⇒ inland, aboveground: this excludes underground BoWs
• (47) floating, halfday ⇒ inland, aboveground, natural, freshwater, perennial: questionable
• (70) natural, flat ⇒ aboveground
• (71) inland, aboveground, natural, flat ⇒ standing : questionable

The majority of others seem to be correct and offer a lot of insights:

• (1) lake ⇒ inland, standing, aboveground, natural, perennial
• (2) sea ⇒ aboveground, natural, perennial
• (3) bay ⇒ aboveground, natural, perennial
• (6) floating, aboveground ⇒ inland, freshwater
• (7) standing, aboveground ⇒ inland
• (10) underground, aboveground ⇒ inland : negating premises lead to trivial implications and can be deleted
• (11) floating, natural ⇒ inland, aboveground, freshwater
• (12) standing, natural ⇒ inland, aboveground : questionable, since there could be underwater lakes
• (16) natural ⇒ inland, aboveground, freshwater
• (18) aboveground, saltwater ⇒ perennial
• (19) natural, saltwater ⇒ aboveground, perennial
• (22) aboveground, freshwater ⇒ inland
• (23) waterstreet, inland, aboveground, freshwater ⇒ perennial
• (35) inland, floating, aboveground, freshwater, perennial ⇒ natural
• (37) waterstreet, inland, floating, aboveground, freshwater, natural ⇒ floating
• (38) waterstreet, inland, aboveground, natural, saltwater, perennial ⇒ sea, verydeep
• (39) sea, aboveground, natural, saltwater, perennial ⇒ verydeep
• (42) sea, bay, aboveground, natural, perennial ⇒ saltwater, verydeep
• (43) floating, shortterm  ⇒ inland, aboveground, natural, freshwater
• (44) aboveground, shortterm  ⇒ inland, floating, natural, freshwater
• (45) natural, shortterm  ⇒ inland, floating, aboveground, freshwater
• (46) freshwater, shortterm  ⇒ inland, floating, aboveground, natural
• (48) aboveground, halfday ⇒ natural, perennial
• (49) natural, halfday ⇒ aboveground, perennial
• (56) floating, small ⇒ inland, aboveground, natural, freshwater, flat
• (57) standing, small ⇒ inland, aboveground, natural, freshwater, flat
• (58) aboveground, small ⇒ inland, natural, freshwater
• (59) natural, small ⇒ inland, aboveground, freshwater
• (61) floating, large ⇒ inland, aboveground, natural, freshwater, perennial
• (62) aboveground, large ⇒ inland, floating, natural, freshwater, perennial
• (66) sea, waterstreet, inland, floating, aboveground, natural, freshwater, perennial, large ⇒ verylarge
• (67) verylarge ⇒ sea, aboveground, natural, saltwater, perennial, verydeep
• (68) standing, flat ⇒ inland, aboveground, natural
• (72) saltwater, flat ⇒ aboveground, natural, perennial
• (73) freshwater, flat ⇒ inland, standing, aboveground, natural
• (74) inland, standing, aboveground, freshwater, perennial, flat ⇒ verysmall
• (76) verysmall, flat ⇒ inland, standing, aboveground, natural, freshwater
• (77) small, flat ⇒ inland, standing, aboveground, natural, freshwater
• (79) aboveground, verydeep ⇒ sea, natural, saltwater, perennial
• (80) natural, verydeep ⇒ sea, aboveground, saltwater, perennial
• (81) saltwater, verydeep ⇒ sea, aboveground, natural, perennial
• (83) floating, verynarrow ⇒ inland, aboveground, freshwater
• (84) aboveground, verynarrow ⇒ inland, floating, freshwater
• (85) natural, verynarrow ⇒ inland, floating, aboveground, freshwater, perennial
• (86) freshwater, verynarrow ⇒ inland, floating, aboveground
• (87) floating, verynarrow ⇒ inland, aboveground, natural, freshwater, verysmall
• (88) aboveground, narrow ⇒ inland, floating, natural, freshwater, small, perennial
• (89) natural, narrow ⇒ inland, floating, aboveground, freshwater, small, perennial
• (90) freshwater, narrow ⇒ inland, floating, aboveground, freshwater, small, perennial
• (91) small, narrow ⇒ inland, floating, aboveground, natural, freshwater, perennial
• (92) floating, verywide ⇒ sea, waterstreet, inland, aboveground, freshwater, perennial, large
• (93) aboveground, verywide ⇒ sea, waterstreet, inland, freshwater, perennial, large
• (94) natural, verywide ⇒ sea, waterstreet, inland, aboveground, freshwater, perennial, large
• (95) freshwater, verywide ⇒ sea, waterstreet, inland, aboveground, natural, perennial, large
• (96) large, verywide ⇒ sea, waterstreet, inland, aboveground, natural, freshwater, perennial

In section Attribute implications it is explained that it is necessary to check the implications in an iterative process and to expand the attribute tables until all implications appear acceptable or cannot be further clarified due to the data situation. The implications that have mutually exclusive features on the left and/or right side can be automatically deleted. The conclusions marked as acceptable can be used to formulate rules.