Cleaning up the confusion about thesauri and classifications

To clean up the confusion mentioned earlier, I wrote a short introductionary article on thesauri and classifications. It mainly relies on an excerpt taken from a book of a former information science teacher of mine. -- Here we go:

Introduction

Base assumption is that data, information, knowledge require to be ordered. The data has to be ordered systematically. People who perform indexing on data bases -- so called "indexers" -- use order systems to make content retrievable. Thus, only if a user knows the tools used during database creation she/he can retrieve information from that very database.[1]


Methods of organizing content representation originate from the fields of library science and documentation.[2]


There are two dominant content representation methods in documentation: classification and thesaurus. "A classification is a structured representation of classes and of the notional relationships between the classes."[3]  Any class is represented by a notation, whereby the notation is independent of any natural language (cf. DIN 37205, 2).[4]  "Similarly, a thesaurus also is an organized compilation of terms, but in this case their natural language appellations are used (cf. DIN 1463/1, 2)."[5] 

Systems of Concepts

Business documentation creates an order. This order refers to notions (= concepts) an relations between these notions. "One may assume that this equals to organize business terminology [...] by a notional order."[6] 


Systems of concept differentiate between two main kinds of relationship: associative and hierarchical relationship. There are two variants of the latter: the abstract and the partitive variant.[7]

Abstract relationship means that a "child" term has the same features as its parent plus at least one additional feature, the parent term does not have.[8]  (The features are not stated anywhere; there is nothing but a parent term, a child term, and a relationship link between both of them, representing that very "the child term has the same features as the parent one, but at least one additional other feature, the parent term does not have". -- In fact, the relationship refers to the items not to the notations: The item referred to by the notation has the more/the less features than the item referred to by the parent/child term notation.)

Partitive relationship means that the items referred to by the child terms are part of the item referred to by the parent term.[9]

Associative relationship refers to a relationship existing between a pair of terms that cannot be related hierarchically to each other, although nevertheless there is a relationship between both of the terms. A pair of antonyms cannot be related hierarchically to each other, therefore here the matching kind of relationship is the associative one.[10]


"A classification is a system of concepts"[11]  having notions as classes. The classes are labeled by notations[12],  which are language independent.

Main difference between Thesaurus and Classification

The main difference between thesaurus and classification is that a thesaurus selects natural language words instead of setting up cryptic notations.[13]  Thus, otherwise than classifications, a thesaurus cannot be used language indepently. To avoid confusing originating from synonyms and homonyms a thesaurus applies terminological control, i.e. it keeps homonyms distinct and sets of synonyms referring to the same item constitute a class. The most common one of these synonyms gets picked up and will be treated as the descriptor of that very class, i.e. a handle for the class -- one could say "a natural language notation" or simply "a label". To ensure that any piece of information (which shall be referred to by at least one "word" of the thesaurus[14])  always gets referred by the same "word", the non-descriptor synonyms refer to the descriptor of the class they belong to. Thus, any piece of information gets referred to by descriptors only, and retrieval attempts using non-descriptor synonyms get redirected to the matching descriptors. So, the pieces of information get tagged by descriptor synonyms (for short just "descriptors") and retrievals also use descriptors, so a match between tagging "words" and retrieval "words" becomes much more probable than if allowing to choose from descriptors and non-descriptors as well.[15] 

Accuracy sacrified to administration convenience

Sometimes synonyms and terms referring to similar items as the other synonyms won't be/aren't kept distinct but simply added to a common class[16].  Reason for this often is to keep administration expense low.

---

[1] cf. [Stock 2000], p. 59, par. 1
[2] cf. [Stock 2000], p. 59, par. 2, sentence 1
[3] [Stock 2000], p. 59, par. 3, sentence 2: "Eine Klassifikation ist eine strukturierte Darstellung von Klassen und der zwischen den Klassen bestehenden Begriffsbeziehungen [...]."
[4] cf. [Stock 2000], p. 59, par. 3, sentence 2
[5] cf. [Stock 2000], p. 59, par. 3
[6] cf. [Stock 2000], p. 59, par. 4 (incl headline)
[7] cf. [Stock 2000], p. 60, par. 1
[8] cf. [Stock 2000], p. 61, par. 2, sentences 1–2
[9] cf. [Stock 2000], p. 61, par. 2, sentences 4–5
[10] cf. [Stock 2000], p. 62, par. 1 (incl. bullet list)
[11] [Stock 2000], p. 63, par. 1, sentence 1: "Ein Klassifikationssystem ist ein Begriffssystem [...]."
[12] cf. [Stock 2000], p. 63, par. 1, sentences 1–2
[13] cf. [Stock 2000], p. 76, no. 3.3, par. 2, sentence 2
[14] cf. [Stock 2000], pp. 81–84
[15] cf. [Stock 2000], p. 77, par. 1 (including the example in between)
[16] cf. [Stock 2000], p. 77, par. 1, sentence 2

[Stock 2000]:
Stock, Wolfgang G.
Informationswirtschaft : Management externen Wissens
number of edition unknown
Muenchen, Wien, Oldenbourg, 2000
ISBN 3-486-24897-9<<

---

Updates: 20070624: Tagged the posting. Updated the posting style (layout) to my current style, such as more precise word picks, better grammar.

From the Information Science local point of view (upgraded)

Information Science and its predecessor sciences like documentation or library science tackle one big problem in information: stay able to retrieve pieces of information once stored.

In ancient days the pieces of information mainly were material, i.e. not computer-indexable. For example, books were such a kind of material.

Common approach from the information science point of view is to assign each of the books with a set of keywords: When you want to retrieve one of them later, you go ahead, choose some of the keywords, and lookup them in a catalogue which itself refers to the books
theirselves.

To be able to handle this all, you need at least three kinds of storage:

1. a storage for the books, e.g. a kind of library, organized in a way to stay able to at least locate the shelf a particular book is placed in
2. a storage for the catalogue
3. and, most important: a storage for the keywords.

The keywords theirselves have to be stored somewhere. If you neglect this part of the task, one day you will apply this keyword to the book and another day another keyword, but both meaning the same -- i.e. being synonyms.

So, what result originates from that?

Assumed you associated two books X and Y very similar in content with two synonymous but different keywords, A and B. One day you want to know something about a topic that is covered by both books X and Y, but you don't know about that. You directly go to the catalogue. You pick up a search keyword that accidently fell into your mind. Say B.
You look up the appropriate catalogue card and find Y. -- That there is a closely related X book you don't even get aware of. So you fetch the Y book, but the X book remains in shelf. Possibly it would have been valuable to find X as well.

Therefore keywords get stored theirselves too.

The main goal of a keyword storage is similar to the other pieces of information storages: To stay able to find the wanted contents, i.e. the keywords -- and to find exactly the keywords wanted. "Keywords wanted" are those that might be applied to one or more books.

The appropriate tool for keyword or, more precisely, term (as in "search term") storage is a terminology. It mentions every word that was applied to at least one piece of information -- e.g. book -- of the pieces of information storage -- e.g. library. (In a converse, to keep administrative work load small, there's the suggest to choose only keywords already listed in the terminology, to associate books with.)

In a simple case, a terminology might be an alphabetical list of terms. Even better a taxonomy is: For each of the terms it offers an orientation help: Usually there are broader and narrower variants of terms: a mammal is treated to be broader than a dog or cat or cow or horse or something else which is a mammal. (In fact, taxonomies refer to items but list the labels of the items. Taxonomies are closely related to ontologies.)

So, a taxonomy offers is a relationships to identify the location of a given term in the whole taxonomy. Less common than is a are has a relationships, like the ones applied between car and something like wheels, motor, front window, doors etc. Both of these relationships are called hierarchical relationships.

My diploma thesis was about the thesaurus kind of taxonomy, so I currently I am not sure if this applies for the classification kind as well: There is at least one more kind of relationship -- the associative one. It relates terms at each other that don't belong to a hierarchical order, but somehow have something to do with each other, like bird and bird cage. (Admittedly, they might be related using a has a relationship, but I never came across such an assignment.)

Synonyms in taxonomies get treated a special way

Synonyms in taxonomies get treated a special way: They get collected to a single class. Each of the items of that class is (/treated to be, compare to administrators' cheats above) synonymous with every other item of the class.

In classifications there are just classes related to each other, being representative for the terms belonging to the class. In thesauri there are no such classes. During thesaurus creation sets of synonymous terms get identified. One "most significant/common" term gets chosen to be the term representative for all the other synonyms. This "most significant/common" one is called "descriptor", while the others become non-descriptors.

Why all the fuss about the synonym details?

If you look up a synonym you get redirected to the class or descriptor the synonym belongs to. None of the books/other pieces of information ever gets keyworded by a synonym. That solves the problem of searching the X and Y books mentioned above one day by the A and another day by the B search term: Either A points
to B or B to A or both to a third term, say C. And both the books are associated by the "most significant/common" term, e.g. C. So, either if you chose A or B as your search term, you always find all the relevant pieces of information/books, e.g. X and Y.


So far the common part of terminologies.


But there is a usability problem in the widespread is a approach.<<

Glossary

• to retrieve
  • to find again
• class
  • a set of synonymous words
• synonym
  • a word meaning the same as another word
  • sometimes there is a difference between them both, but the one applying them doesn't notice
  • terminology administrators sometimes think it is not worth the effort to keep "very similar" terms distinct, so they merge them into a single class by claiming the very words were synonyms
• catalogue card
  • associations between keywords are stored on catalogue cards which, as a whole, make up the catalogue itself

---

Updates: 20070624: Tagged the posting. Updated the posting style (layout) to my current style, such as using blockquotes when appropriate, more precise word picks, better grammar.

From the Information Science local point of view

Information Science and its predecessor sciences like documentation or library science tackle one big problem in information: stay able to retrieve pieces of information once stored.

In ancient days the pieces of information mainly were material, i.e. not computer-indexable. For example, books were such a kind of material.

Common approach from the information science point of view is to associate each of the books with a set of keywords: When you want to retrieve one of them later, you go ahead, choose some of the keywords, and lookup them in a catalogue which itself refers to the books itself.

To be able to handle this all, you need at least three kinds of storage:

1. a storage for the books, e.g. a kind of library, organized in a way to stay able to at least locate the shelf a particular book is placed in
2. a storage for the catalogue
3. and, most important: a storage for the keywords.

The keywords theirselves have to be stored somewhere. If you neglect this part of the task, one day you will apply this keyword to the book and another day another keyword, but both meaning the same -- i.e. are synonyms.

So, what result originates from that?

Assumed you associated two contently very similar books X and Y with two synonymuos but different keywords, A and B. One day you want to know something about a topic that is covered by both books X and Y, but you don't know about that. You directly go to the catalogue. You pick up a search keyword that accidently fell into your mind. Say B. You look up the appropriate catalogue card and find Y. -- That there's a closely related X book you don't even get aware of. So you fetch the Y book, but the X book remains in shelf. Possibly it would have been valuable to find X as well.

Therefore keywords get stored theirselves too.

The main goal of a keyword storage is similar to the other pieces of information storages: To stay able to find the wanted contents, i.e. the keywords -- and to find exactly the keywords wanted. "Keywords wanted" are those that might be applied to one or more books.

The appropriate tool for keyword or, more precisely, term (as in "search term") storage is a terminology. It mentions every word that was applied to at least one piece of information -- e.g. book -- of the pieces of information storage -- e.g. library. (In a converse, to keep administrative workload small, there's the suggest to choose only keywords already listed in the terminology, to associate books with.)


In a simple case, a terminology might be an alphabetical list of terms. For each of the terms it offers an orientation help: Usually there are broader and narrower variants of terms: a mammal is treated to be broader than a dog or cat or cow or horse or something else which is a mammal. (In fact, terminologies refer to items but list the labels of these.)

So, a terminology offers is a relationships to identify the location of a given term in the whole terminology. Less common than is a are has a relationships, like the ones applied between car and something like wheels, motor, front window, doors etc. These kinds of relationships are called hierarchical relationships.

My diploma thesis was about the thesaurus kind of terminology, so I currently I am not sure if this applies for the classification kind as well: There is at least one more kind of relationship -- the associative one. It relates terms at each other that don't belong to a hierarchical order, but somehow have something to do with each other, like bird and bird cage. (Admittedly, they might be related using a has a relationship, but I never came across such an assignment.)

Synonyms in terminologies get treated a special way

Synonyms in terminologies get treated a special way: They get collected to a single class. Each of the items of that class is (/treated to be, compare to administrators' cheats above) synonymuos with each other item of the class.

In terminologies called classification there just classes related to each other, being representative for the terms belonging to the class. In thesaurus kind of terminologies there are no such classes. During thesaurus development a sets of synonymous terms get identified. One "most significant/common" term gets chosen to be the term representatively for all the other synonyms. This "most significant/common" one is called "descriptor", while the others become non-descriptors.

Why all the fuss about the synonym details?

If you lookup a synonym you get redirected to the class or descriptor the synonym belongs to. None of the books/other pieces of information ever gets keyworded by a synonym.

That solves the problem of searching the X and Y books mentioned above one day by the and another day by the B search term: Either A points to B or B to A or both to a third term, say C. And both the books are associated by the "most significant/common" term, e.g. C. So, either if you chose A or B as your search term, you always find all the relevant pieces of information/books, e.g. X and Y.

So far the common part of terminologies.

But there is a usability problem in the widespread is a approach.

Glossary

to retrieve - to find again

class - a set of synonymous words

catalogue card - associations between keywords are stored on catalogue cards which, as a whole, make up the catalogue itself

synonym - a word meaning the same as another word

• sometimes there is a difference between them both, but the one applying them doesn't notice
• taxonomy administrators sometimes think it is not worth the effort to keep "very similar" terms distinct, so they merge them into a single class by claiming the very words were synonyms

---

Updates: 20070624: Tagged the posting. Updated the posting style (layout) to my current style, such as using blockquotes when appropriate, more precise word picks, better grammar.