=Paper=
{{Paper
|id=None
|storemode=property
|title=A Hybrid Approach for Learning SNOMED CT Definitions from Text
|pdfUrl=https://ceur-ws.org/Vol-1014/paper_55.pdf
|volume=Vol-1014
|dblpUrl=https://dblp.org/rec/conf/dlog/DistelM13
}}
==A Hybrid Approach for Learning SNOMED CT Definitions from Text==
https://ceur-ws.org/Vol-1014/paper_55.pdf
A Hybrid Approach for Learning S NOMED CT
Definitions from Text
Felix Distel and Yue Ma?
Institute of Theoretical Computer Science, Technische Universität Dresden, Dresden, Germany,
{felix,mayue}@tcs.inf.tu-dresden.de
Abstract. In recent years approaches for extracting formal definitions from natu-
ral language have been developed. These approaches typically use methods from
natural language processing, such as relation extraction or syntax parsing. They
make only limited use of description logic reasoning. We propose a hybrid ap-
proach combining natural language processing methods and description logic
reasoning. In a first step description candidates are obtained using a natural lan-
guage processing method. Description logic reasoning is used in a post-processing
step to select good quality candidate definitions. We identify the corresponding
reasoning problem and examine its complexity.
1 Introduction
Throughout the medical domain the formal representation of knowledge has proven to
be beneficial, allowing for powerful reasoning services for the debugging and querying
of knowledge bases. Among these KR formalisms lightweight Description Logics (DL)
from the EL family such as OWL2EL [12] have proven to be especially successful: they
allow for tractable reasoning while still providing a level of expressivity that is sufficient
for most ontologies in the medical domain. One such ontology is SNOMED which is
now a widely accepted international standard [15].
The downside of formal semantics is the cost associated with creating, maintaining
and extending ontologies. Since the knowledge is represented using logic based syntax
and semantics these tasks require specially trained staff that are both experts in logics
and in the application domain. One approach to facilitate the work of these experts is to
mine the vast expanse of knowledge that is available in textual form, e.g. in PubMed,
textbooks or on the web. A number of natural language based approaches have been
proposed to automatically or semi-automatically convert concept descriptions that are
available in textual form into formal concept descriptions in DL. The descriptions are
obtained through analysis of lexical or linguistic features, without a mechanism that
checks if the resulting descriptions are logically sound.
Typically, one sentence in natural language describes only one aspect of a target
concept, it hardly ever provides a full definition. To obtain full definitions the partial
definitions from different sentences must be compared and the ones with highest quality
must be selected. Typically, the NLP formalism will provide a confidence value for each
sentence which can provide some indication of its quality.
?
Funded by the DFG Research Unit FOR 1513, project B1.
Logic based knowledge representation is not very robust when it comes to errors in
concept descriptions. Even less obvious errors can cause an ontology to yield unwanted
consequences or even to become inconsistent. In this work we propose to use formal
constraints on the target concept in order to ensure at least a certain degree of logical
soundness.
The proposed approach is therefore a hybrid approach. In a first step an NLP formal-
ism is used to obtain candidate definitions from text and in a second step a reasoning
problem is used in order to select a good subset of the candidate definitions. In Section 5
we provide a summary of our text mining algorithm which was first presented in [10,
16]. It uses a multi-class classifier to predict if a given sentence describes an existential
restriction and if so, for which role. In Section 6 the task of selecting good candidates is
formalized as a reasoning problem. The proposed formalization extends an idea from
[9]: the conjunction over the selected set of candidates should satisfy the constraints
while maximizing the accumulated confidence values. In [9] the minimum has been
used to accumulate confidence values, whereas here, we investigate how the choice of
accumulation function influences the complexity of the reasoning problem (Section 7).
2 Related Work
For a long time, algorithms for learning ontologies from natural language have focussed
mainly on learning subclass relationships [20, 14], in a linguistic context sometimes
called hyponyms [5]. These algorithms essentially learn a taxonomy at best.
The generation of complex terminological axioms using more advanced logical
constructors is clearly a much harder task [4, 19]. There are essentially two underlying
ideas for the generation of complex axioms. In works such as [18, 19] the syntax of the
natural language input sentences is parsed, i.e. sentences are broken down into functional
units. These syntax trees are then scanned for predefined patterns which correspond to a
certain type of logical constructor. An interactive approach for ensuring the correctness
of the mined axioms is presented in [13]. The rules transforming lexical and linguistic
patterns to logical syntax are typically manually created [17]. This makes it difficult to
adapt these approaches to new domains as extensive tweaking of the rules is required.
By contrast approaches based on machine learning techniques are easier to adapt.
Some of these [8, 3] work on an instance level and are typically based on Inductive
Logic Programming techniques. On the terminological level there are approaches based
on relation extraction techniques [11]. Here, the patterns themselves are learned from
annotated text. In scenarios where the set of role names is stable, these techniques can
be used to learn simple restrictions of role depth 1 [10, 16].
3 Preliminaries
We consider the lightweight Description Logic EL, whose concept descriptions are built
from a set of concept names NC and a set of role names NR using the constructors
top concept >, conjunction u, and existential restrictions ∃. The semantics of EL is
defined using interpretations I = (∆I , ·I ) consisting of a non-empty domain ∆I and
an interpretation function ·I mapping role names to binary relations on ∆I and concept
descriptions to subsets of ∆I according to Table 1. A concept description C is said to be
atomic if C ∈ NC ∪ {>} or C = ∃r.D for some r ∈ NR and some concept description
D.
Table 1. Syntax and Semantics of EL
Name Syntax Semantics
concept name A AI ⊆ ∆I
role name r rI ⊆ ∆I × ∆I
top concept > > I = ∆I
conjunction C uD (C u D)I = C I ∩ DI
existential restriction ∃r.C (∃r.C)I = x | ∃y : (x, y) ∈ rI and y ∈ C I
primitive definition AvC AI ⊆ C I
full definition A≡C AI = C I
As axioms we allow full definitions and primitive definitions. Full definitions are
statements of the form A ≡ C, primitive definitions are statements of the form A v C
where A is a concept name and C is a concept description. A TBox T is a set of
axioms of these two types. We say that the interpretation I is a model of T if AI = C I
(or AI ⊆ C I ) holds for every full definition A ≡ C (primitive definition A v C.
respectively) from T . A concept description C is said to be subsumed by the concept
D with respect to the TBox T (denoted by T |= C v D) if C I ⊆ DI holds for all
models I of T . It is well-known that subsumption reasoning in EL is tractable, i.e. given
concept descriptions C and D, and a TBox T it can be decided in polynomial time if
T |= C v D [2].
There are two reasons for our restriction to full definitions and primitive definitions
instead of the more expressive GCIs. In our chosen setting we try to learn concept
descriptions for S NOMED CT, which uses only full definitions and primitive definitions.
Second, this restriction allows us to check subsumption on an atomic level according to
the following lemma from [9].
Lemma 1 ([9]). Let T be a TBox containing only primitive definitions. Let C and D be
concept descriptions that can be written as C = C1 u · · · u Cn and D = D1 u · · · u Dm ,
where Ci , Dj are atoms for all 1 ≤ i ≤ n, 1 ≤ j ≤ m. Then T |= C v D iff for every
atom Di , 1 ≤ i ≤ m, there is an atom Cj , 1 ≤ j ≤ n, such that T |= Cj v Di .
4 Task Description
Our approach for learning S NOMED CT-descriptions from text is based on the following
observations. When comparing legacy versions of S NOMED CT to the current one, it is
obvious that the set of role names has remained relatively stable while the number of
concepts has increased. In our approach we therefore assume the set of role names to be
fixed.
In our approach the knowledge engineer is allowed to pick a target concept, i.e. a
concept name that is not already defined in S NOMED CT. Our goal is, by benefiting
from DL reasoning, to facilitate the knowledge engineer’s design task by automatically
generating a definition for the target concept from the natural language text corpus.
We proceeds in two steps. First, in the text mining step the text corpus is searched
for sentences containing the target concept. A relation mining algorithm then decides
for each of these sentences whether it describes a primitive definition of the target
concept. For example the sentence “Baritosis is pneumoconiosis caused by barium dust.”
describes the primitive definition
Baritosis v ∃causative_agent.Barium_Dust.
The right hand side of the primitive definition, in this case ∃causative_agent.Barium_Dust
is then added to the set of candidate descriptions. For each candidate description, the
text mining algorithm also provides a numerical value, indicating confidence in the
correctness of the candidate.
Second, for the reasoning step, a set of formal constraint is obtained beforehand, e.g.
from the S NOMED CT design manual [6]. The objective is to select a subset of the set of
candidate description that
– satisfies the constraints, and
– maximizes an accumulation function over the confidence values.
5 Text Mining Step
The text mining step has been previously presented in [10, 16]. We thus only provide a
summary of the approach here. The main idea is to use existing S NOMED CT descriptions
to train a multiclass classifier to recognize sentences describing existential restrictions.
Data Preparation During the data preparation phase each sentence must be scanned
for occurrences of S NOMED CT concepts and these concepts must be mapped to the
corresponding part of the sentence. State of the art annotators are available to perform
this task. In [10] the tool MetaMap [1] is used, and in [16] MetaMap is combined with a
purpose built annotator.
We use existing knowledge from S NOMED CT to create our training set. In order to
make use of implicit knowledge, the set
R = {(A, r, B) | A, B ∈ NC , r ∈ NR , S NOMED CT |= A v ∃r.B}
is considered. Now, if a sentence is annotated with both A and B for some triple (A, r, B)
in S NOMED CT, then the sentence is labelled with the class r. That is, perhaps a bit
naively, whenever two concepts A and B occur in the same sentence and S NOMED CT
entails the subsumption A v ∃r.B then in the training phase the sentence is assumed to
describe this relation. An example for the data preparation of a sentence is shown in the
first two lines of Table 2.
Table 2. Text Alignment and Features
Annotated “Baritosis/Baritosis_(disorder) is pneumoconiosis caused by bar-
Sentence ium dust/Barium_Dust_(substance).”
S NOMED CT Baritosis_(disorder) | Causative_agent | Barium_Dust_(substance)
relationship
Features left type between-words right type
disorder “is pneumoconiosis caused by” substance
BoW {is, pneumoconiosis, caused, by}
Word 2-grams {is, pneumoconiosis, caused, by, is pneumoconiosis, pneumoco-
niosis caused, caused by }
Char. 2-grams {i, s, p, n, e, u, m, o, c, a, d, b, y, is, pn, ne, eu, um, mo, oc, co, on,
ni, io, os, si, ca, au, us, se, ed, by}
Training Phase To train a multiclass classifier features need to be extracted from each
of those sentences that have been assigned a class in the data preparation step. In [10]
in addition to the between words (words occurring between the parts of the sentence
annotated with the two S NOMED CT-concepts) the types of the two S NOMED CT
concepts are considered. The between words are represented as character n-grams. In
[16] only the between words are considered and a comparison between representations
as word n-grams, character n-grams and bag of words is made.
Test Phase Based on the weights learned in the training phase, the multiclass classifier
can predict for new annotated sentences whether they describe an existential restriction.
These existential restriction are then added to the candidate set. Typically, the classifier
will also give a confidence value, indicating how likely it assumes the prediction to be
correct. The exact nature of the confidence value depends on the classifier used.
An experimental evaluation of the approach can be found in both [10] and [16].
Due to limited availability of high quality text the evaluations consider only the three
roles causative_agent, associated_morphology, and finding_site. These roles have been
selected because they are relatively frequently used in S NOMED CT and occur in the
same fragment of S NOMED CT, the fragment describing diseases. In [10] the influence
of the quality of the text corpus is examined by comparing text from Wikipedia to
text obtained using the tool Dog4Dag [20]. It is shown that the latter, which is less
noisy, yields a visible increase in the quality of the extracted EL descriptions. In [16] a
comparison between different state of the art supervised learning algorithms is made:
logistic regression, support vector machines, multinomial naive Bayes, and random
forests. It appears that support vector machines provide the best quality results, reaching
an f-measure of up to 82.9% for the role causative_agent.
Both evaluations indicate that while the quality of the results is decent, especially
considering the relatively naive generation of the training set, there is still room for
improvement. We propose to use DL reasoning in a post-processing step to improve the
results.
6 Reasoning Step
In this section we formally define the reasoning problem used to select a good set of
candidate descriptions. We assume that a set of constraints can be obtained as described
Section 6.1. In our setting constraints are simply GCIs or negated GCIs where either the
left-hand side or the right-hand side is a concept variable X. We distinguish constraints
of the following four types:
D vX (1) D 6v X (3)
X vD (2) X 6v D (4)
In these constraints D can be a complex concept description. X must be a concept
name not occurring in T , D, or another constraint. For a complex concept description
C in which no concept variables occur, we say that C satisfies the positive constraint
D v X or X v D of X if T |= D v C or T |= C v D, respectively. C satisfies the
negative constraint D 6v X or X 6v D if T 6|= D v C or T 6|= C v D, respectively.
In its simplest form the task is now straightforward: for a given set of description
candidates and a given set of constraints, find a subset of the candidates whose conjunc-
tion satisfies the constraints. For complexity considerations we restate this as a decision
problem.
Problem 1 (Concept Selection (CS)). Input: A set of atomic candidate concept descrip-
tions S, a (possibly empty) ontology T and a setd of constraints C of the forms (1)–(4).
Question: Is there a subset S 0 of S such that S 0 satisfies all the constraints in C?
In the text mining step, each candidate in S has been assigned a confidence value
between 0 and 1. In practice one would like to give preference to solutions of CS that
contain candidates that have been assigned higher confidence values in the text mining
approach. We therefore need to decide upon a function for accumulating the confidence
values of all the selected candidates. Intuitively, such a function should be associative,
commutative, non-decreasing and have 1 as unit, in other words it should be a triangular
norm or t-norm [7]. The most well-known t-norms are
– the minimum t-norm, also known as Gödel t-norm, x ⊗ y = min(x, y),
– the product t-norm, x ⊗ y = x · y, and
– the Łukasiewicz t-norm, x ⊗ y = max{0, x + y − 1}.
In the following we shall only consider t-norms whose computation takes only poly-
nomial time in the size of the inputs x and y. The idea behind the following decision
problem is to maximize the accumulated confidence of the selected candidates with
respect to a given t-norm ⊗.
Problem 2 (Accumulated Confidence Concept Selection (⊗-CS)). Input: A set of atomic
candidate concept descriptions S, a real number m ∈ [0, 1], a (possibly empty) ontology
T and a set of constraints C of the forms (1)–(4), together with a confidence function
wt : C → [0, 1].
Is there a subset S 0 of S such that S 0 satisfies all the constraints in C
d
Question:
and {wt(S) | S ∈ S 0 } ≥ m?
N
While the minimum t-norm typically has the best computational properties, it can
in practice make sense to use one of the other two t-norms. Consider a situation where
several candidates have the same confidence value wt(C) = c < 1. For the minimum
t-norm it makes no change whether the selection contains one or many of these candi-
dates, while the other two t-norms give preference to smaller selections, thereby likely
increasing the precision of the solution.
6.1 Obtaining the Constraints
In an ideal scenario constraints of types (1)–(4) should come directly from the knowledge
engineers. Since this, however, would simply shift the cost from the creation of concept
descriptions to the creation of constraints, other ways of obtaining the constraints need
to be examined.
In the case of S NOMED CT, a surprisingly large number of constraints can be
found in its design manual, the S NOMED CT User Guide [6]. For instance, it restricts
the range of the role finding_site to Body_Structure. One can thus add a restriction
X 6v ∃finding_site.C for all those concepts describing disjoint main branches to
Body_Structure, such as Disorder, Substance, etc. This ensures that the learned concept
does not become unsatisfiable. A similar approach can be used for domain restrictions.
Furthermore, a great number of definitions in S NOMED CT are only partial defini-
tions. If the task is to enrich this partial definition using text mining, then the existing
definition can be used as type (2) constraints.
Finally, one could also consider an interactive way for obtaining constraints. In a first
approximation, one would simply use the conjunction over the complete set of candidates
as the definition of the target concept. This definition would be temporarily added to the
ontology. Both the original and the extended ontology would then be classified. The user
is then presented with a set of subsumptions between concept names that are entailed by
the extended ontology but not by the original one. He can then choose to mark each of
them as intended or unintended. The former are added as positive constraints, while the
latter are added as negative constraints. The concept selection problem is solved again.
The entire process is repeated with the new selection of candidates until the user no
longer marks any new consequences as unintended.
Example 1. Take Baritosis as the target concept. In an experiment described in [9], the
following description candidates were extracted with high weights:
∃causative_agent.Barium_Compound, 0.92140
∃causative_agent.Dust, 0.97038
∃finding_site.Lung_structure, 0.99999
∃causative_agent.Barium_Dust, 0.99997
Due to the high weights returned by the learning approach, it is impossible to exclude
any of the candidates based on weights alone. However, the concept selection framework
can discard the incorrect candidate ∃Causative_agent.Barium_compound, since the
S NOMED CT User Guide states that the range of causative_agent is restricted to exclude
Chemical, translating to the constraint
X 6v ∃causative_agent.Chemical.
Barium_Compound which is subsumed by Chemical can thus be discarded.
7 Complexity
In [9] we have looked at the complexity of CS and ⊗-CS but only for the case where ⊗
is the minimum t-norm (called min-CS in the following). The results from [9] state that
both CS and min-CS are NP-complete, but become tractable when only constraints of
types (1)–(3) are allowed. NP-hardness is thus caused by constraints of type (4).
In Section 7.1, we argue that for min-CS tractability is still maintained if constraints
of all four types are allowed but we require that in every type (4) constraint X 6v D the
concept D must be atomic. Conversely, in Section 7.2 we show that for the product and
Łukasiewicz t-norm ⊗-CS is NP-complete even if only (2) constraints are used.
7.1 Tractable Variants
In [9] it is shown that CS and min-CS are tractable when only constraint types (2) to (3)
are used. In this section we recall the argument and show that it can easily be extended
to constraints of type
X 6v D, where D is atomic, (4’)
using Lemma 1. Throughout this subsection we assume that full definitions have previ-
ously been expanded and the TBox contains only primitive definitions. The reason is,
that in the presence of full definitions one could simply add a full definition for D to the
TBox and thus emulate type (4) restrictions using type (4’) restrictions.
We first consider the variant that restricts to constraint types (2) and (3).
Restriction to (2) and (3): First, let an instance (T , S, C) of CS be given. Notice that
if a concept C satisfies a constraint X v D or D 6v X and E is a concept d 0 description
satisfying E v C then E also satisfies the constraint. In particular, if S satisfies all
constraints for some S 0 ⊆ S then S also satisfies them. Hence, if only constraint types
d
(2) and (3) occur, then there is a solution to the CS problem iff S itself is a solution. The
latter can be verified in polynomial time since subsumption reasoning in EL is tractable.
The same argument shows that there is a solution to the min-CS problem (T , S, k, C, wt)
iff {S ∈ S | wt(S) ≥ k} is a solution. Again, this can be verified in polynomial time.
Hence, CS and min-CS are tractable if we restrict to types (2) and (3).
Restriction to (1)–(3) + (4’): We now assume T contains primitive definitions only, i.e.
d have been expanded. Consider now a constraint D v X ∈ C of type
all full definitions
(1). A concept S 0 for S 0 ⊆ S satisfies this constraint if and only if T |= D v S for
all S ∈ S 0 .
For a constraint X 6v D ∈ C of type (4’) we can use Lemma 1 to exploit the fact that
D is atomic and that T only d contains primitive definitions. The Lemma then states that a
set S 0 ⊆ S satisfies T |= S 0 v D iff there is some S ∈ S 0 such that T |= S v D. In
other words S 0 satisfies the constraint X 6v D ∈ C iff T 6|= S v D for all S ∈ S 0 .
This shows that there is a solution S 0 of (T , S, C) iff there is a solution S00 of
(T , S0 , C0 ) where
S0 = {S ∈ S | ∀(D v X) ∈ C : T |= D v S}
∩ {S ∈ S | ∀(X 6v D) ∈ C : T 6|= S v D} (5)
C0 = {c ∈ C | c of type (2) or (3)}.
Notice that C0 can be obtained in linear time and S0 can be computed in polynomial
time since subsumption reasoning in EL is tractable. This shows that restrictions of type
(1) can be dealt with in a polynomial time preprocessing step. Tractability of CS and
min-CS for constraints of types (1)–(3) then follows immediately from this fact and
tractability of CS and min-CS when restricted to (2) and (3).
This shows that while constraints of type (4’) may still be problematic in the context
of full definitions, tractability is maintained for TBoxes where the expansion of full
definitions does not lead to an exponential blowup.
7.2 NP-hard variants
For all variants of CS and ⊗-CS containment in NP is easy to see, since one can simply
guess a subset of S and verify in polynomial time if it is a solution (remember that
subsumption reasoning in EL is tractable). Notice, that in ⊗-CS we require that the
t-norm itself can be computed in polynomial time.
In [9] it is shown using a reduction from SAT that CS with all 4 types of constraints
is NP-hard. This implicitly proves NP-hardness for ⊗-CS, since CS can be considered to
be a special case of ⊗-CS with all confidence values set to 1.
In this work, using a reduction from the well-known NP-complete problem Vertex
Cover, we show that for the product and Łukasiewicz t-norm ⊗-CS is NP-hard, even
when restricted to constraints of type (2).
Problem 3 (Vertex Cover). Input: A natural number k and an undirected graph G =
(V, E) consisting of a set of vertices V and a set of edges E.
Question: Is there a subset O ⊆ V satisfying O ∩ e 6= ∅ for every edge e ∈ E and
|O| ≤ k.
We describe the reduction for the product t-norm. Starting with an instance of Vertex
Cover consisting of a graph G and a number k, we construct an instance of ⊗-CS as
follows. We choose NR = {r}, NC = {Ae | e ∈ E}. For every node v ∈ V we
introduce a candidate concept description
l
Sv = ∃r. Ae . (6)
v∈e
The confidence function wt is set to be constantly 0.5 (any value strictly between 0 and
1 serves the purpose). For every edge e ∈ E a constraint
X v ∃r.Ae (7)
is added to the set of constraints. We then define m = 0.5k and we let the TBox T be
empty.
d Consider a subset S 0 ⊆ {Sv | v ∈ V }. Since the TBox T is empty the conjunction
0
S satisfies the constraintdX v ∃r.Ae iff Sv v ∃r.Ae holds. From (6) this is equivalent
to v ∈ e. This shows that S 0 satisfies all constraints iff {v | Sv ∈ S 0 } ∩ e 6= ∅ for all
e ∈ E, i.e. iff {v | Sv ∈ S 0 } is a Vertex Cover of G. Furthermore, since wt is constantly
0.5, we have O 0
m = 0.5k ≤ {wt(Sv ) | Sv ∈ S 0 } = 0.5|S | (8)
iff k ≥ |S 0 | = |{v | Sv ∈ S 0 }|. This shows that {v | Sv ∈ S 0 } is a solution cover to the
given instance of Vertex Cover iff S 0 is a solution to the constructed instance of ⊗-CS.
Since Vertex Cover is known to be NP-complete this shows that ⊗-CS is NP-hard. Since
we already know that it is contained in NP we obtain NP-completeness.
Lemma 2. If ⊗ is the product t-norm then ⊗-CS is NP-hard, even when restricted to
constraints of type (2).
In the case of the Łukasiewicz t-norm nearly the same reduction can be used. The
k
only modification that needs to be made is that wt should to be constantly k+1 and
1
m = k+1 . Then again
1 O k
m= ≤ {wt(Sv ) | Sv ∈ S 0 } = max{0, · |S 0 | − (|S 0 | − 1)}
k+1 k+1
1
= max{0, 1 − |S 0 |} (9)
k+1
holds iff k ≥ |S 0 | = |{v | Sv ∈ S 0 }|.
Lemma 3. If ⊗ is the Łukasiewicz t-norm then ⊗-CS is NP-hard, even when restricted
to constraints of type (2).
The complexity results are summarized in Table 3.
8 Conclusion and Future Work
In this paper we have extended the framework from [9] for obtaining EL-concept
descriptions from text in natural language. The framework proposes a hybrid approach
where in a first text mining step description candidates are mined, and good candidates
Table 3. Summary of Complexity Results
without full definitions with full definitions
(1)–(3) (1)–(3) + (4’) (1)–(4) (1)–(3) (1)–(3) + (4’) (1)–(4)
CS, min-CS P P NP P NP NP
Ł-CS, Π-CS NP NP NP NP NP NP
are selected in the reasoning step by solving a constraint problem. We have theoretically
analyzed the complexity of the concept selection problem ⊗-CS. If the minimum t-norm
is used, the problem remains tractable for a restricted type of constraints. We have further
shown that its complexity increases from P to NP when the product or Łukasiewicz
t-norm are used instead of the minimum. Regardless of the accumulation function, the
problem is intractable for the full set of constraints.
In this paper, we have not presented an experimental evaluation. However, in an
earlier work such an evaluation has been performed for the case of the minimum t-norm
[9]. In this evaluation, a text corpus obtained from the web was used. The experiment
was restricted to frequent roles from the disease branch of of S NOMED CT. Concepts
were removed and then relearned from the text corpus using the presented approach. The
hybrid approach showed significant improvements when constraints were used over a
pure text mining approach: for most concepts the precision increased to 100%.
In the present work, the definition of the target concept was always obtained as a
conjunction over the selected candidates. This is a good approach when the candidates
are too general to describe the concept. It is, however, also possible that a candidate is
too specific. In this case it could make sense to consider the least common subsumer
(lcs) of the selected candidates, in order to generalize. For the least common subsumer
we do not expect the concept selection problem to be tractable, or even in NP, since even
the size of the lcs of a set of concept descriptions can be exponential in the size of the
set.
We also plan to investigate, how well the approach performs for ontologies other
than S NOMED CT. For the approach to be applicable, these ontologies need to satisfy a
number of requirements. For the text mining step it is necessary that
– the set of roles is small and stable, and
– there is already a large set of concept definitions available, which can be used to
train a classifier.
For the reasoning step a source for the constraints, such as a design manual, is required.
Finally, since for many classifiers the confidence values returned by the text mining
step are probabilistic in nature, one might investigate the use of probabilistic extensions
to EL.
References
1. A. R. Aronson and F.-M. Lang. An overview of metamap: historical perspective and recent
advances. Journal of the American Medical Informatics Association, 17(3):229–236, 2010.
2. F. Baader, S. Brandt, and C. Lutz. Pushing the EL envelope. In Proceedings of IJCAI’05,
2005.
3. M. Chitsaz, K. Wang, M. Blumenstein, and G. Qi. Concept learning for EL++ by refinement
and reinforcement. In Proceedings of PRICAI’12, pages 15–26, 2012.
4. P. Cimiano. Ontology learning and population from text - algorithms, evaluation and applica-
tions. Springer, 2006.
5. M. A. Hearst. Automatic acquisition of hyponyms from large text corpora. In Proceedings of
Coling’92, pages 539–545. Association for Computational Linguistics, 1992.
6. International Health Terminology Standards Development Organisation. SNOMED CT user
guide. http://www.snomed.org/ug.pdf, 2013.
7. E. P. Klement, R. Mesiar, and E. Pap. Triangular Norms. Springer-Verlag, 2000.
8. J. Lehmann and P. Hitzler. Concept learning in description logics using refinement operators.
Machine Learning, 78(1-2):203–250, 2010.
9. Y. Ma and F. Distel. Concept adjustment for description logics. In Proceedings of K-Cap’13,
2013. to appear.
10. Y. Ma and F. Distel. Learning formal definitions for Snomed CT from text. In Proceedings of
AIME’13, 2013. to appear.
11. M. Mintz, S. Bills, R. Snow, and D. Jurafsky. Distant supervision for relation extraction
without labeled data. In Proceedings of ACL/AFNLP’09, pages 1003–1011, 2009.
12. B. Motik, P. F. Patel-Schneider, and B. Parsia. OWL 2 web ontology language structural
specification and functional style syntax. W3C Recommendation, October 2009.
13. S. Rudolph, J. Völker, and P. Hitzler. Supporting lexical ontology learning by relational
exploration. In Conceptual Structures: Knowledge Architectures for Smart Applications,
pages 488–491. Springer, 2007.
14. D. Sánchez and A. Moreno. Automatic generation of taxonomies from the www. In Practical
Aspects of Knowledge Management, pages 208–219. Springer, 2004.
15. S NOMED Clinical Terms. Northfield, IL: College of American Pathologists, 2006.
16. G. Tsatsaronis, A. Petrova, M. Kissa, Y. Ma, F. Distel, F. Baader, and M. Schroeder. Learning
formal definitions for biomedical concepts. In Proceedings of OWLED’13, 2013. to appear.
17. L. Vasto Terrientes, A. Moreno, and D. Sánchez. Discovery of relation axioms from the
web. In Knowledge Science, Engineering and Management, volume 6291 of Lecture Notes in
Computer Science, pages 222–233. Springer, 2010.
18. J. Völker. Learning expressive ontologies. PhD thesis, Universität Karlsruhe, 2009.
19. J. Völker, P. Haase, and P. Hitzler. Learning expressive ontologies. In Proceeding of OLP’08,
pages 45–69, 2008.
20. T. Wächter, G. Fabian, and M. Schroeder. DOG4DAG: semi-automated ontology generation
in OBO-Edit and protégé. In Proceedings of SWAT4LS’11, pages 119–120, 2011.