Natural languages exhibit in general phenomena such as hyponymy/hypernymy, synonymy, heteronymy, etc. These linguistic phenomena lead to some problems in the IR field, e.g. term mismatch. In this talk, I show how these phenomena or their related problems make the direct word-based document-query comparison insufficient, which is the case in most classical IR models. More precisely, I talk about using concepts instead of words to represent the content of documents and queries. Consequently, we do matching in a more informative space rather than the word space, where the effects of some linguistic phenomena are limited. I mainly talk about two problems that appear when moving to the conceptual space, namely the document length deformation and inter-concept relations quantification. This presentation is mainly based on my talk at University of Lugano (Università della Svizzera italiana).

1. Karam Abdulahhad
GESIS - Cologne
karam.abdulahhad@gesis.org
karam.abdulahhad@gmail.com
Beyond Classical Information Retrieval (IR)
Conceptual IR

2. Linguistic phenomena & IR problems
20-12-2018GESIS - K.Abdulahhad2
How have “fiddles” changed over time
Violins
Like most technological breakthroughs, today's
violin is an evolutionary product. So far as we
know, there were no violins in 1500. A century
later, there were several types and probably
thousands of specimens north and south of the
Alps, and from England to Poland. A marvel of
craftsmanship and acoustical engineering, the
violin produced more sound than any stringed
instrument to date. Almost immediately,
composers, players and collectors liked what
they heard and saw. Italian and non-Italian
makers proliferated.
……….

3. Linguistic phenomena & IR problems
20-12-2018GESIS - K.Abdulahhad3
Historical information about “sugar
river bank”
History and Mission Statement
…………
The Bank continues to grow at a healthy pace.
We have continued to do well and be a leader
in our industry. Our main branch was expanded
in 1982 and we now have branches in Sunapee,
New London, Warner, Grantham and Concord.
We at Sugar River Bank are proud of our
history and growth. It is the responsibility of
each and every member of our Bank's family to
insure continued growth in the future.
…………
www.sugarriverbank.com

4. Linguistic phenomena & IR problems
20-12-2018GESIS - K.Abdulahhad4
Historical information about “sugar
river bank”
The Life-Sustaining Sugar River
…………
The west branch of the Sugar River historically
supported a native trout population, but had
suffered from sedimentation, overgrazing of its
banks and warming water. “Restoration efforts
in the Dane County portion of the watershed
reduced nonpoint source pollution, installed
riverbank vegetative filter strips, improved in-
stream habitat, restricted cattle access to
streams, and improved management of animal
waste from barnyards,” says Hansis.
…………
northwestquarterly.com

5. Linguistic phenomena & IR problems
20-12-2018GESIS - K.Abdulahhad5
Part-Whole
Hand Body
Heteronyms
Bank(com) Bank(geo)
Hyponym / Hypernym
B-cell Lymphocyte
Synonyms
Violin Fiddle
Co-hyponym
Cat Dog

6. Observations
1. Inadequacy of the term-independence assumption,
which leads to the term-mismatch problem
20-12-2018GESIS - K.Abdulahhad6

7. Observations
1. Inadequacy of the term-independence assumption,
which leads to the term-mismatch problem
2. Retrieval process has an inferential nature, where the
classical word-based document-query comparison
paradigm is insufficient
20-12-2018GESIS - K.Abdulahhad7

8. 20-12-2018GESIS - K.Abdulahhad8
Conceptual approach

9. Conceptual approach
20-12-2018GESIS - K.Abdulahhad9
 Concepts are categories encompassing all synonymous
terms

10. Conceptual approach
20-12-2018GESIS - K.Abdulahhad10
 Concepts are categories encompassing all synonymous
terms
Atrial fibrillation
Auricular fibrillation
C0004238
Ticker
Watch
S04563183
Cancer
Malignant neoplastic disease
S14263400
WordNet
Snake
Serpent
Ophidian
S01729333
UMLS
Skin cancer
Melanoma
Malignant neoplasm of skin
C0004238

11. Conceptual approach
20-12-2018GESIS - K.Abdulahhad11
 Concepts are categories encompassing all synonymous
terms
Using concepts IDs
instead of terms
Atrial fibrillation
Auricular fibrillation
C0004238
Ticker
Watch
S04563183
Cancer
Malignant neoplastic disease
S14263400
WordNet
Snake
Serpent
Ophidian
S01729333
UMLS
Skin cancer
Melanoma
Malignant neoplasm of skin
C0004238

12. 20-12-2018GESIS - K.Abdulahhad12
Part I: Relative Concept Frequency
[1] K. . Abdulahhad et al., Revisiting the Term Frequency in concept-Based IR Models. DEXA 2013
[2] K. . Abdulahhad et al., MRIM at ImageCLEF2012. From Words to Concepts: A New Counting Approach.
CLEF 2012

13. Relative Concept Frequency (problem)
20-12-2018GESIS - K.Abdulahhad13
 Text to concepts mapping

14. Relative Concept Frequency (problem)
20-12-2018GESIS - K.Abdulahhad14
 Text to concepts mapping
 Using MetaMap & UMLS concepts

15. Relative Concept Frequency (problem)
20-12-2018GESIS - K.Abdulahhad15
 Text to concepts mapping
 Using MetaMap & UMLS concepts
Precision
Recall

16. Relative Concept Frequency (problem)
GESIS - K.Abdulahhad16
Word-space Concept-space
𝑑 =‘lobar pneumonia x-ray’
𝑑 = 3 𝑑 =? ?
 Document length
20-12-2018

17. Relative Concept Frequency (idea)
 Use all concepts but maintaining word-based document
length
 Structure based redistribution of word-based document
length on concepts
GESIS - K.Abdulahhad17 20-12-2018

18. Relative Concept Frequency (how)
 Computing relative frequency
GESIS - K.Abdulahhad18 20-12-2018

19. Relative Concept Frequency (how)
 Computing relative frequency
 Hypothesis 1: concepts of
larger phrase receive larger
count (more specific meaning)
GESIS - K.Abdulahhad19 20-12-2018

20. Relative Concept Frequency (how)
 Computing relative frequency
 Hypothesis 1: concepts of
larger phrase receive larger
count (more specific meaning)
 Hypothesis 2: the bigger the
set of concepts is for a phrase,
the less important count its
concepts receive (ambiguity)
GESIS - K.Abdulahhad20 20-12-2018

21. Relative Concept Frequency (how)
 Computing relative frequency
 Hypothesis 1: concepts of
larger phrase receive larger
count (more specific meaning)
 Hypothesis 2: the bigger the
set of concepts is for a phrase,
the less important count its
concepts receive (ambiguity)
 Hypothesis 3: maintaining
word-based 𝑑
GESIS - K.Abdulahhad21 20-12-2018

22. Computing Relative Concept Frequency
(Step 1)
 Step 1: map text to concepts (via e.g. MetaMap)
GESIS - K.Abdulahhad22
Sub-phrases Concepts
𝑇1:‘lobar pneumonia’ 𝐶1 = 𝐶0032300, 𝐶0155862
𝑇2:‘pneumonia x-ray’ 𝐶2 = 𝐶0581647
𝑇3:‘lobar’ 𝐶3 =
𝐶1511010, 𝐶1428707,
𝐶0796494
𝑇4:‘pneumonia’ 𝐶4 =
𝐶0024109, 𝐶1278908,
𝐶0032285, 𝐶2707265,
𝐶2709248
𝑇5:‘x-ray’ 𝐶4 =
𝐶0034571, 𝐶0043299,
𝐶0043309, 𝐶1306645,
𝐶1714805, 𝐶1962945
‘lobar pneumonia x-ray’
MetaMap
20-12-2018

23. Computing Relative Concept Frequency
(Step 2)
 Step 2: build hierarchy
GESIS - K.Abdulahhad23
𝑇𝑖, 𝐶𝑖 < 𝑇𝑗, 𝐶𝑗 ⇔ 𝑇𝑖 ⊂ 𝑇𝑗
 11,CT
R
 22 ,CT
 33,CT  44 ,CT  55,CT
Virtual node
Sub-phrases Concepts
𝑇1:‘lobar pneumonia’ 𝐶1 = 𝐶0032300, 𝐶0155862
𝑇2:‘pneumonia x-ray’ 𝐶2 = 𝐶0581647
𝑇3:‘lobar’ 𝐶3 =
𝐶1511010, 𝐶1428707,
𝐶0796494
𝑇4:‘pneumonia’ 𝐶4 =
𝐶0024109, 𝐶1278908,
𝐶0032285, 𝐶2707265,
𝐶2709248
𝑇5:‘x-ray’ 𝐶4 =
𝐶0034571, 𝐶0043299,
𝐶0043309, 𝐶1306645,
𝐶1714805, 𝐶1962945
20-12-2018

24. Computing Relative Concept Frequency
(Step 3)
 Step 3: compute relative frequency 𝑟𝑓𝑖
 Breadth first search
 The relative frequency 𝑟𝑓𝑖 of 𝑐 ∈ 𝐶𝑖 must be
propositional to 𝑇𝑖 (Hypothesis 1), and inversely
propositional to 𝐶𝑖 (Hypothesis 2)
 Maintaining 𝑑 by distributing it on the concepts of 𝑑
(Hypothesis 3).
GESIS - K.Abdulahhad24
Sub-phrases Concept 𝒓𝒇𝒊
𝑇1:‘lobar pneumonia’
𝑇1 = 2, 𝐶1 = 2
𝐶0032300
𝐶0155862
𝑇2:‘pneumonia x-ray’
𝑇2 = 2, 𝐶2 = 1
𝐶0581647
𝑇3:‘lobar’
𝑇3 = 1, 𝐶3 =3
𝐶1511010
𝐶1428707
𝐶0796494
𝑇4:‘pneumonia’
𝑇4 = 1, 𝐶4 =5
𝐶0024109
𝐶1278908
𝐶0032285
𝐶2707265
𝐶2709248
𝑇5:‘x-ray’
𝑇5 = 1, 𝐶5 =6
𝐶0034571
𝐶0043299
𝐶0043309
𝐶1306645
𝐶1714805
𝐶1962945
 11,CT
R
 22 ,CT
 33,CT  44 ,CT  55 ,CT
20-12-2018

25. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad25
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
3
20-12-2018

26. Computing Relative Concept Frequency
(Step 3)
 Step 3: computing relative weight
 For each node 𝑇𝑖, 𝐶𝑖 we compute three values
 𝛼𝑖 the amount that should be distributed on the concepts of the current node 𝑇𝑖, 𝐶𝑖 and
its children
 𝛼𝑖 = 𝛿 𝑝𝑎𝑟𝑒𝑛𝑡 × 𝑇𝑖𝑝𝑎𝑟𝑒𝑛𝑡𝑠
 𝛿𝑖 the portion of one single word of the input amount 𝛼𝑖
 𝛿𝑖 =
𝛼 𝑖
𝑇 𝑖 + 𝑇 𝑐ℎ𝑖𝑙𝑑𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛
 𝛽𝑖, or equivalently 𝑟𝑓𝑖, the relative frequency of each concept 𝑐 ∈ 𝐶𝑖
 𝛽𝑖 =
𝛿 𝑖× 𝑇𝑖
𝐶 𝑖
GESIS - K.Abdulahhad26 20-12-2018

27. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad27
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅 = 3
𝛼 𝑅 3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
𝛿1
𝛽1
𝛼2
𝛿2
𝛽2
𝛼3
𝛿3
𝛽3
𝛼4
𝛿4
𝛽4
𝛼5
𝛿5
𝛽5
𝛿 𝑅 =
𝛼 𝑅
𝑇𝑅 + 𝑇1 + 𝑇2
=
3
4
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
𝛼 𝑅 = 3
20-12-2018

28. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad28
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅
3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
3
2
𝛿1
3
8
𝛽1
3
8
𝛼2
𝛿2
𝛽2
𝛼3
𝛿3
𝛽3
𝛼4
𝛿4
𝛽4
𝛼5
𝛿5
𝛽5
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
𝛼1 = 𝛿 𝑅 × 𝑇1 =
3
2
𝛿1 =
𝛼1
𝑇1 + 𝑇3 + 𝑇4
=
3
8
𝛽1 =
𝛿1 × 𝑇1
𝐶1
=
3
8
20-12-2018

29. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad29
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅
3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
3
2
𝛿1
3
8
𝛽1
3
8
𝛼2
3
2
𝛿2
3
8
𝛽2
3
4
𝛼3
𝛿3
𝛽3
𝛼4
𝛿4
𝛽4
𝛼5
𝛿5
𝛽5
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
𝛼2 = 𝛿 𝑅 × 𝑇2 =
3
2
𝛿2 =
𝛼2
𝑇2 + 𝑇4 + 𝑇5
=
3
8
𝛽2 =
𝛿2 × 𝑇2
𝐶2
=
3
4
20-12-2018

30. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad30
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅
3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
3
2
𝛿1
3
8
𝛽1
3
8
𝛼2
3
2
𝛿2
3
8
𝛽2
3
4
𝛼3
3
8
𝛿3
3
8
𝛽3
1
8
𝛼4
𝛿4
𝛽4
𝛼5
𝛿5
𝛽5
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
𝛼3 = 𝛿1 × 𝑇3 =
3
8
𝛿3 =
𝛼3
𝑇3
=
3
8
𝛽3 =
𝛿3 × 𝑇3
𝐶3
=
1
8
20-12-2018

31. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad31
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅
3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
3
2
𝛿1
3
8
𝛽1
3
8
𝛼2
3
2
𝛿2
3
8
𝛽2
3
4
𝛼3
3
8
𝛿3
3
8
𝛽3
1
8
𝛼4
3
4
𝛿4
3
4
𝛽4
3
20
𝛼5
𝛿5
𝛽5
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
𝛼4 = 𝛿1 × 𝑇4 + 𝛿2 × 𝑇4 =
3
4
𝛿4 =
𝛼4
𝑇4
=
3
4
𝛽4 =
𝛿4 × 𝑇4
𝐶4
=
3
20
20-12-2018

32. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of the phrase ‘lobar pneumonia
x-ray’ on its concepts
GESIS - K.Abdulahhad32
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅
3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
3
2
𝛿1
3
8
𝛽1
3
8
𝛼2
3
2
𝛿2
3
8
𝛽2
3
4
𝛼3
3
8
𝛿3
3
8
𝛽3
1
8
𝛼4
3
4
𝛿4
3
4
𝛽4
3
20
𝛼5
3
8
𝛿5
3
8
𝛽5
1
16
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
𝛼5 = 𝛿2 × 𝑇5 =
3
8
𝛿5 =
𝛼5
𝑇5
=
3
8
𝛽5 =
𝛿5 × 𝑇5
𝐶5
=
1
16
20-12-2018

33. Computing Relative Concept Frequency
(Step 3)
 We distribute the 𝑑 = 3 of phrase
‘lobar pneumonia x-ray’ on its concepts
GESIS - K.Abdulahhad33
 11,CT
R
 22,CT
 33,CT  44,CT  55,CT
𝛼 𝑅
3
𝛿 𝑅
3
4
𝛽 𝑅
𝛼1
3
2
𝛿1
3
8
𝛽1
3
8
𝛼2
3
2
𝛿2
3
8
𝛽2
3
4
𝛼3
3
8
𝛿3
3
8
𝛽3
1
8
𝛼4
3
4
𝛿4
3
4
𝛽4
3
20
𝛼5
3
8
𝛿5
3
8
𝛽5
1
16
𝑇𝑅 = 0
𝐶 𝑅 = 0
𝑇2 = 2
𝐶2 = 1
𝑇1 = 2
𝐶1 = 2
𝑇4 = 1
𝐶4 = 5
𝑇5 = 1
𝐶5 = 6
𝑇3 = 1
𝐶3 = 3
20-12-2018
Sub-phrases Concept 𝒓𝒇𝒊
𝑇1:‘lobar pneumonia’
𝑇1 = 2, 𝐶1 = 2
𝐶0032300 3
8𝐶0155862
𝑇2:‘pneumonia x-ray’
𝑇2 = 2, 𝐶2 = 1
𝐶0581647
3
4
𝑇3:‘lobar’
𝑇3 = 1, 𝐶3 =3
𝐶1511010
1
8
𝐶1428707
𝐶0796494
𝑇4:‘pneumonia’
𝑇4 = 1, 𝐶4 =5
𝐶0024109
3
20
𝐶1278908
𝐶0032285
𝐶2707265
𝐶2709248
𝑇5:‘x-ray’
𝑇5 = 1, 𝐶5 =6
𝐶0034571
1
16
𝐶0043299
𝐶0043309
𝐶1306645
𝐶1714805
𝐶1962945

34. Sub-phrases Concept 𝒓𝒇𝒊
𝑇1:‘lobar pneumonia’
𝑇1 = 2, 𝐶1 = 2
𝐶0032300 3
8𝐶0155862
𝑇2:‘pneumonia x-ray’
𝑇2 = 2, 𝐶2 = 1
𝐶0581647
3
4
𝑇3:‘lobar’
𝑇3 = 1, 𝐶3 =3
𝐶1511010
1
8
𝐶1428707
𝐶0796494
𝑇4:‘pneumonia’
𝑇4 = 1, 𝐶4 =5
𝐶0024109
3
20
𝐶1278908
𝐶0032285
𝐶2707265
𝐶2709248
𝑇5:‘x-ray’
𝑇5 = 1, 𝐶5 =6
𝐶0034571
1
16
𝐶0043299
𝐶0043309
𝐶1306645
𝐶1714805
𝐶1962945
Computing Relative Concept Frequency
(Step 3)
GESIS - K.Abdulahhad34 20-12-2018

35. Sub-phrases Concept 𝒓𝒇𝒊
𝑇1:‘lobar pneumonia’
𝑇1 = 2, 𝐶1 = 2
𝐶0032300 3
8𝐶0155862
𝑇2:‘pneumonia x-ray’
𝑇2 = 2, 𝐶2 = 1
𝐶0581647
3
4
𝑇3:‘lobar’
𝑇3 = 1, 𝐶3 =3
𝐶1511010
1
8
𝐶1428707
𝐶0796494
𝑇4:‘pneumonia’
𝑇4 = 1, 𝐶4 =5
𝐶0024109
3
20
𝐶1278908
𝐶0032285
𝐶2707265
𝐶2709248
𝑇5:‘x-ray’
𝑇5 = 1, 𝐶5 =6
𝐶0034571
1
16
𝐶0043299
𝐶0043309
𝐶1306645
𝐶1714805
𝐶1962945
Computing Relative Concept Frequency
(Step 3)
 From this table, we can see that the
concepts of less ambiguous and
longest phrase have the highest
frequency
GESIS - K.Abdulahhad35 20-12-2018

36. Sub-phrases Concept 𝒓𝒇𝒊
𝑇1:‘lobar pneumonia’
𝑇1 = 2, 𝐶1 = 2
𝐶0032300 3
8𝐶0155862
𝑇2:‘pneumonia x-ray’
𝑇2 = 2, 𝐶2 = 1
𝐶0581647
3
4
𝑇3:‘lobar’
𝑇3 = 1, 𝐶3 =3
𝐶1511010
1
8
𝐶1428707
𝐶0796494
𝑇4:‘pneumonia’
𝑇4 = 1, 𝐶4 =5
𝐶0024109
3
20
𝐶1278908
𝐶0032285
𝐶2707265
𝐶2709248
𝑇5:‘x-ray’
𝑇5 = 1, 𝐶5 =6
𝐶0034571
1
16
𝐶0043299
𝐶0043309
𝐶1306645
𝐶1714805
𝐶1962945
Computing Relative Concept Frequency
(Step 3)
 From this table, we can see that the
concepts of less ambiguous and
longest phrase have the highest
frequency
 Concepts of most ambiguous and
shortest phrase have the lowest
frequency
GESIS - K.Abdulahhad36 20-12-2018

37. Sub-phrases Concept 𝒓𝒇𝒊
𝑇1:‘lobar pneumonia’
𝑇1 = 2, 𝐶1 = 2
𝐶0032300 3
8𝐶0155862
𝑇2:‘pneumonia x-ray’
𝑇2 = 2, 𝐶2 = 1
𝐶0581647
3
4
𝑇3:‘lobar’
𝑇3 = 1, 𝐶3 =3
𝐶1511010
1
8
𝐶1428707
𝐶0796494
𝑇4:‘pneumonia’
𝑇4 = 1, 𝐶4 =5
𝐶0024109
3
20
𝐶1278908
𝐶0032285
𝐶2707265
𝐶2709248
𝑇5:‘x-ray’
𝑇5 = 1, 𝐶5 =6
𝐶0034571
1
16
𝐶0043299
𝐶0043309
𝐶1306645
𝐶1714805
𝐶1962945
Computing Relative Concept Frequency
(Step 3)
 From this table, we can see that the
concepts of less ambiguous and
longest phrase have the highest
frequency
 Concepts of most ambiguous and
shortest phrase have the lowest
frequency
GESIS - K.Abdulahhad37 20-12-2018
𝑟𝑓𝑖 = 3

38. Relative Concept Frequency (results)
 Corpora
GESIS - K.Abdulahhad38 20-12-2018
104.26

39. Relative Concept Frequency (results)
GESIS - K.Abdulahhad39 20-12-2018
(*) indicates to statistically significant (𝛼 < 0.05) improvement w.r.t. classical concepts frequency
TF

40. Relative Concept Frequency (results)
GESIS - K.Abdulahhad40 20-12-2018

41. Relative Concept Frequency (conclusion)
 Dealing with the document length deformation
 Encouraging results
 Increase recall
 Maintain or even increase the precision
 Can be used with classical IR models
 Change the (TF) component
GESIS - K.Abdulahhad41 20-12-2018

42. 20-12-2018GESIS - K.Abdulahhad42
Part II: Concept Embedding
[3] K. Abdulahhad, Concept embedding for information retrieval. ECIR 2018

43. Concept embedding (problem)
20-12-2018GESIS - K.Abdulahhad43
fiddle violinS04544161
C0004238 skin cancermelanoma

44. Concept embedding (problem)
20-12-2018GESIS - K.Abdulahhad44
fiddle violinS04544161
C0004238 skin cancermelanoma
B-celllymphocyte C0004561 C0024264

45. Concept embedding (problem)
20-12-2018GESIS - K.Abdulahhad45
fiddle violinS04544161
C0004238 skin cancermelanoma
B-celllymphocyte C0004561 C0024264
is-a

46. Concept embedding (problem)
20-12-2018GESIS - K.Abdulahhad46
fiddle violinS04544161
C0004238 skin cancermelanoma
B-celllymphocyte C0004561 C0024264
is-a
Relation-based concept similarity is problematic

47. Concept embedding (problem)
20-12-2018GESIS - K.Abdulahhad47
fiddle violinS04544161
C0004238 skin cancermelanoma
B-celllymphocyte C0004561 C0024264
is-a
Relation-based concept similarity is problematic
fiddle violin
B-cell lymphocyte
handbody
is-a
part-of
Relations have different
semantics & properties
synonymous

48. Concept embedding (idea)
20-12-2018GESIS - K.Abdulahhad48
 Concepts as vectors
 Still using concepts to reduce mismatch effect
 Avoiding the complexities of relation-based inter-
concept similarity

49. Concept embedding (idea)
20-12-2018GESIS - K.Abdulahhad49
 Concepts as vectors
 Still using concepts to reduce mismatch effect
 Avoiding the complexities of relation-based inter-
concept similarity
Check adaptability of concept-embedding-based
similarity to IR
Goal

50. Concept embedding (approaches)
20-12-2018GESIS - K.Abdulahhad50
 Flat embedding
⋯
𝑐 = 𝐹 𝑤1, ⋯ , 𝑤 𝑛 𝑐
𝑤1 𝑤 𝑛

51. Concept embedding (approaches)
20-12-2018GESIS - K.Abdulahhad51
 Hierarchical embedding
⋮
⋮⋯
𝑐
⋯
⋯
⋮
⋮
⋮
⋮
𝑤1 𝑤 𝑛
𝑠1 𝑠 𝑚
𝑡1 𝑡 𝑘
𝑠𝑖 = 𝐹 𝑤1
𝑖
, ⋯ , 𝑤 𝑛
𝑖
𝑡𝑗 = 𝐹 𝑠1
𝑗
, ⋯ , 𝑠 𝑚
𝑗
𝑐 = 𝐹 𝑡1, ⋯ , 𝑡 𝑘

52. Concept embedding (approaches)
20-12-2018GESIS - K.Abdulahhad52
 Weighted embedding
𝑐 = 𝐹 𝛼1 𝑤1, ⋯ , 𝛼 𝑛 𝑤 𝑛 𝑐
𝑤1 𝑤 𝑛⋯

53. Concept embedding (experiments)
20-12-2018GESIS - K.Abdulahhad53
 Experiments consist of two parts
 Generating concept embedding vectors
 Testing a vector-based concept similarity for ad-hoc IR

54. Concept embedding (experiments)
1. Generating concept embedding vectors
20-12-2018GESIS - K.Abdulahhad54
 Word embedding
 PubMed Central collection (1177879 vocabularies)
 Word2Vec
 Vector size 500
 Continuous bag of words
 Window size 8
 Negative sampling 25

55. Concept embedding (experiments)
1. Generating concept embedding vectors
20-12-2018GESIS - K.Abdulahhad55
 Concept embedding
 UMLS2017 concepts (only English content)
 For each concept, we build the corresponding set of words
 Flat embedding
 Replace F by avg
 Hierarchical embedding
 Replace F by avg
 Weighted embedding
 Replace F by weighted-avg
 The weight 𝛼 𝑤 of a word w is: 𝛼 𝑤 = ln
𝑁+1
𝑛
 N the number of documents in PubMed Central
 n is the document frequency of w in PubMed Central

56. Concept embedding (experiments)
1. Generating concept embedding vectors
20-12-2018GESIS - K.Abdulahhad56
 Concept embedding (missing words)
 Fixed random vectors
 Several experiments for weighting missing words
 The word is too popular n = N (poor idf)
 The word is too rare n = 1 (high idf)
 Or in between n = N/2

57. Concept embedding (experiments)
2. Testing a vector-based concept similarity for ad-hoc IR
20-12-2018GESIS - K.Abdulahhad57
 Corpora
 clef11 & clef12
 Text to concepts mapping
 MetaMap
 UMLS concepts

58. Concept embedding (experiments)
2. Testing a vector-based concept similarity for ad-hoc IR
20-12-2018GESIS - K.Abdulahhad58
 IR model and concept similarity
𝑅𝑆𝑉 𝑑, 𝑞 = 𝑤𝑒𝑖𝑔ℎ𝑡 𝑞 𝑐 × 𝑠𝑖𝑚 𝑐, 𝑐∗ × 𝑤𝑒𝑖𝑔ℎ𝑡 𝑑 𝑐∗
𝑐∈𝑞

59. Concept embedding (experiments)
2. Testing a vector-based concept similarity for ad-hoc IR
20-12-2018GESIS - K.Abdulahhad59
 IR model and concept similarity
 Weight(c): BM25 and Pivoted Normalization
𝑅𝑆𝑉 𝑑, 𝑞 = 𝑤𝑒𝑖𝑔ℎ𝑡 𝑞 𝑐 × 𝑠𝑖𝑚 𝑐, 𝑐∗ × 𝑤𝑒𝑖𝑔ℎ𝑡 𝑑 𝑐∗
𝑐∈𝑞

60. Concept embedding (experiments)
2. Testing a vector-based concept similarity for ad-hoc IR
20-12-2018GESIS - K.Abdulahhad60
 IR model and concept similarity
 Weight(c): BM25 and Pivoted Normalization
 Concept similarity
𝑅𝑆𝑉 𝑑, 𝑞 = 𝑤𝑒𝑖𝑔ℎ𝑡 𝑞 𝑐 × 𝑠𝑖𝑚 𝑐, 𝑐∗ × 𝑤𝑒𝑖𝑔ℎ𝑡 𝑑 𝑐∗
𝑐∈𝑞
𝑠𝑖𝑚 𝑐𝑖, 𝑐𝑗 =
0 cos 𝜃 ≤ 0
𝛽 × cos 𝜃 2 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

61. Concept embedding (experiments)
2. Testing a vector-based concept similarity for ad-hoc IR
20-12-2018GESIS - K.Abdulahhad61
 IR model and concept similarity
 Weight(c): BM25 and Pivoted Normalization
 Concept similarity
 For comparison (Leacock)
𝑅𝑆𝑉 𝑑, 𝑞 = 𝑤𝑒𝑖𝑔ℎ𝑡 𝑞 𝑐 × 𝑠𝑖𝑚 𝑐, 𝑐∗ × 𝑤𝑒𝑖𝑔ℎ𝑡 𝑑 𝑐∗
𝑐∈𝑞
𝑠𝑖𝑚 𝑐𝑖, 𝑐𝑗 =
0 cos 𝜃 ≤ 0
𝛽 × cos 𝜃 2 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

62. Concept embedding (experiments)
2. Testing a vector-based concept similarity for ad-hoc IR
20-12-2018GESIS - K.Abdulahhad62
 Results
(*) indicates to statistically significant (𝛼 < 0.05) improvement w.r.t.“NoEmb-NoSim”
(†) indicates to statistically significant (𝛼 < 0.05) improvement w.r.t.“NoEmb-Leacock”

63. Concept embedding (conclusion)
20-12-2018GESIS - K.Abdulahhad63
 Three approaches to build concept vectors
based on word embedding
 Promising results to use vector-based concept
representation and similarity
 Concepts and words are represented in the
same vector space
 they are comparable
 Improve approaches like MetaMap

64. 20-12-2018GESIS - K.Abdulahhad64
Conclusion

65. Conclusion
 Dealing with the two observations
 Inadequacy of the term independence assumption
 Retrieval process has an inferential nature
 Conceptual IR
 Document length deformation
 Inter-concept relations quantification
20-12-2018GESIS - K.Abdulahhad65

66. 20-12-2018GESIS - K.Abdulahhad66
Thank you …