Search for Appropriate Textual Information Sources Adam ALBERT 1 , - - PowerPoint PPT Presentation

Search for Appropriate Textual Information Sources

Adam ALBERT1, MARIE DUŽÍ1, Marek MENŠÍK1, Miroslav PAJR2, Vojtěch PATSCHKA1

1VSB-Technical University Ostrava, Department of Computer Science FEI

• 17. listopadu 15, 708 33 Ostrava, Czech Republic

2Silesian University in Opava, Institute of Computer Science,

Bezručovo nám. 13, 746 01 Opava, Czech Republic

Problem to be solved

• One aspect of globalization is the dissemination of knowledge
• There is a huge amount of information in the textual resources
• Search for relevant information resources in the labyrinth of input

textual data

• For instance, by googling ‘cat’ we obtain:
• Approximate number of results 3 180 000 000, of these types:
• Computer-assisted translation
• Well-known excavator brand
• Animal
• Too much information  information overload

How to deal with the problem

• Our system generates explications of the concept in question extracted

from many textual resources

• Background theory – Transparent Intensional Logic (TIL)
• Procedural semantics
• Concepts are defined as meaning procedures
• Explication of a concept is a molecular procedure defining the object in question;
• “Cat is a feline animal”  ‘Cat = wt x [[‘Feline ‘Animal]wt x]

explicandum explication

• Based on a chosen explication the system computes and recommends the

most relevant textual resources

• By applying a data mining method of association rules

Search for Appropriate Textual Information Sources

• Input: an atomic concept (explicandum) + textual resources
• Extraction and TIL formalization of sentences that mention the concept in

question (explicandum)

• Generating Carnapian explications
• Machine-learning methods applied to the formalized sentences
• Results -- molecular concepts, i.e. closed TIL constructions, that explicate the atomic

concept

• Evaluation of the results (relevant documents can be overlooked in large

amount of data):

• Checking inconsistencies and/or looking for similarities, etc.
• Based on associations between the constituents of the molecular concepts the

algorithm computes and recommends other relevant resources

Machine learning (generating explications)

• Symbolic method of supervised machine learning
• Based on positive / negative examples - inserting or adjusting

constituents of a molecular concept

• Three heuristic methods:
• Negative example  Specialization inserts negated concepts.
• Positive example  Refinement inserts new constituents into the molecular

construction learned so far

• Generalization adjusts the constituents.

Example; explication of Wild Cat

[‘Typ-p wt x [[‘ [‘Weightwt x] ‘11]  [‘ [‘Weightwt x] ‘1.2]] [‘Wild ‘Cat]]  [‘Req ‘Mammal [‘Wild ‘Cat]]  [‘Req ‘Has-fur [‘Wild ‘Cat]]  [‘Typ-p wt x [[‘ [[‘Average ‘Body-Length]wt x] ‘80]  [‘ [[‘Average ‘Body-Length]wt x] ‘47]] [‘Wild ‘Cat]]  [‘Typ-p wt x [‘= [[‘Average ‘Skull-Size]wt x] ’41.25] [‘Wild ‘Cat]]  [‘Typ-p wt x [‘= [[‘Average ‘Height]wt x] ’37.6] [‘Wild ‘Cat]]

Association rule 𝐵 ⟹ 𝐶

• Association between items occurring in a dataset that satisfies a predefined

minimal support and confidence.

• Support is an indication of how frequently the itemset appears in the dataset.

𝑡𝑣𝑞𝑞(𝐵) = 𝑢 ∈ 𝐸: 𝐵 ⊆ 𝑢 𝐸

• Confidence is an indication of how we can rely on the validity of the rule

𝑑𝑝𝑜𝑔 𝐵 ⟹ 𝐶 = 𝑡𝑣𝑞𝑞 𝐵 ∪ 𝐶 𝑡𝑣𝑞𝑞 𝐵

• By computing the rules that are valid at least with user-defined minimal

confidence, the algorithm proposes other textual resources that might be relevant as well.

Incidence matrix

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

e1

1 1 1 1 1 1 1 1

e2

1 1 1

e3

1 1 1 1 1 1

e4

1 1 1 1 1 1

e5

1 1 1 1

e6

1 1 1 1

e7

1 1 1 1 1 1 1

e8

1 1 1

• 1. ′𝑁𝑏𝑛𝑛𝑏𝑚
• 2. ′𝐼𝑏s−fur
• 3. 𝜇𝑥 𝜇𝑢 𝜇𝑦 ′≤ ′𝑋𝑓𝑗𝑕ℎ𝑢𝑥𝑢 𝑦

′11

• 4. 𝜇𝑥 𝜇𝑢 𝜇𝑦 ′≥ ′𝑋𝑓𝑗𝑕ℎ𝑢𝑥𝑢 𝑦

′1.2

• 5. 𝜇𝑥 𝜇𝑢 𝜇𝑦 ′≥

′𝐵𝑤𝑓𝑠𝑏𝑕𝑓 ′𝐶𝑝𝑒𝑧−𝑀𝑓𝑜𝑕𝑢ℎ 𝑥𝑢 𝑦 ′47

• 6. 𝜇𝑥 𝜇𝑢 𝜇𝑦 ′≤

′𝐵𝑤𝑓𝑠𝑏𝑕𝑓 ′𝐶𝑝𝑒𝑧−𝑀𝑓𝑜𝑕𝑢ℎ 𝑥𝑢 𝑦 ′80

• 7. 𝜇𝑥 𝜇𝑢 𝜇𝑦 ′=

′𝐵𝑤𝑓𝑠𝑏𝑕𝑓 ′𝑇𝑙𝑣𝑚−𝑇𝑗𝑨𝑓

𝑥𝑢 𝑦 ′41.25

• 8. 𝜇𝑥 𝜇𝑢 𝜇𝑦 ′=

′𝐵𝑤𝑓𝑠𝑏𝑕𝑓 ′𝐼𝑓𝑗𝑕ℎ𝑢

𝑥𝑢 𝑦 ′37.6

Simple example of the computed results

Wild-cat: atomic concept that has been explicated

• eight resources and thus eight explications
• User voted for the first one (e1), biological explication (mammal,

weight, body length, skull size, etc.)

• Confidence = 0.66
• The system computed s4 and s7 describing behaviour of wild cats

{′𝑁𝑏𝑛𝑛𝑏𝑚,′𝐼𝑏s-fur} ⟹𝑓1 {𝜇𝑥 𝜇𝑢 𝜇𝑦 [[′𝑈𝑓𝑠-𝑁𝑏𝑠𝑙𝑗𝑜𝑕𝑥𝑢 𝑦 ′𝐷𝑚𝑏𝑥𝑗𝑜𝑕 ] ∨ [′𝑈𝑓𝑠-𝑁𝑏𝑠𝑙𝑗𝑜𝑕𝑥𝑢 𝑦 ′𝑉𝑠𝑗𝑜𝑏𝑢𝑗𝑜𝑕 ] ∨ [′𝑈𝑓𝑠-𝑁𝑏𝑠𝑙𝑗𝑜𝑕𝑥𝑢 𝑦 ′𝑀𝑓𝑏𝑤𝑓𝑡-𝐸𝑠𝑝𝑞𝑞𝑗𝑜𝑕𝑡 ]]}

Thank you for your attentio ion