Plagiarism Detection as a Problem of Machine Learning

1

• Plagiarism Detection as a Problem of Machine
  Learning

Academician Yuri I. Zhuravlev Correspondent
member of RAS Konstantin V. Rudakov Gleb V.
Nikitov Computing Center of Russian Academy of
Sciences Forecsys Corporation
2
About the problem

• Detect citing in students papers
• Do it quickly and conveniently
• Do it qualitatively and with substantiation

3
Decisions

• Turnitin
• Mydropbox
• www.antiplagiat.ru

4
Working scheme
Collection of documents
Instructor
Antiplagiat
5
Searching domain
Internet
6
Already using

• Higher School of Economics
• Moscow Institute of Economics, Management and Law
• Moscow Pedagogical State University
• Moscow Municipal Psychological and Pedagogical
  Institute
• Nizhni Novgorod State University
• Academy of Budget and Treasury of the Russian
  Ministry of Finance

7
Non-educational use

• Higher Certifying Commission
• Russian State Library (ex-named after Lenin)

8
Negotiations

• Moscow State University
• Moscow Physical and Technical Institute
• Russian Academy of Justice
• International Academy of Enterprise

9
Quality and Performance

• Leading positions in operating speed not
  affecting quality of the results
• 70 thousands of registered users
• Generating about 20 thousands originality reports
  every day
• Continual improvement of searching algorithms and
  expanding functionality

10
Plagiarism, what is it?
11
Formulation of problem

• Permissible objects
• Descriptive functions
• Fixed set of functions

12
Formulation of problem

• The problem
• Initial information
• Final information

13
Formulation of problem

• Precedent information
• Precedent conditions

14
Formulation of problem

• Transitive and reflective relation
• Example

15
Formulation of problem

• Additional conditions

16
Criteria

• Solvability criteria
• For existence of the correct algorithm A it is
  necessary and sufficient that the following
  conditions are met

17
Criteria

• Regularity
• Definition (according to Zhuravlev). The problem
  Z is regular if all the problems with arbitrary
  final information are simultaneously solvable
• Regularity criteria
• For a problem to be regular it is necessary and
  sufficient that the following conditions are met

18
Criteria

• Monotonous solvability criteria
• For monotonous solvability of the problem it is
  necessary and sufficient that the conditions of
  solvability criteria are met and are also met the
  following conditions

19
Criteria

• Monotonous regularity criteria
• For monotonous solvability of the problem it is
  necessary and sufficient that are met the
  following conditions

20
Criteria

• Supercompleteness
• The family of algorithms M is called
  supercomplete in the described class of problems
  if for each problem Z from the set of solvable
  problems there exist in M at least one
  correct algorithm.

21
Criteria

• Completeness
• The family of algorithms M is called complete in
  the described class of problems if for each
  problem Z from the set of regular problems
  there exist in M at least one correct algorithm.

22
Criteria

• Supercompleteness criteria
• For the family of algorithmic operators to be
  supercomplete it is necessary and sufficient that
  the following conditions are met

23
Criteria

• Completeness criteria
• For the family of algorithmic operators to be
  complete it is necessary and sufficient that the
  following conditions are met