Timothy Mwangi

1
Timothy Mwangi Andrew Averhart
2
Motivation

• Aging
• It is well known that as we age our abilities to
  carry out activities of daily living begin to
  diminish.
• If you dont use it, you lose it!
• The main goal is to develop a tool to help impede
  the loss of function that is adaptable to the
  user.

3
Cognition

• What is Cognition?
• The American Heritage Dictionary defines it as,
  The mental process of knowing, including aspects
  such as awareness, perception, reasoning, and
  judgment.
• 6 Main Functions

4
6 Cognitive Functions

• Spatial Orientation - the ability to visualize
  and mentally rotate abstract figures in two and
  three dimensional space.
• Verbal Ability active and passive vocabulary
• Inductive Reasoning the ability to determine a
  rule or principle from individual instances
  probably involved in most human problem solving
• Number Fluency the ability to engage rapidly
  and correctly in a variety of computational
  operations
• Memory the ability to be able to freely recall
  information, usually measured by list learning
  with words or numbers.
• Processing Speed the ability to rapidly and
  accurately identify visual details, similarities
• and differences.

5
Cognitive Training

• Last summer
• Mental Aerobics
• Only had 3 functioning games
• Games only targeted 2 of the 6 cognitive
  functions

6
Personal Cognitive Trainer
7
Ubiquity

• The Cognitive Cross Trainer game is fully
  integrated with the operating system
• Most games of this nature are not used in a
  consistent manner
• Various games, activities and hints are
  introduced at different
• time intervals within the system

8
Training

• The Cognitive Cross Trainer introduces personal
  coaching within the game to help the player
  improve specific areas of cognitive function
• For example, in the Number Memory mini game, as
  the numbers continue to grow larger and larger,
  chunking is introduced. This is a tip that helps
  the player memorize larger numbers

9
Training

• Try to memorize the following number
• 741852963

Now try to memorize the same number in
chunks 744 852 963
10
Personalization

• Positive thinking improves cognition
• Can be triggered by pictures, interests,
  memories, etc.
• Research shows by incorporating such items,
  cognitive performance can be greatly improved.
• Therefore the purpose of personalization is to
  determine what can be learned about the user.

11
Personalization Requirements

• Minimal disruption
• Simple and efficient
• We dont need to know everything
• Privacy and security
• The program only look at the directory
• The information will not be accessible by any
  other program
• Run Offline

12
Personalization Strategy

• Word Frequency
• The program scans through the directory and
  identifies the frequency at which words appear
• A list of the 20 most frequently used words are
  collected for the game
• These words are then compared in WordNet for
  meaning and relatedness

13
WordNet

• Is a lexical database of the English language
• Nouns, verbs, and adjectives are grouped into
  synonyms or synsets that represent an individual
  or distinct idea
• The networking of these synsets into a
  hierarchical structure allows this tool to be
  used for computing linguistics and language
  processing.
• Internet accessible and freely downloadable

14
(No Transcript)
15
WordNet
16
WordNet

• The structure of WordNet allows us to measure the
  similarity and relatedness of different words and
  senses. This is known as semantic relatedness.
• This is useful in determining the relationships
  between words found in the directories.
• Semantic distance is a measurement of the
  dissimilarity between two words
• If we had a method to accurately assess this
  measurement, then we will be able to determine
  which words are related.

17
Semantic Distance

• Three Approaches of Measurement
• Path length
• Distance up a path from one word and down the
  path to another word
• Characterized by the distance and number of
  direction changes
• Direction changes occur when we change from the
  is a hierarchy to a part of hierarchy
• For example, cat is feline is a mammal is an
  animal which has a reptile which has a snake and
  fangs are part of snake.
• Ex. Six Degrees of Separation

18
(No Transcript)
19
Semantic Distance

• Three Approaches of Measurement
• Information Content
• Lowest Common Super Class
• Shared knowledge
• The relatedness of the two words is dependent on
  the information content super class.
• Also based on the probability of finding the
  super class word in a corpus
• For example, appliance carries more information
  than entity.

20
(No Transcript)
21
Semantic Distance

• Three Approaches of Measurement
• Combined Approach
• Incorporates characteristics of Path length
• Incorporates characteristics of Information
  Content
• Common super class, probability of parent class
  as well as child node.
• Most implementations are usually a variation of
  this method.

22
Jiang-Conrath

• According to Budanitsky Hirst, the
  Jiang-Conrath implementation was most successful
  in evaluating semantic distance.
• They found that when compared to human-assessed
  relatedness, of the five different
  implementations Jiang-Conraths method proved
  most successful.
• Their implementation is based off the idea of
  information content, as well as conditional
  probability. This is the probability that the
  child synset will appear in a corpus given the
  parent synset appears.

23
Directory Scan

• After scanning the directory for words and their
  frequencies, we then applied the Jiang-Conrath
  method to evaluate their relatedness.
• The values we got are larger if the words are
  very related and closer to 0 if unrelated.

24
(No Transcript)
25
What is it that we gain from this?

• We get an idea of the words the user uses
• From these words we develop relationships between
  words
• Different groups of words imply different
  meanings
• Ex. knife wood craft carving
• knife spoon fork cutlery
• We can include these interests to the game
• When they are playing a word game, we can have
  the game use words that are related to their
  different interests to trigger positive feelings
  and improve performance.

26
Temporary Shortcomings

• File names may not necessarily be real words.
  Some names may even be abbreviated, which cannot
  be directly interpreted by the computer program.
• The amount of usable words vary between different
  PCs
• Important words that describe the players true
  interests may not be frequently used. If such
  words a part of the 20 most significant words,
  they cannot be used in personalizing the game.

27
Present Approach and Future Plans

• Create at least two additional programs, one
  that will be able to interpret abbreviations and
  acronyms, and the other able to tokenize words
  without spaces in between them.
• Implement new data mining techniques that not
  only determine the frequency of significant
  words, but also analyze the data structure for
  key words not frequently used.

28
References
Budanitsky, Alexander and Graeme Hirst. An
Experimental, Application-Oriented Evaluation of
Five Measures. Department of Computer Science.
University of Toronto Conrath, David and Jay
Jiang. Semantic Similarity Based on Corpus
Statistics and Lexical Taxonomy. Proceeding of
International Conference Research on
Computational Linguistics. 1997. Taiwan.
29
Questions and Suggestions