MemeStore - a Knowledge Base
Posted by Tsert.Com
The MemeStore
concept consists in using
traits,
as
in
personality traits,
in
the storing of data or memories, to build a knowledge
base as a weighted
graph. The criteria
used in the storage of memories are
traits,
as
well
as,
semantics,
and
relevance
based on frequency.
MemeStore - a Knowledge Base with Traits
Posted by Tsert.Com ThinkTank
The MemeStore concept is based on the way humans and
animals store
memories.
MemeStore is an adaptive knowledge base, which
may be
given a personality, through
the use of traits.
The whole concept is based on animals, man included, who evolved with the imperative
to survive and reproduce. These imperatives are controlled through hormones
and memories,
creating attractions and aversions -- related
to danger and safety, danger being anything to do
with predation
or any other type of harm; and safety being food, shelter, play, and reproduction--
which are stored as memories, and then drive
certain types of
behaviour; or predispose
the
animal to
certain types of behaviour.
A trait
is associated with the concept of like
or dislike,
acceptability
or unacceptability,
affinity
or conflict,
appealing
or unappealing,
attraction
or aversion.
Traits
are incorporated
into
the framework of the knowledge
base or meme
store, as a single word, a sentence which includes
the words or
concepts of like,
dislike,
affinity,
acceptability,
attraction,
aversion,
repulsion,
philia,
phobia,
hazard
or
safety;
such
as
'I
like
vegetables', or 'I
like
technology', a sentence which indicates a like
or/and
dislike, such as 'I
am
a
vegan', or a sentence which indicates a
modification in
behaviour when processing commands, or responding to sensory
stimuli,
such as 'be
lenient
when
...', or 'be
stubborn
when
...'.
The MemeStore memory framework is based on a weighted
graph; where each
node/vertex
in
the
knowledge
base
represents
a
word
or
concept;
and each edge represents the weighted relationship
between the vertices/nodes. The relevance of
each edge is indicated by the numeric value of
its weight.
The weight
of the relationships, between the vertices, are made to
vary, based on
the data or information, made of text, audio, and images,
that is
processed by
the knowledge
base or meme store.
The criteria
used in
attributing a weight to an edge are: first its semantic
value, that is the value computed by examining the
semantic
type of
the edge. Said types are isa,
hasa,
ako,
likes, etc..
The second criterion is the similarity
in
concept
where synonyms
are attributed a higher weight than antonyms.
The
third
criterion
is
relevance
re-enforcement; either, by a person browsing the meme store,
by the meme
store
itself, coming across the same related concept identified by
the edge
between the nodes, when reading or receiving sensory inputs.
The weight
of the relationships, between the nodes, are also made to
vary
according to the list of traits
that is assigned to the knowledge
base or meme store.
As data is
processed, any concept relating to a like
trait is attributed a higher weight; and any
concept relating to
a dislike
trait is attributed a lower weight when it comes to
search-engine
relevance.
All
traits
have an associated degree
of
relevance, which dictates how extracted concepts,
which are
related with said traits,
are
processed
and
stored.
For
adaptive
search-engines used in systems such as robots;
both
likes
and
dislikes are
treated equally
when it comes to relevance; since their relevance
level is
associated with the concept of safety and danger.
The MemeStore is made of several modules. The modules are
a character
recognition
(OCR)
module, a scanning/reading
module, a personality
trait/attribute
module, a memory
store
module, a graph
traversal
module, a path
overlay/memory
trail module, a query/command
module, a conversation
module, , a search
engine
module, a semantic
command
mapping
module, a visualization
module. and a reward
system
module.
The MemeStore uses our statistics-based text
scanning and heuristics algorithm (CETE),
using type
disambiguation
(patent
pending), to read
text
as
a
person, and build it's own knowledge base, with the
appropriately weighted links/edges. Rules
on how to
read dictionaries and thesauri are first incorporated into
the reading
module
of the
MemeStore.
The query
module, creates a small graph of the entered
keywords and their
relationships. Concepts are extracted from the graph. The
keyword graph
is signatured, using its set of nodes, edges, and weights.
The
signatures are kept in a database, and are used when
satisfying
searches. Keyword graphs, capturing the same concept or
concepts, will
have similar signatures.
Path
overlays (patent
pending) are the set of nodes and edges, which are
traversed
when a query is issued. They are signatured and stored in a
database.
They are used when satisfying searches through the knowledge base.
The reward
module, helps to reinforce the weight/relevance
of a given memory. It uses sensory inputs, and the concept
of emotions,
to implement a feedback system, where retrieval of a memory triggers a
set of behaviours; which in turn may trigger an emotional
response. An emotional response serves in reinforcing a memory; and,
can be triggered directly from the retrieval of said memory. In
nature, emotions are associated with hormonal triggers and sensory
inputs; in the MemeStore,
they are simply viewed as a graph consisting of links between memories,
behaviours, sensory inputs, and signals or triggers, akin to
hormonal triggers, pointing to a set of behaviours to express -- A lion taking a bite of a zebra's thigh, while endorphins flood the lion's brain, to re-inforce the chase behaviour.
The MemeStore has two types of memory stores. As for
humans and animals, the types of memory stores are a permanent
and a temporary
one. Memories that are kept in the temporary store, can
either be moved into the permanent store or deleted altogether. The
criteria used to decide, when memories are moved to the permanent store or
deleted, are based on how often the concepts they capture, are re-enforced
by user interaction, assigned
traits, or information
processed
by reading or sensors (i.e. heat, light, temperature, visual,
etc.). All extracted concepts which are
tagged, as dislikes,
are always moved first to the temporary store, depending on their
level of relevance.
The engine is built using propositional calculus,
-- A
proposition is a statement which may
be true or false -- and modeled using graph theory, and De Brujin
sequences.
A level of fuzzy
is added, consistng of the two meters dealing with like
and dislike.
The
propositional
calculus
engine, of the MemeStore,
can ultimately be
burned
onto an integrated circuit.
Our MemeStore
PCG engine cannot be
circumvented,
by
building a similar propositional calculus engine, using Bayes
theorems, formal grammar
or graph
theory
methodologies.
An adjunct to this
patent
is
the ability MemeStore
provides, which allows the examination of the evolution of
the knowledge base
over time, through the use of path
overlays, or memory
traces
or
trails; the paths that are traversed, through the
knowledge base, when retrieving a memory.
A second adjunct to
this
patent
is
the data-mining ability MemeStore
provides, which allows the extraction of relevant
information, by
examining the log/database of network
path
and semantic
signatures and associated data.
A third adjunct to this
patent
is the use of path
overlays, or memory
traces
or
trails; by the MemeStore
to satisfy search-queries made with keywords or natural
language. The memory
traces
or
trails are associated with
scanned web pages
or documents.
A fourth adjunct to
this patent
is the use of path
overlays, or memory
traces
or
trails; by the MemeStore
to allow the visualization of search-queries made with
keywords or
natural language
A fifth adjunct to
this patent
is
the ability MemeStore
has in retro-fitting old inference rule engines into its
framework, and
allow their sales as valuable knowledge bases. Medical, legal,
and
engineering knowledge bases can
thus be retro-fitted.
A sixth adjunct to
this patent
is
the ability MemeStore
has in using traits
to restrict certain updates in the knowledge base.
For example,
adding a stubborn
trait to MemeStore,
can prevent updates that are contrary to a primary trait.
A primary trait, is a trait, that has a large percentage of associated links,
based on a given threshold, in the knowledge base. An example of a stubborn trait is 'be
stubborn when
updating anything related to medical diagnosis',
or 'be
stubborn when
updating traits related to eating preferences'. These types of restrictions are
usually referred to as constraints.,
constraints
on
behaviour,
on
searches,
on
responses,
etc..
A seventh adjunct
to this patent
is
the ability MemeStore
has in categorizing and storing sensory inputs as experiential
memories; which are information, related to unusual or
relevant sensor
data, emanating from the particular task or activity being
performed by
an actor (i.e. robot, computer desktop, search engine, etc.)
using a MemeStore.
The
relevance
of
experiential
memories
can
also
be
categorized,
based
on
the list of assigned traits.
An eighth adjunct to this
patent
is
the capability, the MemeStore
has, of using 3-dimensional indexing in capturing and
storing the information associated the extracted concepts
and their relationships. The 3-dimensional index can then be
used to extract information based on a degree of similarity
of the semantics associated with a particular query.
Patent
Pending
Tsert.inc/Tsert.Com