Machine learning (ML) is a core branch of AI that aims to give computers the ability to learn without being explicitly programmed (Samuel, 2000)…. [620277]

Machine Learning
Machine learning (ML) is a core branch of AI that aims to give computers the ability
to learn without being explicitly programmed (Samuel, 2000).
From: Artificial Intelligence in Behavioral and Mental Health Care , 2016
Related terms:
Schizophrenia, Artificial Intelligence , Data Mining, Learning Algorithm , Neural Net-
work, Neuroimaging, Neurons
View all Topics
Machine Learning in Transportation
Data Analytics
Parth Bhavsar, … Dimah Dera, in Data Analytics for Intelligent Transportation Sys-
tems, 2017
12.1 Introduction
Machine learning is a collection of methods that enable computers to automate
data-driven model building and programming through a systematic discovery of
statistically significant patterns in the available data. While machine learning
methods are gaining popularity, the first attempt to develop a machine that mimics
the behavior of a living creature was conducted by Thomas Ross in 1930s [1]. In,
1959 Arthur Samuel defined machine learning as a “Field of study that gives computers
the ability to learn without being explicitly programmed” [2]. While the demonstration
by Thomas Ross, then a student: [anonimizat], included a Robot Rat that can find a way through artificial maze
[1], the study presented by Arthur Samuel included methods to program a computer
“to behave in a way which, if done by human beings or animals, would be described as
involving the process of learning.” With the evolution of computing and communica-
tion technologies, it became possible to utilize these machine learning algorithms to
identify increasingly complex and hidden patterns in the data. Furthermore, it is now
possible to develop models that can automatically adapt to bigger and complex data

sets and help decision makers to estimate impacts of multiple plausible scenarios in
a real time.
The transportation system is evolving from a technology-driven independent system
to a data-driven integrated system of systems. For example, researchers are focusing
on improving existing Intelligent Transportation Systems (ITS) applications and
developing new ITS applications that rely on quality and size of the data [3]. With
the increased availability of data, it is now possible to identify patterns such as flow
of traffic in real time and behavior of an individual driver in various traffic flow
conditions to significantly improve efficiency of existing transportation system
operations and predict future trends. For example, providing real-time decision
support for incident management can help emergency responders in saving lives
as well as reducing incident recovery time. Various algorithms for self-driving cars
are another example of machine learning that already begins to significantly affect
the transportation system. In this case, the car (a machine) collects data through
various sensors and takes driving decisions to provide safe and efficient travel
experience to passengers. In both cases, machine learning methods search through
several data sets and utilize complex algorithms to identify patterns, take decisions,
and/or predict future trends.
Machine learning includes several methods and algorithms, some of them were de-
veloped before the term “machine learning” was defined and even today researchers
are improving existing methods and developing innovative and efficient methods.
It is beyond the scope of this book to provide in-depth review of these techniques.
This chapter provides brief overview of selected data preprocessing and machine
learning methods for ITS applications.
> Read full chapter
Resource Sharing
Ryan Litsey, in Resources Anytime, Anywhere , 2017
Predictive analytics and resource sharing
Machine learning and predictive analytics are a natural progression for libraries in
general and resource-sharing unit specifically. While a progression, for sure, the
ideas and processes are at least a decade away from being fully realized. Whenever
there are opportunities for progress currently, the methods of machine learning and
predictive analytics can provide a much needed insight and opportunity for viewing
the workflow in an entirely new way. To better understand how these cutting-edge

techniques can be incorporated into a resource-sharing unit, it is necessary to
understand what is meant by the terms machine learning and predictive analytics.
One thing to cover at the outset of the discussion. Machine learning is not artificial
intelligence. Machine learning is one aspect of the larger whole that could be un-
derstood as artificial intelligence. A good example of machine learning is: machine
learning is the books one would read, and artificial intelligence would be the paper
that is written after the books are read. The books serve as data that can be cataloged
and tracked by a learning machine. Artificial intelligence would take liberties with the
data and reassemble it into new information. Machine learning allows a machine to
interpret a set of data and learn from those interpretations to apply them to another
similar set of outputs. Artificial intelligence is a holistic way of computerized thinking
that can confront undetermined parameters and execute actions. The key distinction
in this example is awareness. Machine learning is a series of algorithms designed to
interpret data. Artificial intelligence is an awareness of self. Machine learning is the
focus for this new type of resource sharing. Resource-sharing units should think of
machine learning as an opportunity to analyze something libraries are already very
good at collecting, and that is large amounts of data. For a more concise definition
of machine learning, we can look at the work of Jason Bell. In the book, the author
defines machine learning as
Using computing, we design systems that can learn from data in a manner of being
trained. The systems might learn and improve with experience, and with time, refine
a model that can be used to predict outcomes of questions based on previous
learning.Bell (2015, p.2)
Machine learning, then, based on this definition, is the refinement of an algorithm
used to assess a large amount of data. Practically put, think of a filtration system: at
the first stage of any filter is a process to filter out large particles. The second stage
filters out smaller particles and each stage after that is able to filter finer and finer
materials until some pure remains. This is a type of machine learning. An algorithm
is constructed that is then applied to a data set. From the initial result of the first
data set the algorithm is run again and those two results are then compared to one
another to see if a refinement needs to be made. From that comparison, the model
is adjusted by the machine that is the learning component of machine learning.
The different successive models are often called generations. With each passing
generation, the algorithm refines its processes until the final result is a pure as it
possible with the data being collected. How can this apply to resource sharing?
Imagine if a resource-sharing unit were to run a predictive analytic, it would look
something like this. The machine would catalog all of the requests for a given time
frame. These requests would be grouped into generations, say a fall semester or a
spring semester. From that grouping, the machine would use predictive statistical

modeling to determine the probability that the same book would be requested
the next generation. That prediction would take the form of a percentage. For
example, the machine is 90% sure that the book would be requested next year. The
machine would also output a confidence interval; the confidence interval indicates
just that: the confidence the system has in its prediction. If the confidence interval
is low and the prediction is high that means the machine has determined a high
likelihood that the item would be requested the following semester. While having
one or two semesters or generations worth of data would not build a very confident
prediction, having several years or a decade’s worth the prediction would refine
itself to a narrower and narrower accuracy. From that, the system could opt to
put a recommendation of what the resource-sharing professional could do next.
It could suggest that the request be made ahead of time for example given the
probability that the item will be requested. If the model is accurate the requested
item would effectively have been ordered and be in the library around the time the
patron thought to request it. The resource-sharing unit would have essentially made
a request for an item before a patron knew they wanted it. It is less science fiction
and more of a mathematical process. While these are some of the more cutting edge
aspects of resource sharing, examining the case studies below will demonstrate how
these processes have been implemented and some preliminary results have been
gathered.
> Read full chapter
Inductive Logic
Ronald Ortner, Hannes Leitgeb, in Handbook of the History of Logic , 2011
1.1 Introduction
Machine Learning is concerned with algorithmic induction. Its aim is to develop
algorithms that are able to generalize from a given set of examples. This is quite
a general description, and Machine Learning is a wide field. Here we will confine
ourselves to two exemplary settings, viz. concept learning and sequence prediction.
In concept learning, the learner observes examples taken from some instance space X
together with a label that indicates for each example whether it has a certain property.
The learner's task then is to generalize from the given examples to new, previously
unseen examples or to the whole instance space X. As each property of objects in X
can be identified with the subset C X of objects that have the property in question,
this concept C can be considered as a target concept to be learned.

EXAMPLE 1. Consider an e-mail program that allows the user to classify incoming
e-mails into various (not necessarily distinct) categories (e.g. spam, personal, about
a certain topic, etc.). After the user has done this for a certain number of e-mails, the
program shall be able to do this classification automatically.
Sequence prediction works without labels. The learner observes a finite sequence
over an instance set (alphabet) X and has to predict its next member.
EXAMPLE 2. A stock broker has complete information about the price of a certain
company share in the past. Her task is to predict the development of the price in the
future.
In the following, we will consider each of the two mentioned settings in detail.
Concerning concept learning we also would like to refer to the chapter on Statistical
Learning Theory of von Luxburg and Schölkopf in this volume, which deals with
similar questions in a slightly different setting.
> Read full chapter
Advanced Topics
Colleen McCue, in Data Mining and Predictive Analysis (Second Edition) , 2015
15.3.1 Automated Feature extraction
Machine learning algorithms increasingly are able to process imagery and extract
features including structures, water, vegetation, and debris fields, which enables
very rapid processing of large amounts of imagery in support of real-time or near
real-time insight. In addition to the clear implications for enhanced situational
awareness, the ability to leverage these capabilities in support of change detec-
tion can facilitate rapid, informed decision-making regarding resource allocation.
Moreover, these results can be used as independent variables in geospatial predic-
tive analysis, or integrated with other derived content to include calls for service
in support of informed emergency response and resource allocation. Conversely,
sometimes an absence of behavior is key to understanding risk or need. With this in
mind, identifying communication “void spaces” within a debris field may highlight
greater need than a location associated with a high number of calls for assistance. In
these cases where speed is of the essence, the rapid response enabled by automated
feature extraction may truly change outcomes.
> Read full chapter

TO COGNIZE IS TO CATEGORIZE:
COGNITION IS CATEGORIZATION
STEVAN HARNAD, in Handbook of Categorization in Cognitive Science , 2005
11 Learning algorithms
Machine-learning algorithms from artificial-intelligence research, genetic algo-
rithms from artificial-life research and connectionist algorithms from neural-net-
work research have all been providing candidate mechanisms for performing the
“how” of categorization.
There are, in general, two kinds of learning models: so-called “supervised” and
“unsupervised” ones. The unsupervised models are generally designed on the as-
sumption that the input “affordances” are already quite salient, so that the right
categorization mechanism will be able to pick them up on the basis of the shape of
the input – from repeated exposure and internal analysis alone, without need of any
external error-correcting feedback.
By way of an exaggerated example, if the world of shapes consisted of nothing but
boomerangs and Jerry-Fodor shapes, an unsupervised learning mechanism could
easily sort out their retinal shadows on the basis of their intrinsic structures alone
(including their projective geometric invariants). But with the shadows of newborn
chick abdomens, sorting them out as males and females would probably need the
help of error-corrective feedback. Not only would the attempt to sort them on the
basis of their intrinsic structural landscape alone be like looking for a needle in a
haystack, but there is also the much more general problem, that the very same things
can often be categorized in many different ways. It would be impossible, without
error-correcting supervision, to determine which way was correct in a given context.
The right categorization can vary with the context: for example, we may want to
sort baby chicks sometimes by gender, sometimes by species, and sometimes by
something else [Harnad (1987)].
In general, a nontrivial categorization problem will be “underdetermined.” Even if
there is only one correct solution, and even if it can be found by an unsupervised
mechanism, it will first require a lot of repeated exposure and processing. The
figure/ground distinction might be something like this: How, in general, does our
visual system manage to process the retinal shadows of real-world scenes in such
a way as to sort out what is figure and what is ground? In the case of ambiguous
figures such as Escher drawings, there may be more than one way to do this, but,
in general, there is a default way to do it that works, and our visual systems usually
manage to find it quickly and reliably for most scenes. It is unlikely that our visual
systems learned to do this on the basis of having had error-corrective feedback from

sensorimotor interactions with samples of endless possible combinations of scenes
and their shadows.
> Read full chapter
To Cognize is to Categorize
Stevan Harnad, in Handbook of Categorization in Cognitive Science (Second Edi-
tion), 2017
2.11 Learning Algorithms
Machine-learning algorithms from artificial-intelligence research ( Michalski, Car-
bonell, & Mitchell, 2013 ), genetic algorithms from artificial-life research ( Mitchell &
Forrest, 1994) and connectionist algorithms from neural-network research (Schmid-
huber, 2015) have all been providing candidate mechanisms for performing the
“how” of categorization: how our brains learn to abstract the invariant features that
distinguish the members of each category from the nonmembers, allowing us to do
the right thing with the right kind of thing.
There are, in general, two kinds of learning models: so-called “supervised” and
“unsupervised” ones. The unsupervised models are generally designed on the as-
sumption that the input “affordances” are already quite salient, so that the right
categorization mechanism will be able to pick them up on the basis of the shape of
the input landscape—from repeated exposure and internal correlations and analysis
alone, with no need of any external error-correcting feedback.
By way of an exaggerated example, if the world of shapes consisted of nothing but
boomerangs and Jerry Fodor shapes, an unsupervised learning mechanism could
easily sort out their retinal shadows on the basis of their intrinsic structures alone
(including their projective geometric invariants). But with the shadows of newborn
chick abdomens, sorting them out as males and females would need the help of
error-corrective feedback. Not only would the attempt to sort them on the basis of
their intrinsic structural landscape alone be like looking for a needle in a haystack,
but there is also the much more general problem of underdetermination ( Stanford,
2009), which is that the very same things can often be categorized in many different
ways. It would be impossible, without error-correcting supervision, to determine
which way was correct in a given context. The right categorization can vary with the
context: e.g., we may want to sort baby chicks sometimes by gender, sometimes by
subspecies or size, and sometimes by something else ( Harnad, 1987).

In general, a nontrivial categorization problem will be “underdetermined.” Even if
there is only one correct solution, and even if it can be found by an unsupervised
mechanism, it will first require a lot of repeated exposure and processing. For the
figure/ground distinction ( Kelly & Grossberg, 2000; Peterson & Salvagio, 2010 ), how
does our visual system manage to process the retinal shadows of real-world scenes in
such a way as to sort out what is figure and what is ground? In the case of ambiguous
figures such as Escher drawings, there may be more than one way to do this, but,
in general, there is a default way to do it that works, and our visual systems usually
manage to find it quickly and reliably for most scenes. It is unlikely that our visual
systems learned to do this on the basis of having had error-corrective feedback from
sensorimotor interactions with samples of endless possible combinations of scenes
and their shadows.
> Read full chapter
The Arts and The Brain
Matthew Sachs, … Hanna Damasio, in Progress in Brain Research , 2018
2.3 Advantages of Newer Analytical Techniques
Machine learning algorithms , used primarily in the field of computer science, have
more recently been applied to the analysis of neuroimaging data. The goal with
such methods is to identify patterns within the recorded fMRI signal and relate
these patterns to the cognitive states being experienced by the participant. Termed
multivoxel pattern analysis (MVPA; Norman et al., 2006 ), this method evaluates
the fMRI signal recorded from multiple voxels at once, which can provide a more
nuanced picture of how information is spatially encoded/distributed across the
brain. By classifying cognitive states based on the spatially distributed patterns of
neural signal associated with them, this type of analysis does not identify brain
regions that are “activated” by a stimulus, but, rather, brain regions that carry key
information regarding some identifying property of that stimulus ( Mitchell et al.,
2004).
MVPA has been used to classify emotional states induced by music based on patterns
of fMRI data. Kragel and LaBar (2015) found that distinct patterns of activity within
several cortical and subcortical brain regions could predict one of seven discrete
emotional states induced by music. These included the precuneus, cingulate, insula,
thalamus, amygdala, and prefrontal cortex. Another study using music with either
positive or negative valence corroborated those findings, reporting that patterns
of activity within the precuneus, cingulate, thalamus, and prefrontal cortex could

successfully predict the valence of the musical piece that was presented ( Kim et al.,
2017).
Recently, in our own work ( Sachs, Habibi, Damasio, & Kaplan, 2018 ), we showed
that emotional categories of musical sounds expressing happiness, sadness, and
fear, produced by a violin and clarinet, could be reliably distinguished in fMRI data,
based on the patterns of the neural signal detected in bilateral auditory cortices,
insulae, parietal opercula, post- and precentral gyri, inferior frontal gyri, and right
medial prefrontal cortex. Neural patterns within the primary and secondary auditory
cortices, insulae, and parietal opercula could also reliably distinguish the same three
emotions when training the classifier on data using a different set of nonmusical,
nonverbal vocalizations, e.g., the sound of a person crying, screaming, or laughing
(see Belin et al., 2008), suggesting that these patterns of activity are not unique to
the processing of emotions conveyed through musical instruments. In combination,
the MVPA results appear to support the conclusions drawn from fMRI studies with
univariate data, namely that feelings induced by music evoke responses similar to
those associated with feelings induced by other stimuli ( Koelsch, 2014).
In addition to MVPA, other model-free methods have recently been employed with
neuroimaging to assess brain responses to more naturalistic, ecologically valid
stimuli. One such approach involves recording neural signal continuously (using
either EEG or fMRI) during the presentation of a full-length piece of music. This
type of approach does not rely on a predefined model to evaluate the changes in
neural signal. Instead, it involves calculating correlations between the neural signal
in different brain regions or between the neural signal in the same brain region
across different participants. Changes in these correlations over time can then be
related to specific changes in the music. In this way, the fluctuating patterns of
brain synchrony can be linked to an emotional experience, shared among different
subjects, in response to the music presented. Studies using such an technique have
shown that continuous ratings of valence and arousal while listening to a music piece
were associated with synchronized activity in the amygdala, insula, ACC, and caudate
nucleus across participants ( Trost et al., 2014). Furthermore, within-network analyses
showed that synchrony of activity within regions of the limbic system was correlated
with changes in valence ratings during a listening period ( Singer et al., 2016). While
these methods have only been used with musical stimuli recently and sparsely, the
early findings provide further support for the hypothesis that music-evoked feelings
involve the continuous interaction of multiple brain regions over time, highlighting
the importance of capturing the temporal parameters of emotional responses to
music.
> Read full chapter

An Introduction to Artificial Intelli-
gence in Behavioral and Mental Health
Care
David D. Luxton, in Artificial Intelligence in Behavioral and Mental Health Care , 2016
Machine Learning and Artificial Neural Networks
Machine learning (ML) is a core branch of AI that aims to give computers the
ability to learn without being explicitly programmed ( Samuel, 2000). ML has many
subfields and applications, including statistical learning methods, neural networks-
, instance-based learning, genetic algorithms, data mining, image recognition,
natural language processing (NLP), computational learning theory, inductive logic
programming, and reinforcement learning (for a review see Mitchell, 1997).
Essentially, ML is the capability of software or a machine to improve the performance
of tasks through exposure to data and experience. A typical ML model first learns
the knowledge from the data it is exposed to and then applies this knowledge
to provide predictions about emerging (future) data. Supervised ML is when the
program is “trained” on a pre-defined set of “training examples” or “training sets.”
Unsupervised ML is when the program is provided with data but must discover
patterns and relationships in that data.
The ability to search and identify patterns in large quantities of data and in some
applications without a priori knowledge is a particular benefit of ML approaches. For
example, ML software can be used to detect patterns in large electronic health record
datasets by identifying subsets of data records and attributes that are atypical (e.g.,
indicate risks) or that reveal factors associated with patient outcomes ( McFowland,
Speakman, & Neill, 2013; Neill, 2012 ). ML techniques can also be used to automat-
ically predict future patterns in data (e.g., predictive analytics or predictive modeling )
or to help perform decision-making tasks under uncertainty. ML methods are also
applied to Internet websites to enable them to learn the patterns of care seekers,
adapt to their preferences, and customize information and content that is presented.
ML is also the underlying technique that allows robots to learn new skills and adapt
to their environment.
Artificial neural networks (ANNs) are a type of ML technique that simulates the
structure and function of neuronal networks in the brain. With traditional digital
computing, the computational steps are sequential and follow linear modeling
techniques. In contrast, modern neural networks use nonlinear statistical data
modeling techniques that respond in parallel to the pattern of inputs presented
to them. As with biological neurons, connections are made and strengthened with

repeated use (also known as Hebbian learning; Hebb, 1949). Modern examples of
ANN applications include handwriting recognition, computer vision, and speech
recognition (Haykin & Network, 2004; Jain, Mao, & Mohiuddin, 1996 ). ANNs are also
used in theoretical and computational neuroscience to create models of biological
neural systems in order to study the mechanisms of neural processing and learning
(Alonso & Mondragón, 2011 ). ANNs have also been tested as a statistical method
for accomplishing practical tasks in mental health care, such as for predicting
lengths of psychiatric hospital stay ( Lowell & Davis, 1994 ), determining the costs
of psychiatric medication ( Mirabzadeh et al., 2013 ), and for predicting obsessive
compulsive disorder (OCD) treatment response ( Salomoni et al., 2009 ).
ML algorithms and neural networks also provide useful methods for modern expert
systems (see Chapter 2). Expert systems are a form of AI program that simulates
the knowledge and analytical skills of human experts. Clinical decision support
systems (CDSSs) are a subtype of expert system that is specifically designed
to aid in the process of clinical decision-making ( Finlay, 1994). Traditional CDSSs
rely on preprogrammed facts and rules to provide decision options. However,
incorporating modern ML and ANN methods allows CDSSs to provide recom-
mendations without preprogrammed knowledge. Fuzzy modeling and fuzzy-genetic
algorithms are specific ancillary techniques used to assist with the optimization of
rules and membership classification (see Jagielska, Matthews, & Whitfort, 1999 ).
These techniques are based on the concept of fuzzy logic ( Zadeh, 1965), a method
of reasoning that involves approximate values (e.g., some degree of “true”) rather
than fixed and exact values (e.g., “true” or “false”). These methods provide a useful
qualitative computational approach for working with uncertainties that can help
mental healthcare professionals make more optimal decisions that improve patient
outcomes.
> Read full chapter
Categorization by Humans and Ma-
chines
Raymond J. Mooney, in Psychology of Learning and Motivation , 1993
VI Conclusions
Recent results in both machine learning and cognitive psychology demonstrate that
effective category learning involves an integration of theory and data. Theories can
bias induction and alter the representation of data, and conflicting data can result
in theory revision. This paper has reviewed two recent machine learning systems

that attempt to integrate theory and data. IOU uses a domain theory to acquire part
of a concept definition and to focus induction on the unexplained aspects of the
data. either uses data to revise an imperfect theory and uses theory to add abstract
features to the data. Recent psychological experiments reveal that subjects perform
many of the same operations as these machine learning systems. Like IOU, people
separate category definitions into explanatory and nonexplanatory components,
acquire explanatory components earlier, and have more confidence in explanatory
aspects. Like either, people use background theories to derive abstract features from
the data, and revise portions of their theories to account for conflicting data.
Nevertheless, in many ways, current machine learning systems are not nearly as
adept as people at integrating theory and data in learning. Particular areas requiring
further research concern revising probabilistic and relational theories. Most current
integrated learning systems are restricted to theories expressed in propositional
logic. Consequently, they are incapable of reasoning about their confidence in their
theories and conclusions, and cannot handle complex, relational descriptions that
require the expressive power of first-order predicate logic. These areas of machine
learning are just beginning to be explored ( Fu, 1989; Pazzani, Brunk & Silverstein,
1991; Richards & Mooney, 1991 ). In general, the interaction between theory and data
in learning has just begun to be investigated.
From a machine-learning perspective, methods for integrating theory and data
in learning can greatly improve the development of intelligent systems. Standard
methods for building knowledge bases by interviewing experts are laborious and
error-prone. Standard machine learning methods for learning from examples are
also inadequate since one rarely has enough data to induce a complete and correct
knowledge base from scratch. In addition, machine-induced knowledge fails to
make use of existing human concepts and is therefore frequently unable to provide
comprehensible explanations for the conclusions it warrants. Theory revision, on the
other hand, allows a system to accept an incomplete, approximate knowledge base
and refine it through experience. People acquire expertise through a combination of
abstract instruction and experience with specific cases, and machine learning sys-
tems that integrate theory and data are trying to successfully emulate this approach.
From a psychological perspective, methods for integrating theory and data can
hopefully improve our understanding of human category learning. Artificial learning
problems that minimize the role of prior knowledge are not representative of the cat-
egorization problems that people confront every day. Machine learning algorithms
that can simulate psychological data on the effect of prior knowledge on learning
can provide valuable insight into how people learn in more natural settings. In turn,
understanding the specific ways in which theory and data interact in human learning
can hopefully lead to the development of more effective educational methods for
combining the presentation of abstract rules and principles with concrete examples.

> Read full chapter
Bioinformatics of Behavior: Part 1
Kyle H. Ambert, Aaron M. Cohen, in International Review of Neurobiology , 2012
10 Knowledge Mining
One alternative to using machine learning for assisting manual database curation
is that of automated mining from document databases. Because the financial and
time costs associated with developing a large curated document collection are often
prohibitive, researchers will sometimes perform automated association mining, in
which textual features are extracted from a large collection of input documents and
used either to further one's understanding of the relationships between the docu-
ments themselves or to develop hypotheses that can be investigated on their own.
Voytek and Voytek (2012) , for example, used co-occurrences of brain region men-
tions, cognitive functions, and brain-related diseases to demonstrate that known
relationships can be extracted in an automated and scalable way by using clustering
algorithms. Importantly, they were able to extend this approach to semi-automat-
ically generate hypotheses regarding “holes” in the literature associations between
brain structure and function, or function and disease which are likely to exist, but
lack support in the literature. For example, they discovered that the structure striatum
and the term migraine were strongly related to the term serotonin (they co-occurred
in nearly 3000 publications for each relationship), yet the striatum and migraine had
only 16 shared publications themselves, indicating that this association may exist
but be understudied.
French and Pavlidis (2012) used knowledge mining to automatically map neu-
roanatomical identifiers found in a large volume of journal abstracts from the Journal
of Comparative Neurology (JCN) to connect over 100,000 brain region mentions to
8225 normalized brain region concepts in a database. In this work, they used an
annotated collection of abstracts from JCN and other neuroscience journals (French,
Lane, Xu, & Pavlidis, 2009 ), expanding all abbreviations in the text, and manually
identified the brain region mentions they contained. They also put together a dictio-
nary of 7145 brain regions having formal unique identifiers from the NeuroNames
vocabulary (Bowden et al., 2007 ), NIFSTD/BIRNLex ( Bug et al., 2008), Brede Data-
base (Nielsen, Hansen, & Balslev, 2004 ), Brain Architecture Management System
(Bota & Swanson, 2008 ), and Allen Mouse Brain Reference Atlas ( Dong, 2008).
In total, they used five different techniques to link the free-text neuroanatomical
mentions to the compiled set of terms: exact string matching, bag of words, stem-
ming, bag of stems (similar to gap-edit global string matching; Srinivas, Cristianini,

Jones, & Gorin, 2005 ), and the Lexical OWL Ontology Matcher, which allows for the
specification of specific types of entities ( Ghazvinian, Noy, & Musen, 2009 ). Scientists
interested in using these resources could incorporate their annotated data (freely
available at http://www.chibi.ubc.ca/WhiteText ) into a classification system like the
ones described in the previous section.
> Read full chapter
ScienceDirect is Elsevier’s leading information solution for researchers.
Copyright © 2018 Elsevier B.V. or its licensors or contributors. ScienceDirect ® is a registered trademark of Elsevier B.V. Terms and conditions apply.