Conference PaperPDF Available

An Evaluation of Unstructured Text Mining Software


Abstract and Figures

Five text mining software tools were evaluated by four undergraduate students inexperienced in the text mining field. The software was run on the Microsoft Windows XP operating system, and employed a variety of techniques to mine unstructured text for information. The considerations used to evaluate the software included cost, ease of learning, functionality, ease of use and effectiveness. Hands on mining of text files also led us to more informative conclusions of the software. Through our evaluation we found that two software products (SAS and SPSS) had qualities that made them more desirable than the others.
Content may be subject to copyright.
An Evaluation of Unstructured Text Mining Software
Micah J. Crowsey, Amanda R. Ramstad, David H. Gutierrez, Gregory W. Paladino, and K. P. White,
Jr., Member, IEEE
Abstract— Five text mining software tools were evaluated
by four undergraduate students inexperienced in the text
mining field. The software was run on the Microsoft
Windows XP operating system, and employed a variety of
techniques to mine unstructured text for information. The
considerations used to evaluate the software included cost,
ease of learning, functionality, ease of use and effectiveness.
Hands on mining of text files also led us to more informative
conclusions of the software. Through our evaluation we
found that two software products (SAS and SPSS) had
qualities that made them more desirable than the others.
Unstructured data exists in two main categories: bitmap
objects and textual objects. Bitmap objects are non-language
based (e.g. image, audio, or video files) whereas textual
objects are “based on written or printed language” and
predominantly include text documents
. Text mining is the
discovery of previously unknown information or concepts
from text files by automatically extracting information from
several written resources using computer software
. In
text mining, the files mined are text files which can be in
one of two forms. Unstructured text is usually in the form of
summaries and user reviews whereas structured text consists
of text that is organized usually within spreadsheets. This
evaluation focused specifically on mining unstructured text
Many industries are relying on the field of text mining to
solve specific applications using a variety of software.
Currently, there does not exist a comprehensive review or
comparison of the top software suites. Our research was
performed so users would have an unbiased reference for
looking at the different software features and the pros, cons,
and methods for how to most efficiently use them. In
comparing the software, we looked at the features that the
software had as well as the ease of use and learning of the
software. We used a test corpus that we developed for this
Manuscript received April 5, 2007. This work was supported in part by
the University of Virginia under a grant provided by an anonymous
M. J. Crowsey is with the University of Virginia, Charlottesville, VA,
22902. (phone: 434-924-5393. e-mail:
A. R. Ramstad is with the University of Virginia, Charlottesville, VA,
22902. ( e-mail:
D. H. Gutierrez is with the University of Virginia, Charlottesville, VA,
22902. (e-mail:
G. W. Paladino is with the University of Virginia, Charlottesville, VA,
22902. ( e-mail:
K. P. White is with the University of Virginia, Charlottesville, VA, 22902.
The five software suites we reviewed were Leximancer,
SAS Enterprise Miner, several products of SPSS,
Polyanalyst, and Clarabridge. Most of the software was
acquired through an anonymous consulting firm. We also
obtained software through the University of Virginia and by
requesting online software demos.
The evaluators of the software were four fourth year
Systems Engineering students at the University of Virginia
who had some previous experience with data mining, but
little experience with text mining. This inexperience proved
useful in determining the ease of use and learning of the
software, though it sometimes posed challenges when
attempting to find out how the software operates.
I. T
A. General Information and Pricing
The chart below shows the company, product, version, and
cost of the software. Unfortunately, certain vendors were
unwilling to disclose cost information for their products.
Table 1
Company Product Version Cost
SAS Enterprise
4.3 not provided
due to company
SPSS Text mining for
Text Mining
Server (1 CPU)
Megaputer Polyanalyst 6.0 Professional
Server $80,000
Client $5,000
Leximancer Leximancer pro $2500
Clarabridge Clarabridge
Content Mining
2.1 $75,000
B. Learnability
The criteria used to assess the ease of learning were
based on whether software had the following:
A demo version
User’s manual
Sample Solutions
Online help
The functionality results of the software were recorded using
a spots and dots methodology. The ease of learning results
can be seen in Table 2.
The project team attempted to learn how to use each of the
software suites solely by referencing the help files, tutorials,
and other learning tools that were provided along with the
software package. These materials provided information on
the general process each software suite uses to mine text and
on the specific features offered at different steps in the text
mining process.
Table 2: Learnability
Tutorial User’s
Clarabridge X X
Clementine X X
Polyanalyst X X X X
Leximancer X X X X
As we were unable to obtain working copies of
Clarabridge and Polyanalyst, we were unable to gain
experience using these software suites. The evaluation of
this software was done by looking at product features, going
through live internet demonstrations, and performing
research on the software companies’ websites.
Overall, the help documentation which accompanied the
SAS, SPSS, and Leximancer software was sufficient to learn
basic processes and features employed by each suite. SAS
and SPSS both offer traditional help files which serve as a
good starting point for learning the process that each
software suite uses to mine text. These resources provide
both an introduction to text mining as well as the process
flows that each of the software suites use to accomplish
different text mining functions such as text extraction, text
link analysis, and categorization. At a more detailed level,
the help documentation of both software suites provide
information on how a user can manipulate various features
at different nodes in the text mining process in order to
affect the results yielded. Finally, the documentation also
provides information on how to view, edit, and make use of
SPSS help documentation presents basic information
which provides as user with a quick start to text mining.
Example projects also are provided to show how different
nodes can work together to accomplish various text mining
SAS documentation presents information on the text
mining process and its unique methods and features in much
more detail. Step-by-step examples of how to achieve a few
text mining functions are also very helpful.
Leximancer presents its help documentation in a slightly
different fashion than SAS and SPSS. Leximancer provides
several help topics describing how to accomplish certain text
mining functions within its architecture. Leximancer’s text
mining process consists of a preset stream of nodes which a
user cannot alter, and information on how a user can
manipulate the features of these nodes to achieve different
results resides within the nodes themselves.
The fact that Leximancer presents information on the
features offered by different nodes in the text mining
process within the nodes themselves makes for quick
referencing, and the content is very helpful in general.
Leximancer also allows the user to adjust the level of
complexity of the features it offers.
Although help documentation and tutorials are sufficient
for the beginning of the learning curve with these software
suites, the group found that more advanced knowledge of
how to get the most out of each product is best achieved
through an iterative process of hands on experimentation.
Manipulating the unique features of each software suite in
various ways provides the best knowledge of how to achieve
desired results. Also, experimenting with a variety of nodes
in sequence allows a user to see how these nodes can
interact to achieve more advanced text mining functions.
C. Data Preparation
The importance of data preparation in text mining cannot
be stressed enough. Given the importance of proper data
preparation to the success of a data mining effort, it is
advised that the user perform some “cleaning” on the data to
put it into a semi-structured form. Although text mining
seeks to find relationships between concepts in unstructured
data, we found through our evaluation that mining
technology does not eliminate the need for data preparation.
If the user wishes to achieve a high level of reliability and
extract useful concepts from the data, then structuring the
data, even in small ways, is helpful.
To achieve useful information for our evaluation, we ran
text files through the software in order to see how they were
processed. We also looked at the quality of the results after
the mining was completed. For this task, we used HTML
documents that were gathered from the University of
Virginia Cavalier Daily newspaper website, A software program that copies
entire websites was used to gather approximately 1200
HTML pages from the website, and these formed the corpus
that we ran through the software.
D. Software Pros and Cons
Leximancer is a software suite that focuses on extracting
concepts and showing the relationships between those
concepts, along with the relationship strength. Although its
Java-based user interface is somewhat different from the
other software suites evaluated, Leximancer still offers
many of the same features that allow a user to manipulate
stages in the text mining process. Leximancer’s results
browser is very effective at presenting several pieces of
information at once and allowing a user to browse extracted
concepts and concept relationships. Leximancer’s results
browser is shown in Figure 1 below.
Figure 1: Leximancer concept map and list of linked
SAS Enterprise miner uses a unique mining approach
called SEMMA (Sampling, Exploration, Modification,
Modeling, and Assessment). This package offers features
that compare the results of the different types of modeling
through statistical analysis and in business terms. The
integrated environment allows for the statistical modeling
group, business managers and the information technology
department to work together. Although SAS supports
loading text in a variety of file formats, all externally stored
text files must first be converted to a SAS data set via the
use of a prewritten macro, adding additional complexity and
time consumption to this step in the text mining process.
SAS’ user interface is less intuitive than other software,
but it still offers many of the same features as other products
which affect how text is parsed. In addition to offering these
basic features, SAS also offers a user the ability to affect
how its algorithm is run which represents parsed documents
in a structured, quantitative form. SAS’ term extraction
generally yields a larger number of terms than other
software, but its results browser allows for easy browsing
and filtering of these terms. SAS also simplifies the text
mining process somewhat by automatically clustering
documents and identifying links between terms when a term
extraction is executed. Figure 2 below shows SAS’ text
mining browser for documents, extracted terms, and
Figure 2: SAS text mining browser
SPSS is flexible in that it supports many text formats,
including: plain text, PDF, HTML, Microsoft Office, and
XML text files. It has open architecture which allows the
program to join together with other text analytics
applications including Text Mining Builder, LexiQuest
Categorize, and all other SPSS data mining and predictive
analytics applications. Clementine’s text extraction does
well to offer the use several methods for limiting the scope
of an extraction and therefore tends to yield the most
informative terms and phrases in its results. Figure 3 below
shows Clementine’s text extraction results browser.
Figure 3: Ranked list of extracted concepts in
Clementine also includes a text link analysis which has
two main functions. The first is that it recognizes and
extracts sentiments (i.e. likes and dislikes) from the text with
the help of dictionary files created in Text Mining Builder.
Some of these dictionary files are already provided in the
software package while others can be developed by the user.
The second is that it detects correlations between things
such as people/events, or diseases/genes. SPSS also allows a
user to manually define a customized taxonomy with the
help of LexiQuest Categorize, which uses training data to
learn the characteristics of the taxonomy predict the
classification of documents.
Polyanalyst, manufactured by Megaputer, can process
large scale databases. The software also has a low learning
curve and step by step tutorials. The integration of an
analysis performed on both structured and unstructured text
is available. The results of the analysis can be incorporated
into existing business processes.
Clarabridge provides analysis of data and is used in
conjunctions with commercially available business
intelligence tools which are used to view and interpret the
results of this analysis. It also allows parallel processing of
large amounts of data. During the processing, entities,
relationships, sections, headers, and topics, as well as
proximal relationships, tables, and other data are recognized.
This information is stored into the capture schema thus
maintaining metadata and linking it back to its source.
Clarabridge can contain large amounts of data and maintain
high throughput. The GUI requires a minimal amount of
coding from users and processes can be done without human
Table 3 shows the different functions that the software
have. For some of these functions, such as extraction, all of
the software possesses some form of the function. For
others, such as clustering, not all of the software have the
feature. A table of this form is useful for a quick visual
software functionality comparison.
Table 3: Functionality
FUNCTIONS SPSS SAS Clarabridge Polyanalyst Leximancer
Extraction X X X X X
Summarization X
Categorization X X X X X
Clustering X X X
Linking X X X X X
Trees X X X
Linkage X X X X
Reasoning X X X X
Regression X X X
data for
analysis in
Time Series X X X
data for
analysis in
Unfortunately we were unable to run the Cavalier Daily
files through Clarabridge because this software package
requires additional business intelligence tools in order to
achieve readable results.
After running the sample corpus through the other four
software suites and comparing the subjective ease of use of
the software, two products rose to the top: SAS and SPSS.
The immediate advantage of these pieces of software is
that they were developed for large projects, so while
processing 1200 documents took a significant amount of
time, they were able to display meaningful results. The
choice between these products, however, rests on the
particular application in which they are used.
Because of the non-intuitive interface and steep learning
curve, SAS is best used in situations where the user already
has a general understanding of text mining. It is also an
excellent choice for processing large amounts of documents,
however, it only gives truly meaningful information if the
input has been pre-processed and made to be semi-
When running the files through, SPSS proved to be the
quickest at mining the files. SPSS also is a good choice for
processing large amounts of documents and provides more
useful results if the input has been pre-processed and made
to be semi-structured. Another benefit that SPSS has over
SAS is that SAS extracts a large amount of useful terms.
All of the software products tested primarily extracted
sport-related concepts from the given corpus in the
explorative analysis that was done. This indicates that in the
Cavalier Daily newspaper, sports are the main topics that are
reported on. Again, because we were unable to obtain
working copies of Clarabridge and Polyanalyst, we were
unable to test them using our sample corpus. Further results
could be obtained with a deeper analysis, but as we were
using our corpus to get only a preliminary idea of the
features of the software, we did not pursue a more advanced
Future work with text mining software is already
underway. While the test corpus that was used to evaluate
the software in this report was large, the problem that was
attempted to be solved was not well-defined. Therefore, a
new problem has been proposed that is well-defined, and
work is underway to analyze and solve it.
There is a current project underway in which a group is
attempting to extract relationships between the results from
social security disability claims in the court systems and the
content of the claims that are filed with the courts. This is a
problem that is semi-structured and well-defined, and is
perfect for further testing of the SAS and SPSS suites.
The data for these cases are being gathered from various
state and federal websites that have cases on record having
to do with social security disability claims. This data will be
collected, parsed, inputted into a database table, and then
processed by SAS and SPSS in order to extract relationships
in the data.
The hope is that this processing will lead to discoveries
about what types of claims are most often approved or
rejected by the courts, if there is such a relationship. For
example, it might be the case that if a person mentions “Lou
Gehrig’s disease” in their claims, that they are almost always
approved for their claim. If such a relationship were true,
then text mining software like SAS and SPSS should be able
to extract it through predictive capabilities.
The following goals were achieved by the conclusion of this
Identified the common needs of users of text
mining tools through researching the industries that
use text mining and the applications for which text
mining is employed.
Addressed the current state in text mining through
background research in the field and hands on
Evaluated and compared text mining software. This
goal can be improved upon in future projects by
considering an expanded set of evaluation criteria.
This Appendix provides a glossary of terms commonly
used in discussions of text mining software.
KDD-knowledge discovery and data mining
Queries-a common way of extracting information from
Tuples-finite sequence or ordered list of object
Ease of Learning-how easy or hard it is to learn how to use
the software
Ease of Use-once the software is learned, how easy or hard
it is to use the software
Clustering-Clustering algorithms find groups of items that
are similar. For example, clustering could be used by an
insurance company to group customers according to income,
age, types of policies purchased and prior claims experience.
Decision tree-A tree-like way of representing a collection of
hierarchical rules that lead to a class or value.
Regression tree-A decision tree that predicts values of
continuous variables.
Time series model-A model that forecasts future values of a
time series based on past values.
Extraction-locating specific pieces of data and extracting it
from the document
Summarization- summarization extracts the most relevant
phrases or even sentences from a document.
Concept Linking- Usually comes in the form of some web-
like visualization in which the links between extracted
concepts are shown based on their co-occurrence and
proximity within documents.
Document Linkage – The ability to view in the results where
in the documents the concept occurs. Results link back to
input documents.
Categorization- Organization of documents into predefined
categories based on existence of specified indicator concepts
within the documents.
Memory-based Reasoning- MBR uses training records to
train a neural network to learn to predict certain
characteristics of new documents
The Capstone team would like to thank their technical
advisor, Professor K. Preston White for guiding them
through the capstone project. Also, the team would like to
acknowledge Elder Research Incorporated for allowing them
to participate in such a rewarding project through funding it.
The team would like to also thank Debbie and Jordan for
everything they have done for the team.
[1] Weglarz, G. (2004). Two worlds of data – unstructured and structured.
DM Review, 14(9), 19-22.
[2] J. Elder et al., An Evaluation of High-end Data Mining Tools for
Fraud Detection. Available: eval articles/ smc98
abbot mat eld.pdf.
[3] Megaputer, (2002), Polyanalyst Powerpoint.
[4] S. Grimes, The Word on Text Mining. Available:
[5] Saving Lives, Cutting Patient Costs: Louisville Hospitals Advance
with SAS Text Miner, SAS, 2006. Available:
[6] R. Rao, From Unstructured Data to Actionable Intelligence, IT Pro,
9202(03), 2003, pp. 1-7.
[7] P. Fule, J. Roddick, Detecting Privacy and Ethical Sensitivity in Data
Mining Results, School of Informatics and Engineering, Flinders
University, South Australia.
[8] Text Mining Terms, SPSS White Paper, Sept. 2006.
[9] K. Michel, J. Elder, Evaluation of Fourteen Desktop Data Mining
Tools, 1998.
[10] M. Goebel, L. Gruenwald, A Survey of Data Mining and Knowledge
Discovery Software Tools, SIGKDD Explorations, pp. 20-33, 1999.
[11] Apte, C. (2002). Business applications of data mining.
Communications of the ACM, 45(8), 49-53.
[12] Blumberg, R. (2003). The problem with unstructured data. DM
Review, 13(2), 42-4.
[13] Grimes, S. (2005). The Developing Text Mining Market. [White paper,
electronic version]. Retrieved October 12, 2006 from
[14] Hand, D., Mannila, H., Smyth, P. (2001). Principles of Data Mining.
MIT Press: Cambridge, MA.
[15] Marti Hearst. “What is Text Mining?” 17 October 2003. (29 October
... Robb [7] claims that unstructured data, " most of it in the form of text files, typically accounts for 85% of an organization's knowledge stores, but it's not always easy to find, access, analyze or use. " Crowsey et al. [2] conducted study using five text mining software tools with four undergraduate students at the University of Virginia who were had some experience in data mining but little experience in text mining. The text mining software used in study by Crowsey et al. [2] included a discussion of Megaputer PolyAnalyst® and SAS® Enterprise Miner but did not include SAS® Text Miner or a working copy of Megaputer PolyAnalyst®. ...
... " Crowsey et al. [2] conducted study using five text mining software tools with four undergraduate students at the University of Virginia who were had some experience in data mining but little experience in text mining. The text mining software used in study by Crowsey et al. [2] included a discussion of Megaputer PolyAnalyst® and SAS® Enterprise Miner but did not include SAS® Text Miner or a working copy of Megaputer PolyAnalyst®. The evaluation by Crowsey et al. [2] concluded using their available resources that the two software products of SAS® and SPSS had qualities that made them more desirable than others. ...
... The text mining software used in study by Crowsey et al. [2] included a discussion of Megaputer PolyAnalyst® and SAS® Enterprise Miner but did not include SAS® Text Miner or a working copy of Megaputer PolyAnalyst®. The evaluation by Crowsey et al. [2] concluded using their available resources that the two software products of SAS® and SPSS had qualities that made them more desirable than others. ...
Full-text available
This paper presents text mining using SAS® Text Miner and Megaputer PolyAnalyst® specifically applied for hotel customer survey data, and its data management. The paper reviews current literature of text mining, and discusses features of these two text mining software packages in analyzing unstructured qualitative data in the following key steps: data preparation, data analysis, and result reporting. Some screen shots are presented for web-based hotel customer survey data as well as comparisons and conclusions.
... In addition, the pedagogical concerns have been more or less marginally addressed in many of the studies by considering features/functionality (e.g., [34]) and usefulness/utility (e.g., [35]) of the given technology. ...
Full-text available
Conference Paper
Usability is important to consider when selecting technology for educational use in a learning environment. It can influence learning and teaching experience, and consequently adoption and retention of educational technology. The objective of this study is to (1) analyze and systematize usability methods and attributes that scholars have considered in the context of educational technology selection in higher education, (2) analyze pedagogical criteria, and (3) the learner/teacher perspective in the identified approaches, as well as to (3) identify future research perspectives in this regard. Therefore, a systematic research review has been conducted and 45 papers have been selected and analyzed. Analytical methods, particularly expert assessment, have been more often reported than empirical methods. The most frequent usability attribute is ease of use. Most of the studies have addressed pedagogical criteria to some extent. Almost a half of the studies have attempted to consider the learner/teacher perspective by involving actual or prospective users. There is a need for a relatively simple and efficient, yet effective enough, usability evaluation that fits well into the educational technology selection. However, it is mostly unverified how well the reported approaches meet this need. Notwithstanding, there are positive examples that deserve attention and further research.
... In this paper, Information retrieval method is put forward based on the vector space model [9]. This model is referred as multi class vector space model. ...
Full-text available
Conference Paper
A new model of information retrieval algorithms, multi class vector space model, is proposed in this paper based on traditional vector space model. Web document has semi structured characteristic. The keyword or terms that are used for indexing purpose in any location, so content of this location represent important information in the web documents. Vector space model ignores the importance of these terms with respect to their position while calculating the weight of the indexing terms. The experimental result shows that this method can further improve the performance of vector space model, save storage space and speed up the retrieval speed with high precision and recall rate. General Terms Web information retrieval,stemming,Extraction.
... Comments with similar positive or negative magnitudes may then be grouped by subject to provide examples of attributes and detriments (Holcomb & Daum, 2012 SIOP Symposium). Additionally, text analysis software such as SPSS Text Mining or SAS Enterprise Miner can apply categorization, regression tree, neural network reasoning, and concept linkage algorithms to identify broad themes and relationships (Crowsey et al., 2007), although such programs do not remove the need for the researcher to actually read the data (Angelique et al., 2005). At the comment or document level, text summarization approaches may attempt to generate a representation of the subject matter and then score sentences on importance or relevance to that representation. ...
... In this paper, Information retrieval method is put forward based on the vector space model [9]. This model is referred as multi class vector space model. ...
... Relevant terms are then grouped together under a lead term called a concept. Pattern matching algorithms are applied to the extracted concepts in order to identify relationships across all documents (SPSS, Crowsey et al., 2007). Text link analysis can be grouped under the text mining topic of 'sentiment classification'. ...
Conference Paper
One of the most challenging knowledge management tasks is to obtain, summarize, and present tacit knowledge. It is important to develop approaches for creating insightful summaries from the knowledge obtained. In this paper, we present different information extraction methods for summarizing interview transcripts. Manual, semi-automatic, and automatic text analysis are evaluated to transform tacit knowledge into explicit form and to substantially reduce the time required to perform this transformation. These approaches are described in the context of a real application.
... The core of these expert texts is a set of contextually related words and phrases (terms in the fields). The retrieval of "key words" and relationships among them is usually done through text mining methods and algorithms [2], [3], [4], [5], [6]. The process of verification and correction [7], [8] is needed to guarantee storing correct expert data in the database. ...
Full-text available
Conference Paper
Specialized applied fields in natural sciences - medicine, biology, chemistry etc. require building and exploring of information systems based on related text forms (words, phrases). These forms represent expert information and knowledge. The paper discusses the integration of two basic approaches - relational for structuring complex related texts and object oriented for data analysis. This conception is implemented for building of information system "Crop protection" in Bulgaria based on the complex relationships between biological (crops, pests) and chemical (pesticides) terms in textual form. Analogy exists between class objects in biology, chemistry and class objects and instances of object oriented programming. That fact is essential for building flexible models and software for data analysis in the information system. The presented example shows the potential of object oriented modelling to define and resolve complex tasks concerning effective pesticides use. © 2011 IFIP International Federation for Information Processing.
Student rating instruments are recognised to be valid indicators of effective instruction, providing a valuable tool to improve teaching. However, free-form text comments obtained from the open-ended question component of such surveys are only infrequently analysed comprehensively. We employed an innovative, systematic approach to the analysis of text-based feedback relating to student perceptions of and experiences with a recently developed university program. The automated nature of the semantic analysis tool Leximancer enabled a critical interrogation across units of study, mining the cumulative text for common themes and recurring core concepts. The results of this analysis facilitated the identification of issues that were not apparent from the purely quantitative data, thus providing a deeper understanding of the curriculum and teaching effectiveness that was constructive and detailed.
Full-text available
Full-text available
Conference Paper
Knowledge discovery allows considerable insight into data. This brings with it the inherent risk that what is inferred may be private or ethically sensitive. The process of generating rules through a mining operation becomes an ethical issue when the results are used in decision making processes that effect people, or when mining customer data unwittingly compromises the privacy of those customers.Significantly, the sensitivity of a rule not be apparent to the miner, particularly since the volume and diversity of rules can often be large. However, given the subjective nature of such sensitivity, rather than prohibit the production of ethically and privacy sensitive rules, we present here an alerting process that detects and highlights the sensitivity of the discovered rules. The process caters for differing sensitivities at the attribute value level and allows a variety of sensitivity combination functions to be employed. These functions have been tested empirically and the results of these tests are reported.
Full-text available
Conference Paper
Fourteen desktop data mining tools (or tool modules) ranging in price from US$75 to $25,000 (median <$1,000) were evaluated by four undergraduates inexperienced at data mining, a relatively experienced graduate student, and a professional data mining consultant. The tools ran under the Microsoft Windows 95, Microsoft Windows NT, or Macintosh System 7.5 operating systems, and employed decision trees, rule induction, neural networks, or polynomial networks to solve two binary classification problems, a multi-class classification problem, and a noiseless estimation problem. Twenty evaluation criteria and a standardized procedure for assessing tool qualities were developed and applied. The traits were collected in five categories: capability, learnability/usability, interoperability, flexibility, and accuracy. Performance in each of these categories was rated on a six-point ordinal scale, to summarize their relative strengths and weaknesses. This paper summarizes a lengthy technical report (Gomolka et al., 1998), which details the evaluation procedure and the scoring of all component criteria. This information should be useful to analysts selecting data mining tools to employ, as well as to developers aiming to produce better data mining products
Full-text available
1 Data mining tools are used widely to solve real-world problems in engineering, science, and business. As the number of data mining software vendors increases, however, it has become more challenging to assess which of their rapidly-updated tools are most effective for a given application. Such judgement is particularly useful for the high-end products due to the investment (money and time) required to become proficient in their use. Reviews by objective testers are very useful in the selection process, but most published to date have provided somewhat limited critiques, and haven't uncovered the critical benefits and shortcomings which can probably only be discovered after using the tool for an extended period of time on real data. Here, five of the most highly acclaimed data mining tools are so compared on a fraud detection application, with descriptions of their distinctive strengths and weaknesses, and lessons learned by the authors during the process of evaluating the products. ...
The growing interest in data mining is motivated by a common problem across disciplines: how does one store, access, model, and ultimately describe and understand very large data sets? Historically, different aspects of data mining have been addressed independently by different disciplines. This is the first truly interdisciplinary text on data mining, blending the contributions of information science, computer science, and statistics. The book consists of three sections. The first, foundations, provides a tutorial overview of the principles underlying data mining algorithms and their application. The presentation emphasizes intuition rather than rigor. The second section, data mining algorithms, shows how algorithms are constructed to solve specific problems in a principled manner. The algorithms covered include trees and rules for classification and regression, association rules, belief networks, classical statistical models, nonlinear models such as neural networks, and local "memory-based" models. The third section shows how all of the preceding analysis fits together when applied to real-world data mining problems. Topics include the role of metadata, how to handle missing data, and data preprocessing.
y illustrative of the tremendous potential of KDD technology. 1.1 Risk Management and Targeted Marketing Insurance and direct-mail retail are examples of businesses that rely on effective data analysis in order to make profitable business decisions. For example, insurers must be able to accurately assess the risks posed by policyholders in order to set insurance premiums at competitive levels. Overcharging low-risk policyholders would motivate such policyholders to seek lower premiums elsewhere. Undercharging high-risk policyholders would attract more high-risk policyholders because of the lower premiums. In both cases, costs would increase and profits would decrease. Effective data analysis leading to the creation of accurate predictive models is essential in order to address these issues. In the case of direct-mail targeted marketing, retailers must be able to identify subsets of the population that are likely to respond to promotions in order to offset mailing and printing costs.
The article focuses on application of data mining to business processes. The traditional approach to data analysis for decision support has been to couple domain expertise with statistical modeling techniques to develop handcrafted solutions for specific problems. More recently as of August 1, 2002, several trends have emerged to challenge this approach. One is the increasing availability of large volumes of high-dimensional data occupying database tables with millions of rows and thousands of columns. Another is the competitive demand for the rapid construction and deployment of data-driven analytics. A third is the need to give end users analysis results in a form they readily understand and assimilate, helping them gain insights they need to make critical business decisions. Moreover, knowledge discovery in databases techniques emphasizing scalable, reliable, fully automated, explanatory structures have shown that in data analysis, such structures supplement, and sometimes supplant, human-expert-intensive analytical techniques for improving decision quality.
There's content everywhere, but not the information you need. Content analysis can organize a pile of text into a richly accessible repository. This article explains two key technologies for generating metadata about content - automatic categorization and information extraction. These technologies, and the applications that metadata makes possible, can transform an organization's reservoir of unstructured content into a well-organized repository of knowledge. With metadata available, a company's search system can move beyond simple dialogs to richer means of access that work in more situations. Information visualization, for example, uses metadata and our innate visual abilities to improve access. Besides better access, metadata enables intelligent switching in the content flows of various organizational processes - for example, making it possible to automatically route the right information to the right person. A third class of metadata applications involves mining text to extract features for analysis using the statistical approaches typically applied to structured data. For example, if you turn the text fields in a survey into data, you can then analyze the text along with other data fields. All these metadata-powered applications can improve your company's use of its information resources.