Note: Descriptions are shown in the official language in which they were submitted.
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
A METHOD AND SYSTEM FOR A DOCUMENT SEARCH SYSTEM
USING SEARCH CRITERIA COMPRISED OF RATINGS
PREPARED BY EXPERTS
CROSS-REFERENCE TO RELATED APPLICATIONS
This application claims the benefit of U.S. Provisional Application No.
60/220,398 filed July 24, 2000.
This application relates to co-pending provisional applications:
(1) serial number 60/216,469, filed July 6, 2000, titled "System and
Method for Matching Psychological Profile Information with Target
Information";
(2) serial number 60/215,492, filed July 6, 2000, titled "System and
Method for Anonymous Transaction In A Data Network and User Profiling of
Individuals Without Knowing Their Real Identity.";
(3) serial number 60/252,868, filed November 21, 2000, titled "Interactive
Assessment Tool.";
which are incorporated fully herein by reference.
COPYRIGHT NOTICE
A portion of this patent document contains material which is subject to
copyright protection. The copyright owner has no obj ection to the facsimile
reproduction by anyone of the patent document or the patent disclosure, as it
appears in the Patent and Trademark Office patent file or records, but
otherwise
reserves all copyright rights whatsoever.
TECHNICAL FIELD
This invention relates to the field of electronic computer related systems.
More particularly, the present invention relates to a method and system for
the
automated search of document files using search criteria based upon document
identifiers generated by expert reviewers in lieu of key words in context, and
relates to a method and system for indexing such documents.
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
BACKGROUND ART
A technical problem presently exists in the attempt to use modern day
search engines for searching for documents on the world wide web (the "web").
Generally the problems facing users is that almost all search engines search
for
key words in all or portions of the documents. The problem with key word
searches is that an extremely large number of documents are usually returned
by
the search engine, all of which typically must be read or scanned to find
those
few documents or that one document that contains the desired information.
LexisTM, AltavistaTM, YahooTM, are examples of such key-word based search
systems. Some specialized databases , such as the database of U. S. issued
patents, contained at the site www.delphion.com and at the U.S. Patent Office
web site www.uspto.gov permit customized searches with known parameters in
lieu of key words, such as Inventor name, assignee name, patent agent name,
etc.,
but also include key-word searches. These searches also suffer from the same
malady: returning many documents which must generally be read to find the
pertinent ones.
An article titled "The Search Engine as Cyborg" By LISA GUERNSEY,
The New York Times, June 29, 2000 further describes the problem. The article
explains that "To cope, many search engines have concluded that simply
indexing
more pages is not the answer. Instead, they have decided to rely on the one
resource that was once considered a cop-out: human judgment. Search engines
have become more like cyborgs, part human, part machine." For example, a
highly ranked search service is AskJeevesTM, which prods people to narrow
their
queries by picking from a list of questions and answers written by the
company's
employees.
Both GoogleTM and Northern LightTM rely on computers and software to
scan and index the Web, but human judgment is part of the mix. At Google, Web
pages that are linked from authoritative Web sites are deemed most relevant.
At
Northern Light, librarians constantly fine-tune their directory structure and
come
up with names of categories used for sorting Web sites. Similarly, some music
sites appear to have songs indexed with ratings by distributors or listeners
as to
2
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
genre, type such as vocalist, instrumental, folk, jazz, hip-hop, etc. so that
selections by these criteria can be made. See for example, www.listen.com.
Some other efforts have been made to solve this problem. For example
Manning & Napier Information Services Inc.TM of Rochester, New York has
several products whose technologies are based on research and development in
information retrieval (1R) and artificial intelligence (AI), including natural
language processing (NLP), information extraction, agents, link analysis,
question-answering, data visualization, data fusion, knowledge discovery,
knowledge management, genetic algorithms, neural nets, and cross-language
information retrieval (CLIR). This system is built around a process whereby
the
searcher is requested to give the system much more data than just a few key
words
(a paragraph, for example, to attempt to describe the document contents). The
system then constructs a linguistic vector based upon the paragraph given as
the
search argument and attempts to find equivalent vectors in its document
databases. This is not a general Inteinet search engine system but rather a
proprietary one that has its own databases of documents which have been
previously processed to produce linguistic vectors which characterize the
documents, based on the word contents of the documents.
Another approach to solving the basic key word search problem has been
developed by Dr. William Woods, at Sun MicrosystemsTM, Inc. Laboratories. Dr.
Woods has addressed the problem wherein the articulation of the desired
subject
matter is different that that used by the authors of the documents being
searched.
This is sometimes referred to as the "synonym problem" although Dr. Woods
characterizes the problem in a broader connotation by referring to it as the
"paraphrase problem" and his general solution approach is called "conceptual
indexing" and more specifically as "subsumption technology." Subsumption
technology is used to automatically integrate syntactic, semantic, and
morphological relationships among concepts that occur in the material, and to
organize them into a structured conceptual taxonomy that is efficiently
useable by
retrieval algorithms and also effective for browsing. Dr. Woods conceptual
indexing approach is described in a number of papers including
3
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
"Natural Language Technology in Precision Content Retrieval" by Jacek
Ambroziak and William A. Woods, Proceedings of the International Conference
on Natural Language Processing and Industrial Applications, August 18-21,
1998,
Moncton, New Brunswick, Canada, and
"Knowledge Management Needs Effective Search Technology," by
William A. Woods, Sun Journal, March, 1998
both of which are incorporated fully herein by reference.
As these papers describe, the Sun Microsystems Laboratories' Conceptual
Indexing Project was created to address the problems cited above and to
improve
the convenience and effectiveness of online -information access. A central
focus
of this project is the "paraphrase problem," in which the words in a query are
different from, but conceptually related to, those in material one needs. This
project developed techniques that use knowledge of word and phrase meanings
and their inter-relationships to find correspondences between the words one
uses
in their request and concepts that occur in text passages.
In this solution to the problem, they use taxonomic subsumption
algorithms that exploit generality, or subsumption, rather than synonymy. That
is,
when a concept is more general than another, the more general concept is said
to
subsume the more specific one and concepts are organized around the notion of
conceptual subsumption rather than synonym classes. This relates more general
concepts to more specific ones without losing information and enables a
retrieval
algorithm to automatically find subsumed concepts. The algorithms do not
automatically explore more-general terms, so the level of generality is
controlled
by the searcher's choice of query terms. For example, if one asked for "motor
vehicles," he would get trucks, buses, cars, etc., whereas if he asked for
"automobiles," he would get cars and taxicabs but not trucks and buses. The
algorithm can let one know about more-general concepts that subsume the
searcher's query, in case he wants to generalize his request, but it does not
make
this decision without the user's knowledge and consent.
This approach is further taught in U. S. Patent No. 5,724,571 issued March
3, 1998 ( Woods) titled "Method and apparatus for generating query responses
in
4
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
a computer-based document retrieval system" which is also incorporated fully
herein by reference.
The key concepts in the Woods and Manning & Napier approaches are
that a two step process is required: First a linguistic vector or structured
conceptual taxonomy must be constructed by the indexing engine when the
material is indexed, and second a special retrieval algorithm is used to find
either
equivalent linguistic vectors or combinations of morphological and semantic
subsumption relationships that connect concepts in the request with concepts
that
occur in the indexed material. While both approaches appear to provide
significant efficiency over key word searches, and while the Wood approach
appears to be the more efficient of the two, both have the same disadvantages.
Both systems require first a baseline database of target documents and second
a
powerful lexical computing engine to create the linguistic vectors or
combinations
of morphological and semantic subsumption relationships. Only then can the
search technologies of the two be used.
However these systems as well as the earlier described databases
containing popularity-based ratings use fixed, pre-determined indexing
algorithms
to mathematically combine words and phrases in a description vector which can
be matched with a similarly computed vector based on search criteria inputted
by
the user.
What is needed is a database system with individual document ratings
from experts in the field where these expert ratings are based on an accepted
taxonomy of attributes for the specific field rather than an unrelated
mathematical
algorithm. It would be these expert ratings that would be the basis of a
search
rather than an algorithmatic computation built around the words in the
document.
And similarly needed is a search engine capable of mapping inputted search
attributes to this expert ratings attribute indexed database.
Biomedicine is largely a knowledge industry. While a physical product,
the medicine, does have to be developed, tested, manufactured and delivered,
the
knowledge of how to do so and the knowledge of which product works best in
particular cases contributes most of the value.
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
A second characteristic of biomedical knowledge is that it is highly
dynamic. At the research level, significant advances in our understanding of
biomedical phenomena happen on a weekly basis. Therefore, biomedical
professionals have an ongoing need to keep up with the advances relevant to
their
own specialty area. Such needs have become particularly acute in health-care,
because patients can now use the Web to learn about the latest developments
themselves; as a result, they demand increasingly detailed and timely
information
from health-care professionals.
NEEDS RELATING TO CENTRALIZATION
There is as yet no centralized source of biomedical information on the
web. The information one seeks may be available somewhere on the web. The
hard part is finding it. There are thousands of biomedical Web pages, ranging
from individual sites to corporate sites. These sites generally fall into the
following categories:
~ Government research center sites
~ University biomedical sites
~ Commercial firm sites (including vendor firms)
~ Biomedical journal sites
~ Individual researcher/professor sites (usually only a few pages
with papers and links)
A list of the major Web sites can be found in an Appendix in the recently
published book, "From Alchemy to IPO; The Business of Biotechnology," by
Cynthia Robbins-Roth, Perseus Books Group, 2000, ISBN 0-738202533, which
is incorporated herein by reference.
NEEDS RELATING TO SEARCH STRATEGIES
Despite the availability of an enormous amount of information, this
information is not indexed or summarized for easy consumption.
1. Existing human-edited directories, such as Yahoo, do not have the
skilled biomedical personnel or the time to adequately index biomedical pages.
6
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Human-edited directories, such as Yahoo, generally index only a small fraction
of
the Web, because of the cost of having human workers look at each page.
2. Existing search engines that mechanically index pages, such as Alta-
Vista, also have limitations as indicated above: the number of irrelevant
pages
generated; and the poor quality of links generated.
NEEDS RELATING TO CONTEXTUALIZATION
Another problem caused by specialized content is incomplete
understanding. No individual is a specialist in all subsets within a
particular
discipline. Thus, there are always parts of the content that are more
understandable than others. This is particularly so when the user is a non-
specialist and the content is, say, a biomedical research paper. There is a
need to
provide information in a form such that the user can quickly grasp the
essentials
of concepts underlying the content.
NEEDS RELATING TO PERSONALIZATION
An additional issue of importance to the effective dissemination of
biomedical content is the manner in which content is served to the user.
Virtually
all content on the web today is served in a one-size-fits-all mode.
Nevertheless,
studies have shown that people learn better when content is presented in a
manner
more suited to their own individual cognitive style.
NEEDS RELATED TO MULTIDIMENSIONAL TAXONOMIES.
Another problem with presently known search approaches is that they
address taxonomies which are, basically, hierarchical i.e. one-dimensional.
However in many domains, in the biomedical arena for example, an n-
dimensional taxonomy is more appropriate. That is, a biomedical development
might be considered mundane from a technical standpoint, yet highly
significant
from a social or business viewpoint. While it is true that this "significance"
issue
might be expected to be handled by the way the query is structured (i.e. from
the
technical viewpoint or from the social or business viewpoint), systems such as
the
Sun and Manning & Napier systems cannot handle these issues because of the
pre-defined mathematical indexing algorithms they use.
7
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
The solution to these technical problems therefore is to provide a method
for analyzing a database of documents wherein a multi-dimensional taxonomy of
attributes for a specific domain can be developed and used to tag the related
documents with significance rating indicia, which can then be searched by a
qualitative matching engine.
SUMMARY OF THE INVENTION
The present invention provides a solution to the needs described above
through a system and method for creating and maintaining a Biomedical
document database, wherein the documents have been reviewed by biomedical
and other experts, who have assigned taxonomic based attribute indicia to each
document wherein a specialized search engine can rapidly retrieve relevant
documents based upon the commonly known taxonomy.
A method is disclosed for creating and maintaining a biomedical reference
database in which each reference is associated with rating attributes across a
mufti-dimensional taxonomy. "Reference" is defined as a URL or literature
citation associated with an original research article or similar content. The
input
interface components for this database are:
1. An Interface for expert input of references: Expert input may be
provided by the editorial staff of a leading trade journal, by selected
leading
practitioners in the field, or by other similar expert mechanism. For the
purposes
of biomedical research expert knowledge, "leading practitioners," or "domain
experts" are defined as those within the top echelon of researchers within the
biomedical sub-specialty in question, as judged by publication record, public
recognition through competitive awards (such as the Nobel prize) and peer
evaluations.
2. An Interface for expert ratings of references: Expert input (as defined
in the previous section) regarding each reference is collected under a
multidimensional rating taxonomy. Each reference receives ratings under each
relevant taxonomic category from appropriate expert sources. A composite
rating
8
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
may be computed from the mean of multiple ratings received under a single
taxonomic category.
3. An Optional interface for expert commentary: Expert ratings may be
accompanied by text commentary on each reference. Such commentary received
from multiple expert sources can provide additional insight into the relevance
of a
particular reference.
4. An Optional interface for acquiring profiles of experts: Input from
each expert source may be normalized for certain variables, based on
attributes
measured for that expert source. For example, mean ratings and distributions
collected and analyzed from each expert source may allow that expert's rating
input to be expressed as standard deviations from the mean.
Still other embodiments of the present invention will become apparent to
those skilled in the art from the following detailed description, wherein is
shown
and described only the embodiments of the invention by way of illustration of
the
best modes contemplated for carrying out the invention. As will be realized,
the
invention is capable of modification in various obvious aspects, all without
departing from the spirit and scope of the present invention. Accordingly, the
drawings and detailed description are to be regarded as illustrative in nature
and
not restrictive.
9
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
BRIEF DESCRIPTION OF THE DRAWINGS
The features and advantages of the system and method of the present
invention will be apparent from the following description in which:
Figure 1 illustrates an exemplary Internet distributed system
configuration.
Figure 2 illustrates a representative general purpose computer server
configuration.
Figure 3 illustrates a block diagram of a preferred embodiment of the
process of rating documents and storing them in a database.
Figure 4 illustrates a block diagram of a preferred embodiment of the
process of search and retrieval from the database.
Figure 5 illustrates a preferred embodiment of an exemplary data
structure for the indicia storage related to a document
Figure 6 illustrates a preferred embodiment of an exemplary data
structure of a multidimensional taxonomy structure.
Figure 7 illustrates a preferred embodiment of an exemplary data structure
of a taxonomy structure for use with the present invention.
Figure 8 illustrates a preferred embodiment of an exemplary input screen
showing how to rate significance in a first dimension.
Figure 9 illustrates a preferred embodiment of an exemplary input screen
showing how to rate significance in subcategories of first dimension.
Figure 10 illustrates a preferred embodiment of an exemplary input screen
showing how to rate significance in a second dimension.
Figure 11 illustrates a preferred embodiment of an exemplary input screen
showing how to rate significance in a third dimension.
Figure 12 illustrates a preferred embodiment of an exemplary input screen
showing how to rate significance in subcategories of third dimension.
Figure 13 illustrates a preferred embodiment of an exemplary input screen
showing how to rate significance in a fourth dimension.
Figure 14 illustrates a preferred embodiment of an exemplary input screen
showing a summary review significance pattern.
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Figure 15 illustrates a preferred embodiment of an exemplary input screen
showing how to capture a document from the PubMed database.
Figure 16 illustrates a preferred embodiment of an exemplary input screen
showing haw to submit a reference to the database of the invention.
Figure 17 illustrates a preferred embodiment of an exemplary input screen
showing how to submit a critique of a reference to the system of the
invention.
11
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
DETAILED DESCRIPTION OF THE INVENTION
The present invention provides a solution to the needs described above
through a system and method for creating and maintaining a Biomedical
document database, wherein the documents have been reviewed by biomedical
and other experts, who have assigned taxonomic based indicia to each document
wherein a specialized search engine can rapidly retrieve relevant documents
based
upon the commonly known taxonomy.
A method is disclosed for creating and maintaining a biomedical reference
database in which each reference is associated with rating attributes across a
multi-dimensional taxonomy. "Reference" is defined as a I1RL or literature
citation associated with an original research article or similar content. The
input
interface components for this database are:
1. An Interface for expert input of references: Expert input may be
provided by the editorial staff of a leading trade journal, by selected
leading
practitioners in the field, or by other similar expert mechanism. For the
purposes
of biomedical research expert knowledge, "leading practitioners," or "domain
experts" are defined as those within the top echelon of researchers within the
biomedical sub-specialty in question, as judged by publication record, public
recognition through competitive awards (such as the Nobel prize) and peer
evaluations.
2. An Interface for expert ratings of references: Expert input (as defined
in the previous section) regarding each reference is collected under a
multidimensional rating taxonomy. Each reference receives ratings under each
relevant taxonomic category from appropriate expert sources. A composite
rating
may be computed from the mean of multiple ratings received under a single
taxonomic category.
3. An Optional interface for expert commentary: Expert ratings may be
accompanied by text commentary on each reference. Such commentary received
12
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
from multiple expert sources can provide additional insight into the relevance
of a
particular reference.
4. An Optional interface for acquiring profiles of experts: Input from
each expert source may be normalized for certain variables, based on
attributes
measured for that expert source. For example, mean ratings and distributions
collected and analyzed from each expert source may allow that expert's rating
input to be expressed as standard deviations from the mean.
As appropriate, input interfaces may be designed to collect information
relating to time of input, and other relevant attributes. Keywords, summaries,
abstracts, slides, audio, editorials, interactive modules, educational content
and
other items related to a reference may also be stored in the database.
The computational and search functions associated with this database are
built around the unique attributes stored for each reference. Any conventional
quantitative search algorithm may be utilized. For example, a search query may
utilize taxonomic category information (example: cancer), date of reference
(example: significant developments in the last 3 months), bias of party making
the
search (for example, an individual may be interested in cancer, but
particularly
interested in new clinical treatment modalities- not in basic science research
advances) or any other pertinent attributes. The relative contribution of each
attribute comparison to the final match reported may be manipulated by any
prediction algorithm. The result returned by the search algorithm may be a
score.
Results of a search may be displayed as a list of references sorted by score,
or it
may be searched further by additional criteria.
Each reference may be linked to derivatives such as keywords, summaries,
abstracts, slides, audio, editorials, interactive modules, educational content
and
other items related to a reference.
The biomedical reference rating database may be created and searched,
using the elements described in the preceding paragraphs. The salient features
of
this system are its multi-dimensional taxonomy of rating categories, and its
strict
dependence on expert ratings. An expert panel-based mechanism, or other
similar
13
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
peer-review-based mechanism, is the only traditionally credible method for
assigning significance in the biomedical domain. For this reason it remains
quite
distinct from "popularity contest" rating systems wherein the source of the
rating
is not known to be expert. One value of a credible rating system is that the .
database and search functions described above generate results which are
credible
and trusted by biomedical professionals and, by extension, those non-
professional
or non-specialist audiences that rely on the judgement of biomedical
professionals.
The invention is described in more detail below by way of descriptions of
a preferred embodiment, however those skilled in these arts will recognize
that
various specific implementations of the functions described will accomplish
the
results of the invention.
OPERATING ENVIRONMENT
The environment in which the present invention is used encompasses the
general Internet-based systems haxdware and infrastructure along with well
known electronic transmission protocols both conventional and wireless.
Some of the elements of a typical Internet network configuration are
shown in Figure 1, wherein a number of client machines 105 possibly in a
remote
local office, are shown connected to a gateway/hub/tunnel-server/etc. 106
which
is itself connected to the Internet 107 via some Internet service provider
(ISP)
connection 108. Also shown are other possible clients 101,103 similarly
connected to the Internet 107 via an ISP connection 104, with these units
communicating to possibly a central lab or office via an ISP connection 109 to
a
gateway/tunnel-server 110 which is connected 111 to various enterprise
application servers 112,113,114 which could be connected through another
hublrouter 115 to various local clients 116,117,118. Any of these servers 112,
113,114 could function as a database server for the storage of the indexed
documents and messages of the present invention as well as the server for the
search engine of the present invention, as more fully described below.
An embodiment of portions of the Document Search System of the present
invention can operate on a general purpose computer unit which typically
includes
14
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
generally the elements shown in Figure 2. The general purpose system 201
includes a motherboard 203 having thereon an input/output ("I/O") section 205,
one or more central processing units ("CPU") 207, and a memory section 209
which may have a flash memory card 211 related to it. The I/O section 205 is
connected to a keyboard 226, other similar general purpose computer units 225,
215, a disk storage unit 223 and a CD-ROM drive unit 217. The CD-ROM drive
unit 217 can read a CD-ROM medium 219 which typically contains programs 221
and other data. Logic circuits or other components of these programmed
computers will perform series of specifically identified operations dictated
by
computer programs as described more fully below.
The preferred Embodiment
Some of the distinguishing factors in the invention over the prior art are:
1. The rating taxonomy which, in our case, is multidimensional and
developed over the biomedical domain. The multidimensional nature of the
classification of knowledge is a key element because those skilled in these
arts
have been able to come up with a satisfactory systematization of biomedical
knowledge "significance" without it. That is, significance or relevance (both
in its
individual or in its broader societal sense) is dimensional. For example a
particular research paper in the area of cancer will be rated for significance
in a
particular way if the dimension is, say, drug development, but in a completely
different way if the dimension is basic science impact, technological impact,
or
societal impact.
Once again, an n-dimensional taxonomy is used wherein each dimension
is independent. That is, the same item may eventually appear under several
dimensions of the taxonomy (see Figure 6). This is novel and important for
developing a comprehensive taxonomy in the biomedical arena. Most traditional
taxonomies are hierarchical. For example the music rating taxonomy of
Listen.com would have various branches and the final categories would all be
distinct.
The distinction is important because people look at knowledge in a multi-
dimensional way. Significance means different things, even to the same person,
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
depending on the dimension. For example, a biomedical development might be
considered mundane from a technical standpoint, yet highly significant from a
social or business standpoint.
2. An input interface for tagging items with a significance rating
according to the taxonomy (this includes the peer-review mechanisms described
below, which are different from the art in that they employ expert panels).
The
rating is done by acknowledged experts in the dimension of relevance, in a
peer-
review process analogous to grant proposal review, or pre-publication
manuscript
review. It is this process that brings credibility to the ratings.
3. A profiling interface (which may itself contain several embedded
technologies) for creating profiles of users. Again, a multidimensional
taxonomy
of user profiles is used, which is also believed to be distinct from the art.
For
example, one dimension may be knowledge taxonomic domains, the other
dimension may be some or all of the following profile layers:
a. User ID
b. predicted preferences (based, for instance on inferred cognitive style)
c. reported preferences (by user)
d. reported experience (by user)
e. locally documented experience (such as assessment test results,
purchase records)
f. composite profile layer
4. A personalized search strategy based on each individual's weighted
preferences in the categories of our multidimensional taxonomy. A person might
want to know the most significant recent developments in the field of cancer,
but
what does that really mean? If that person's preferences are for developments
of
translational interest or, even more complicating, developments of
translational
interest which offer opportunities for investment, the present system of
keyword
searches could never satisfy this requirement. Thus, our rating taxonomy
allows
for novel ways of searching for "significant" knowledge. The search is
personalized with respect to what would be considered significant.
16
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
An alternative embodiment may include predictional algorithms based on
prior research findings. These may optionally be used to compute the
"composite
profile" of a user. Calculating a composite profile may be useful in
situations
where information about a user's interest in specific taxonomic categories
might
be derived from several different sources such as known professional
experience,
self reported experience, self reported interest, or actually recorded
performance
in educational assessments or other interactive modules. In such cases, it may
be
desirable to compute and use a composite profile calculated from these
disparate
sources of input information. The actual method of computation may be an
arithmetic mean, a weighted mean based on hypothesis or previous findings, or
any other suitable computation. A detailed example of how composite profiles
may be constructed from psychometric data is described in Applicant's co-
pending US provisional patent application serial number 60/216,469, filed July
6, 2000, titled "System and Method for Matching Psychological Profile
Information with Target Information" which is incorporated fully herein by
reference.
In alternative embodiments, any appropriate quantitative matching
algorithm and search engine may be used. For example, Liquid EnginesTM Inc.
has developed a generally matching algorithm which may be used for this search
purpose.
FUNCTIONAL IMPLEMENTATION
A method is disclosed for creating and maintaining a biomedical reference
database in which each reference is associated with rating attributes across a
multi-dimensional taxonomy.
In the preferred embodiment, the search function of this invention will
serve professional users in a far more sophisticated manner than currently
available search methods, with directory/search engine features more suited to
current needs.
17
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
CONTENT SORTING AND SUMMARIZATION: SEARCH
STRATEGIES
In the preferred embodiment, the primary goal is to make relevant
biomedical content as easy as possible to find - and to present it in a form
that is
easy to digest.
To select references for rating two methods can be used. First the existing
brand-name credibility and editorial capabilities of the undisputed opinion-
leading
trade journals, academic institutions, and science journalists in a biomedical
specialty may be used; these include journals such as NatureTM and ScienceTM,
institutions such as U.C. BerkeleyTM and StanfordTM, and journalistic
enterprises
like SoundVisionTM and NovaTM. Secondly, to determine content relevance,
online expert panels organized under subject categories, can be used. The
process
is analogous to peer review of grant proposals, manuscripts and scientific
presentations, and can generate substantial community interest. A systematic
method of processing ratings and editorial comments on each month's most
significant developments in the biomedical field has been developed by the
inventor. Users will sort content based on these ratings and their own
interest
profiles.
Each search result lists a variety of derivative products. For example,
instead of providing the original research articles, a link to summaries and
short
editorial commentary relating to such articles may be provided. Each summary
is
also linked to the original content.
CONTEXTUALIZATION
In the preferred embodiment, each derivative page is hypertextually linked
to a central biomedical knowledge database, allowing for quick educational
reference to the underlying concepts. This adds value to the consumer's need
to
understand the content more fully, in the shortest amount of time.
In any academic field, not only is the original research important, but also
the comments and feedback from others. Was the study properly done? Was it
refuted later? What does it really mean, in the big picture of things? Today,
18
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
systematic editorial comment on research topics is difficult to find. In this
respect, comments from colleagues can be invaluable, and the site of the
preferred
embodiment will also be a community where biomedical professionals can share
(unpublished) comments and opinions of the type normally exchanged in faculty
dining halls. As indicated above, software-based technologies that facilitate
and
promote such exchanges among physicians and among biotech researchers have
been developed.
PERSONALIZATION
In order to customize the display of content to each user, some
information will be stored and served in two formats (expert and non-expert).
A
biomedical researcher with a Ph.D. might see a different version of selected
content than a non-specialist.
These characteristics of the preferred embodiment are now described in
additional detail. The overall functional operation of the system in general
will
now be described.
Refernng to Figure 3, in the preferred embodiment, documents are
selected for special significance to this community of users 303. The expert
reviewers supply ratings for each document 305. Expert input may be provided
by the editorial staff of a leading trade journal, by selected leading
practitioners in
the field, or by other similar expert mechanism. For the purposes of
biomedical
research expert knowledge, "leading practitioners" are defined as those within
the
top 10% of researchers within the biomedical sub-specialty in question, as
judged
by publication record, public recognition through competitive awards (such as
the
Nobel prize) and peer evaluations.
Rating values are assigned to search elements 307 according to the
taxonomy indicia (the taxonomy is described in more detail below). Expert
input
(as defined in the previous section) regarding each reference is collected
under a
multidimensional rating taxonomy. Each reference receives ratings under each
relevant taxonomic category from each appropriate expert source. A composite
rating may be computed from the mean of multiple ratings received under a
single
taxonomic category. This rating process and calculation is explained in more
19
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
detail below. When completed the annotated document is stored in the database
309. A generalized data structure for a document in the database is shown in
Figure 5 although those skilled in the art will understand that there are a
multitude of ways to structure such an index.
Referring now to Figure 4, a generalized proprietary database search and
access system 400 is described. Input to such a system 400 may be directly or
through the Internet from a client machine which may be a Personal Computer
(PC) or from a Personal Data Assistant (PDA) device such as a 3ComTM hand-
held device, and may use any number of communications protocols such as
HTML, XML, WAP, WML, etc. In Figure 4, a user contacts the system through
its web page 403. The system checks the user's password and ID 40S in order to
determine whether the user is a subscriber to the service. If not 407 the user
is
requested to become a subscriber and if not the system exits 409, 411. If the
user
is a subscriber 413 he is given a search format page wherein he can enter the
specific search criteria 41S in which he is interested. The database is
searched
417 for matches to his input criteria, and a page of pointers to relevant
documents
is returned 419. The user may request another search if he desires 421, 427 or
he
may terminate the search 423 at which time his search time and costs are
calculated and saved 42S for periodic billing of the user. Alternative
billing/subscription schemes may be used wherein the subscriber is billed a
flat
fee per period.
Within this general framework of document selection, coding anc~ indexing
with subsequent search and retrieval, the unique aspects of the invention are
now
described in more detail.
The Taxonomy
A taxonomy for biomedical knowledge for use in a preferred embodiment
of the present invention is shown in Figure 7.
Rating accordin,~ to the Taxonomy
The rating is done by acknowledged experts in the dimension of relevance,
in a peer-review process analogous to grant proposal review, or pre-
publication
manuscript review. It is this process that brings credibility to the ratings.
One
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
cannot simulate this with reviews written by online users. When one reads a
review for a book on Amazon that says: "...most useful book I have read in the
past two years..." one doesn't know what qualifications that individual has,
whether in fact he or she has read more than one book in the past two years,
or
whether he/she is a nut.
In the preferred embodiment, document reviewers perform the reviews as
follows. Referring to Figure 7, a document reviewer initially selects the
Enterprise Domain 701 type B 703, type T 705, type C 707 or type S 709. The
reviewer then selects a type 713 in the Disease Group 711, if applicable, and
a
type 717 in the Underlying Concepts Group 715 if applicable.
If the reviewer selects B 703 or T 703 in the Enterprise Domain 701, he
then selects a type Reductionist 721 or Abstractive 723 in the Investigative
Emphasis group 719. If B 703 was selected in the Enterprise Domain 701, the
reviewer then selects a topic type 727 from the 1B1 Topic Group 725. If T 703
was selected in the Enterprise Domain 701, he selects a topic type 731 from
the
1T1 Topic Group 729. If C 707 was selected in the Enterprise Domain 701, the
reviewer then selects a topic type 735 from the 1 C 1 Topic Group 733. If S
709
was selected in the Enterprise Domain 701, the reviewer then selects a topic
type
739 from the 1 S 1 Topic Group 737.
Thus, a particular document might be coded
1B, 2G, 3L, 4R, 1B1-03
for a document that was deemed to be about
"Basic Science, describing a Genetic Disease, with Scientific Certainty,
using a Reductionist investigative emphasis, wherein the gene/DNA structure is
described."
Similarly, a searcher for such a document would specify in the search
criteria
1. - basic science;
2. - genetic disease;
3. - scientific certainty;
4. - reductionist; and
21
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
5. - gene/DNA structure.
and the search engine can convert this to
1B, 2G, 3L, 4R, 1B1-03
In some cases, a search request for
1. - basic science;
2. - genetic disease;
3. - any;
4. - any; and
5. - gene/DNA structure
would produce any documents matching
1B, 2G, any, any, 1B1-03.
Those skilled in these arts will recognize that various other specific
codings for such taxonomic structures may be used without deviating from the
spirit of the invention.
User Profiles
A personalized search strategy based on each individual's weighted
preferences in the categories of our multidimensional taxonomy. For example,
the
attributes 1B, 2G, 3L, 4R, 1B1-03 of the above example might carry associated
weightings assigned based on prior observation, or by hypothesis. Thus, in one
example, one might assign importance to each attribute as in: 1B/9.4, 2613.4,
3L/3.1, 4R/2.5, 1B1-03/1.1, with each weighted value reflecting the user's
personalized level of interest in matching each category. Such preference
information might, for example, have been collected at the time of initial
user
registration at the web site. A person might want to know the most significant
recent developments in the field of cancer, but what does that really mean? If
that
person's preferences are for developments of translational interest or, even
more
complicating, developments of translational interest which offer opportunities
for
investment, the prior art systems of keyword searches could never satisfy this
requirement. Thus, our rating taxonomy allows for novel ways of searching for
"significant" knowledge. The search is personalized with respect to what would
be
considered significant by a given individual.
22
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Search/Qualitative matching Engine
As an example, a search function that takes x inputs on the user profile,
and y inputs on each database item and returns a list of documents, each with
a
mathematical probability of a match, may be used. Any appropriate quantitative
matching algorithm may be used. For example, Liquid EnginesTM, Inc. has
developed a generally applicable matching algorithm which may be used for this
purpose.
An Exemplary Functional System Description
An exemplary preferred embodiment of the system of the invention could
encompass two types of discussion groups, or "bulletin boards":
~ Public Groups: Wherein users can create a discussion group, and add
comments to it and share comments with their "friends" or other members.
A group creator "manages" the group, and can add, remove, and manage
members.
~ Subscription Groups: Wherein users pay a fee to see comments and
feedback of articles and content from experts within a field. Articles and
comments are rated within a set of categories by experts and are then
searchable by date or relevance for those subscribing to those groups.
What are the most significant things that have happened within this field
within the last month?
The ability to search content by significances according to Taxonomy
criteria as described above.
Features
Both sets of groups share some common functionality, and as such the
preferred embodiment design would be such that the core of the system is
common. For the purposes of this spec, the two group types will be
distinguished
by "Public" groups and "Subscription" groups.
23
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Feature Public Subscription
No, only by system
admin.
' Group Creation Yes, by anyone Groups also only
"
"
Taxonomy
correspond to
_ fields
l ~ ' ~
Add Members # Yes, by owner ' ~yone can join if
they
."_.. ~.............._ ...
paY
~
w If you don't pay,
you are
Remove Members ' Yes, by member, ' removed. A user can
or by
owner remove themselves and
forfeit their subscription
No~ only by system
Group Deletion ' Yes, by owner
administrator
_ _ a. .. _. ._ . m ~.~A . . .
..m~o
Post Messages up y ~Yone in the Only by "elite" members
~ s
o
Rate Messages nncategories' No Multiple members can
rate
the same message
Y
Delete Messages Yes, by owner 3 Only by system
administrator
In the preferred implementation below, the system will generate a
framework in which both types of groups can exist in the same suite of tables
in a
database, with different front end implementations.
Definitions
We can define four roles played in the group system:
~ Public Groups
~ Public Owner - This user created and manages a group
~ Public Member - This user has been added to a public
group.
~ Subscription Groups
~ Elite Member - Posts messages, and rates postings by
other members .
~ Subscription Member - Observes postings by elite
members and can search database by taxonomy criteria.
24
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Database Implementation
The implementation consists of additions of a user membership tables,
group tables, and finally a messaging table for determining messages contained
within a particular group.
Membership Table "tbl Membership"
The membership table simply indicates whether the user is a member of a
group or not. It also contains the time at which point a users membership
expires
(if at all), as well as a billing reference (customizable depending on what
billing
service
is used.)
Column CondensedNullableDescription
Name Ty a
;. ~ .. ~~. ~~ .. ~~~.........,............~_.....
.._ .. . . .
uniqueidentiNOT
UserID User 1D who is a member
Eer NULL
a
~lqueidentiNOT ' Group ID that UserID is a
member in...
GroupID fier NULL ~ This column is indexed.
NOT Flags defining what type of
member this
Perms integer NULL user is in a group
ExpireTi datetime . NULL When their membership expires
~
me
,
W
,
of the
last
~ The billing or tracking ID
_ varchar(64)' NULL commerce payment associated
BillinglD with this
payment. (Type may change)
For all "public" group, the "ExpireTime" and the "BillinglD" will be
NULL for Public Owners and Public Members.
For a "subscription" group, the "Expire Time" and "BillinglD" will be
NULL for Elite Members, but not for Subscription Members.
These constraints should NOT be used to determine what type of group the
user is included in.
This table contains a PRIMARY KEY which is (UserlD, GroupID)
In addition, a UNIQUE constraint is placed upon (LJserID, GroupID) to
prevent multiple inclusions of a user in a group.
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
The Perms is a bitfield (where any value can be on or off) which described
the type of privileges a user has within a group.
Perms ~ ~ Meaning . _
Value v~
Dec Binary,. ' Categorywhen 0 when 1
, ~
1 0...00000001 User can not User may
add users to group add users to group
User H . ,
n User can
2 0...00000010List User ca remove other users
not from
within remove users from
group
Grou group
p
4 0...00000100 User may not User may
view user list for view user list for
group group
_... .
.
6 User may not fr User may
8 0...00001000
~ post messages to group~ post messages
to group
Message User may not User may
16 0...00010000 List delete messages
delete messages from from
group
within ou
~' p
Group ~ User may
32 0...00100000 User may not view messages from
view messages from
group
... ........
up
64 0 ...00100000, ~' - ~ -
Additional flags will be added as needed.
26
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Group Table "tbl Groups"
The group table outlines the details of a particular group:
Column Condensed ''
~ NullableDescription
Name Type ,
GroupID ~lqueidentif~ NOT The unique ID for this Group
ier NULL
OwnerID ~ ~lqueidentifALL The Userm of the owner (NULL
in
s ier subscription groups)
Creation datetime NOT Time of group creation
Time NULL
Name varchar(64)~ The "short" name of the group
L
Descripti varchar(1024~ ~L The "long" description of this
group.
on )
Flags Integer ~ NOT Flags determining type and membership
NULL access to group (see below
BillingU varchar(1024ALL Optional billing URL for subscription
RL ) groups.
BillinglD ~ varchar(64)a ~,L ~tional billing information for
commerce
systems (Type may change)
The "Flags" field is a bitfield (so multiple values may be on or off) which
maps to the following values:
Flag Meaning
Value m u,~~e~~..-
a ~
~
~
Dec Binary ~ when 0
~ when 1
1 0...00000001 Public Group Subscription Group
Appears in list of groups all Does not
appear in list
2 0 00000010 of groups all users can
users can see ~ .. . _..
see
~
~
4 0...00000100 Requires inmtation
Anyone can join group
_ ~ _. . __._ ma.. . .~m.~. ... _....~ .._.~a..~,_._
. .._. _~ ~'om Owner .
8 0...00001000 - ~ -
16 0...00010000 - -
More flags will be added as they are needed.
27
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Messages Table "tbl Messages"
The Messages Table contains messages posted to all groups.
4 _.. ~....... ,.
Column CondensedNullable: Description
i
Name Ty a
H
Messagem uniqueidentiNOT ~ The m
of this message (unique)
fier ~ NULL
Userm ~ uniqueidentu'NOT The User who posted this message
fier NULL
ToUserm ~lqueidenti~L An optional "To" user for private
fier ~ messages (e.g. to group owners,
etc.)
Groupm fie queidenti~ The group m this message was
' posted to
L
PostTime datetime ~ The time of the post
L .....,....._,...,~.",~,~,........,.,.
~.. ". . ~.. .. .....
Subject fr ~ ~ch~(256T ' The subject of this post
' N
0
L
~
N
1
Message v~char(204NOT The message or comment from
the user
8) NULL about the content (if any)
pe ntentTy ~teger ~,L Tna type of information being
commented
o
ContentUR varchar(102 The URL or information being
. ~L
L 4 commented upon
The ContentType is the type of information being commented about:
Content Type ~ ,Meaning
0 E No content
___. ,..e. m..~..,a "~"",~,..~..~ ~..~~c...
..,~~. .~~e.~.., , .~.~e_ ...a.~.~_
" BioExpertise Site
Family i
1 URL (within family
of
sites)
Advertisement URL
2 (comment on
advertisement)
External URL (outside
of
sites)
~~
Dead URL
(URL has
4 been checked and
no
longer works)
28
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
An index should be created for GrouplT? to accelerate searching for all
messages within a group.
Ratings Table (tbl Ratings)
The Ratings Table contains ratings for subscription based groups. Ratings
are based upon the relevance of a particular reference to a taxonomic
category, as
described above.
Column
~ Condensed ~ Description
Type t Nullable
Name
Messagem uniqueidentifierE ALL The m of this message
~ (unique)
Userm uniqueidentifierNOT ~ The User who is rating
this
NULL message
Group>D uniqueidentifier' NOT The relevance of this
message to
NFL . his group
RatingTime datetime AL The time of this rating
L
RatingA smallint AL The rating in category
A
~ _. L
RatingB smallint AL ~ The rating in category
B
L
RatingC smallint 4 ALL ~ The rating in category
C
RatingZ smallint ~ The rating in category
Z
L
a
Given~a MessagelD that is to be rated and a UserID who is rating a
message, the "rating" script will do:
SELECT * FROM tbl Groups WHERE (Flags & 1 =1);
The expert user will rate the relevance to these references according to the
Taxonomy shown in Figure 7, which will then generate a list of ratings. For
some
29
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
"references" there will be no rating, in which case no value will be added to
the
database. So, for example, the user is presented with:
BioCritique.com
Article posted July 5th, 2000 by "cancerexpert" to group "Cancer" in
Biocritique:
Date: July 5th, 200012:23AM GMT
Content: (See content below)
(citation)
Subject:
Rate this Reference (Select Categories)
E You are logged in as "acureisoutthere". Please rate cited article on
each of the following attributes (scale of 1-10)
Category Rate in
Category
GENOMES, POPULATIO_NS__ _ ANDEVOLUTION 6
GENE l DNA STRUCTURE ~ 2
-_ .. _. __ ~ __._.
RNA SYNTHESIS, STRUCTURE, FUNCTION
y,;~~ . ~,..~ " ,,.., ".".. ~..
PROTEIN SYNTHESIS, STRUCTURE,
FUNCTION
METABOLIC PATHWAYS
MACROMOLECULAR TRANSPORT ~~
Rate Message i
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
When the user clicks "Rate Message in Checked Categories" the user then
goes through a process of rating as follows:
"~;~,a___....~.__~_....,..._..._......._._._.__........___....._.
~__~....~_.__._..__..._.~~.~.._..7....~..~.~_~.....w..~..._...___.,", _ _
BioCntique com ..~.... ., ...... ...... . ...rt.... ....-,:.,Y"~"~"",.........
.....~,~"i~i
Article posted July 5th, 2000 by "cancerexpert" to group "Cancer" in
Biocritique:
Date: July 5th, 2000 12:23AM GMT
Content: (See content below)
(citation)_ .. ... .. ... _...
' Rate by "Investigative Method"
You are logged in as "acureisoutthere". Please rate article's relevance to
"GENOMES, POPULATIONS, and EVOLUTION" in the following
categories:
Investigative Method
Category ... . Rating
Not relevant Relevant
;~
Reductionist C ~ O O O O .~"".~ t"'"..~
...................................w......................~"~,. .
Abstractive ~ O ~'a ~:'_ ~:'.~ C ~, O O
....,~.iiii~y~~uSV.-~,..., iii i. i..~...yl ... .iyiJiii17Je4'4Wi .,
Next Section
. , .., . . .. _ _ ,. _
Finally, rating this message based on tbl Ratings database table is updated
with 2 rows:
reference ID a UserID of # Groupm of "GENOMES~,~~' Current
of article '' acureisoutthere POPULATIONS AND Time ' ~ i 0 3 ' .
EVOLUTION"
GroupID of "PROTEIN
Reference ID ~ UserID of ~ SYNTHESIS, ' Current 0 p 2
of article acureisoutthere STRUCTURE, Time ~ '
FUNCTION"
31
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
The document classification sub-system measures the user's perception
and evaluation of the document. A set of questions in the form of analysis
screens
is posed to the user regarding the relevance of the document being rated,
which
the user responds to by selecting an option box ("Highly irrelevant,"
"irrelevant,"
"Slightly irrelevant," "Neutral," "Slightly relevant," "relevant," and "Very
relevant.") Sample rating categories include: "Basic Science Impact";
"Technology Impact"; "Business Impact"; "Societal Impact"; and "Clinical
Impact." Each category is scored on a seven-point scale (-3 to +3) where a
score
of-3 is "Highly Irrelevant" and +3 is "Very Relevant."
An Exemulary Embodiment
A present exemplary embodiment of the invention is found at the web site
www.biocritique.com. Once a user has logged on to the system she is able to
post
an article, rate an article, critique an article or search for articles that
correlate
most closely with the specific interests of the user. Those users who are
permitted
to post and rate an article are members of a selected "expert panel" who have
agreed to participate in this system.
Expert Panels and Article Selection
In the exemplary embodiment, the work of the expert panels in the
BioCritique Forums rapidly forms a database of rated articles and reviews from
which users can obtain significant information relevant to their specific
needs, i.e.
the latest important developments in their own and related fields. Panels of
15 top
experts in each topic area post, map and critique articles deemed to have a
high
impact in their field. Every month, each of the 1 S panelists on the Bioethics
Forum will attempt to post two articles they consider to be of major
importance.
They will then anonymously rate and critique two articles from their postings
or
those of other panelists, thereby providing a first level of peer (i.e. panel-
specific )
review. User comments and ratings, tracked and reported separately, will
provide
a second level of validation. Each article selection will thus be subjected to
review
first by other members of the panel, then by members of the professional
32
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
community. This process protects the credibility of the Forum with regard to
article selection.
Knowledge-Mapping and Pattern-Matching
In the exemplary preferred embodiment, Panelists will map articles using
BioCritique's multi-dimensional taxonomy of significance (see below). Users
store quantitative profiles of their interests rated across the same taxonomy.
Pair-
wise correlations are performed to sort the database for each user. A search
of the
BioCritique database is thus based on an intuitive "pattern-matching" concept
akin to the way human beings relate to information in the real world. Users
can
store multiple profiles, thereby customizing them to different needs.
The fundamental ingredients of the mapping (rating) taxonomies
employed within Biocritique are dimensionality and analog scaling,
specifically
designed for biomedical knowledge. Unlike keyword taxonomies, which are
binary and one-dimensional, in the exemplary embodiment BioCritique looks at
each obj ect in its database in 6 primary dimensions, rating each on a scale.
For
example, the Attributes dimension scores the following parameters on a scale
of 1
to 9:
0405 HOW NOVEL (HYPOTHESES AND FINDINGS)
0410 HOW CONTROVERSIAL (HYPOTHESES AND
CONCLUSIONS)
0415 HOW RIGOROUS (EXPERIMENTAL DESIGN AND
EXECUTION)
0420 HOW READABLE (ARTICLE)
Similarly, the Perspectives dimension is scored across the following
categories:
0101 BASIC SCIENCE IMPACT
0102 TECHNOLOGICAL llVIPACT
0103 BUSINESS IMPACT
0104 SOCIETAL IMPACT
0105 CLINICAL IIVVIPACT
33
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
Rating any one of these categories triggers the appearance of
subcategories. For example, a rating in the CLINICAL IMPACT category would
reveal the following subcategories:
0905 CLINICAL INTERVENTIONS
0910 DIAGNOSIS
0915 DISEASE PREVENTION
0920 EPIDEMIOLOGY l PUBLIC HEALTH
0925 DISEASE ETIOLOGY
0930 CLINICAL TOXICOLOGY
The expertise dimension contains a large number of biomedical specialties
ranging from Biostatistics to Toxicology. Disease dimension categories are
subcategorized. A rating under Cardiovascular would trigger the following
subcategories:
ACUTE CORONARY SYNDROMES
ARRHYTHMIAS
CVD: PREVENTION & RISK MANAGEMENT
VASCULAR BIOLOGY
HEART FAILURE
STRESS TESTING l NON-INVASIVE CARDIOLOGY
CVD: CLINICAL TRIALS
HYPERTENSION
CVD: INTERVENTIONAL & DEVICES
In all, there are several hundred rating categories in this scheme. Because
the ratings are quantitative (rather than binary, as in the case of keyword-
based
systems), it is possible to perform pair-wise correlation analyses between
rating
iapatternsle. Measures of proximity in this multi-dimensional space thus
permit
mapping and relational retrieval of disparate data types (research articles,
drugs,
company programs, book reviews) using a common syntax. The purpose of
building mufti-dimensional coordinates for each object in the database is to
permit
sorting of research articles according to each user's personal mix of
interests and
values.
34
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
An exemplary preferred embodiment of the invention is now described
with respect to Figures 8-17.
In Figure 8 a first rating screen 800 is illustrated, indicating that the user
is to rate the significance in a first dimension 801. Shown on the screen are
five
basic categories 807. The user chooses one or two categories by indicating
which
categories are not relevant 803. In this example "clinical impact" has been
chosen. For each category chosen the user chooses a relevancy category on a
scale of 1-9. Here the user has selected an indication which has a value of 5.
805.
On completion of the rating process the user clicks on the button 809 to go to
the
next section. These selections would produce a basic "Impact Domain" selection
vector, for example, that would look like this:
BSI 0
TI 0
BI 0
SI 0
CI 5
In Figure 9 the next section shows the subcategories of first dimension
related to the "Clinical Impact" selection 900. Once again, the user selects
the
categories shown that are not relevant 903 and for the categories deemed
relevant,
a rating selection is made 905. In the case indicated, the selection vector
would
be:
CI 6
D 3
DP 0
EPH 0
DE 4
CT 0
The user is then asked to rate the significance in a second dimension
related to "clinical impact" 1001 in Figure 10. Once again the categories in
this
second dimension which are deemed not relevant are selected 1003. For each
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
category deemed relevant a rating in indicated 1005. In the case indicated,
the
selection vector would be:
AAI 0
A 0
CM 8
CC 4
D 0
E 0
EM 4 .
EN 0
EP 0
GA 0
GE 0
Referring now to Figure 11, the user is requested to rate the significance
in a third dimension ("disease states") 1101. As before the user elects those
categories not relevant 1103 and for the ones deemed relevant a rating is
selected
1105. In this case the selection vector would look like:
AL 0
AR 0
AS 0
CA 0
CAR 8
DI 5
MH 0
OB 6
OS 0
GH 0
ID 0
Referring now to Figure 12, the user is asked to rate the significance in
subcategories of third dimension 1201, which in this instance relate to
"cardiovascular subcategories." Here the user again elects the categories that
are
36
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
"not relevant" 1203 and for each category deemed relevant, a rating is
selected
1205. In this case the selection vector would look like:
ACS 5
AR 0
CVDP 8
VB 0
HF 0
ST 0
CVDC 0
HY 0
CVDI 0
Finally, referring to Figure 13, the user in this example would be asked to
rate General Attributes, which in our rating scheme would be rating
significance
in a fourth dimension 1301. In this case all categories are deemed relevant
1303
and selections made for each 1305. Here the significance vector would look
like
this:
IMP 5
NOV 7
CON 2
RIG 6
REA 1
SPE 8
Given a combination of all of these significance selections, a summary is
shown in Figure 14. As indicated, this exemplary rating of a given document by
a user would produce the significance vectors as indicated above and these
would
be stored with the document. Another user who would review the same document
and provide a similar rating could very well select different categories and
different significance ratings even for the same categories. In a preferred
embodiment, these different significance vector values are averaged and the
resulting vector with the averages is saved with the document along with a
"number of raters" value which is used to compute the new average. For
37
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
example, looking at the significance vector associated with the description of
Figure 9 above, if we had a second review of this document the system would
execute the acts to produce a significance vector with averages for each
category
as follows:
1st Rater 2nd Rater New Vector
CI 6 4 5
D 3 2 2.5
DP 0 1 0.5
EPH 0 0 0
DE 4 8 6
CT 0 5 2.5
Raters=1 2
As can be seen, the panel of Expert Reviewers will, over time, review the
documents submitted multiple times and this rating structure permits a
valuation
of a general consensus of the experts opinions relative to a given document.
Additional personal critiques of the documents can be inputted and saved with
links to the specific document, wherein these critiques can be found whenever
a
document is displayed to a searcher. Figures 15-17 disclose an exemplary set
of
screens which facilitate downloading a reference (Fig.15), submitting the
reference to the BioCritique database (Fig.16), and adding personal comments
to
a reference (Fig.17).
Searching for Documents
Registered users of the system containing the invention (hereinafter
"BioCritique") can create and save a set of significance vectors to be used
regularly thereafter whenever they sign on, or a special one-time set of
significance vectors can be generated. This set of significance vectors are
the
same as those indicated above, and are generated by the user going through the
same set of screens as shown in Figures 8-14.
When a user wants to find a set of documents matching hislher
significance vectors, in the preferred embodiment, the search is conducted by
doing a pair-wise correlation between the user specific significance vectors
and
38
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
the significance vectors stored for each document. This pair-wise correlation
is
performed using the Pearson correlation coefficient method (which is explained
below). A Pearson correlation coefficient ("r") is calculated for each
document
and the documents then sorted with those having the highest "r" value first.
The
first 10 such documents meeting a minimal "r" value level are then presented
to
the searching user.
Pearson Correlation Coefficient
The Pearson Correlation indicated above is described in standard statistical
textbooks such as
Romesburg, H.C. (1984). Cluster Analysis for Researchers. Lifetime
Learning Publications. Pp. 334.
Sharma, S. (1996) Applied Multivariate Techniques John Wiley & Sons.
Pp. 493.
Srivastava, M.S. and E.M. Carter. (1983) An Introduction to Applied
Multivariate Statistics. North-Holland. Pp. 394.
Tabachnick, B.G. and L. S. Fidell. (1996) Using Multivariate Statistics.
Harper Collins. Pp. 860,
but for completeness is described generally as follows.
The correlation between two variables reflects the degree to which the
variables are related. The most common measure of correlation is the Pearson
Product Moment Correlation (called Pearson's correlation for short). When
measured in a population the Pearson Product Moment correlation is designated
by the Greek letter rho (cp). When computed in a sample, it is designated by
the
letter "r" and is sometimes called "Pearson's r." Pearson's correlation
reflects the
degree of linear relationship between two variables. It ranges from +1 to -1.
A
correlation of +1 means that there is a perfect positive linear relationship
between
variables. A correlation of -1 means that there is a perfect negative linear
relationship between variables. It would be a negative relationship because
high
scores on the X-axis would be associated with low scores on the Y-axis. A
correlation of 0 means there is no linear relationship between the two
variables.
39
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
The formula for Pearson's correlation takes on many forms. A commonly
used formula.is shown below. The formula looks a bit complicated, but taken
step
by step as shown in the numerical example below, it is really quite simple.
A numerical example is as follows:
X Y
1 2
2 5
3 6
~~- ~X~Y
r= N
(~XZ - (~X)a )(EYZ - (~Y)2 )
N N
~XY = (1)(2) + (2)(5) + (3)(6) = 30
~X=1+2+3=6
~XZ=12+22+32=14
~Y=2+5+6=13
~Y2=22+52+62=65
N=3
~XY-~X~Y/N = 30 - (6)(13)/3 = 4
~XZ - (~X)Z/N =14 - 62/3 = 2
r = 4 / (2)(8.6667) = 4 / 4.16333
~Y2 - (~Y)2/N = 65 - 132/3 = 8.667
_ .9608
This value, 0.9608, would say that the numbers in the X column are highly
correlated with the numbers in the Y column (a value of +1.0 meaning the
numbers were perfectly correlated).
CA 02415608 2003-O1-09
WO 02/08946 PCT/USO1/23058
In our example here, if the X column numbers were derived from a user's
inputted answers to three types of ratings, and the Y column were numbers
associated with a document's significance vector, then this high correlation
(0.9600 would characterize this document as highly likely to be of significant
interest to this user and his given search criteria.
Having described the invention in terms of a preferred embodiment, it will
be recognized by those skilled in the art that various types of general
purpose
computer hardware may be substituted for the configuration described above to
achieve an equivalent result. Similarly, it will be appreciated that
arithmetic logic
circuits are configured to perform each required means in the claims for
performing the various features of message recognition, message creation,
message storage and connection to a mobile telephony system. It will be
apparent
to those skilled in the art that modifications and variations of the preferred
embodiment are possible, such as different mobile telephony systems may be
used, different communications media such as wireless communications, as well
as different types of PCDs may be used by addressees and or senders of various
types of electronic messages, all of which fall within the true spirit and
scope of
the invention as measured by the following claims.
41
