Use-Centered Content

Goeffrey Adams

David Marques

Purpose

Labs.elsevier.com is a site where new ideas are put for customers and potential customers to test drive, evaluate and help us refine.

• To help innovate and test new ideas.
• To give participants a chance to influence Elsevier's direction for adding value.
• Focus is on end-user value.

Expectations

Of Elsevier

• 6 (minimum) to 12 (we hope!) new prototypes per year
• Email technical support, but very light
• Feedback on suggestions

Of Participants

• Test, evaluate, comment on prototypes
• Focus on end-user trials and evaluation

Present Status

• Hardware and systems software installed and operational in Oxford
• 12 prototype applications already loaded and running
• 6 Participants already enrolled
  • CSIRO - Australia
  • University of Toronto - Canada
  • Oxford University - United Kingdom
  • KAIST - Korea
  • Yale University - United States
  • Academia Sinica - Taiwan

Next Steps

• Define procedures and mechanics in more detail
• WEBEX conferences
• Involvement of additional key third parties
• Longer term outlook

For more information

• labs-admin@elsevier.com
• David Marques
  • Chief Technology Officer
  • d.marques@elsevier.com
• Geoffrey Adams
  • Director of IT Solutions
  • g.adams@elsevier.com

Maximized Efficiency of Discovery

Goal: Added value through powerful discovery tools

• Search
• Classification
• Clustering
• Thesaurus browsing

2004 Priorities

• Thesaurus term expansion
• Question answering
• Result set visualization
• Advanced find similar
• Analysis and data mining
• True, seamless combined search across all content, including MDL

Descriptive Search

Lots of text for initial query

• less precise (too many hits)
• better relevance (top 10 hits usually excellent)
• could be "OR" of text or concept vectors

Find similar: can be based on several dimensions

• Reference overlap (BioMednet used this)
• keywords
• "OR" of full text
• "OR" of concepts
• "AND" of concepts (Scirus)
• Concept vectors (concept plus weighting)

Richest Context of Content

Goal: Add value through multi-dimensional context for research and health information

• Citation linking
• Citation analysis
• Entity recognition

2004 Priorities

• Trend analysis
• Rich Linking
• Data sets, modelling
• Result set visualization

Restructuring Content: Granularity

Goal: Quicker to market, more agile products

• New web-service architecture around XML content
• XML Repository for all content

2004 focus

• XQuery programming for XML document handling
• Synthesized documents
• Permissions management
• Editorial Support through workflow and content management
• Internationalization support in products
• Arbitrary granularity of the content for apropriate contexts
• Exploring trade-off between load and retrieval

User-Centered Design

Goal: Maximize ease of use and user value of electronic products

• Elsevier-wide UCD group in 2001 -- HCI, not web designers
  • Understand users
  • Design user interfaces
  • Evaluate usability and accessability
• Focus groups regularly on large projects
• Scopus did monthly prototyping with customers for over 6 months during envisioning phase
• Field Studies at research labs
  • Case Western RNA research
  • U Toronto Pharmacology

Field Study at University of Toronto

Scope

• Pharmacology, Pharmaceutical Sciences, Clinical Neuroscience
• 5 principal investigators, 1 research librarian, 9 doctoral students, 1 master's student, 2 post-doctoral researchers
• Goals
  • understand information tasks and motivators for everyday research activity
  • gain insights on how to help research and those activities better
• Major activity areas of focus
  • Research projects and experimental activity
  • Manuscript preparation
  • Grant proposal activity
  • Education and course development

Field Study Setup

Method

• 18 semi-structured interviews
• 10 task diaries
• information service walkthroughs
• on-site observations and information use lifecycles

Information tasks

• Finding all studies in a topic area.
• Immediate inquiries on issues from research data.
• Validating exp findings by comparing with other research.
• Locating information unavailable locally on an as-needed basis - especially methods, materials, and procedures.
• Locating background, supporting info for establishing context and fitting findings into literature.
• Analyzing references, and finding authors supporting or arguing against findings. (A publishing strategy activity).

Findings: Roles

Role is a strong predictor of information sources

• Most graduate students used PubMed for literature searching
• Many started from Library website
• PIs used Library site for Ovid MEDLINE
• For known articles, Library's e-Journals
• End goal: Full text PDF format articles

Faculty much more efficient in searching; students very inefficient

• students spent 30-180 minutes just searching; average 99 min
• PIs 5-10 minutes searching
  • delegation of long searches
  • higher level of tool skill
  • more precise questions they are trying to answer

Tasks differ

• Doctoral students need full text, eg experimental procedures
• Post-docs: alerting and keeping on top -- by proactive searches, scientists' web pages
• Faculty: managing research information -- references, data, tasks

Findings: information use by activity

Grant proposal

• PubMed/MEDLINE
• Literature review
• Scientists' web pages -- be sure to list articles of key figures in the field
• Trade names for drugs
• Clinical trials

Course Development

• Intranet, BlackBoard
• quick review for treatments
• "perfect" papers
• slides, figures

Unmet needs

Complex citation analysis: cited for discussion, cited for method, cited for findings support, cited as exception

Case studies database/clearinghouse

MethodMatch: methods and protocols database

LitViz: visualization of summaries and trends in the literature

PiXearch: image database for searching

Integration with Customer Systems

Goal: Open, industry-standard web services interfaces for integration into customer applications and through new devices

• Some Web services available today
  • Search
  • Content retrieval
  • Journal TOC browse
  • Syndication
  • Metadata (citation information)
  • Attachments
• Key standards bodies
  • Member of Medbiquitous: donated Elsevier web services architecture
  • XQuery working group
  • NISO, DOI, OASIS, W3C

2004 focus

• RSS possible future
• More-Like-This
• Linking (reference, chemical, genomic)
• TabletPC pilots -- no compelling use cases yet

Questions and Answers

Problem: Content granularity

• Clinicians have specific questions with multiple constraints
• Clinicians at point-of-care need small, granular content answers
• The answers are contained in books and 70-page pamplets

Solution

• Clinical questions are of roughly 15 types/formats (Stanford)
• Content can be (semi-automatically at this time) coded to the finite set of questions -- at the paragraph/sentence level of granularity
• Content can be retrieved at the right level of granularity

Learning

Goal: Develop course materials and investigate technologies to supplement textbooks, assist instructors and students

• Course Management
• LiveInk

2004 Focus

• Continuing education
• Corporate training
• Learning management and assessment
• Learning content management
• High-end context, with built-in analysis functionality
• LABS: Site for testing new ideas with our customers
• Entity Trawler: Analytics coming to LABS

Entity Extraction

Identify entities (things) in a text corpus

• Examples: authors, universities... diseases, drugs, side-effects, genes... companies, law suits, plaintiffs, defendants...
• Use lexicons, patterns, NLP for finding all instances

Identify relationships

• Through co-occurrence
  • Relationship presumed from proximity
  • Example: author-university affiliation
• Through limited NLP
  • Semantic relations: causes, is-part-of, etc.
  • Examples: drug causes disease... drug is-treatment-for disease... a is-suing b...

Entity Analysis

Sumarize corpus

• Report on entities, relationships, numbers of occurrences

Enhance search

• Tag documents with and search on entities

Ad hoc analysis

• How do all drugs relate to this one disease?
• Over time?
• In the context of this institution?
• How does this institution relate to all other diseases?
• Show me the document evidence...

The Value of Analytics -- connecting the dots

• Goes beyond search: overviews, relationships, synthesized answers, insights
• Can be a true "discovery" tool: what is the value of finding a "hole" in current research?
• Requires exhaustive content, or enough to be representative of a subject

Current work

• Extracted all relationships into a database
• Built a simple web trawling front end (demo later)
• Imported relationships into free visualization tool -- Anacubis
• Trained computational linguist and programmers to support rule writing
• Evaluating other domains for value (economics, Neuroscience)