Transcript Intelligent Interactions with Search Results

Intelligent Interactions with
Search Results
Getting Beyond Those Blue Results Lists
(or Smart Text)
Tom Reamy
Chief Knowledge Architect
KAPS Group
http://www.kapsgroup.com
Program Chair – Text Analytics World
Agenda
 Introduction: Search and Structure: Smart Text
–
Smart Text– foundation of text analytics
 Adding Structure to Unstructured Text
–
Dynamic Sections and more, Better Relevancy Calculations
– Complex Document Summaries, Deeper Personalization
 Case Study
–
Publishing: Processing 700K Proposals
 Beyond Search – Building on the Foundation
 Conclusions
2
Introduction: KAPS Group
 Knowledge Architecture Professional Services – Network of Consultants
 Applied Theory – Faceted & emotion taxonomies, natural categories
Services:
– Strategy – IM & KM - Text Analytics, Social Media, Integration
– Taxonomy/Text Analytics, Social Media development, consulting
– Text Analytics Quick Start – Audit, Evaluation, Pilot
 Partners – Smart Logic, Expert Systems, SAS, SAP, IBM, FAST,
Concept Searching, Attensity, Clarabridge, Lexalytics
 Clients: Genentech, Novartis, Northwestern Mutual Life, Financial
Times, Hyatt, Home Depot, Harvard Business Library, British Parliament,
Battelle, Amdocs, FDA, GAO, World Bank, Dept. of Transportation, etc.
 Program Chair – Text Analytics World – March 29-April 1 - SF
 Presentations, Articles, White Papers – www.kapsgroup.com
 Current – Book – Text Analytics: How to Conquer Information Overload,
Get Real Value from Social Media, and Add Smart Text to Big Data
3
Introduction:
Elements of Smart Text - Text Analytics
 Text Mining – NLP, statistical, predictive, machine learning
 Extraction – entities – known and unknown, concepts, events
–
Disambiguation - Ford
 Fact Extraction - ontology, relationships of entities
 Sentiment Analysis - Positive Negative – products, companies,
 Auto-categorization
–
–
–
–
–
Training sets, Terms
Rules – simple – position in text (Title, body, url)
Boolean– Full search syntax – AND, OR, NOT
Advanced – DIST(#), ORDDIST#, PARAGRAPH, SENTENCE
Based on taxonomy/ontology
4
Enterprise Text Analytics
 Search is still #1 = 30-50% of applications
 New Standard Search – facets (more and more metadata), autocategorization built on taxonomies, clustering
 Trend = Text Analytics/Search as Semantic Infrastructure
–
Platform for Info Apps (Search-based applications)
 SharePoint – Major focus of TA companies – fix problems with
taxonomy/folksonomy
–
Hybrid workflow – Publish document -> TA analysis -> suggestions
for categorization, entities, metadata -> present to author
 External information = more automation, extraction – precision
more important
5
Enterprise Text Analytics
Adding Structure to Unstructured Content
 Beyond Documents – categorization by corpus, by page, sections
or even sentence or phrase
 Documents are not unstructured – variety of structures
 Text indicators to define sections of the document
– Objectives, Abstract, Purpose, Aim – all the “same” section
– Sections – Specific - “Abstract” to Function “Evidence”
– Start of section is easy – where does it end?
 Experiment – clusters / vocabulary to define section
– Textual complexity, level of generality
6
Enterprise Text Analytics
Categorization and Beyond
 Need to develop flexible categorization and taxonomy – tweets to
200 page PDF
 Rules or sample documents?
– Need more precision and granularity than documents can do
– Training sets – not as easy as thought
– Applications require sophisticated rules, not just
categorization by similarity
 Separate logic of the rules from the text
– Stable rules, changing text
 Scores – relevancy with thresholds
– Not just frequency of words
7
8
Enterprise Text Analytics
Document Type Rules
 (START_2000, (AND, (OR, _/article:"[Abstract]",
_/article:"[Methods]“), (OR,_/article:"clinical trial*",
_/article:"humans",
 (NOT, (DIST_5, (OR,_/article:"approved", _/article:"safe",
_/article:"use", _/article:"animals"),
 If the article has sections like Abstract or Methods
 AND has phrases around “clinical trials / Humans” and not words
like “animals” within 5 words of “clinical trial” words – count it and
add up a relevancy score
 Primary issue – major mentions, not every mention
– Combination of noun phrase extraction and categorization
– Results – virtually 100%
9
Case Study
Publishing Project: Reed Construction Data
 700,000 Proposals – Wide Variation
 Process Proposals – extract data – 30-50 types
 Current Manual Process – Internal Teams
–
Expensive and Slow
 Structure Variety of Unstructured Documents
–
–
Generate Table of Contents
Generate Sections and Capture Text
 Semi-automatic extract Key Information
 Save Time & Money, Flexible Hiring, New Offerings
10
Publishing Project: Example Rules
Automated Table of Content
11
12
Publishing Project: Example Rules
Key Data Extraction





Bid Dates/Times
Roles (Architect, Designer, etc.) – names and addresses, etc.
Project Attributes – Cost, Invitation Number, Parking, etc.
Some Easy, Some Hard – Address!
Example:
–
ARCHITECT: MICHEAL KIM ARCHITECTURE
– 1 HOLDEN STREET BROOKLINE, MA 02445
– P: (617) 739-6925 F: (772) 325-2991
 Technique – create broad and stable templates, variation in the
text
13
14
15
Publishing Project: Example Rules
Key Project Data
16
Publishing Project: Process & Approach
17
Smart Search: Metadata, Metadata, Metadata
 Basic Facets: Date, People, Organization, Content-Type
 Advanced Facets: Materials, Methods, Project Attributes, etc.
– Context dependent
 Deep personalization
– Selection of facets by role, community, task, content
 Smart Summarization
– Better conceptual description
– Complex summaries – key data, document sections, etc.
 Smart Search – beyond simple relevancy
18
Building on the Foundation: Applications
Pronoun Analysis: Fraud Detection; Enron Emails
 Patterns of “Function” words reveal wide range of insights
 Function words = pronouns, articles, prepositions, conjunctions,
etc.
– Used at a high rate, short and hard to detect, very social,
processed in the brain differently than content words
 Areas: sex, age, power-status, personality – individuals and
groups
 Lying / Fraud detection: Documents with lies have:
– Fewer, shorter words, fewer conjunctions, more positive
emotion words
– More use of “if, any, those, he, she, they, you”, less “I”
 Current research – 76% accuracy in some context
19
Building on the Foundation: Social Media
Beyond Simple Sentiment
 Beyond Good and Evil (positive and negative)
–
Degrees of intensity, complexity of emotions and documents
 Importance of Context – around positive and negative words
Rhetorical reversals – “I was expecting to love it”
– Issues of sarcasm, (“Really Great Product”), slanguage
–
 Essential – need full categorization and concept extraction
 New Taxonomies – Appraisal Groups – “not very good”
– Supports more subtle distinctions than positive or negative
 Emotion taxonomies - Joy, Sadness, Fear, Anger, Surprise, Disgust
–
New Complex – pride, shame, confusion, skepticism
20
Building on the Foundation: Applications
Behavior Prediction – Telecom Customer Service
 Problem – distinguish customers likely to cancel from mere threats
 Basic Rule / Intention
–
(START_20, (AND, (DIST_7,"[cancel]", "[cancel-what-cust]"),
–
(NOT,(DIST_10, "[cancel]", (OR, "[one-line]", "[restore]", “[if]”)))))
 Examples:
–
customer called to say he will cancell his account if the does not stop receiving
a call from the ad agency.
– cci and is upset that he has the asl charge and wants it off or her is going to
cancel his act
 More sophisticated analysis of text and context in text
 Combine text analytics with Predictive Analytics and traditional behavior
monitoring for new applications
21
Building on the Foundation:
Current Applications
 Survey Analysis – Add analysis of free text
 Automated Essay Scoring – Second Generation
–
Beyond words (polysyllabic) to meaning
 Story Telling – Data Heavy, Sports, Finance
–
90% of news machine written by 2025, books?
 Legal Review / eDiscovery
–
TA- categorize and filter to smaller, more relevant set
– Payoff is big – One firm with 1.6 M docs – saved $2M
 Voice of the Customer / Employee / Voter
–
Analysis of Blogs, Tweets, Social Networks
– Early Identify problems with products and services
– Customer Relationship & Brand Management, Fraud Detection
22
Smart Text : New Directions - Integration
Deep Integration – Text Analytics
 New Forms of Rules – Combine Text Mining and Text Analytics
Incorporate clusters – CLUSTER Operator
– Like SENTENCE but more flexible, dynamic
–
 More Dynamic Sections
–
–
–
Build up from “Categorization” of sentences – based on co-reference
Smaller units – Appraisal Taxonomies for Subjects, Build Larger
Units
Complex Units – Collections of Paragraphs based on meaning
 Sentence Level Sentiment Techniques for Subjects
–
Smarter Relevancy – not frequency – develop new scoring
 Graph Database
–
Enrich queries, improve relevancy
– Graph traversal – browse mechanism.
– Text Analytics to fill the database – “real” semantics
23
Conclusions
 Text Analytics can feed/extend Big Data and Cognitive Science
applications
–
Discover structure in (un)structured text
– Apply text analytics to sections of document – new kinds of relevancy
– Creating multiple views into data inside text – smart search results –
interactive (facets plus)
– Modular design – better search, new applications, Watson
 Future: Cognitive Computing:
–
Learns, discover patterns based on context, highly integrated,
meaning-based, highly interactive
– Text Analytics adds depth of meaning
 Future – Women, Fire, and Dangerous Things
–
Text Analytics and Cognitive Science = Metaphor Analysis, deep
language understanding, common sense?
24
Questions?
Tom Reamy
[email protected]
KAPS Group
Knowledge Architecture Professional Services
http://www.kapsgroup.com