CS 584: Personal Information

Assistants
Lecture 1: Course Overview and Introduction

CS 584: Personal Information Assistants.

1/24/19 1
Spring 2019
 Lecture Plan
• What is IR? (the big questions)

• Course overview and topics

• Logistics

CS 584: Personal Information Assistants.

1/24/19 2
Spring 2019
 Information Retrieval
• Information retrieval is a field concerned with the
structure, analysis, organization, storage, searching,
and retrieval of information
Gerald Salton, 1968

• Compare to Google corporate mission:

Google's mission is to organize the world's
information and make it universally accessible and
useful.

CS 584: Personal Information Assistants.

1/24/19 3
Spring 2019
 Intelligent Information Access:
General Research Areas

Search, Ranking,
Recommendation,
Personalization

Text and Social Data Intent Inference, User

Mining, Classification Attention & Interactions,
Information Extraction Evaluation

CS 584: Personal Information Assistants.

1/24/19 4
Spring 2019
 Intelligent Information Access:
Examples (1)
Question Answering: Search, Ranking,
Recommendation,
Personalization

Text and Social Data Intent Inference, User

Mining, Classification Attention & Interactions,
Text-based
QA Information Extraction Evaluation
Crowds

Unstructured data Semi-structured data KB-based QA Structured data

Contextual (Social) Text Mining:

1. Novel temporal representation of text
c
c1
Word n

Represent
words as concept
vectors

2. Novel temporal text-similarity measurement

Word1 Word 2 Method for computing
semantic relatedness
using the temporal
representation

CS 584: Personal Information Assistants.

1/24/19 5
Spring 2019
 Intelligent Information Access:
Examples (2)
Personal Intelligent Assistants
Conversational Search
• Intelligent Assistants:
in Alexa Prize 2018
• Dialogue-based
Search Interfaces
• Explicit interaction

Search, Ranking,
Recommendation,
Personalization

User Behavior Models:

Text and Social Data Intent Inference, User Inferring User Attention
Mining, Classification Attention & Interactions, Memory Loss (Alzheimer’s)
Information Extraction Evaluation
Interactions (Parkinson’s)

CS 584: Personal Information Assistants.

1/24/19 6
Spring 2019
 Contextual Personalized Search
• Contextual Search on ubiquitous devices
– Grounding search in physical
and social context
– Incorporate sensor data
– Speech-based search: use acoustics,
other sensor input

• Personalized search:
– When, what, and how to personalize?
– Seamlessly access info when needed
– Decision support anytime, any topic
CS 584: Personal Information Assistants.
1/24/19 7
Spring 2019
 Lots of Data

Ex: Large Hadron Collider:

36 Billion events per minute.

How much of this data is

useful information?
CS 584: Personal Information Assistants.
1/24/19 8
Spring 2019
 Fundamental Mismatch
• Computers deal with data
• Humans (and systems) want useful information
• Examples:
– Structured Data (easiest): bitsàschemaàquery
– Text: bits à represented as words
– Audio: bits à speech recognition à words
– Images: bits à image processing à computer vision à
object recognition à <tags>?
– Video: image processing+audio+activity recognition à ?
CS 584: Personal Information Assistants.
1/24/19 9
Spring 2019
 Search Challenges (2002)
UMass CIIR report, 2002

• Global information access: Satisfy human

information needs through natural, efficient
interaction with a system … [over the world’s
data]… in any language.

• Contextual retrieval: Combine search

technologies and knowledge about query and
user context … to provide the most “appropriate”
answer for a user’s information needs.
CS 584: Personal Information Assistants.
1/24/19 10
Spring 2019
 A “classic” IR Task
• Given:
– A corpus of textual natural-language documents.
– A user query in the form of a textual string.
• Find:
– A ranked set of documents that are relevant to the
query.

Contrast with database/SQL queries

CS 584: Personal Information Assistants.

1/24/19 11
Spring 2019
 A search model
TASK Get rid of mice in a
politically correct way
Misconception?
Info Need Info about removing mice
without killing them
Mistranslation?
Verbal How do I trap mice alive?
form
Misformulation?

Query mouse trap

SEARCH
ENGINE

Query Results
Corpus
Refinement
CS 584: Personal Information Assistants.
1/24/19 Spring 2019 12
 Information Retrieval Challenges
• Understand user’s query (information need)

• Interpret and organize data (indexing, classification)

• Rank “documents” by expected utility for user

• Find answers to show to user (selection, presentation)

• Evaluate, improve search, repeat

CS 584: Personal Information Assistants.
1/24/19 13
Spring 2019
 Search Challenges (2012) SWIRL 2012 report
PROMISE 2012 report

• Beyond ranked list: enrich querying & results

• IR for all: empower the user to search & learn
• Capture context: current task, time, etc.
• Beyond document retrieval: complex data &
result integration
• Domain search: Verticals, apps, restricted data
• Evaluation: for new search types, tasks

CS 584: Personal Information Assistants.

1/24/19 14
Spring 2019
 Search Chalenges: SWIRL 2018 (1)
• https://www.damianospina.com/wp-content/uploads/2018/04/swirl3-report.pdf
• Decision Support over Pathways: Understanding and designing systems to
help people in making decisions.
• • Generating New Information Objects: Ad hoc generation, composition,
and summarization of new text, and layouts in response to an information
need.
• • Transparent/Explainable Information Retrieval: Explaining ranking
decisions. Providing reliable and responsible information access.
• • Cognitive-aware IR: Tracking and modeling user behavior and perception.
Modeling political-correctness of decisions. Identifying fake news and
provenance.
• • Societal impact of information retrieval: Understanding the long term
impact of IR on society and the economy.
• • Personal information access: Federated personal information search and
management (e.g. knowledge graphs). Biometrics for affective state.
• • Next Generation Efficiency-Effectiveness Issues: Efficient machine learning
inference. Resource-constrained search.
CS 584: Personal Information Assistants.
1/24/19 15
Spring 2019
 Search Chalenges: SWIRL 2018 (2)
• https://www.damianospina.com/wp-content/uploads/2018/04/swirl3-report.pdf
• Machine Learning and Search: Developing effective machine-learned retrieval models (e.g.
neural networks, reinforcement learning, meta-optimization).
• • Personalized interaction: Diversified and personalized interactions.
• • Conversational information access: Information-seeking conversations. Learning
representations for conversations.
• • New approaches to evaluation: Moving beyond the Cranfield paradigm, topical relevance,
and queries. Controlling for variability. Counterfactual evaluation and off-policy evaluation.
• • New interaction modes with information, multi-device search: Multi-device search.
• • Blending online and physical: Search in the context of mobile, smart environments, and
augmented/virtual reality.
• Task-specific representation learning: Adapting machine learned models for new search
domains.
• • Pertinent Context: Surfacing and using the relevant contextual information for search.
• • Success prediction: Formal models and principles to inform retrieval system design (build
the right bridge instead of build six bridges and see which survives).

CS 584: Personal Information Assistants.

1/24/19 16
Spring 2019
 History of the (IR) World: the Ancients

– Vanevar Bush: “As we may think”:

“Memex”, 1945 à proto-IR à desktop search

– Gerald Salton: founder of IR as field

CS 584: Personal Information Assistants.
1/24/19 17
Spring 2019
 History: the mainframe era (50s-70s)
• High-level programming
languages
• Fortran, Ada, Cobol, …
• Data structures
• Sorting
• Searching
• Routing
• Graph theory
• Internet (ARPA-net)
• Salton develops IR

CS 584: Personal Information Assistants.

1/24/19 18
Spring 2019
 IR System (80’s, 90’s)

Document
corpus

Query IR
String System

1. Doc1
2. Doc2
Ranked 3. Doc3
Documents .
.

CS 584: Personal Information Assistants.

1/24/19 19
Spring 2019
 History: Web 1.0 (90s)
• Distributed computing
• Mosaic, Netscape, IE
• Web search
– Alta Vista, Excite…

CS 584: Personal Information Assistants.

1/24/19 20
Spring 2019
 Google architecture (circa 1998)

CS 584: Personal Information Assistants.

1/24/19 21
Spring 2019
 Web 2.0: 2000’s

CS 584: Personal Information Assistants.

1/24/19 22
Spring 2019
 Now (2010à)
• Knowledge Graphs

• Deep Learning for IR

• Personal Assistants

CS 584: Personal Information Assistants.

1/24/19 23
Spring 2019
 And in this course…

CS 584: Personal Information Assistants.

1/24/19 24
Spring 2019
 Search Engine Architecture: Indexing

CS 584: Personal Information Assistants.

1/24/19 25
Spring 2019
 Search Engine Architecture: Querying

CS 584: Personal Information Assistants.

1/24/19 26
Spring 2019
 Course Logistics
• Lectures:
– Tuesday & Thursday: 1pm-2:15pm
– Following dates will be rescheduled or canceled:
• Office hours (tentatively):
– Tuesday,, Thursday, 3-4pm, Emerson E500
• Communication/website:
https://canvas.emory.edu/courses/39249

CS 584: Personal Information Assistants.

1/24/19 27
Spring 2019
 Course structure
• The course covers roughly 4 topics, with some overlap:
– Information retrieval and recommender systems (e.g., Bing,
Netflix)
– Personalization and contextualization (e.g., personal ranking,
local suggestions)
– Explicit and implicit feedback (ratings, clicks, engagement)
– Personal Information Assistants (1+2+3+domain specifics)
• Grading:
– Two implementation homeworks: 40%
– Paper presentations/leading discussions: 15%
– Final project: 45% (proposal, implementation, presentation,
report).

CS 584: Personal Information Assistants.

1/24/19 28
Spring 2019
 Homework projects

• HW1: Lucene or Whoosh over local collection e.g.,

movies, semi-structured search, e.g., "movies about
robots"
• HW2: (personalized) recommendation, e.g., movie
or song or book, can be local and/or remote
dataset. platform TBA.

CS 584: Personal Information Assistants.

1/24/19 29
Spring 2019
 Texts
• CMS: Information Retrieval in Practice
Croft, Metlzer Strohman
Free online: https://ciir.cs.umass.edu/irbook/
• MRS: Introduction to Information Retrieval,
Manning, Raghavan and Schütze, Cambridge University
Press (free online):
http://nlp.stanford.edu/IR-book/
• SUI: Search User interfaces, Marti Hearst, CUP, free
online:
http://searchuserinterfaces.com/
• Additional readings will be posted online.

CS 584: Personal Information Assistants.

1/24/19 30
Spring 2019
 Ideas for final projects
• Personalized ”Skill” for Alexa platform, e.g.,
personalized news briefing
• Personalized reminder/assistant App
• TREC 2018/2019 challenges, esp. conversational
search: https://trec.nist.gov/pubs/call2019.html

CS 584: Personal Information Assistants.

1/24/19 31
Spring 2019