605.744: Information Retrieval, Spring 2007

Lecturer

Paul McNamee, <paulmac@apl.jhu.edu>,

Textbooks

• Introduction to Information Retrieval by Manning, Raghavan, and Schutze, Cambridge University Press, 2007. Not available in print; however the text is currently available in digital form at Stanford.
• Students who desire to own a hardcopy IR reference should consider Managing Gigabytes by Witten, Moffat, and Bell, or Information Retrieval: Algorithms and Heuristics by Grossman and Frieder.

Course Times and Location

• Lecture: Tuesday, 7:15pm - 10:00pm, Kossiakoff Center Room K-2.
• Rather than holding fixed office hours, I will meet with students by appointment; however, email is expected to be the primary means of out-of-class communication.

Course Overview

This course covers the storage and retrieval of unstructured digital information. Topics include automatic index construction, retrieval models, textual representations, efficiency issues, search engines, text classification, and multilingual retrieval.

Grading Policy

Grades will be given based on the A,B,C, etc..., scale, per university policy. Work for the class will include homework assignments, an independent research project, a midterm exam, and classroom participation (e.g., quizzes, oral presentations). Refer to the course outline for details.

Academic Integrity

Work for this class is expected to be the result of individual effort; however, unless explicitly prohibited, it is perfectly acceptable to make use of published examples and source code from the literature or public domain - but only if attribution is given. Furthermore, while it is permissible to discuss the general nature of lecture material and assignments with your peers, this does not extend to discussing or revealing solutions or source code. Students are expected to uphold the academic integrity of the university. Students using without reference, published material or copying the work (i.e., particulary source code) of another individual will face consequences such as receiving a zero on the assignment and having the matter referred to the dean. Contact me if you have any questions, no matter how slight, about this policy, or if you have questions about a particular assignment.

Assigned Readings

• 1/23/07, Michael Lesk, The Seven Ages of Information Retrieval, 1995.
• (optional, not assigned): If you liked that paper, try How Much Information Is There in the World?", also by Lesk, the author of "Practical Digital Libraries" by Morgan Kaufmann.
• 1/23/07, IIR, Chapters 1 & 2. Key ideas: Boolean document representations and query evaluation in chapter 1; word normalization in chapter 2.
• 1/30/07, Damashek's N-gram paper (only available in hard copy)
• 1/30/07, IIR, Chapters 3 & 4. Key ideas: Coping with mispellings; wildcard search; building an inverted file (pay particular attention to section 4.1 of the text)
• 2/6/07, IIR, Chapters 5 & 6. Key ideas: Dictionary and Index Compression; Term weighting
• 2/20/07, Term Weighting paper by Salton and Buckley (hardcopy)
• 2/20/07, IIR, Chapters 7-9 Key ideas: Vector Space model, Evaluation (of accuracy), Query Expansion
• 2/27/07, IIR, Chapters 11, 12 Key ideas: Probabilistic model, statistical language models
• 3/6/07, IIR, Chapters 13-15 Key ideas: Text classification, Naive Bayes (NB), feature selection, k-nearest neighbor (kNN), support vector machines (SVMs). Ignore section 14.3, and don't worry about the math in Chap 15 - get a conceptual understanding.
• 3/6/07, Thorsten Joachims, Text Categorization with Support Vector Machines, A tech report that also appeared at ECML-98.
• 3/20/07, IIR, Chapters 19-21 Key ideas: Characteristics of Web documents, Hyperlink-sensitive methods, Crawling.
• 3/20/07, Nature and Science articles by Lawrence and Giles (hardcopy).
• 3/27/07, Michael Wesch's, The Machine is Us/ing Us on YouTube.
• 4/3/07. Hull and Grefenstette, 'Querying across languages: A dictionary-based approach to multilingual information retrieval'. SIGIR-96 paper that also appeared in Readings in IR. Sent as softcopy by email.
• 4/10/07, Mark Sanderson, 'Retrieving with Good Sense'.

Handouts

• Course Outline
• Lecture 1 Notes
• Lecture 2 Notes
• Lecture 3 Notes
• Lecture 4 Notes
• Lecture 5 Notes
• Lecture 6 Notes
• Lecture 7 Notes
• Lecture 8 Notes
• Why You Should Use N-grams for Multingual Information Retrieval" slides
• Lecture 9 Notes
• Lecture 10 Notes
• Lecture 11 Notes

Assignments

• The electronic text of Les Miserables is 1.3 MB in size (gzipped) and about 3.5 MB uncompressed. You'll need this data for the first assignment.
• Homework #1: Corpus Statistics. Due 6 February 2007.
• Homework #2: Indexing. Due 20 February 2007.
• Reading List. Note: students must select a paper to review for the class.
• Class Project
• Homework #3: Retrieving. Due 20 March 2007. Text: time.txt.gz Queries: queries.txt
• Homework #4: Classification and Web Query Logs. Due 10 April 2007. Logfile (removed from website) querylog.gz
• Homework #5: Language ID and NLP/IR Due 24 April 2007. Training Data: English: en.train.gz French: fr.train.gz Spanish: es.train.gz Test Data: English: en.tst.gz French: fr.tst.gz Spanish: es.tst.gz Link to WordNet. The Sanderson paper is available here.

Course related web-links

• Sources for on-line papers: CiteSeer (scientific and CS articles) ACL Anthology TREC Publications ACM Digital Library
• IR Textbooks: Managing Gigabytes, Readings in Information Retrieval (at amazon), Foundations of Statistical Natural Language Processing text
• IR Evaluations: TREC, CLEF, NTCIR FIRE
• Organizations that distribute corpora: LDC, ELRA
• IR Journals: JASIST, IP&M, IR
• IR-related conferences: SIGIR, CIKM, KDD, ACL WWW-2005
• On-line magazines: Search Engine Watch, D-Lib Magazine
• Berkeley Primer: Finding Information on the Internet
• HLT Central Repository
• Parallel text processing: bibliography
• Discrete Mathematics Primer
• Web Protocols: HTML 4, Z39.50 (Information Retrieval)

Software Resources

• Cornell's SMART system
• Managing Gigabytes mg system
• Lemur language modelling IR toolkit
• Very nice list of NLP, IR, CL, resources (i.e. parsers, taggers) at Stanford.
• University of Michigan tool suite: Clairlib
• Trigrams-n-Text (TnT) toolkit, a visible markov model tagger written by Thorsten Brants (now of Google).
• QTag a probabilistic POS-tagger.
• On-line translators: Systran, FreeTranslation.com
• WordNet, a lexical database for English
• Andrew McCallum's MALLET toolkit, a Java-based API for machine learning applications using Conditional Random Fields
• Wget
• Perl LWP library (at CPAN).
• Public Domain Speech Recognition Software (from Miss State Univ).

Frivolity

• Journal of Machine Learning Gossip: A tongue-in-cheek (yet curiously informative) career guide for young researchers

Cool Demos

• A 'meta' search engine: Dogpile
• A question-answering system: START
• An online joke recommendation system that demonstrates collaborative filtering: JESTER
• A search engine for speech: PodZinger
• A faux computer science paper generator, SCIgen, from MIT

IR Test collections

• Reuters 21578
• CACM collection
• The University of Glasgow has a list of available collections

Several Web search engines

• Google (an exemplary engine)
• altaVista
• NorthernLight
• All The Web
• Looksmart

JHU on-line resources

• JHU on-line library catalogs: Homewood & Gibson (APL).
• The Johns Hopkins University's Center for Language and Speech Processing
• The Whiting school's part-time program home page.