Introduction to Information Retrieval

1
Introduction to Information Retrieval

• Martin Volk
• Computational Linguistics
• Stockholm University
• volk_at_ling.su.se

2
Documents

• Textual documents
• Manuals (Software manuals, airplane manuals,
  power plant manuals)
• Patent texts
• Law texts
• Medical texts
• Audio and video documents

3
IR System
IR System
4
Information Retrieval

• Terms (JM p. 647)
• Document a unit available for retrieval
• Collection the set of all documents
• Term lexical item in a document
• Query formulation of the users information
  need ( question)
• Index a list of terms with pointers to their
  source

5
Questions

• How can the relevant information be found?
• How can the relevant facts be retrieved?
• How can the found facts be used for inferencing?
• How can the information be used for
  classification?
• How can information be found across languages?

6
Relevance

• Relevance is a subjective judgment and may
  include
• Being on the proper subject.
• Being timely (recent information).
• Being authoritative (from a trusted source).
• Satisfying the goals of the user and his/her
  intended use of the information (information
  need).

7
Keyword Search

• Simplest notion of relevance is that the query
  string appears verbatim in the document.
• Slightly less strict notion is that the words in
  the query appear frequently in the document, in
  any order (bag of words).

8
Problems with Keywords

• May not retrieve relevant documents that include
  synonymous terms.
• restaurant vs. café
• PRC vs. China
• May retrieve irrelevant documents that include
  ambiguous terms.
• bat (baseball vs. mammal)
• Apple (company vs. fruit)
• bit (unit of data vs. act of eating)

9
IR System Components

• Text Operations forms index words (tokens).
• Stopword removal
• Stemming
• Indexing constructs an inverted index of word to
  document pointers.
• Searching retrieves documents that contain a
  given query token from the inverted index.
• Ranking scores all retrieved documents according
  to a relevance metric.

10
Web Search System
IR System
11
Recall and Precision

• Recall is the ratio of
• The number of correctly found documents to
• The number of all relevant documents.
• ? What percentage of the relevant documents has
  been found?
• Precision is the ratio of
• The number of correctly found documents to
• The number of all found documents.
• ? What percentage of the found documents are
  relevant?

12
The vector space model

• The idea documents (d) and queries (q) are
  represented as vectors of features that stand for
  the terms (t)
• dj (t1,j, t2,j, t3,j, ... tN,j)
• qk (t1,k, t2,k, t3,k, ... tN,k)
• Simple assumption terms are present (1) or
  absent (0)

13
The vector space model

• Then the similarity between the query (vector)
  and the document (vector) is measured by summing
  the number of terms they share.
• But some terms are more important than others.
• Therefore 1/0 is replaced by weights.
• dj (w1,j, w2,j, w3,j, ... wN,j)
• qk (w1,k, w2,k, w3,k, ... wN,k)
• The collection is then a matrix of weights with
  wi,j which stands for the weight of term i in
  document j.

14
The vector space model

• Vectors need to be normalized to unit length.
• Then the dot product between vectors computes
  the cosine of the angle
• Identical documents cosine 1
• Totally disjunct documents cosine 0

15
Search engines ranking principles

• A document D is ranked higher
• If D contains more search terms.
• If the search terms are more frequent in D.
• If the search term occurs less frequently in the
  set of all documents.
• If D is shorter but still contains the search
  terms.
• If the search terms are closer to each other in
  D.
• If the search terms appear earlier (closer to the
  top) in D.

16
Search engines ranking principles

• A document D is ranked higher
• If D is newer !?
• If there are more documents that point to D !?
• If D is accessed more frequently !?
• If Ds owner pays more ... !? ?

17
Increasing Precision

• Information Retrieval (Google)
• Query ? List of documents for human inspection
• Answer Retrieval (SUIS)
• Query ? List of answers (specific parts of
  documents) for human inspection
• Fact Retrieval (??)
• Query ? List of facts for insertion in database

18
Other IR tasks

• Document categorization,
• Document filtering / routing,
• Text mining (deriving new information ie.
  information that is not explicitly mentioned
  from a textual document)
• Text summarization

19
Filtering, Routing

• Classifying documents (e.g. Emails) according to
  predefined criteria.
• Distinguishing real emails from junk mail (
  spam)
• Routing emails to the respective department
  (orders, complaints, product information,
  question, ...)
• Routing ticker new for special customers.
• Filtering yes / no routing

20
Summary

• Modern Information Retrieval systems use
  statistical methods for ranking documents
  according to relevance.
• Precision and Recall are the most frequently used
  methods for measuring the quality of IR systems.
• Routing/filtering, text mining, summarization are
  other IR tasks.

21
Any Questions?