CSM06 Information Retrieval

1
CSM06 Information Retrieval

• LECTURE 6 Tuesday 28th October
• Dr Andrew Salway
• a.salway_at_surrey.ac.uk

2
Lecture 6 Some Current Issues in IR

• Part 1 HCI Issues for IR Interfaces
• Part 2 Information Visualisation
• Part 3 Inference Beyond the Index

3
Part 1 HCI Issues for IR

• Three principles for interface design (adapted
  from Shneiderman)
• Offer informative feedback
• Reduce working memory load
• Alternative interfaces for expert / novice

4
Offer Informative Feedback

• Interface should help user to understand what the
  system has done and why
• For an information retrieval system this means
  helping the user to understand the relationship
  between their query and the returned documents

5
Reduce Working Memory Load

• Information retrieval is interactive interface
  must help user keep track of different strategies
  / return to old queries
• Interface may also suggest search terms and let
  user navigate a list of sources or a hierarchy of
  topics / keywords

6
Alternative Interfaces for Expert / Novice

• Trade-off between simplicity (ease of use) and
  power
• A good interface will allow a user to gradually
  progress from simple queries to more powerful
  ones, i.e. incrementally adding more
  sophisticated features to their searches and
  returning more information about the search
  results

7
Good HCI is important for

• Selecting a document collection
• Query Specification
• Making sense of results set (Information
  Visualisation)

8
Selecting a Document Collection

• An unordered list of collections does not help a
  user to choose - overviews can help
• Manually organised topic categories, e.g. Yahoo
  or in specialist domains like 1200 ACM categories
• Automatic document clustering, e.g. with a text
  description of each cluster (Scatter/Gather) or a
  2D map of the document collection where proximity
  and size of regions indicate the content of the
  document collection (e.g. Kohonen Maps)
• Co-citation analysis identifies which documents
  are central to a collection also applied to
  links between Web pages

9
WEBSOM http//websom.hut.fi/websom/ (based
on Kohonen Maps)
10
Query Specification

• Consider expressing a (faceted) Boolean query
  through..
• Command Language expressive, accurate and quick
  for experts non-intuitive for novices (meaning
  of AND / OR and use of nested structures)
• Form fill-in prompts user with attributes so
  users dont need to learn that part of the
  syntax
• Menu selection constrains users choices where
  there is a finite set of choices

11
Query Specification

• Consider expressing a Boolean query through..
• Direct Manipulation gives users a better feel,
  e.g. Venn diagrams for Boolean queries
  filter-flow visualisation shows results at each
  stage
• Natural Language requires extracting concepts
  and logical connectives in the disambiguation of
  queries may use question templates so users can
  ask naturally phrased questions, cf. Ask Jeeves

12
Part 2 Information Visualistion for Making Sense
of the Results Set

• Three systems considered here
• TileBars
• InfoCrystal
• Mappuccino

13
Making Sense of the Results Set

• TileBars shows which query terms are where in
  which documents
• One row per query term
• One column per passage of text
• Shading indicates frequency of term in passage
• Users can look for terms co-occurring in passages
  and occurring throughout long documents
• http//www.sims.berkeley.edu/hearst/tb-overview.h
  tml

14
(No Transcript)
15
Making Sense of the Results Set

• InfoCrystal displays the results of a faceted
  query simultaneously
• Limited to four query terms
• Shows number of documents in the intersections /
  unions of the query terms
• Users can look judge relative influence of facets
  on results set
• http//www.scils.rutgers.edu/aspoerri/InfoCrystal
  /Ch_7.html

16
(No Transcript)
17
(No Transcript)
18
Making Sense of the Results Set

• Mapuccino shows a sub-set of a WWW-site in nodes
  and links view
• Positive weighting given to the outlinks of
  relevant pages
• The graphical depiction is of link structure
  however this may reflect semantic content
• http//www.alphaworks.ibm.com/tech/mapuccino

19
Further Reading

• Baeza-Yates and Ribeiro-Neto (Chapter 10)

20
Part 3 Inference Beyond the Index

• Scientific Citation Analysis measuring impact
• Discovering latent knowledge
• Web-adjacency Analysis identifying authorities
  and hubs

21
Inference Beyond the Index

• Main focus in IR research (and in this module)
  has been on the Index i.e. the mapping between
  documents and descriptive keywords
• Other kinds of relationships can help a user
  searching for information, e.g. relationships
  between documents / authors / keywords (cf.
  Belews Adaptive IR system)
• Here considering in particular relationships
  between documents (citation analysis)

22
Relationships between Documents citations

• When a new text is produced it may be woven into
  the larger fabric of other texts, e.g.
  conversational threads, legal arguments,
  scientific papers and web-pages
• Interesting information can be gained from
  analysing which documents (or parts of documents)
  cite / are cited by which other documents (or
  parts of documents)

23
Scientific Citation Analysis

• Belew pp. 185-190
• Bibliometrics studies the graphs created by
  bibliographic citation links
• Can analyse the impact of a scientific paper,
  e.g. how many times it is cited (by other
  authors) measuring its in-degree
• Can measure the similarity of two scientific
  papers by comparing their bibliographies
• Web of Science makes this kind of analysis
  relatively easy now http//wos.mimas.ac.uk/

24
Scientific Citation Analysis

• Can analyse citation structure over time, e.g.
  Figure 6.3
• This shows 200 hundred papers, numbered in the
  order that they were published, and indicates
  which other papers were cited in each
• The graph highlights classic papers (large
  impact), review papers and the rate at which
  ideas are developing (the width of the research
  front)

25
(No Transcript)
26
Discovering Latent Knowledge

• Belew pp. 234-238
• Scientists tend to specialise, sometimes so much
  so that they are not aware of works in other
  areas that could help solve problems in their own
  area
• Swanson considered the scientific literature as a
  web of logical connections some connections
  are made explicit, e.g. through citations, but
  others may be implicit
• Swansons idea was to discover latent knowledge
  by automatically identifying implicit logical
  connections in the literature

27
Discovering Latent Knowledge

• Belew pp. 234-238
• Query-cluster to identify factors related to
  medical condition / disease as reported in one
  literature
• Then, query-cluster to identify something related
  to those factor in another literature
• See Figure 6.22
• Still a challenge to automatically label the
  logical relationships between cause and
  effect

28
(No Transcript)
29
Web-adjacency Analysis

• Belew pp. 195-199
• For some web-queries they may be many thousands
  of relevant pages perhaps we need to consider
  another measure in order to provide a user with
  the best results
• From among all the relevant pages, select those
  that are most authoritative

30
Web-adjacency Analysis

• Kleinberg and colleagues proposed a method for
  identifying authoritative web-pages
• Identify set of relevant pages (as normal)
• Identify those with a large in-degree, i.e. lots
  of pages point to them (cf. impact)
• Ensure that the authorities selected are referred
  to by a number of the same hubs, i.e. those with
  a large out-degree

31
Web-adjacency Analysis

• Hubs and authorities exhibit what could be
  called a mutually reinforcing relationship
  (Kleinberg 1998, in Belew)
• Computing authority and hub values for web-pages
  is an iterative process over a graph, where each
  node is a web-page
• Two weights are given to each node relating to
  in-degree and out-degree total in-degree weights
  and total out-degree weights are kept constant
• Weights are modified each iteration depending on
  weights of connected nodes

32
Further Reading

• Swanson and Smalheiser (1997), An Interactive
  System for Finding Complementary Literatures,
  Artificial Intelligence 91, pp. 183-203.
• Lawrence, Giles and Bollacker (1999), Digital
  Libraries and Autonomous Citation Indexing, IEEE
  Computer 32(6).
• (Available through librarys eJournals, and on
  the shelf).

33
Reading

• Belew pages given in previous slides.
• Kleinberg (1998), Authoritative Sources in a
  Hyperlinked Environment, Journal of the ACM.
  http//citeseer.nj.nec.com/87928.html