Gravitation-Based Model For Information Retrieval

1
Gravitation-Based Model For Information Retrieval

• Shuming Shi, Ji-Rong Wen, Qing Yu, Ruihua Song,
  Wei-Ying MaMicrosoft Research Asia, 49 Zhichun
  Road, Beijing, 100080, P.R. China

• Department of Computer Science, Beijing Institute
  of Technology, Beijing, P.R. China

2

1. Introduction
2. Background and related work
3. Gravitation-based model
4. Experiments

3
Introduction

• There are some problems faced by information
  retrieval models
• Most IR models fail to satisfy even some basic
  intuitive heuristic constraint.
• Many IR models are commonly lack intuitive
  interpretations, especially physical
  interpretations.

4

• The relationship between a query and a document
  is modeled as the attractive force between them.

5
Background and related work

• Newtons theory of gravitation

6
Information retrieval perspectives and models

• Vector space model
• Pivoted normalization weighting model formula
• s is between 0.0 to 1.0

7

• Probability model
• Okapis BM25 formula
• w(t) has the potential negative IDF issue, and
  we will use ln((N 1) / df(t))
• Language model

8
Structured document retrieval

• A document is said to be structured when it
  contains multiple fields.
• The most commonly used approach for structured
  document retrieval may be score/rank (linear)
  combination, which treats each field as a
  separate document and computes scores/ranks for
  them.

9
Gravitation-based model

• The GBM tries to understand the IR problem within
  the physical framework.
• Particle
• Particle has two attributes with it type, and
  mass.
• Two particles of the same type have some kind of
  gravitation force between them
• ?

10

• Term object
• Two shapes will be discussed in this paper the
  sphere, and the ideal cylinder

11

• Document object
• ? explicit terms, include all the terms in
  the content of the document. (D)
• ? implicit terms represent the hidden
  meaning of the document. (H(D))
• The mass and the diameter of a document is the
  mass and diameter of all its seen and hidden
  terms.

12

• A query is modeled as an object composed of its
  terms.
• Relevance as gravitation force.

13
The discrete version of the model
14

• The relevance of a document given a query is
  defined by the attractive force between them when
  the document is in its optimal-term placement
  state.
• ?
• ?

15

• We give some straight-forward and natural
  simplicity assumptions on which the estimation is
  based.
• m(t) the global importance of term t in the
  collection.

16

• Combine assumption 1 and assumption 2
• ?

17

• ?
• ?
• ß1/(1?)

18

• e(D) is the diameter per term when document D has
  average length.

m0
?
19
The continuous version of the model

• The model description for the discrete GBM model
  also applies to the continuous version.

20
(No Transcript)
21

• A family of effective ranking formulas

22
Exploiting document structures

1. Terms will compete for places and the ones have
   larger gravitation forces will get nearer to the
   query terms.
2. The relevance of a document given a query is
   defined by the maximal gravitation force between
   them.

23
(No Transcript)
24
Experiments

• We use mean average precision to evaluate the
  query results.

25
Term weighting experiments
26
Field structure experiments
27
Conclusion and future work

• In this paper, a gravitation-based IR model was
  proposed by adopting a physical perspective on
  information retrieval.
• Its interesting to study the relationship and
  possible combination between GBM and existing
  models.
• This model has potential to include term
  proximity and static document ranking.