Text Summarization

1
Text Summarization

• Dr. Ying Xie

2

• This lecture note primarily describes the work of
  Gunes Erkan and Dragomir R. Radev
• LexRank Graph-based Centrality as Salience
  in Text Summarization, Journal of Artificial
  Intelligence Research 22, 2004.

3
Text Summarization

• Text summarization is the process of
  automatically creating a compressed version of a
  given text that provides useful information for
  the user

4
Applications of Text Summarization

• put a book on the scanner, turn the dial to 2
  pages, and read the result...
• ...download 1000 documents from the web, send
  them to the summarizer, and select the best ones
  by reading the summaries of the clusters...
• ...forward the Japanese email to the summarizer,
  select 1 par, and skim the translated summary.

5
Applications of Text Summarization(2)
Cited from Dr. Radevs slides NLP/IR tools
developed at the CLAIR Lab
6
Two types of summarization

• -- Topic-oriented summarization focus on a users
  topic of interest, and extract information
  related to the specified topic.
• -- General summarization covers as much of the
  information content as possible and preserves the
  general topical organization of the original text

7
Two types of summarization techniques

• There are two types of summarization methods
• - Extractive summarization produce summaries
  by choosing a subset of the sentences in the
  original document(s)
• - Abstractive summarization information in
  the text is rephrased in a abstractive way.

8
Sentence Centrality

• Assess the centrality of each sentence in a
  cluster and extract the most important ones to be
  included in the summary.
• Different ways to evaluate sentence centrality
• - Centroid-based
• - Graph-based

9
Each sentence is represented as a vector of terms
10
Centroid-based centrality

• Conduct clustering analysis on all the sentences.
  The sentences closed to the center of each
  cluster is the most salient sentences.

LexRank
11
Graph-based centrality

• Applying social networking to summarization
• Social networking is a mapping of relationships
  between interacting entities.
• Social networking is widely applied in computer
  networks and information retrieval (Recall
  pageranking?)
• Here social network is used to represent the
  similarity relations between sentences.

12
Graph-based centrality (2)

• The assumption is sentences that are similar to
  many of the other sentences in a cluster are more
  central to the topic.
• Two problems to solve
• - how to evaluate the similarity between
  sentences
• - how to compute the overall centrality of a
  sentence given the similarity relationship.

13
Similarity Between Sentences
LexRank
14
LexRank
15
LexRank
16
Degree Centrality

• Degree Centrality the degree of the
  corresponding sentence in the graph.
• This method takes advantage of the graph in the
  simplest way.

17
LexRank

• Since we have a graph, why not use the idea of
  pageRank?
• The idea is, each sentence has a centrality value
  and distributes its centrality to its neighbors.
• So, a sentence linked by many sentences with high
  centrality will have high centrality as well.

18
LexRank (2)
LexRank
19
LexRank (3)

• LexRank takes into consideration of the
  information subsumption among sentences.
• In other words, if the information content of a
  sentence subsumes other sentences in a cluster,
  it is more preferred to be contained in the
  summary.

20
lexRank Demo
LexRank