1. Summarizing Entities: A Survey Report
Shruti Chhabra ; and Srikanta Bedathur
徐丹云

2. Definition Resources Summarizing entities task Various documents
Knowledge bases
Summarizing entities task
Extract information associated with entities from these sources
Produce summaries

3. Categories Output Summary Type Number of Input Entities Single Entity
Entity Pair
Multiple Entities
Textual
Paragraph (biographical, definitional, domain-specific), Wiki-style
Textual relations, Evidence
trails
Evidence trails
Structural
Property-value pair ranking,
Structure extraction
Relation extraction, Structure extraction

4. Textual-Single Entity
Entity description
“... Reuters reported that the radical Muslim group Hamas has claimed responsibility for the bombing act...“
noun phrases (NPs) Islamic Resistance Movement Hamas
Paragraph summaries
Wiki-style summaries
Wikipedia

5. Textual-Single Entity-paragraph
Generic Summarizaiton
Biographical Summarization
“Who is X”
Use a top-down approach to search for patterns specific for biographies
Utilize bottom up approach to further filter sentences which are biographical in nature
Definitional queries
“Who is X” and “What is X”
Domain-specific
Financial(company-specific entities) and geo-locational
Distinctive characteristics

6. Textual-Single Entity-paragraph
Generic Summarizaiton
Topic signature[57], Conroy et al.[27]
Graph based LexRank[34]
Topic-sensitive version of LexRank, Otterbacher et al.[62]

7. Textual-Single Entity-paragraph
net centroid value of the sentence
position of the sentence in the document
similarity with the first sentence
LexRank
QueryPhraseMatch
sentence length
tf.idf score
similarity with the headline of the document
Weighted frequency of entity
alias or mention as pronoun
Frequency of cue words associated with facets of a person
frequency of words usually found with query entity

8. Textual-Single Entity-paragraph

9. Textual-Single Entity-Wiki Style
distinctly presenting the various aspects of entity
written in well structured manner
Techniques
learning templates
Select content to fill these templates(Sauper and Barzilay [67], Autopedia system[92])
Selecting facts similar to infoboxes(Liu et al. [58])

10. Textual-Entity Pair
find direct and indirect relations between entity pair
Textual Relations(direct)
Patty, Open IE
the entities with corresponding relational pattern
providing contextual evidence for the relationship between entities.
Evidence Trails(indirect)
sequence of sentences

11. Textual-Multiple Entity
a connected graph for any number of inputs(Srihari et al. [70])
a concept chain graph (CCG) was built by processing a corpus
graph matching technique finding minimum steiner tree in CCG.

12. Structural-Single Entity
Property-Value pair Ranking(1-hop)
Structure Extraction
System
technique
Frequency
informativeness
diversity
popularity
Cheng et al.[23]
Random walk 
Thalhammer et al. [81]
Sydow et al. [78]
Greedy algrithom
Sydow et al. [75]

13. Structural-Entity Pair
Relation Extraction
facts relating given entities
rule based, feature based and kernel based methods
Structure Extraction
paths or subgraphs

14. Structural-Entity Pair-Structure Extraction
System
Main idea
Technique
Graph Theory
Faloutsos et al. [36]
a connected subgraph,
edge-weighted undirected graph,
model the graph as a circuit
maximizing the goodness function
greedy heuristic
Relational Databases
(keyword search)
Agrawal et al.[1]
Symbol table: analogous to inverted lists in IR
BANKS[12]
Current single source shortest path algorithms from each source;
Find root nodes connecting all the source nodes
Knowledge Discovery
Halaschek et al. [44]
Paths between two entities
rank list of relationship paths
Anyanwu
and Sheth [8]
a path from each of these entities which meet at a common node
a m-length path from entity such that the parent class of each node in one path is same as the parent class of corresponding node in other path

15. Structural-Multiple Entities
mine an important subgraph for query nodes
Tong and Faloutsos [82]
retrieved nodes in a graph closely connected to query node and thus
mined an important subgraph for query nodes.
Kasneci et al. [52]
Obtain steiner tree from the subgraph using STAR [53] algorithm

16. Comparison Textual Summaries Structural Summaries source
Textual - documents
Structural – knowledge bases
information
Rich in context
poor
cleanliness
noisy
Noise-free
processable
difficultly
easily
understandable
Leverage the advantages of both text and structure
Use knowledge bases to obtain connections among entities forming structural summaries
augment with text to form textual summaries

17. Thanks