Relatedness-based Multi-Entity Summarization
Download as PPTX, PDF1 like461 views
The document proposes a new approach called relatedness-based multi-entity summarization (MES) to generate concise summaries of related entities. It formulates MES as a quadratic multidimensional knapsack problem (QMKP) to select important and diverse intra-entity features while also selecting inter-entity features that indicate relatedness. It presents an algorithm called REMES based on the grasp heuristic to solve the QMKP formulation. A user study shows REMES outperforms other entity summarization methods at multi-entity summarization tasks.
Relatedness-based Multi-Entity Summarization
- 1. Relatedness-based Multi-Entity Summarization Kalpa Gunaratna1, Amir Hossein Yazdavar1, Krishnaprasad Thirunarayan1, Amit Sheth1, and Gong Cheng2 1Kno.e.sis Center, Wright State University, Dayton OH, USA 2National Key Laboratory for Novel Software Technology, Nanjing University, China
- 2. Reading news online Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs.
- 3. Text enriched by entity linking with knowledge bases Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs.
- 4. Text enriched by entity linking with knowledge bases Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs. Apple Inc. is an American multinational technology company headquartered in Cupertino, California that … such as a textual pop-up from Wikipedia
- 5. Text enriched by entity linking with knowledge bases Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs. or a structured pop-up from DBpedia or Wikidata founders:Steve_jobs product: IPod location: California industry: Consumer_electronics
- 6. An entity can have several hundred property-value pairs (called features). … …
- 7. An entity can have several hundred property-value pairs (called features). … … An entity summary is a subset of k features (to fit in a pop-up).
- 8. Entity summarization (previous research) • Problem statement (k=3, in this example)
- 9. Entity summarization (previous research) • Problem statement (k=3, in this example) • Research challenge • To select important and diverse features
- 10. Text enriched by entity linking with knowledge bases Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs. founders:Steve_jobs product: IPod location: California industry: Consumer_electronics
- 11. Text enriched by entity linking with knowledge bases Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs. founders:Steve_jobs product: IPod location: California industry: Consumer_electronics after: Tim_Cook knownFor: Microcomputer_revolution title: Apple_Inc. birthPlace: California
- 12. Text enriched by entity linking with knowledge bases Within one month of the iPod nano and iTunes phone special event, Apple Computer announced today another special event to be held on October 12. It is to be held at the California Theater in downtown San Jose, California. The invitation reads, “One more thing…”, the teasing tagline of Steve Jobs. founders:Steve_jobs product: IPod location: California industry: Consumer_electronics after: Tim_Cook knownFor: Microcomputer_revolution title: Apple_Inc. birthPlace: California
- 13. Multi-entity summarization • Problem statement (n=3, k=3, in this example) • New research challenge • To select intra-entity important and diverse features • To select inter-entity related features i.e., context-dependent entity summarization (context: other entities)
- 14. Multi-entity summarization • Problem statement (n=3, k=3, in this example) • New research challenge • To select intra-entity important and diverse features • To select inter-entity related features i.e., context-dependent entity summarization (context: other entities)
- 15. Multi-entity summarization • Problem statement (n=3, k=3, in this example) • New research challenge • To select intra-entity important and diverse features • To select inter-entity related features i.e., context-dependent entity summarization (context: other entities)
- 16. • Multi-entity summarization as a Quadratic Multidimensional Knapsack Problem (QMKP), to jointly generate summaries for n entities Problem formulation
- 17. • Multi-entity summarization as a Quadratic Multidimensional Knapsack Problem (QMKP), to jointly generate summaries for n entities Problem formulation Profit earned by selecting a feature f: Intra-entity importance of f
- 18. • Multi-entity summarization as a Quadratic Multidimensional Knapsack Problem (QMKP), to jointly generate summaries for n entities Problem formulation Profit earned by selecting a feature f: Intra-entity importance of f Profit earned by selecting two features fi and fj describing the same entity: Intra-entity dissimilarity between fi and fj
- 19. • Multi-entity summarization as a Quadratic Multidimensional Knapsack Problem (QMKP), to jointly generate summaries for n entities Problem formulation Profit earned by selecting two features fi and fj describing the same entity: Intra-entity dissimilarity between fi and fj Profit earned by selecting a feature f: Intra-entity importance of f Profit earned by selecting two features fi and fj describing different entities: Inter-entity relatedness between fi and fj
- 20. • Multi-entity summarization as a Quadratic Multidimensional Knapsack Problem (QMKP), to jointly generate summaries for n entities Problem formulation Profit earned by selecting two features fi and fj describing the same entity: Intra-entity dissimilarity between fi and fj Profit earned by selecting a feature f: Intra-entity importance of f Profit earned by selecting two features fi and fj describing different entities: Inter-entity relatedness between fi and fj n constraints: Selecting at most k features for each of the n entities
- 21. Solution • A GRASP algorithm for QMKP • Measures • Importance of a feature (i.e., property-value pair): Informativeness of feature * popularity of value • Dissimilarity between two features: Negative semantic similarity • Relatedness between two features: Semantic similarity
- 22. Solution • A GRASP algorithm for QMKP • Measures • Importance of a feature (i.e., property-value pair): Informativeness of feature * popularity of value • Dissimilarity between two features: Negative semantic similarity • Relatedness between two features: Semantic similarity WordNet-based similarity between properties + graph-embedding-based similarity between values
- 23. User study • 2 datasets on news, linked with entities in DBpedia • 3 entity summarizers • REMES: our multi-entity summarizer • FACES & RELIN: two state-of-the-art entity summarizers
- 24. Take-home messages • Multi-entity summarization (MES) • finds applications (e.g., news item presentation) • faces new research challenges (e.g., inter-entity relatedness) • needs specialized approaches • Future work • Improved methods for relatedness-based MES • Extended methods for novel applications of MES