Topic/S – A Topic and Trend Recognition Approach in News-Media, I-Semantics13
0 likes797 views
The document outlines a project focused on automatic topic discovery from vast amounts of data generated by news agencies and in-house productions, addressing challenges like data overload. It details the workflow for information extraction, language recognition, named entity extraction, and topic detection, with a live demo showcasing results. Key findings include the need for improved models for non-English recognition and a dedicated user interface for optimal topic prediction.
![Storage of Semantic Data
Using Oracle 11gR2
Pros
Already available, existing knowledge
Integrated querying of relational and
semantic data
Cons
Inference
Incomplete SPARQL 1.1 support
Limited custom rule support
Benchmark of triple stores [Voigt2012]
Friday, 06.09.2013
Topic/S
Slide 11](https://image.slidesharecdn.com/topics-isem2mv-1-131206091922-phpapp01/85/Topic-S-A-Topic-and-Trend-Recognition-Approach-in-News-Media-I-Semantics13-12-320.jpg)
Topic/S – A Topic and Trend Recognition Approach in News-Media, I-Semantics13
- 1. Michael Aleythe, Martin Voigt, Peter Wehner Sächsische AufbauBank Forschung und Entwicklung - Projektförderung Projektnummer - 99457/2677
- 2. Structure Motivation, Problems, and Goals Topic/S Workflow Demo Conclusion Friday, 06.09.2013 Topic/S Slide 1
- 4. Problem In-house production Archive News agencies Web, social media Online DPA Twitter Reuters Facebook KNA Blogs … … Overwhelming amount of data e.g., WAZ 5000 articles/day from agencies and inhouse production Friday, 06.09.2013 Topic/S Slide 3
- 5. Vision Automatic topic discovery using Named Entities and other keywords (Semantic Items, SemItem) Investigation of trending topics Media Assets Named Entities Topics E1 E2 E3 MA1 E4 Push them to the editor T1 T2 E5 MA2 E6 T3 E7 Pre-Processing Friday, 06.09.2013 Topic/S Slide 4 Post-Processing
- 6. Structure Motivation, Problems, and Goals Topic/S Workflow – – – – Overview Information Extraction Storage Topic Detection Demo Conclusion Friday, 06.09.2013 Topic/S Slide 5
- 8. Workflow: Preprocessor Language Recognition (Ger/Eng) Rule based Named Entity Extraction word list + statistics Source: onelanguageoneposter.com Keyword Extraction Lemmatization, word list Categorisation Source based Friday, 06.09.2013 Topic/S Slide 7
- 9. Structure Motivation, Problems, and Goals Topic/S Workflow – – – – Overview Information Extraction Storage Topic Detection Demo Conclusion Friday, 06.09.2013 Topic/S Slide 8
- 11. Semantic Facts Named Entities required but no lists available SemItem Number (with alt. names) Person 1.504.341 (2.499.962) Organization 63.332 (98.127) Place 89.702 (95.178) Keyword 1351 Stored preferred and alternative names ID: http://www.topic-s.de/topics-facts/id/person/Rene_Muller Names: Rene Muller, Rene Müller, René Muller, René Müller Triples without SemItems: 27,6 Mio. Friday, 06.09.2013 Topic/S Slide 10
- 12. Storage of Semantic Data Using Oracle 11gR2 Pros Already available, existing knowledge Integrated querying of relational and semantic data Cons Inference Incomplete SPARQL 1.1 support Limited custom rule support Benchmark of triple stores [Voigt2012] Friday, 06.09.2013 Topic/S Slide 11
- 13. Structure Motivation, Problems, and Goals Topic/S Workflow – – – – Overview Information Extraction Storage Topic Detection Demo Conclusion Friday, 06.09.2013 Topic/S Slide 12
- 14. Workflow: Topic Detection Clustering Friday, 06.09.2013 Topic/S Slide 13
- 15. Workflow: Topic Detection Clustering Friday, 06.09.2013 Topic/S Slide 14
- 16. Workflow: Topic Detection Clustering Obama Merkel Politics Audi Highway Traffic Friday, 06.09.2013 Topic/S Slide 15
- 17. Workflow: Topic Detection Clustering (Top Cluster 25.08.2013) Article Name 43 Bundesliga, Fußball, Spieltag , 1. FC Union Berlin, SC Paderborn 07 eV, FC Augsburg, FSV Frankfurt Yes 25 Euro, SPD, Berlin, Griechenland, FDP, CDU, Deutschland Yes 19 Bericht, Diplomat, Google Inc , Anbieter, Berlin, Deutschland, Auto Yes 18 Veranstaltung, Bernd Lucke, Angreifer, Berlin, Polizei, Angriff, Deutschland Yes 15 Friday, 06.09.2013 HotTopic Gericht, Prozess, Bo Xilai, Christian Wulff, Anklage, Verfahren, Mord Yes Topic/S Slide 16
- 18. Structure Motivation, Problems, and Goals Topic/S Workflow Demo Conclusion Friday, 06.09.2013 Topic/S Slide 17
- 20. Structure Motivation, Problems, and Goals Topic/S Workflow Demo Conclusion Friday, 06.09.2013 Topic/S Slide 19
- 21. Sum it up! Result Identifying topics and pushing them to the editor Lessons learned NER: bad for non-English, combination required model needs to be optimized for queries dedicated user interface required Outlook prediction of topics with causal/temporal relations Friday, 06.09.2013 Topic/S Quelle: ooltapulta.com Quelle: business-strategy-innovation.com Slide 20
- 22. Thanks! Questions? Sächsische AufbauBank Forschung und Entwicklung - Projektförderung Projektnummer - 99457/2677
- 23. Workflow: Preprocessor Named Entity Recognition word list Tool: LingPipe + Extension Quelle: churchthought.com Sources: LOD (DBPedia, Geonames, YAGO2, GND) Advantages: controlled vocabulary, guarantied recognition of entities statistics Tool: Stanford NLP Source: pre-trained model Advantage: Recognition of unknown entities Friday, 06.09.2013 Topic/S Slide 22
- 24. Workflow: Preprocessor Categorization Categoriser Reuters Politics Article DPA Categoriser DPA IPTC Media Topic Categoriser OTS Friday, 06.09.2013 Topic/S Slide 23
- 25. Workflow: Preprocessor Categorization - Quality News-Agency KNA 80,3 % DPA 94,4 % EPD 80,3 % Reuters 90,8 % OTS 93,5 % AFP 86 % Method accuracy One cat. for all agencies 85 % One cat. per agency Friday, 06.09.2013 accuracy 87,5 % Topic/S Slide 24
- 26. Workflow: Preprocessor Keywords Lemmatization Quelle: hugdaily.org Developing a word list Extraction using the word list Bonus: frequent terms of an article Friday, 06.09.2013 Topic/S Slide 25
- 27. Disambiguation Quelle: de.wikipedia.org Quelle: fansshare.com Quelle: lounge.espdisk.com Friday, 06.09.2013 Topic/S Slide 26
- 28. Disambiguation Identification of Entity Cluster Michael Jackson Internal Facts Beer Michael Jackson Beer Whiskey Michael Jackson External Facts (DBpedia, etc.) Music King of Pop Problem: not all SemItems available in the LOD Friday, 06.09.2013 Topic/S Slide 27