Web search - Political news analysis tool
- 1. Presentation Web Search Xander Koning Justin Verhulst
- 2. Problem ● Which political parties appear in the news? Is there an influence on electoral support? ● Interesting topic for political communication scholars ● Manual content analysis too labor-intensive ➢ Solution: a system that enables users to analyse political news coverage
- 3. Chosen approach ● Entity prominence → How often are political parties mentioned? ● Time → When are political parties mentioned in news articles? ● Focus on Dutch news sites NOS and nu.nl
- 4. Will the party that is covered most frequently in the news win the elections?
- 5. System: data acquisition ● Crawling of politics section of Nu.nl and NOS website. ● Python’s Scrapy library ● Extract: - Name of news site - Article Title - Full text - Date of publication
- 6. NOS.nl
- 13. Create the index using mappings PUT politics { "mappings": { "article": { "_all": { "enabled": false }, "properties": { "newssite": { "type": "keyword" }, "title": { "type": "text" }, "text": { "type": "text" }, "date": { "type": "date", "format": "dd-MM-yy HH:mm||date_time_no_millis" } } } }}
- 14. System: data processing and indexing ● Index json files with curl, using elasticsearch’s bulk API: curl -s -XPOST 'http://localhost:9200/_bulk' --data-binary @nos.json curl -s -XPOST 'http://localhost:9200/_bulk' --data-binary @nu.json ● 559 NOS.nl articles, 392 nu.nl articles ● Period: 02-01-2017 - 12-03-2017
- 19. Example analysis Order of party news coverage (# of mentions) 1. VVD 2. PVDA 3. PVV 4. D66 5. CDA 6. SP 7. GroenLinks 8. ChristenUnie 9. Denk 10. 50+ 11. SGP 12. Partij voor de Dieren 13. FvD Order of party size (# of seats) after elections 1. VVD 2. PVV 3. D66 4. CDA 5. SP 6. Groenlinks 7. PVDA 8. ChristenUnie 9. Partij voor de Dieren 10. 50+ 11. Denk 12. SGP 13. FvD
- 20. Conclusion ● System enables the analysis of large quantities of party-political news ● We showed a clear correlation between news exposure and election results → demonstrates the usefulness of this analysis tool ● Much more to be researched!
- 21. Future work ● Expand timeframe ● Add more news sites ● Retrieve more information from articles, e.g. sentiment ● Evaluate interface with real users (media / political communication scholars)
- 22. Thank you!