Text Mining 101

Most of the data that exists is unstructured text data. Prominent examples of this are social media posts, scientific publications or news. In order for classical data mining techniques to make any use of it, this data needs to be structured. Text Mining is the science that tries to structure texts and to use the gained insights to synthesize new knowledge and make new predictions. This talk will give an introduction to the problem text mining is tackling, the standard approaches to text mining, the application possibilities of machine learning algorithms within text mining systems and possible practical applications of text mining.

Transcript

1. Philipp Blohm, 24.03.2015
6. Alzheimer‘s: 107.050 Articles 8.467 Average Words per article 250 Words per Minute (Average Reading Speed) 8 Hours per Day 249 Workdays per year
7. Alzheimer‘s: 107.050 Articles 8.467 Average Words per article 250 Words per Minute (Average Reading Speed) 8 Hours per Day 249 Workdays per year Ca. 30,3 years of just reading
8. Alzheimer‘s: 777,8 Articles per Month (current publication rate) 283.069,9 New articles in 30,3 years Ca. 80,2 more years of just reading
9. Text Mining
10. Text Mining Computer Linguistics Semantics NLP Cognitive Computing AI
11. Untangling Text Data Mining Marti Hearst, 1999 “a process […] that leads to the discovery of heretofore unknown information, or to answers for questions for which the answer is not currently known. ” “Mining implies extracting precious nuggets of ore from otherwise worthless rock.”
14. You need a tool for the screw.
15. You need a tool for the screw. screw tool
16. You need a tool for the screw. screw tool Screw you, you tool!
17. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER
18. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER
19. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER endothelin-1-mediated vasoconstriction endothelin-1 -mediated vasoconstriction
20. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER Labeling each token in the sentence with its part-of-speech VBG DT NN IN DT NN IN PRP$ NN
21. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER
22. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER
23. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER
24. Information Extraction Pipeline Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER Machine Learning Machine Learning Machine Learning Machine Learning Machine Learning Machine Learning
25. Swanson‘s ABC Model Picture taken from Evans, J. and Rzhetsky, A., Machine Science, Science, 2010.
26. Raynaud‘s Syndrome
27. Dietary Fish Oil
28. Swanson‘s ABC Model
29. Swanson‘s ABC Model
30. Swanson‘s ABC Model Magnesium Insufficiency Migraine
33. Error Propagation Sentence Splitting Tokenization POS Tagging Sentence Analysis Event Extraction NER
34. Did you mean Microsoft Word?
47. Time flies like an arrow. NN/NNP NNS/VBZ DET VB/IN NN/NNP
48. ... ... ... ... ... ... Shared Representation POS Tagger Sentence Analyzer Named Entity Recognizer Deep Multitask Learning
49. Collobert et al., Natural Language Processing (Almost) From Scratch, 2011.
50. Collobert et al., Natural Language Processing (Almost) From Scratch, 2011.
51. Thanks for your attention.
52. Bonus material: Since some people asked, here is the unsupervised feature learning on youtube videos paper: http://static.googleusercontent.com/media/research.google.com/de//ar chive/unsupervised_icml2012.pdf