Data Science for Digital Humanities: Extracting meaning from Images and Text
summary
Claim: Python pipelines make large-scale image and text work practical. Concrete tools and results: scikit-learn, Keras, gensim, ImageNet 1.2M images, pretrained CNN vectors (single-layer 4096‑d features) reused for transfer learning, and LDA on 2.5M articles to produce topic distributions. Built examples include an OCR+matching pipeline that links parliament visitors to lobbyists and large-scale visualizations of blog networks and news clusters from 2.5M articles and 3,346 novels. Failures include Google Flu Trends overprediction and hashtag-driven misinterpretation of Gezi activity.
AI-written summary · about
chapters (20)
- 0:00Data science in humanities
- 1:55Sentiment analysis applications
- 3:02Statistics with a twist
- 4:51Political blogosphere divides
- 7:04Digital humanities and social science
- 9:55Small world experiments
- 12:26Defining digital humanities
- 13:50Early electronic text projects
- 14:12Topic modeling
- 17:02News content analysis
- 19:58Literary history digitally
- 21:45Text analysis limits
- 23:11Google Flu Trends pitfalls
- 24:36Computer vision in DH
- 28:54Recommended books
- 30:27Neural networks and cognition
- 32:12Publishing interdisciplinary challenges
- 33:33Programming in humanities
- 34:37Image and event classification
- 36:03Q&A
related talks
37mUsing Exploratory Data Analysis to Discover Patterns
30mKick-Start your Understanding of Machine Learning with Python
44mTeaching Computers ABCs: A Quick Intro to Natural Language Processing
04mUsing Computer Vision to Understand Big Visual Data
01h 18mComputer Vision after the Victory of Data
01h 28mTricks, tips and topics in Text Analysis
34mDeep Learning at Scale
35mHow does text become data?