Temporal Analysis on Topics Utilizing Word2Vec
summary
Claims a reproducible pipeline that detects semantic change over time by training separate Word2Vec embeddings per time slice, measuring drift with cosine similarity, and visualizing trajectories in a Django web app. Applied to news, sports, COVID-19 and a nursing‑home corpus (~21,000 documents), using 100‑dim vectors and window size 5, it surfaces convergences (Olympics → sports terms) and CDC/unofficial divergence during COVID, works but shows noise on Twitter, and benefits from larger or modern embeddings for sturdier results.
AI-written summary · about
chapters (22)
- 0:00Temporal topic analysis intro
- 1:12Nursing home report goals
- 2:59User-centric semantic trends
- 5:26Word embeddings methodology
- 7:13Visualizing word relationships
- 13:10Olympics dataset validation
- 16:22Key semantic insights
- 17:49Future embedding methods
- 19:53Interactive demo visualization
- 21:18Q&A sentiment analysis
- 22:01Gymnastics trend anomalies
- 23:47Sarcasm in embeddings
- 24:50Live analysis demo
- 27:18Reverse term retrieval
- 27:59Nursing home dataset analysis
- 30:09Embeddings over time
- 32:37LDA limitations
- 34:41Word movement across topics
- 35:49Q&A seasonality issues
- 37:53Initialization and counting fixes
- 38:14Nursing home data applications
- 38:55Q&A closing
related talks
33mUsing the Amazing World
01h 28mTricks, tips and topics in Text Analysis
30mWords in Space
37mData Science for Digital Humanities: Extracting meaning from Images and Text
34mTopic modeling with BERTopic
35mUsing Scattertext and the Python NLP Ecosystem for Text Visualization
01h 55mVocabulary Analysis of Job Descriptions
24mBertTopic, From Free-Text feedbacks to Calls for Action