Review Analysis: an Approach to Leveraging User-Generated...
summary
Pipeline extracts product-specific decision features via tf-idf + word2vec and an entity graph, runs spellchecking and POS filters, then scores per-feature sentiment with MIT VADER (renormalized). Results aggregate into radar charts comparing products; a usable signal emerges from surprisingly few reviews — one example uses 11 reviews vs 80 and still flags battery life as strongly negative while weight is positive. Work is semi‑unsupervised, requires synonym tuning and bias correction (site ratings average ~4) and is being evaluated for online refresh.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 0:45Outline
- 1:29Customer
- 1:50Computer Options
- 2:21Customer Criteria
- 3:13Online Reviews
- 5:14Current Landscape
- 7:51Algorithm Pipeline
- 9:41Future Space Computation
- 11:49Sentiment Analysis
- 13:51Examples
- 16:02Bringing it all together
- 16:56Final result
- 17:49Example
- 18:27Radar Charts
related talks
43mTopRater.com: Machine Understanding of Millions of Reviews
34mActionable Insights with Real-time Streaming Analytics of Customer Reviews
01h 22mWhen Rotten Tomatoes Isn’t Enough: Analyzing Twitter Movie Reviews
46mProcessing Hotel Reviews with Python
43mDetecting fake reviews at scale using Apache Spark and John Snow Labs
02h 03mBuilding on AI - Delhi meet-up
30mProductionizing an unsupervised machine learning model to...
32mLeveraging recommender systems to personalise search results