bitliner/polar-intelligence

📖 My reading of academic papers and research notes

📖 polar-intelligence

My reading of academic papers and research notes.

Main topics: sentiment analysis and automatic intelligence.

Sentiment Analysis

Surveys

• 2016 - http://people.few.eur.nl/frasincar/papers/TKDE2016/tkde2016.pdf

2017

"Sentiment Neuron" or "Learning to Generate Reviews and Discovering Sentiment"

• Topic:
  • Description of an unsupervised neural network reaching 90% accuracy in sentiment classification. Based on 1 month of training, using 4 GPUs. It does not need any labeled dataset.
• Authors: Alec Radford, Rafal Jozefowicz, Ilya Sutskever (article published by Open AI)
• Link: Blog Article, Paper
• Date: Apr 2017

Learning when to skim and when to read

• Topic:
  • simple models VS complex models
• Authors: Medamind Blog
• Link: Article

🔖 Emergence of Grounded Compositional Language in Multi-Agent Populations

• Topic:
  • language understanding via agents that learn words in combination with how they affect the world, rather than spotting patterns in a huge corpus of text. Supporting Gary Marcus's ideas.
• Authors: Igor Mordatch (OpenAI), Pieter Abbeel (OpenAI)
• Link: Paper - Blog Article

🔖 A Roadmap for Natural Language Processing Research in Information

• Topic:
  • Analysis of NLP research in 2004/2015, in order to identify state of NLP research, trends and define a roadmap
• Link: https://scholarspace.manoa.hawaii.edu/bitstream/10125/41285/1/paper0136.pdf

Feb 2016

Exploiting New Sentiment-Based Meta-level Features for Effective Sentiment Analysis

• Topic:
  • Sentiment-based meta-level features for sentiment analysis on short text.
  • Meta-level means: e.g. derived from bag-of-words representation.
  • Example: sentiment distribution of the k-nearest neighbor of a document; document polarity of neighbors given by unsupervised lexical-based methods
• Performance:
  • Tested on 19 datasets.
  • Superior by 16% to bag-of-word representation; largely superior to the best lexicon-based method.
• Benefits:
  • Transform original feature space into a smaller and more informed one
• Link: http://homepages.dcc.ufmg.br/~fabricio/download/wsdm377-canutoA1.pdf

Oct 2015

Sentiment Embeddings with Applications to Sentiment Analysis

• Topic:
  • Sentiment Embeddings - Encode text not only with context, but also with sentiment -
• Performance:
  • Outperforms on multiple datasets the context-based embeddings
• Benefits:
  • Inspire same encoding in other specific tasks
  • Apply encoding to improve sentiment
• Link: http://ieeexplore.ieee.org/abstract/document/7296633/

2014

A Rule-Based Approach to Aspect Extraction from Product Reviews

• Topic:
  • Rules for aspect extraction
• Performance:
  • High precision and recall (>90% on certain datasets)
• Benefits:
  • TBD
• Link: http://aclweb.org/anthology/W/W14/W14-5905.pdf

Oct 2013

👍 Rule-based Information Extraction is Dead! Long Live Rule-based Information Extraction Systems!

• Topics:
  • Gap on IE techniques between market and research community, and a roadmap to fill the gap. Gap caused by different perception of usefuless of rule-based systems by the research community
• Link: https://aclweb.org/anthology/D/D13/D13-1079.pdf or Link

TODO

• Find papers from intersection of language acquisition, semi-supervised learning, and sentiment analysis