1. Social Media Analytics: Introduction

• Overview: RICHER + BIGGER DATA -> Opportunities
• Applications
• Challenges

2. Preprocessing social media texts

Properties of Social Media Textual Data Social Media Texts: a Foe of NLP? Standard NLP tools Perform Poorly on Social Media Textual

2.1. POS Tagging

2.2. Lexical Normalization

PRINCIPLE + example An Example of Token-Based Approaches

2.3. Bonus

Feature Representation Using Dependencies? SVM Training Data Generation Detecting Ill-Formed Words Evaluation Results Ex for project: Adapt NLP to data: CMU Twitter POS tagger

2.4. Summary: NLP tools and Noisy Input

• Lexical Normalization • Token based approaches • Distributional similarity approaches • NLP Tools adaptation (POS, NER, Parsing) • CMU ARK TweetNLP http://www.ark.cs.cmu.edu/TweetNLP/ : Twokenizer, POS tagger, TweeboParser • Gate Twitter Pos Tagger https://gate.ac.uk/wiki/twitter-postagger.html

3. Sentiment Analysis

1. Reminder on text classification and evaluation measures
2. Sentiment analysis: Introduction
   • Negation
   • Coreference
   • Slang and writing errors
   • Comparative
   • Domain Dependent Opinion
   • Many more challenges
   • Subjectivity
   • What is an Opinion? (document / sentence level)
     • Mathematical definition
   • Entity and Aspect Level
   • Structured the Unstructured
   • Document Level sentiment analysis
   • Aspect Based Sentiment analysis
3. Affective lexicons = Dictionaries of well-known sentiment words

4. Methods for sentiment analysis

1. Machine Learning approaches
   • ML algorithms + linguistic features
2. Lexicon based approaches
   • Sentiment lexicons (pre-compiled terms)
   • Corpus-based
3. Hybrid approaches: combine both
4. Most frequent algorithms
   • Naïve Bayes
   • Maximum Entropy
   • Logistic Regression
   • SVM
   • Recently: Deep Learning: RNN (LSTMs), CNN & word embeddings

5. Methods for ABSA

ABSA: Fine-grained opinion annotation • Determine sentiments about different aspects of entities (e.g. movies, restaurants, cell phones,…) • Aspects are features of an entity (service, food in a restaurant; screen, battery of a cell phone,…)

4. Fake news and stance detection

Learn the representation Word2Vec [Mikolov et al. 2013] Word2Vec Fun with Word Embeddings Drawbacks of word embeddings

5. Practical Hints

6. Sources

• Lecture notes of Naver Labs
• Wikipedia