AI News, BOOK REVIEW: Detecting Sarcasm with Deep Convolutional Neural Networks

Detecting Sarcasm with Deep Convolutional Neural Networks

By Elvis Saravia, Affective Computing & NLP Researcher Overview This paper addresses a key NLP problem known as sarcasm detection using a combination of models based on convolutional neural networks (CNNs).

This allows for detection of contradiction between the objective polarity (usually negative) and the sarcastic characteristics conveyed by the author (usually positive).

In the example, “I love the pain” provides knowledge of the sentiment expressed by the author (in this case positive), and “breakup” describes a contradicting sentiment (that of negative).

Other challenges that exist in understanding sarcastic statements is the reference to multiple events and the need to extract a large amount of facts, commonsense knowledge, anaphora resolution, and logical reasoning.

In the proposed framework, personality-based features, sentiment features, and emotion-based features are incorporated into the sarcasm detection framework.

Each set of features are learned by separate models, becoming pre-trained models used to extract sarcasm-related features from a dataset.

CNN Framework CNNs are effective at modeling hierarchy of local features to learn more global features, which is essential to learn context. Sentences are represented using word vectors (embeddings) and provided as input.

We can observe that when the models (specifically CNN-SVM) combine sarcasm features, emotion features, sentiment features, and personality traits features, it outperforms all the other models with the exception of the baseline model (B).

The table below shows comparison results of the the state-of-the-art model (method 1), other well-known sarcasm detection research (method 2), and the proposed model (method 3).

Detecting Sarcasm with Deep Convolutional Neural Networks

OverviewThis paper addresses a key NLP problem known as sarcasm detection using a combination of models based on convolutional neural networks (CNNs).

This allows for detection of contradiction between the objective polarity (usually negative) and the sarcastic characteristics conveyed by the author (usually positive).

In the example, “I love the pain” provides knowledge of the sentiment expressed by the author (in this case positive), and “breakup” describes a contradicting sentiment (that of negative).

Other challenges that exist in understanding sarcastic statements is the reference to multiple events and the need to extract a large amount of facts, commonsense knowledge, anaphora resolution, and logical reasoning.

Each set of features are learned by separate models, becoming pre-trained models used to extract sarcasm-related features from a dataset.

(See diagram of the CNN-based architecture below) To obtain the other features — sentiment (S), emotion (E), and personality (P) — CNN models are pre-trained and used to extract features from the sarcasm datasets.

We can observe that when the models (specifically CNN-SVM) combine sarcasm features, emotion features, sentiment features, and personality traits features, it outperforms all the other models with the exception of the baseline model (B).

The table below shows comparison results of the the state-of-the-art model (method 1), other well-known sarcasm detection research (method 2), and the proposed model (method 3).

How to Do Sentiment Analysis - Intro to Deep Learning #3

In this video, we'll use machine learning to help classify emotions! The example we'll use is classifying a movie review as either positive or negative via TF Learn ...

Sentiment Analysis in 4 Minutes

Link to the full Kaggle tutorial w/ code: Sentiment Analysis in 5 lines of ..

Nate Silver (full) | Conversations with Tyler

Nate Silver joins Tyler Cowen for a conversation on data, forecasting, My Bloody Valentine, the social value of gambling, Donald Trump and the presidential field ...

re:publica 2017 | Day 2 - Livestream Stage 1 - English incl. Translation

Watch all sessions of stage 1 in our Livestream. For watching without translation look here: 10:00 - 11:00: »Flachsinn - über gute ..