EMNLP 2018 study: GRU model ties n-grams on headline popularity
TL;DR
- A 2018 EMNLP paper by Lamprinidis, Hardt and Hovy builds a multi-task GRU to predict news headline popularity, with news-section prediction and part-of-speech tagging as auxiliary tasks.
- Pre-trained word embeddings and the two auxiliaries both help, yet the model's performance ends up very similar to a logistic regression over character n-grams.
- Feature analysis flags forward-looking deictic expressions and second person pronouns as structural patterns associated with popular headlines.
A 2018 EMNLP paper by Sotiris Lamprinidis, Daniel Hardt and Dirk Hovy models news headline popularity with a multi-task GRU network, pairing the main objective with two auxiliary tasks: news-section prediction and part-of-speech tagging. Pre-trained word embeddings help. Combining the two auxiliaries helps too.
The deflating finding is in the abstract itself. The authors report that 'performance is very similar to that of a simple Logistic Regression model over character n-grams.'
The feature analysis is the keeper. The paper says it finds 'structural patterns of headline popularity, including the use of forward-looking deictic expressions and second person pronouns' — the 'you, tomorrow' end of headline grammar, surfaced as a learnable signal. Published in the EMNLP 2018 proceedings, pages 659–664. Two researchers from our Who's Who directory have shared the link.
Shared on Bluesky by 2 AI experts
-
#TBT #NLProc 'Predicting News Headline Popularity' by Lamprinidis, Hardt, @dirkhovy.bsky.social (2018) shows neural networks perform similar to Logistic Regression in prediction. aclanthology.org/D18-1068
View on Bluesky →
Originally reported by aclanthology.org
Read the original article →Original headline: Predicting News Headline Popularity with Syntactic and Semantic Knowledge Using Multi-Task Learning