A technology that can filter out ‘click-bait headlines’, which are articles with exaggerated titles on irrelevant contents to induce user clicks, was developed. It can be used as an extension to portals or search engines.
KAIST revealed on the 21st that its professor Meeyong Cha of department of computer engineering and SNU professor Kyomin Jung of department of electrical and computer engineering co-developed a deep learning technique which can separate articles with irrelevant titles and implemented a click-bait headlines filtering model.
This model is based on a deep learning model ‘Recurrent Neural Network (RNN)’. It uses only the content of the article to check if the subject is consistent and understands the relationship with the title of the article depending on the paragraph structure and word sequences.
When processing multiple data, RNN remembers the state of the previous step to make use of it in the next step. Therefore, it is useful in analyzing articles which creates a meaning with sequences of words.
The research team applied the ‘attention technique’ to the model to increase the accuracy. This technique can figure out in which part of the article to focus on to increase the model accuracy. The model was trained with a training dataset of up to 2 million articles.
The model accuracy is over 90%. It can filter out almost all click-bait articles in a pool of articles without external assistance. The main application will be an automatic article analysis service as an extension in portal and search engines. It can also be used in academic areas.
The research team will present its results at a top conference in the field of artificial intelligence ‘AAAI-19’ on the 27th in Hawaii, the United States.
They are also working on additional research. The final goal of the research will be to achieve 100% accuracy. The team is striving to make the model work as well as human intuition.
Professor Cha said, “This research has the advantage of conveniently filtering out click-bait articles only using internal elements of articles. It can provide a service that can help the public to completely trust the articles they read.”
Translated by Kyungjin Lee, English Editor of Department of Electrical and Computer Engineering, email@example.com