The Google Brain Team says it has made strides in teaching computers to summarize text, and has developed a machine-learning algorithm that can write "very good" headlines.
"We've observed that due to the nature of news headlines, the model can generate good headlines from reading just a few sentences from the beginning of the article," says the Google Brain Team's Peter Liu.
He notes the team used Google's TensorFlow software library as a basis for the model, which has generated several passable headlines using extracts from articles. For example, the algorithm converted "Australian wine exports hit a record 52.1 million liters worth 260 million dollars in September, the government statistics office reported on Monday" into "Australian wine exports hit record high in September."
Liu says the model was trained on data from John Hopkins University's massive Annotated English Gigaword dataset, and Google used a deep-learning model called sequence-to-sequence learning to mimic humans' text summarization.
Liu acknowledges the technique is less effective when producing a summary requires reading an entire document. However, he says the method can function as a baseline, and Google has made the models publicly accessible on GitHub.
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found