The comment section on e-newspapers is crucial in all interaction strategies with readers. Therefore, publishers need to manage, process, and edit well to ensure that contents displayed in this section are valuable, morally appropriate, and adherent to the law and the paper’s credo.
The Automate Comment Review solution, developed by FPT Online, will help make management of online websites and e-newspaper easier and more effective, at the same time reducing human’s workload by utilizing new technologies for higher accuracy. Recently, Automate Comment Review had also become one out of four solutions awarded as FPT’s best Digital Transformation.
Challenges in comment review
As the deployer of online services like e-publisher, online ads over popular online news sites of VnExpress.net, Ngoisao.net, and iOne.vnexpress.net, editors at FPT Online has to process around 15,000 comments from readers on a daily average. Some days, this number even reach 30,000.
The role of reader comments is starting to show in many online news sites. Each popular article can garner up to hundreds or even thousands of comments, each with thousands of “likes”. Comments under articles had also become a useful interaction channel between e-newspapers and their reads. Before, readers have to write to publisher to express their opinion regarding one or many articles, and need to wait for response. Today, their comments can appear just moments after an article is published.
The challenge in comment review is not exclusive to VnExpress or Ngoisao.net, but rather is a mutual matter to many Vietnamese publisher today. In fact, many news companies had faced problems and legal issues when their readers comment misdirecting, agitative, and hateful contents aimed at individual/organizations, or even borderline illegal. Per the law, this make these sites subjective to being sued by the aforementioned individuals/organizations, as when the publisher allows the comments to appear, those will be regarded as opinions of the publisher as well. However, most publishers are severely lacking in reviewing staff, while journalists and editors who partake in reviewing are inexperienced or neglectful, thus allowing unconstructive comments.
Automate Comment Review with machine learning
The Automate Comment Review solution, with its automatic filtering system, will be able to classify comments based on set criteria. At the same time, machine learning technology will be used to solve the comment filtering problem.
In particular, based the analysis of comment principles commonly used in news publishers, the system can identify these features for filtering purposes: Comments with sensitive wordings, comments without punctuations, comments of inappropriate lengths (under 3 words), comments with repetitive contents, comment with external links not in the news’ link network.
Those passing the above initial filtering will then be deep analyzed and evaluated by machine learning. Then, based on the consideration points issued by machine learning, the filter system will decide to remove or retain the comments. As each news category will have different comment review principles, machine learning can be adjusted accordingly for higher accuracy.
At the moment, Automate Comment Review is being tested on FPT Online’s e-newpapers. After 2 months of utilization, the solution had received high regards and positive feedbacks from the editors. According to them, the amount of accumulated comments had reduced to 5% from the usual 40%, and online new sites are no long overloaded with unreviewed comments. Machine learning also helped to review around 30% of comments every day, with accuracy up to 80%. With Automate Comment Review, situations like lacking staff for comment review and having to divide human resources for the task also lessen considerably.
In the near future, FPT Online will continue to develop this filter system, as well as adjust algorithms and principles for each category for better practical usage. Hopefully, the solution will soon be commercialized to eliminate the problem of comment review for Vietnamese news publishers.
Xuan Viet – Thao My