(Original title: Instagram has pushed two new AI filter tools, yes! The hero behind it is Deep Text) According to foreign media "Linkage" magazine, Instagram released two new tools on Thursday. One is an automatic comment filter, and the other is a spam filter. The technical foundation of both filtering tools is based on Facebook's artificial intelligence system. The company said that the two tools can be used to reduce the amount of spam, while shielding offensive comments in content and video. The automatic comment filter has been in existence since last September, and users can choose to start the program automatically. The program can use machine learning to identify potentially offensive comments. If offensive comments still appear with comment filters on, users can report directly to Instagram just as they did before. Instagram said comment filters currently only support English, but will support other languages ​​in the future. On the other hand, the spam filter has been kept secret from the outside world since it went live last October. However, it has not been discovered by the users in September and it is not a good thing or a bad thing for Instagram. Currently, this feature automatically clears spam in 9 languages ​​including English, Spanish, Portuguese, Arabic, French, German, Russian, Japanese, and Chinese. It is reported that Facebook acquired Instagram for $1 billion in 2012 and migrated its internal technology to Facebook’s data center. According to reports, the Instagram comment filter was published using the "Deep Text" system established by Facebook's AML laboratory. On Facebook, it can combine machine learning aids to handle more than 4 billion translations a day. In addition, the system can understand the contents of thousands of emails in more than 20 languages ​​in less than one second. In June last year, Deep Text went online. It was initially targeted as an internal tool to help Facebook engineers quickly sort large amounts of text and create classification rules. After the Instagram executives learned more about the system, they immediately saw the opportunity to use it to fight spam. Because for Instagram users, spam is almost an annoyance that greatly affects the user experience. After deciding to use the system, Instagram’s first step was to hire a group of people to comment on the platform and categorize the comments as “spam†and “non-spamâ€. In fact, this type of work is very common in the technology industries involved in social media. First use humans to train machines and let them perform monotonous or even boring work. As a result, the machine will gradually become smart and smart. According to the person in charge of Instagram, 3/4 of the data was eventually sent to Deep Text through continuous data integration. Based on this, Instagram engineers can create algorithms and classify spam correctly. After more than four months of continuous testing and research, the Instagram team quietly launched spam filtering in October of last year. Instagram CEO Kevin Systrom is very satisfied with the effect of this feature. He decided to use Deep Text to deal with more complex issues - eliminating opinions or comments that are contrary to the Instagram community guidelines. To this end, Instagram also publicly released a 1200-word long text to explain the spirit of its community. Similar to the development of spam filtering, this time, Instagram has hired a large number of people and repeats one thing every day - look at the comments and determine if the review is appropriate. Then classify whether it involves fraud or racism or sexual harassment. All these staff must have two languages. After a centralized process, these people analyzed a total of about 2 million comments. At the same time, Instagram’s employees took the lead in testing the system on their mobile phones to help companies adjust their algorithms. Like the spam algorithm, the system analyzes the relationship between the person who posted the post and the commenter (and his published historical commentary) based on the semantics of the text. Through the training of neural networks and testing of models using real data for a period of time, Deep Text has been able to detect very subtle semantic differences between texts. Until today, Instagram has finally officially announced the launch of two major tools. Of course, for the current two major tools, there are still some algorithmic flaws. For example, when asked about some specific sentences, the system cannot give a specific response. After the above description of the development process of the two tools of Instagram, we can also see that Instagram's AI still relies on human strength to train machine learning systems. Just as Richard Allen, Facebook’s vice president of public policy for Europe, Middle East, and Africa, stated, Facebook “will take a long time to rely on machine learning and artificial intelligence to handle the complexity of evaluating hate speech.†36V Bms,Bms For E-Bike,Bms For Scooter,10S 36V 15A Bms HuiZhou Superpower Technology Co.,Ltd. , https://www.spchargers.com
燑br>