[ad_1]
The Naver News Algorithm Review Committee (hereinafter referred to as "review above") badyzed the Naver News service automation algorithm and concluded that it was useful in addressing equity issues and reliability.
This is an explanation of the fact that the intervention of the manager is in principle impossible.
The review also positively evaluated the development of fairness and transparency algorithms across the press service, and then asked for more research and development to improve the algorithms.
The Naver News Algorithm Review Committee has published a presentation on Naver News' algorithms review over the last six months at the Seoul Press Center in Korea on April 29.
As a result of the review of the Naver News algorithm, "in the field of news search services, a focus of quality badessment has been established on the basis of clear guidelines and it will be important to continue to put in place such a system ".
Myeong-Hyun Chair of Myeong Naver News Algorithm Review
The Naver News Algorithm Review Committee was launched in May as 11 experts in three areas: IT, Information Science and Communications. The purpose of the review was to ensure objectivity and transparency of the service by badyzing the Naver News algorithm academically and technically, providing advice on technical excellence and recommending directions for the future development of the service.
First, we divided the Naver News service into a news service (ranking algorithm), a recommendation of news AiRS (a personalized news service), an entertainment news service and sports.
Process of securing data for each service, algorithm determining actual search results and reflected qualities, process in which the service is actually disclosed to the user and adequacy of established business processes for the entire process.
During the examination, the examiners shared the details of the technical procedures in each area and the details of the learning data samples on the Naver side.
As a result of this review, he stated: "The orientation of the quality badessment has been established on the basis of clear guidelines in the field of news search services, and it will be important to continue to put in place such a system in the future ".
"The Naver news ranking learning algorithm uses SVMRank and is not suitable for the use of learning algorithms. Therefore, Naver 's. will strive to look for better qualities rather than replace algorithms and strive to improve the research results. "He said.
"Since the results of the news search are automatically arranged with the help of algorithms, it is impossible for the managers to intervene in principle," he said. "All historical decisions for the development of the news search – the operation – the improvement is recorded and the comments of the users are collected and reflected." We confirmed that we were constantly evaluating the current users' satisfaction with search quality. "
In addition, in the review referral service area, Naver decided that Naver automatically collected feedback data from the information users without publisher intervention and secured the training data. We also determined that we used algorithms that combined collaborative filters and quality models to compute custom reference scores that take into account the preferences and quality of the user's articles.
In particular, they indicated that they minimized the problem of the "filter bubble" by making it possible to recommend articles in fields other than the current interests of the user.
Given the fact that the number and type of articles to be collected in the general press articles and the period of validity of the information are different from each other in the service of recommendation of articles of entertainment and of On the basis of the information provided, we found that we provide a non – personalized news recommendation service.
In addition, compared to the two previous services, the service is operated according to a rules-based algorithm rather than a data-based machine-learning because it reflects the characteristics of the domain in which the formal characteristics exist. However, the qualities that play a key role in the recommendation algorithm for entertainment and sports articles are used in a systematic and appropriate way, but it is suggested to check the validity of the qualities as their usefulness may change over time.
Overall, the review indicated that the automation of Naver's News services was an effective alternative to address issues of fairness and reliability. It is also the first attempt in the world to discuss how to improve the fairness and transparency of algorithms in information services, and that is very encouraging.
We will further shorten the construction period of other learning data, cover different age groups, strive to improve algorithms in R & D, establish common standards for good articles, "develop and periodically implement methods of evaluating the quality of articles, Documented and shared with the entire service.
Mr. Maeng Sung-hyeon, chairman of the review committee, said: "Naver's Naver News Algorithm Review Board, which is unprecedented in the world, is an information portal on which the opinions of the committee must pursue the public values and values of the company.I hope this will be a valuable base for us to move from a competitive advantage to a platform that is not shaken. . "
Source link
