peshkov - stock.adobe.com
South Korean internet giant Naver has been shoring up its artificial intelligence (AI) capabilities in recent years to counter the dominance of the US and China in the field.
In 2017, the company, which operates the top search engine in South Korea, the Line messaging app and other online services such as e-commerce, set out to build a global research network by acquiring the Xerox Research Centre in France.
Since then, it has expanded its research footprint to Japan, Vietnam and Hong Kong. More recently in September 2021, it appointed Kim Young-bum as chief scientist of the newly formed Naver Search US (NSU), the epicentre of its global research network.
An established AI researcher, Kim led the dynamic ranking science group at Amazon, where he developed key technologies to understand a user’s context and intentions for the Alexa AI platform.
Speaking to Computer Weekly from the US, Kim talks up the work of NSU, how Naver, which wants to reach one billion users globally in five years, is applying AI models to deliver good user experiences and the role of open innovation.
How is Naver implementing AI in its products and what kinds of problems are you trying to solve?
Kim Young-bum: We’re trying to develop state-of-art technologies for different global products that feature AI capabilities. We are currently researching and developing AI products and we are trying to bring separate Naver services into one group so that there are organic relationships between them.
Could you share more about the AI capabilities? For example, how is AI used to deliver personalised search experiences for users?
Kim: The goal is to provide the best optimised experience for individuals. So, when you say “experience”, it’s not limited to one single service, it’s for many different products, services and platforms. Our AI technology aims to help users in the Naver ecosystem achieve their goals.
What kinds of data do you use from users to deliver that experience? Would it be things like search history, where the users come from and their behaviour on different services? What are some of the key data points that shape how Naver’s algorithms work?
Kim: What’ve you mentioned are some of the data points we use. The data we are using is anonymised and authorised by users who have agreed to provide us with their personal and sensitive information. There are many different services in Naver that are connected to our search service, and we look at the interaction data of users across those services.
Queries will be important. So, for example, if a user uses certain queries to search for something, we see if the user scrolls down the results page, clicks on specific things or uses the “back” button to go back to where they were.
But it’s not limited to search – we also look at similar types of data from Naver services like online shopping. For Naver Shopping, for example, we look at whether the user has viewed or purchased an item, whereas for Naver Map, we look at their relative and absolute locations. All of that data is being used to develop our AI models.
Can you share how Naver improves its AI models over time? What sorts of feedback do you get from users?
Kim: We use both explicit and implicit feedback. I think a major trend for in-house developed AI services is to take into account user feedback. Customers are the ones who know what’s the ideal environment for them. So, to create this environment, we need to get feedback directly or indirectly to optimise the system and become more customer-centric.
There have been a lot of conversations in the industry about the need for explainable AI. What is Naver’s take on that?
Kim: From our perspective, explainable AI is very critical and important. We are adopting the concept in areas such as privacy. In the past, we focused on getting as much data as possible and understanding the correlation between the data. Now, we are training our AI models with common sense and assumptions so that we can look at causal relationships and increase the efficiency of data use.
We are also addressing the bias issue. For example, if there is a deep-learning model with a particular neuron that is creating gender bias, then we will work to deactivate that neuron.
When implementing AI services, there are promises we have to keep for users. We need to win trust from users, and we aim to utilise the interpretability of our system to improve this.
Can you give me a sense of the role of NSU in the broader scheme of things? Was it set up to gain access to talent?
Kim: Yes, we set up our office in the US to attract global talent so we can enhance our technology and competitiveness on the global level. Our team members used to work for other companies, but because they trusted me and the vision of NSU, they decided to join.
For example, we have professor Yoon Kim who is currently working at the Massachusetts Institute of Technology. He’s the number one deep-learning expert. We also have Rutgers University professor Karl Stratos, who is a global natural language processing expert.
Kim Young-bum, Naver Search US
Our direction is to create a global service and we want everybody to use Naver services in their daily lives. So, as the first step, we need to develop global-level technologies that can lead the search engine industry, and to improve Naver Search’s capability. We also plan to build an ecosystem by connecting various Naver services and as our ecosystem expands, we can get closer to our users.
But instead of following research trends, we are going to focus on the practicality of our services. So, whether something delivers better performance is not important. What matters is whether the user experience is better, not the technology or performance. I believe if you provide the best services and products, it will naturally lead to further development of the technology. That’s our purpose and vision.
Is Naver making its tools available to third-party developers that might benefit from the technology you’ve developed?
Kim: That’s a very important point you’ve made. To reinterpret your question, I think you are asking if we are going to allow open innovation.
When you are building a global scale project, mainstream adoption is necessary. If that’s the reality, then there’s a clear limit in our own in-house technology. So, I believe our services and technologies should be shared, so that third-party technologies can be included and interoperable with ours.
Currently, we are going to focus on making our technology more sophisticated and we are reviewing the plan to make it accessible from an open innovation point of view. At the same time, we need to strike a balance between quality and openness, because we have to be able to control quality. But as we are aiming for global usage, once our technology is sophisticated enough, we might disclose it to third-party developers. I think your question was very sharp.
Do you see the technology you are developing being applied in other domains or used by businesses?
Kim: Yes, and we can do that through a technology called Ocean, which stands for Optimal Composable Experience Action Network. Technologically speaking, it was developed to create the best experience based on user behaviour and it’s going to connect all Naver products and services into a network.
The importance of this particular technology is that user experience doesn’t come from one single service. And once you have the Ocean model implemented, you can remove many of the limitations that you used to have because the model’s data is based on actual behaviour. This can also be used by businesses with multiple platforms and provide multiple services.
Has Naver has submitted its algorithms, for example, those used for image recognition, for third-party benchmarking?
Kim: What’s important to us now is to focus on long-term rather than short-term value. And when I say long-term value, it’s not about getting more advertising income. We are going to focus on optimising user engagement. And once we improve user satisfaction, our user base and engagement will increase, and we will be able to monetise that and ultimately achieve our goal of reaching one billion users.
Read more about AI in APAC
- APAC organisations region are expected to spend more on AI systems over the next three years to improve employee efficiency and speed up decision-making, among other goals.
- TNT Surveillance’s technology is being deployed by bus and port operators to reduce driver fatigue and prevent accidents using cameras and radar sensors.
- Singapore launches two national AI programmes to bolster adoption of AI by financial institutions and the public sector.
- InnerVision Engineering’s smart parking solution uses machine vision and other technologies to deliver a frictionless parking experience for drivers in Australia.