Discover New Tastes With
“AI Gift Explorer”
Remember those times when you almost forgot a family member or your friend’s birthday? Choosing a gift that considers the tastes of the person you are celebrating is not an easy task. Kakao Commerce has put significant thought into resolving these concerns for users. With these considerations, the goal was to create an environment where users can shop conveniently—anytime, anywhere—leading to the service rollout of KakaoTalk Gifts.
Kakao Commerce has not stopped there. It continues to offer various e-commerce services, such as called Shopping and Live Shopping, to allow users to shop what they want comfortably and at reasonable prices. In response to the rapid development of artificial intelligence (AI), Kakao Commerce has been making various attempts to expand AI applications within the commerce sector.
In this 8th volume of Tech Ethics, we will focus on the AI recommendation service applied to one of Kakao Commerce’s flagship KakaoTalk Gifts service, and discuss how Kakao Commerce services will continue to evolve.
How does KakaoTalk Gifts recommend products to users?
A key advantage of the popular KakaoTalk Gifts service is that users can send gifts as easily as sending a message, without needing the recipient’s address or phone number. The service also features a recommendation system that helps users quickly find products that the recipient is likely to appreciate.
First, let’s explore the recommendation process of "You May Also Like" slot. As users browse for gifts, they will see nudge text like "You May Also Like" as they scroll to the bottom of the screen, together with various related products to the product they clicked. Such recommendation helps users find suitable products, especially when the selected item isn't exactly what they wanted or if they wish to explore other products. By recommending similar products that many users have chosen and rated highly, the system aids users in discovering the right gift.
What kind of technology is used to recommend in the "Customers Also Bought" section that is shown after a user adds a product to their cart? When a user browses for gifts and adds to their cart, they will see related products that pair well as additional gifts in the "Customers Also Bought" product recommendation section. Such recommendation helps users explore and purchase related additional products by showing items frequently bought together by other customers.
"My Favorite Categories" organizes and displays categories clearly by analyzing not only the user’s clicks on gift products, but also their purchase patterns through an AI system. This feature provides users with their preferred categories at a glance, eliminating the need to search for categories each time they are unsure of what gift to choose. Browsing through the "My Favorite Categories" section can help users find suitable products and categories for friends as well as for themselves.
AI-based gift-giving experience... The recommendation technology behind 'AI Gift Explorer
Beyond the technology used for recommendation explained earlier, Kakao Commerce has continually explored ways to enhance the value users get from the KakaoTalk Gifts service. Building on the efforts, Kakao Commerce officially launched the all-new “AI Gift Explorer” feature in September 2023, leveraging trending AI technologies to further ease users’ concerns when choosing gifts. The goal of AI Gift Explorer is to help users quickly find gifts that will satisfy the recipient.
Users have two options to access the AI Gift Explorer feature. They can tap on the “AI Gift Explorer” banner on the home screen or select the "Gift Explore" tab in the top tab bar on the home screen. When accessing through the tab, products are recommended immediately. However, if users approach through the banner, they are directed to a screen that allows them to input conditions for the recipient (e.g., recipient, gender, age, birthday, type of gift, packaging options). An example is shown in the image below.
If users choose to enter through the banner, they then select conditions according to their needs and tap the "Start Gift Explorer" button. The screen transit to the Explore Gifts tab, where gifts matching the chosen conditions are recommended. For instance, as shown in the image below, selecting options such as “Female,” “30s,” and “Romantic” results in a message displaying “Romantic recommendations for women in their 30s,” along with a list of suitable gift items. Through the AI Gift Explorer, users can reduce the stress of choosing a gift for someone and increase the likelihood of satisfying the recipient.
So, how does AI Gift Explorer recommend products to users in a way that enhances both user and recipient satisfaction while reducing the hassle of searching for gifts?
When users typically enter the KakaoTalk Gifts service and consider gift options, they review wishlists or past gift exchanges to decide on what they believe is the most suitable gift. With AI Gift Explorer, the recommendation AI provides a curated list of suitable gifts for the recipient, without the user needing to go through the effort of checking wishlists or making detailed considerations. The process proceeds in the following steps:
First, a candidate pool for the recommended product list is generated based on service usage logs related to the gift recipient. The recipient’s “public wishlist” and gift exchange history with the sender are considered. This information reflects the actual interests of the recipient, increasing the likelihood that they will be highly satisfied with the recommended items. Similar products to the wishlist or past gifts, as well as popular wishlist items that users may find difficult to discover on their own, are also generated to expand the candidate pool for the recommended product list. Users will receive more product recommendations compared to searching for gifts themselves.
The second step is the personalized sorting of the generated product list. Even with the same recommended product list, the sequence in which products are arranged varies for each user. The AI Gift Explorer tool calculates the scores for each product by determining weights based on the selected recipient conditions (e.g., gender, age), gift types (e.g., luxury, health-oriented, holiday gifts), and the user’s service usage history. Based on these scores, products are arranged in a customized order for each user.
Finally, “recommendation reasons” are attached to the items in the personalized recommended product list. The “recommendation reason” information assists users in making their final selection when choosing a gift. For instance, products desired by the recipient or by people with similar tastes are sorted and presented to the user, accompanied by “recommendation reasons.” Providing both the recommended products and reasons helps users make quick decisions, naturally enhancing user satisfaction.
Expanding Personalized AI Application in the KakaoTalk Gifts
Kakao Commerce plans to continually expand the application of AI technology on KakaoTalk Gifts to enhance user convenience when searching for products within the service. One area under review for potential expansion by Kakao Commerce is “Personalized Search Results.”
Gift searches differ in purpose from general searches. While general searches are primarily used to explore detailed information on a topic, gift searches focus on finding specific items or event-related products for purchase. With this, user satisfaction increases when search results reflect user preferences such as product price, reviews, purchase options, and popular items, not just shopping product outcomes. To achieve this, efforts are being made to personalize the sorting of search results using user behavior and purchase information.
There are also plans to personalize the “Life Themes” section. Currently, the home screen of the gift service features a section called “Life Themes,” which classifies everyday themes related to gifts rather than product categories. This section currently displays the same themes and detailed products to all users uniformly. However, this approach has the drawback of not reflecting user service behavior, which makes it difficult to quickly find highly satisfying products. To address this, 12 themes and the detailed product arrangements within those themes will be personalized through a recommendation AI model for each user.
By considering directions that enhance users’ shopping flow and the exploration convenience, Kakao Commerce plans to continue expanding the application of recommendation AI in the gift service with a focus on user “personalization.”
Experimentation and validation, continuing to address the changing concerns of users
As user preferences, trending products, and lifestyle patterns continually change, the recommendation AI model must also evolve to keep pace with these changes. Otherwise, user satisfaction with the service is likely to decline.
To address this, Kakao Commerce regularly monitors user’s feedback information Through A/B tests, the cycle of “design-experiment-validation” is repeated to optimize the recommendation AI model. This series of processes ensures that the recommendation AI models applied throughout the service consistently improve usability and user satisfaction, contributing to the growth of the service.
Ensuring the safety of service user data
Developing recommendation AI model, various types of user data are utilized, such as behavioral data and service operation logs.In the development process, it is essential to carefully consider how to safely utilize user data within the scope of its intended purpose without exposing it to external parties.
Before developing recommendation models, Kakao Commerce thoroughly ensures that the data used for model training and inference is secure from a user privacy protection standpoint. We thoroughly conduct review processes with privacy officers to determine whether the collected user data can be used for recommendation purposes. Only the data reviewed as usable is reflected in recommendation model training and inference. During the recent development of the 'AI Gift Explorer' recommendation service, data considered difficult to use for recommendations was excluded after review with data protection officers, ensuring a secure service development.
Kakao Commerce continues to make numerous efforts to provide users with a better online shopping experience. With constantly considering what information users want and what kind of experience they hope to have, Kakao Commerce is striving to find ways to address the concerns. Recommendation AI is seen as a solution to these challenges.
For this reason, Kakao Commerce plans to apply recommendation AI not only to the gift service but also to other services, such as called Shopping and Live Shopping. This aims to solve problems that were once difficult for people to handle, thereby enhancing user satisfaction.
Wrapping up, here is a statement of Kakao Commerce's commitment stated by CTO Hocheol Yang.
"Kakao Commerce has evolved its data platform, originally designed for large-scale data processing, into an MLOps platform. Additionally, it has enhanced the platform by expanding its applications to areas such as MAB-based personalized recommendations, CRM targeting, and more."
Moreover, as mentioned in the previous example of the gift explorer, we have developed systems and processes that adhere to privacy compliance when applying personalized recommendations
As a result, many users are safely utilizing services powered by AI technology. As of the time of writing, the Gross Merchandise Value (GMV) generated by AI-powered recommendations accounts for 9% of the total GMV in the Kakao gift service.
As a result, many users are safely utilizing services powered by AI technology. As of the time of writing, the Gross Merchandise Value (GMV) generated by AI-powered recommendations accounts for 9% of the total GMV in the Kakao gift service.
Moving forward, we will continue our research and development efforts to provide unique and enhanced customer experiences. This will be done through the advancement of FMOps, the expanded application of LLMs, and personalized recommendation searches.”
- if(kakao) How does Kakao make daily life with AI?
#AI#AINative#ifkakao
