Upstage announced that it has been awarded the Best Paper Award - Honorable Mention at the 16th ACM International WSDM Conference, held in Singapore from February 27 to March 3.

WSDM is an annual conference organized by the Association for Computing Machinery (ACM) that is widely recognized as one of the world's most prestigious gatherings in the fields of web search and data mining. This year's conference received 690 submissions covering a wide range of topics, including recommendation systems, click-through rate prediction, and semantic search. Only 123 papers, or 17.8 percent, passed the rigorous screening process, making Upstage's recognition all the more impressive.

Upstage's award-winning paper, "Efficiently Leveraging Multi-level User Intent for Session-based Recommendation via Atten-Mixer Network," was the result of a collaborative research effort between the company's research team in Hong Kong. The paper explores session-based recommendation, a technique that predicts a user's next action based on recent behavior. It proposes a method to model user intent in multiple levels and improve recommendation performance through a new network structure called the Atten-Mixer (Multi-Level Attention Mixture Network).

One of the main challenges in developing effective recommendation systems is managing the exponential growth in model complexity that comes with incorporating various graph neural networks (GNNs). Upstage overcame this challenge by dissecting part of the neural networks and focusing on the most important elements, thus optimizing the search space and achieving optimal recommendations.

The team also considered both the features and relationships of the item to predict which item the user is likely to select, thereby improving the quality of recommendations.

For instance, if a user searches for a wedding dress and then looks for furniture to purchase, Upstage's proposed method can deduce that the user's intention is related to marriage and home decoration, and recommend relevant products accordingly, thus boosting the accuracy of recommendations.

Upstage's research was highly regarded by the WSDM program committee and participants, and was selected as a Best Paper Honorable Mention Award recipient. This award is given only to the top 0.5 percent of all submitted papers, and recognizes the outstanding research capabilities and achievements of the Upstage-Hong Kong research team.

The WSDM committee commended Upstage's research for proposing a fresh approach to session-based recommendation by exploring multi-level inference on user intent. They also praised the team's effectiveness in simplifying complex GNNs and designing more effective models to greatly reduce the search space in session-based recommendation, while demonstrating improved empirical performance through various online and offline experiments.

The proposed techniques developed by Upstage have already been launched on a large-scale e-commercial online service since 2021, and have led to significant improvements in top-tier business metrics, enhancing the performance of the company's RecSys AI Pack.

Upstage's AI Pack utilizes numerous models that have demonstrated their performance in various competitions such as the award-winning paper and Kaggle, and selects recommendation models tailored to each industry and client's service environment to provide optimal recommendation results.

With the AI Pack, Upstage assists customers in customizing three AI technologies simultaneously: OCR technology, recommendation technology that takes into account customer information and product/service characteristics, and natural language processing (NLP) that enables meaning-based search. The Upstage AI Pack not only simplifies data processing, AI modeling, and indicator management, but also supports continuous updates and makes it easy to stay up-to-date with the latest AI technology.

Jae Boum Kim, Upstage's AI Product who spearheaded the research, expressed his excitement saying, "It is an honor to receive recognition from the academic community for our recommendation technology through a collaboration with our Hong Kong research team. We are proud to have gone beyond academic research to study commercial models that are actually used in the field." He further added, "We will continue to create better AI Packs that are tailored to the unique needs of various industries and client's service environments, and contribute to their success."