Core Concepts
The author introduces InteraRec, a novel recommendation framework that leverages screenshots and large language models to provide personalized recommendations, diverging from traditional approaches.
Abstract
InteraRec is an innovative recommendation system that captures high-frequency screenshots of user interactions on web pages in real-time. By utilizing multimodal large language models, it generates tailored recommendations based on user behavior summaries extracted from the screenshots. This approach overcomes the limitations of traditional recommendation systems by offering more personalized and effective suggestions to users.
Weblogs are rich sources of data for understanding user preferences, but interpreting this data can be challenging. InteraRec simplifies this process by using screenshots for a clearer representation of user actions. The framework integrates visual data, language models, and optimization tools to deliver real-time personalized recommendations to users.
By systematically capturing screenshots, processing them with MLLMs, and executing optimization tools based on user behavior summaries, InteraRec provides valuable insights into user preferences. This method ensures a seamless and satisfying shopping experience for users while maximizing revenue for online platforms.
Stats
Numerous state-of-the-art recommendation systems leverage weblogs for personalized recommendations.
Screenshots are captured at regular intervals during a browsing session.
The MLLM translates user interaction behavior into summary information based on predefined keywords.
The LLM decomposes the summary into constraints for generating optimal recommendations.
The assortment planning problem involves choosing an assortment that maximizes expected revenue.