Topic

The application of big models to rednote recommendations

Introduction：Community recommendation scenarios have rich multimodal content: graphic, video, live broadcast, the previous recommendation system based on user behavior, do recall, sorting, reordering, as much as possible to enhance the user's content consumption experience. In the rednote scenario, the diversified business and user-centered UGC community positioning bring a lot of challenges to content recommendation. Such as efficient and fast cold start, multimodal content understanding, user behavior modeling and reasoning. The booming development of big model technology, which makes full use of world knowledge, better learning and reasoning capabilities, provides some new paradigms for solving the recommendation problem in these community scenarios. On the one hand, the introduction of world knowledge in LLM, combined with the user behavior in the recommendation scenarios, dramatically improves the content understanding ability, which can better solve the problems of fast cold-start of content, accurate target modeling, and decentralized distribution of content; on the other hand, the reasoning ability of LLM, which can better combine the user's portrait, behavior, and consumption scenarios, can be interpreted to achieve the exploration of interests and break the information cocoon. Outline: -Introduction to rednote -Decentralized recommendation -Combination of rednote recommendation and big models