Abstract
Recommender models excel at providing domain-specific item recommendations byleveraging extensive user behavior data. Despite their ability to act aslightweight domain experts, they struggle to perform versatile tasks such asproviding explanations and engaging in conversations. On the other hand, largelanguage models (LLMs) represent a significant step towards artificial generalintelligence, showcasing remarkable capabilities in instruction comprehension,commonsense reasoning, and human interaction. However, LLMs lack the knowledgeof domain-specific item catalogs and behavioral patterns, particularly in areasthat diverge from general world knowledge, such as online e-commerce.Finetuning LLMs for each domain is neither economic nor efficient. In this paper, we bridge the gap between recommender models and LLMs,combining their respective strengths to create a versatile and interactiverecommender system. We introduce an efficient framework called InteRecAgent,which employs LLMs as the brain and recommender models as tools. We firstoutline a minimal set of essential tools required to transform LLMs intoInteRecAgent. We then propose an efficient workflow within InteRecAgent fortask execution, incorporating key components such as a memory bus, dynamicdemonstration-augmented task planning, and reflection. InteRecAgent enablestraditional recommender systems, such as those ID-based matrix factorizationmodels, to become interactive systems with a natural language interface throughthe integration of LLMs. Experimental results on several public datasets showthat InteRecAgent achieves satisfying performance as a conversationalrecommender system, outperforming general-purpose LLMs.