Conversational recommendation systems have garnered increasing attention in recent years, driven by the successful development of large language models. These systems can elicit users’ real-time preferences through interaction, thereby achieving better recommendation results. However, traditional conversational recommendation systems primarily focus on improving accuracy while neglecting the diversity of the results, which can lead to information redundancy and decreased user satisfaction. If properly considering diversity when constructing a conversational recommendation system, it is possible to enhance user satisfaction and improve system performance simultaneously. In this paper, we propose a novel diversified conversational recommendation system that simultaneously increases both the success rate and diversity of recommendations. The system incorporates two key functions: interest exploration and item recommendation. During the interest exploration stage, two diversification mechanisms are employed to capture user preferences. Subsequently, triplet loss is applied to further enhance diversification in the item recommendation stage. Experimental results demonstrate that our proposed method outperforms traditional conversational recommendation systems on widely used datasets.