In this paper, we identify and study the research problem of gradient item retrieval. We define the problem as retrieving a sequence of items with a gradual change to a certain attribute, given a reference item and a modification text. For example, we may want a more floral dress after we see a white one. The extent of “floral” is objective, thus we may ask the system to present a sequence of products with a gradual increasing floral attribute. Existing item retrieval methods mainly focus on whether the target items appeared at the top of the sequence of items ranked by similarities between query and items. However, those methods ignore the demand for retrieving a sequence of products with gradual change on a certain attribute. To deal with this problem, we propose a weakly-supervised method to learn a disentangled item representation from user-item interaction data and ground the semantic meaning to dimensions of the item representation. Our method takes a reference item and a modification as a query. During inference, we start from the reference item, then gradually change the value of certain meaningful dimensions of the item representation to retrieve a sequence of items. We demonstrate our proposed method can achieve disentanglement through weak supervision. Besides, we empirically show our method can retrieve items in a gradient manner and, in item retrieval task, our method outperforms existing approaches on three different datasets.