Formal concept analysis (FCA) has been employed for recommendation systems. However, conditional FCA approaches are confined to binary user-item relationships, failing to incorporate explicit rating information and consequently diminishing recommendation accuracy. In this paper, we develop fuzzy sub-contexts with rating information and generate a fuzzy concept set for recommendation. For fuzzy sub-context, we devise dual (user/item) similarity metrics for sub-context generation from fuzzy formal context. For fuzzy concepts, we develop a construction algorithm using concept size as heuristic information, which generates a fuzzy concept set based on the fuzzy sub-context. For the recommendation, we design user weights to calculate recommendation confidence within the fuzzy concept to formulate personalised recommendations. Experiments have been undertaken on nine real datasets. Results show that our algorithm achieves superior recommendation quality compares to seven algorithms: collaborative filtering algorithms (kNN, IBCF, BMF, GreConD-kNN) and formal concept-based recommendation algorithms (GRHC, CSPR, CSBR).