Aiming at the sparsity problem of cold start and user item matrix in TV and movie personalized recommendation, this paper presents an improved collaborative filtering recommendation algorithm based on user situations and missing values estimation (BUM) applied to smart TV service. First of all, the users are clustered according to the cold start conditions. Then the user similarity of the cold start and non cold start users is calculated, and the neighbor users are selected. For cold start users, we model user attributes by analyzing user scenarios, and select neighbor user by user similarity which defined by scenario dissimilarity. For non-cold start users, we insert the default value based on user preferences into supplement of user-item rating matrix to solve the sparsity, and then calculate the similarity to select neighbor users. Finally, the results are obtained by using the neighbor users through the CF scoring prediction algorithm to estimate the rating. The experimental results show that the proposed algorithm is effective.