Proxy re-encryption (PRE) is a promising cryptographic structure for pervasive data sharing in cloud-based social networks, which enables a semi-trusted proxy to convert a ciphertext for Alice into a ciphertext for Bob without seeing the corresponding plaintext. Since the proxy is semi-trust, a PRE scheme which can resist the collusion attack will be of great practical value. Jiang et al. in 2015 and Kim and Jeong in 2016 have proposed collusion-resistant PRE (CR-PRE) schemes from the lattice by a similar technique, respectively. However, through the analysis of their schemes, the authors find that both of them have defects in the construction of the re-encryption key, which will lead to the re-encryption ciphertext cannot be decrypted or decrypted error with high probability. In this study, the authors first point out the defects in the work of Jiang et al.'s work and Kim and Jeong's work and propose an improved collusion-resistant unidirectional PRE scheme from lattice, based on learning with errors problems. In addition to solving the defects, CR-PRE still has many useful properties similar to the previous schemes, such as unidirectional, collusion resistant, chosen-plaintext attack secure and so on.