LDPC was first proposed in 1962, after that considerable achievements have been made both in decoding algorithms and also in the structure of the decoder. In terms of decoding algorithm, from belief propagation (BP) to min-sum (MS), then normalized MS (NMS) and finally to a more performance oriented variable MS (VMS) algorithm. In terms of the decoder structure, there are mainly two, flooding and shuffled structure (FS and SS), and flooding structure is mainly used in the past and the selection of variable factors in FS VMS has already been studied in the literature. As is known, the horizontal shuffled structure (HSS) are more hardware friendly and now widely deployed, thus to apply VMS algorithm to HSS has an important meaning for the industry. However, there is no effective method for the selection of variable factors in HSS VMS. Based on the study of FS VMS, this paper describes how to apply the generalized mutual information (GMI) based metric to HSS VMS reasonably and proposes the modified GMI based formula. Simulation results on a certain DVB-S2's LDPC code show that, by using modified formula for HSS VMS, we can obtain 0.13dB gain over HSS NMS and the performance is only 0.1dB away from that of HSS BP in terms of bit error rate (BER) at level 1e-7. And simulation results on a certain DVB-T2's LDPC code show that, by using modified formula for HSS VMS, we can obtain 0.04dB gain over HSS NMS and the performance is only 0.06dB away from that of HSS BP in terms of bit error rate (BER) at level 1e-7.