Many decision problems using linguistic approaches to assess preferences are assumed that the linguistic term set is uniform and symmetrical distributed.However, there exist decision making problems whose assessments are unbalanced linguistic term sets.Theretore, a linguistic computational model based on symbolic models is proposed.Firstly, a weighted graph composed of basic labels is constructed.An unbalanced linguistic label is represented by some vertices in the graph.Then the Manhattan distance measure in the graph of any vertices is defined, with is applived to compute the distance between the two unbalanced linguistic terms.Finally, a decision model is designed to solve the TOPSIS problem.A numerical example is given to illustrate that the proposed method can visualise the unbalanced linguistic term set, and the Manhattan distance is better than the Euclidean distance when solving TOPSIS problems.