As the premise and basis of text sentimental analysis, the emotion polarity discrimination of lexicons is particularly important. Existing methods of select basic sentimental lexicons in the study of semantic tendency are mostly based on artificial discrimination and lexicons frequency. Those ways suffer the defects of randomness and subjectivity. And it is difficult to ensure the full coverage of the semantic relations in the dictionary. In the paper, we present a method that treats the candidate basic sentimental lexicons as the vertex and the HowNet acquaintance as edge weight to build sentimental lexicons undirected graph. The betweeness-centrality value of nodes in the graph is used as the reference of basic lexicons selecting. Thus we can ensure the reliability of the selected basic lexicons. Experiments show our method has a high accuracy in the classification of emotional tendencies.