Huge volume of text generated today poses big challenges to text mining applications. A tremendous amount of fragmented short texts are growing at a scale that makes it impossible for human to visually extract. Techniques proposed in text mining such as topic modeling dramatically improve the understanding of those unstructured and noisy texts. Naive while widely used models are word-cloud and word-bags. While we observed that these data models never considered the semantic relationship between the words, which make the results relatively hard to understand. One bright option is to organize those words semantically and generate an output of human understandable sentences. In this paper, we step our first foot into this direction by proposing a new data model: Belief Graph. We also proposed a schema to build belief graph model from short texts.