Data mining the underlying trust in the US Congress

In this paper, we mine the US congress voting records to extract the latent information about the trust among congress members. In particular, we model the Senate as a social network and the voting process as a set of outcomes of the underlying opinion dynamics which we assume follow a corrupted DeGroot model. The transition matrix in the opinion dynamics model is the trust matrix among Senators that we estimate. Our methodology is to first cluster the voting bills into different groups, and then obtain the Senators' opinions about the theme of each cluster, by performing a weighted Bernoulli sampling on the binary voting results. A key characteristic of the US congress is that most of the Senators stick with their own ideology. In view of this, we assign the role of stubborn nodes to some Senators, since their existence can facilitate estimating the trust matrix. In fact, we find the trust matrix by solving a linear regression problem, and then analyze the underlying latent information. Interestingly, our numerical results are quite consistent with the common intuition. More importantly, the trust information extracted can help understand the underlying relationship in the Senate and offer insights for devising political strategies.