With the popularization and rapid development of the Internet, the negative impact caused by negative comments is exacerbated. The increasingly common negative comments on various video websites and social networking sites have exercised a malign influence on public opinion and caused adverse social consequences. This paper uses LightGBM to establish a high-performance negative comment prediction model. The data set selected in this paper is from the Internet movie database IMDB, in which positive and negative comments have been marked for our training model. After Bag of Words processing and TF-IDF processing of the dataset and substituting it into our training model, the researchers obtained the final model with an accuracy of more than 95%, which has a more satisfactory performance compared with the classical integrated model and the traditional binary classification model. Thus it can be seen that this model has an advantage in identifying spiteful comments online.

At the time mobile devices and online payment make people’s life more convenient, they caused an increasing number of fraud cases in recent years. In this paper, we represented trading data as graphs by graph machine learning and setting up a high-performance model which could detect fraudulent transactions automatically. The datasets that the paper used were the fraudulent transactions dataset on Kaggle’s credit cards. By random under-sampling, it was processed and shown as bipartite graphs which were substituted into our training models after being processed by graph embedding algorithm. Finally, the optimal model was found by the coming out results. The result reveals that average embedder algorithm could detect fraud more precisely than the other three algorithms.