Our job is to build a classifier to classify the field of the given academic papers. Each records corresponds to one paper, containing the hashed title, abstract, reference, number of citations, published year and the target label. There are four labels so we use one VS rest technique to construct models. In the feature engineering process, we achieved the unigram, bigram and trigram features in the text and perform a TFIDF (term frequency inverse document frequency) transformation. Finally we tried different classifiers in sklearn which support sparse input. The best classifier we use is the ensemble of 10 different multi-layer perceptron. Rank third at the end.