aiai at the FinSBD-2 task: Sentence, List, and Item Boundary Detection and Items Classification of Noisy Financial Texts Using Data Augmentation and Attention Model

摘要

This paper describes the method that we submitted to the FinSBD2-shared task in IJCAI-2020 to detect the sentence, list, and item boundaries and classify the items from noisy unstructured English and French financial texts. We used the spatial and semantic information of text to augment each tokenized word of text as a fixed-length sentence, and we labeled each word sentence as different boundary types. Then, we proposed the deep attention model based on word embedding to detect the sentence, list, and items boundaries in noisy English and French texts extracted from the financial documents and classified the item sentences into different item types. The experiment shows that the proposed method could be an effective solution to deal with the FinSBD2-shared task.