Automatic Consistency Checking of Table and Text in Financial Documents

A company's financial documents use tables along with text to organize the data containing key performance indicators (KPIs) (such as profit and loss) a quantity linked them. The KPI’s in table might not be equal similarly described KPI's text. Auditors take substantial time manually audit these mistakes this process is called consistency checking. As compared existing work, paper attempts automate task help of transformer-based models. Furthermore, for checking it essential table's KPIs embeddings encode semantic knowledge structural table. Therefore, proposes pipeline that uses tabular model get embeddings. takes input KPIs, generates their embeddings, then checks whether are identical. evaluated on German language comparative analysis cell embeddings' quality from three models also presented. From evaluation results, experiment used English-translated Tabbie generate KPIs’ achieved an accuracy 72.81% task, outperforming benchmark, other