Text-to-Table: A New Way of Information Extraction

Abstract

We study a new problem setting of information extraction (IE), referred to as text-to-table . In text-to-table, given a text, one creates a table or several tables expressing the main content of the text, while the model is learned from text-table pair data. The problem setting differs from those of the existing methods for IE. First, the extraction can be carried out from long texts to large tables with complex structures. Second, the extraction is entirely data-driven, and there is no need to explicitly define the schemas. As far as we know, there has been no previous work that studies the problem. In this work, we formalize textto-table as a sequence-to-sequence (seq2seq) problem. We first employ a seq2seq model finetuned from a pre-trained language model to perform the task. We also develop a new method within the seq2seq approach, exploiting two additional techniques in table generation: table constraint and table relation embeddings.We consider text-to-table as an inverse problem of the well-studied table-to-text, and make use of four existing table-to-text datasets in our experiments on text-to-table. Experimental results show that the vanilla seq2seq model can outperform the baseline methods of using relation extraction and named entity extraction. The results also show that our method can further boost the performances of the vanilla seq2seq model. We further discuss the main challenges of the proposed task. The code and data are available at https://github. com/shirley-wu/text_to_table. 1

A Probe into the Mutual Enhancement between Tertiary Education of Art and Intangible Cultural Heritage in China: A Case Study of Xiamen Bead Embroidery Course by Xiamen Academy of Arts and Design, Fuzhou University

“…The input text is a report of a basketball game. In the existing dominant model (Wu et al, 2022), the output table is serialized into a token sequence during training by concatenating all rows in a top-down order. Here, ⟨s⟩ token is used to separate the cells of each row, ⟨n⟩ token is utilized to separate rows, and ⟨ ⟩ token means an empty cell.…”

“…Here, ⟨s⟩ token is used to separate the cells of each row, ⟨n⟩ token is utilized to separate rows, and ⟨ ⟩ token means an empty cell. Unlike Wu et al (2022), in this work, we model the generation of each table as a table header and then a set of table body rows. Note that these rows can be further decomposed into the first column and data cells wrapped in the red box.…”

“…In this aspect, Wu et al (2022) first propose sequence-to-sequence (seq2seq) text-to-table generation models. They first serialize each table to a token sequence by concatenating all rows in a top-down order.…”

“…Compared with the seq2seq models (Wu et al, 2022), our model has the following advantages: 1) our model is able to not only alleviate the order bias caused by the sequence generation but also reduce the effect of error propagation on the generation of long sequences; 2) our model achieves faster generation speed since table body rows can be efficiently generated in parallel.…”

“…

Recently, the text-to-table generation task has attracted increasing attention due to its wide applications. In this aspect, the dominant model (Wu et al, 2022) formalizes this task as a sequence-to-sequence generation task and serializes each table into a token sequence during training by concatenating all rows in a topdown order. However, it suffers from two serious defects: 1) the predefined order introduces a wrong bias during training, which highly penalizes shifts in the order between rows; 2) the error propagation problem becomes serious when the model outputs a long token sequence.

…”