An Interleaved Model for Chinese Socio-Economic Indicator Extraction

摘要

Automatically extracting socio-economic indicators from text allows us to know the real status of a country or a company. Most conventional methods rely on hand-crafted features or use pipeline or joint deep learning model. They are either effort intensive or do not utilize the dependency between indicator detection and element extraction. In this paper, we use an interleaved model to cope with the challenges in this task. After feature extraction, the input sequence is first fed into the preliminary detection module to test whether it mentions indicators. If the sequence contains indicator mentions, it will then be input into element extraction module to extract indicator elements. If there are no elements, the sequence will be judged containing no indicators. We conduct extensive experiments on the dataset we build, which shows that the proposed model performs better both on f1-score and efficiency.