Structured Semantic Modeling of Scientific Citation Intents

The search for relevant information within large scholarly databases is becoming an unaffordable task where deeper semantic representations of citations could give impactful contributions. While some researchers have already proposed models and categories of citations, this often remains at a theoretical level only or it simply reduces the problem to a short-text classification of the context sentence. In this work, we propose CiTelling: a radically new model of fine-grained semantic structures lying behind citational sentences able to represent their intent and features. After an extensive and multiple annotation of 1380 citations (https://github.com/rogerferrod/CiTelling), we tested the validity and the reliability of the proposal through both qualitative and quantitative analyses. In particular, we were able to 1) extend the current depth of existing semantic representations when used in computational scenarios, 2) achieve high inter-annotator agreement and 3) obtain state-of-the-art classification results with straightforward neural network models.

Automated summary

Searching for relevant information in large scholarly databases is becoming challenging, but deeper semantic representations of citations could be beneficial; CiTelling is a new model that can express the fine-grained semantic structures behind citation sentences, and through extensive annotation, its validity and reliability have been tested.