Cross-Lingual Trends Detection for Named Entities in News Texts with Dynamic Neural Embedding Models

This paper presents an approach to detect real-world events as manifested in news texts. We use vector space models, particularly neural embeddings (prediction-based distributional models). The models are trained on a large ‘reference’ corpus and then successively updated with new textual data from daily news. For given words or multi-word entities, calculating difference between their vector representations in two or more models allows to find out association shifts that happen to these words over time. The hypothesis is tested on country names, using news corpora for English and Russian language. We show that this approach successfully extracts meaningful temporal trends for named entities regardless of a language.

• Type: Journal
• Year: 2016
• Where: ECIR
• Authors: Andrey Kutuzov, Elizaveta Kuzmenko
• Keywords: ECIR 2016, Information Technology