Evaluating the Impact of Knowledge Graph Context on Entity Disambiguation Models

Abstract

Pretrained Transformer models have emerged as state-of-the-art approaches that learn contextual information from text to improve the performance of several NLP tasks. These models, albeit powerful, still require specialized knowledge in specific scenarios. In this paper, we argue that context derived from a knowledge graph (in our case: Wikidata) provides enough signals to inform pretrained transformer models and improve their performance for named entity disambiguation (NED) on Wikidata KG. We further hypothesize that our proposed KG context can be standardized for Wikipedia, and we evaluate the impact of KG context on state-of-the-art NED model for the Wikipedia knowledge base. Our empirical results validate that the proposed KG context can be generalized (for Wikipedia), and providing KG context in transformer architectures considerably outperforms the existing baselines, including the vanilla transformer models.

A Multilingual Dataset for Named Entity Recognition, Entity Linking and Stance Detection in Historical Newspapers

“…Compared to contemporary datasets (e.g. AIDA-CONLL dataset), EL systems achieved around 94 − 95 points in precision [15]. This gap in the performance shows how historical data still a challenge for the EL task.…”

“…The NewsEye dataset is segmented into sentences using the textto-sentence splitter 15 which supports several European languages among them we can find German, French, Finnish, and Swedish. Each corpus is split into ∼ 80% for training (train), ∼ 10% for validation (dev) and ∼ 10% for testing (test).…”

“…Therefore, F1macro is reported instead of F1-micro in this task. 15 https://github.com/mediacloud/sentence-splitter…”

Survey on English Entity Linking on Wikidata: Datasets and approaches

“…Many of the ambiguous mentions reference the most popular one, while also many exact unique matches exist. T-REx, the Kensho Derived ISTEX-1000 [24] 2 [ 24,77] LC-QuAD 2.0 [27] 2 [ 8,100] Knowledge Net [23] 1 [ 69] KORE50DYWC [80] 0…”

“…The vast majority of methods is using machine learning to solve the EL task [8,15,16,18,24,53,60,65,77,78,86,89,105]. Some of those approaches solve the ER and EL jointly as an end-to-end task.…”