Intro to Keyword Generator

Motivation

While uploading dataset to Depositar, find wikidata keyword for a dataset can be trivious and tedious.

In the current Depositar website, users have to fill in wikidata keyword by their own:

Fig. 1 A screenshot of the Depositar interface for inputting Wikidata keywords

Thus, we aims to develop a metadata generator, to help users find the apporporiate wikidata keywords to describe their dataset.

Method

In this seciton, we provide a 2-step pipeline to achieve the goal:

Fig. 2 A pipeline for the keyword generator project

Stpe 1: NER

After obtaining the input metadata for the current dataset, we will utilize Named Entity Recognition (NER) on the input data to selectively extract words that could potentially correspond to Wikidata Q-items.

To achieve this goal, we will utilize the ckiplab/bert-base-chinese-ner NLP task model.

The ckiplab/bert-base-chinese-ner model is part of the CKIP Transformers project, which offers transformer models specifically designed for traditional Chinese language processing.

We have selected the ckiplab/bert-base-chinese-ner model due to its superior F1 score in NER when compared to other models provided by CKIP Lab.

Step 2: Wikidata API Search

Here, we send a search request through the Wikidata API. By using action=wbsearchentities, we can search the Wikidata database using labels and aliases, thus finding the corresponding items that match our input potential keyword.

For more information, check for Media Wiki help: action=wbsearchentities