Açık Seminer 2020 - Turkish NLP Seminar and Workshop

This repo includes the notebooks and slides for the Turkish Natural Language workshop. The implemented modules are:

• Text preprocessing
• Named Entity Recognition with SpaCy
• Unsupervised text classification with K-Means

TWNERTC (Turkish Wikipedia Named-Entity Recognition and Text Categorization Dataset ) by Sahin, et al. is used for Named Entity Recognition. The TWNERTC dataset contains approximately 300K named entities in 77 domains with more than 1000 fine-grained entity types. A subset of the dataset (the astronomy domain) is provided in the repo and the full clean version of the dataset in json format can be downloaded here.

JSON schema

[ {
    TOPIC_1: {
        SENTENCE_1: {
            "entities": [
                [
                    START_INDEX,
                    END_INDEX,
                    ENTITY_LABEL
                ], ...
            ]
        },
        SENTENCE_2: {...}
    TOPIC_2 : {...}
 } ]

A small Turkish news dataset crawled from various news websites is used for text clustering. This dataset contains news in 5 categories (economy, arts, politics, sports, technology) with 100 samples per category.

Clone the repo and install the requirements before running the notebooks:

git clone https://github.com/alaradirik/TR-NLP-workshop.git

pip install -r requirements.txt