First Assignment – Entity Linking

Web Data Processing System Assignment 1 – 2021 – Group 26

• Zhining Bai
• Bowen Lyu
• Tianshi Chen
• Yiming Xu

Description

This is a Python program to Entity Linking by processing WARC files. We recognize entities from web pages and link them to a Knowledge Base(Wikidata). The pipeline for this program as below:

Read WARC

• Use pyspark to read large-scale warc files, so the program supports parallel computing.
• Extract text information from HTML files by using beautifulsoup.

Named entity recognition

• Extract entities by using recognize_entities_bert model from sparknlp.

Disambiguation and NIL

We considered the popularity of the candidate page as well as the semantic similarity between the sentence where the entity is located and the candidate description to achieve Disambiguation.

• Popularity: Calculate popularity rankings using the Elasticsearch scoring algorithm and the number of properties of the mention from the knowledge graph.
• Sentence similarity: Measure the difference between text and description using the Levenshtein distance.

NIL: Retain results with distances < 40.

Prerequisites

Codes are run on the DAS cluster at /var/scratch/wdps2106/wdps_2126, result1 is a conda virtual environment that has been created. Below are the packages installed to run the assignment.

# if you want to use pip(pip for python3) to install the packages, use the following command(python version 3.8)
pip install pyspark==3.1.2
pip install spark-nlp==3.3.3
pip install beautifulsoup4
pip install python-Levenshtein
pip install elasticsearch

# if you want to use conda to install the packages, use the following command(recommended)
conda create -n <env> python=3.8
conda install pyspark
conda install bs4
conda install elasticsearch
pip install python-Levenshtein
pip install sparknlp

Run

To run the program, you can simply use the command below. The parameter Keyname is the name of page ID in WARC files such as WARC_TREC_ID. You need to declare the name of the page ID using this parameter. Be aware that the result file will be renamed as result.tsv.

sh run.sh /path/to/warc/file.warc.gz /path/to/result/ Keyname

If you use DAS cluster, you also need to add this command before running:

export OPENBLAS_NUM_THREADS=10

To check the score of the result file, use the command below.

python3 score.py /sample/annotation/file/sample.tsv /generated/result/file/result.tsv

Result

We tested our entity linking code using sample.warc.gz. Since sample_annotations.tsv only contains the entities that page_id is less than 92, our test results only output entity links with page_id <= 92. The f1 score of the sample data is 0.1122.

GitHub

View Github