This project contains a scripts for preparing datasets for the specific task of building machine learning models to learn the features of academic abstracts that indicate that they are of key interest for a specific academic enquiry.

Status: Partially Functional

Basic Text Features for the Articles Complete 
Query Only Text Features Complete
Keyword Match Features Complete
Article to Query Match Feature Complete

All completed features marked with an asterisk * below


Current Work: Word Embedding Features

The scripts will take a dataset contain information about a set of academic papers that have been returned from an initial database search. This includes: title, abstract, authors.

They will then make use of information that describes the criteria of the academic search. This includes separate files describing inclusion and exclusion criteria, and two sets of keywords that could describe separate core topics (for multi-displinary papers).

All of this information is then used to generate features to help a machine learning algorithm distinguish which of the papers are appropriate for the query.

These are the planned list of features.

• Abstract - Length *
• Abstract - Wordcount *
• Abstract - Mean Word Length *
• Abstract - Max Word Length *
• Abstract - Proportion of Content Words *
• Title - Length *
• Title - Wordcount *
• Title - Mean Word Length *
• Title - Max Word Length *
• Title - Proportion of Content Words *
• Authors - Count *

• Abstract - Keywords 1 Matches *
• Abstract - Keywords 2 Matches *
• Abstract - Jaro Distance - Inclusion Criteria
• Abstract - Levenschtein Distance - Inclusion Criteria
• Abstract - Jaccard index - Inclusion Criteria
• Abstract - Sorensen-Dice - Inclusion Criteria
• Abstract - Ratcliff-Obershelp - Inclusion Criteria
• Abstract - Jaro Distance - Exclusion Criteria
• Abstract - Levenschtein Distance - Exclusion Criteria
• Abstract - Jaccard index - Exclusion Criteria
• Abstract - Sorensen-Dice - Exclusion Criteria
• Abstract - Ratcliff-Obershelp - Exclusion Criteria
• Keywords - Keywords 1 Matches *
• Keywords - Keywords 2 Matches *
• Title - Keywords 1 Matches *
• Title - Keywords 2 Matches *

• Keywords 1 - Count *
• Keywords 1 - Max Length *
• Keywords 2 - Count *
• Keywords 2 - Max Length *
• Inclusion Criteria - Jaro Distance - Exclusion Criteria *
• Inclusion Criteria - Levenschtein Distance - Exclusion Criteria *
• Inclusion Criteria - Jaccard index - Exclusion Criteria *
• Inclusion Criteria - Sorensen-Dice - Exclusion Criteria *
• Inclusion Criteria - Ratcliff-Obershelp - Exclusion Criteria *

All of the text similarity algorithms are explained in this article

In order to provide a concise representation of how well the abstract matches the query we calculate word embeddings for all words in both and then calculate their cosine similarity.

• Abstract - Word Embedding Cosine Similarity - Inclusion Criteria
• Abstract - Word Embedding Cosine Similarity - Exclusion Criteria