University of Novi Sad Faculty of Technical Science

Soft Computing 2015/2016 ( https://github.com/ftn-ai-lab )

Nenad Todorović RA 78/2012 Dragan Vidaković RA 134/2012

Semantic Classification of Sentences

Automatically assigning input sentences to a set of categories. Sentence assigning process is based on it's semantic. Categories are defined with the indicator words.

Data

"Sentence Classification Data Set" downloaded from Machine Learning Repository of Center for Machine Learning and Intelligent Systems - University of California, Irvine (http://cml.ics.uci.edu/). Data Set contains data for classification of sentences in scientific articles into categories:

1. AIMX - The specific research goal of the paper
2. OWNX - The author's own work (methods, results, conclusions...)
3. CONT - Contrast, comparasion or critique of past work
4. BASE - Past work that provides the basis for the work in the article
5. MISC - Any other sentences.

Data Set contains sentences from the abstract and introduction of 90 scientific article from three different domains:

1. PLoS Computational Biology (PLOS)
2. The Machine Learning repository on arXiv (ARXIV)
3. The rsychology journal Judgment and Decision Making (JDM).

Directories

1. "training_set"

• 24 PLOS articles with 893 sentences
• 24 ARXIV articles with 793 sentences
• 24 JDM articles with 814 sentences

2. "test_set"

• 6 PLOS articles with 220 sentences
• 6 ARXIV articles with 197 sentences
• 6 JDM articles with 200 sentences

3. "word_lists"

This directory contains one plaintext file for each of the 4 categories AIMX, OWNX, CONT and BASE. Each plaintext file lists the indicator words for the corresponding category. This directory also contains a stopwords file. The stopwords file contains stopwords that are not likely to be important for the taske of sentence classification (how, show, our...). File contains a set of stopwords that are not likely to be strong features for this task and thus can be safely removed.

Implementation

• Naive Bayes Classifier
• Decision Tree Classifier
• Support Vector Machines

Implementation relied on NLTK (http://www.nltk.org/) and TextBlob (http://textblob.readthedocs.org/en/dev/) Python libraries.

Added "new_labeled_data" directory with 18000+ new sentences from scientific articles, labeled using implemented classifiers.