Clique-based Semantic kernel

Source code for Clique-based Semantic Kernel, NLE journal (2015)

It is possible to reproduce the experiments of the original paper.

Dependencies:

This code is written in R. To use it, you will need:

R version 3.2.2 or higher
RStudio IDE

Project Structure:

The original paper contains two types of experiments which are described in sections 4 and 5 respectively. For convenient use of this package, we built a seperate folder for each experiment:

TextSemanticRelatedness:
- doc_concept.csv: a collection of 50 text documents from (Lee, Pincombe and Welsh 2005). Wikipedia's entites for each document have been recognized by Wikifier (Milne and Witten 2013).
- sim.tbl: the above documents are paired in all possible ways and evaluate using the average human judgments.
- G.net: feature similarity graph contains 2,671 edges between 496 unique concepts.
- utils.R: a set of utilities to calculate correlations and reproduce scatterplots of the original paper.
- semanticrelatedness.R: source code of the text semantic relatedness experiments (section 4).
- Text Semantic Relatedness.Rproj: a project solution in Rstudio which contains a copy of objects.
ConceptSimilarity:
- conceptsim.csv: a collection of 97 WordNet concept pairs which is a benchmark data set in the task of concept similarity (Schwartz and Gomez 2011).
- G.net: a subgraph of WordNet which contains 2,796 vertices and 3,087 edges by starting from 152 unique concepts (conceptsim.csv) and add all neighbors which are reached by all types of semantic relations. This feature similarity graph contains 2,812 maximal cliques.
- utils.R: a set of utilities to calculate correlations and reproduce scatterplots of the original paper.
- conceptsimilarity.R: source code of the concept similarity experiments (section 5).
- ConceptSimilarityExperiments.Rproj: a project solution in Rstudio which contains a copy of objects.

Getting Started:

Navigate to the appropriate folder (TextSemanticRelatedness or ConceptSimilarity) and run .Rproj project file. The project will be opened in RStudio. After that, you can run semanticrelatedness.R or conceptsimilarity.R in order to reproduce experiments in the section 4 and 5 respectively. Feel free to contact me if you need any further queries.

Reference:

If you found this code useful, please cite the following paper:

Jadidinejad, A. H.; Mahmoudi, F.; Meybodi, M. R. Clique-based semantic kernel with application to semantic relatedness, Natural Language Engineering, 21 (5), pp. 725-742, 2015.

@article{NLE:10000595,
    Author = {JADIDINEJAD,A. H. and MAHMOUDI,F. and MEYBODI,M. R.},
    Doi = {10.1017/S135132491500008X},
    Issn = {1469-8110},
    Issue = {Special Issue 05},
    Journal = {Natural Language Engineering},
    Month = {11},
    Numpages = {18},
    Pages = {725--742},
    Title = {Clique-based semantic kernel with application to semantic relatedness},
    Url = {http://journals.cambridge.org/article_S135132491500008X},
    Volume = {21},
    Year = {2015},
    Bdsk-Url-1 = {http://journals.cambridge.org/article_S135132491500008X},
    Bdsk-Url-2 = {http://dx.doi.org/10.1017/S135132491500008X}
}

License