valeriobasile / deko

Default Knowledge about Objects

Default Knowledge about Objects

DeKO is a repository of default knowledge about common object.

DeKO is built by parsing natural language text with KNEWS, extraccting instances of frames from the text, and the clustering the frames.

Frame clustering

Clustering is based on some kind of distance (or conversely, similarity) measure. For DeKO, we defined a similarity measure between frame isntances. It is implemented in the script framesimilarity.py.

Instructions:

1. script: Code works for any of the following scripts, given as partially or fully. 'src.py -a -t <Frame_similarity_Algo> -e <element_similarity_Algo> -r <true,false> '

'src.py --alpha --F_Sim_Algo <Frame_sim_Algo> --E_sim_Algo <element_similarity_Algo> --role <true,false> '

Parameters values:

1. -a or --alpha : default : 0.5

2. -t or --F_Sim_Algo : default : WUP

3. -e or --E_sim_Algo : default : WUP

4. -r or --role : default : false

The input files are two text files containing RDF triples, defining one frame instance each.

*Test input files are run from the folder named "frame_files" *Any similarity calculating algorithm can be implemented instead of WUP for either between frame_types or frame_elements. Just add the module and call that similarity. Comments are mentioned in the code at place where new algo can be called. *Necessary comments are mentioned in the code. *For analysing results at each level of calculation, uncomment the print statements and run. *Data structures like list, dictionary and tuples are used so that the code can be easily used for multiple frames if require at some point.