Skip to content

Latest commit

 

History

History
72 lines (43 loc) · 2.2 KB

README.org

File metadata and controls

72 lines (43 loc) · 2.2 KB

ADW.py

A python port of ADW [1].

Requirements

  • Python 2.x
  • NLTK, with the WordNet corpus
  • Semantic Signatures database (see Installation)
  • Offset2ID map (see Installation)

Installation

  1. Download the original Semantic Signatures database (for all the 118K concepts in WordNet 3.0, size ~ 1.4 GB) using the following link:

    http://lcl.uniroma1.it/adw/ppvs.30g.5k.tar.bz2

  2. Extract the folder ppvs.30g.5k from the downloaded tarball:

    $ tar zxf ppvs.30g.5k.tar.bz2

  3. adwpy.config expects this database (a) to be called ‘ppvs’ and (b) to be located in the adwpy/data/ directory. You could either move the downloaded folder to the expected location:

    $ mv ppvs.30g.5k adwpy/data/ppvs

    Or create a symlink:

    $ cd adwpy/data $ ln -s /path/to/ppvs.30g.5k ppvs

    Or edit adwpy.config.py to point to your prefered location.

  4. Similarly with offset2ID.map.tsv from the ADW github page:

    https://raw.githubusercontent.com/pilehvar/ADW/master/resources/offset2ID.map.tsv

    The default location is also in adwpy/data:

    $ cd adwpy/data $ ln -s /path/to/offset2ID.map.tsv

Quick start

adwpy ports the two disambiguation methods and three Semantic Signature comparison measures described in Pilehvar et al., 2013 [2], and implemented in [1]:

  • Disambiguation methods:
    • NONE (i.e., no disambiguation)
    • ALIGNMENT_BASED (i.e., alignment-based disambiguation)
  • Semantic Signature comparison measures
    • Cosine
    • WeightedOverlap
    • Jaccard

The only input format supported by adwpy is “surface text”.

The main front-end function in the library is api.pair_similarity. ADWTest.py gives an example of this function in use. Essentially:

dm = config.DisambiguationMethod.ALIGNMENT_BASED
sc = config.SignatureComparison.WEIGHTED_OVERLAP
score = pair_similarity(text1, text2, dm, sc)

Where score is a floating point number between 0.0 and 1.0.

References

[1] ADW https://github.com/pilehvar/ADW

[2] M. T. Pilehvar, D. Jurgens and R. Navigli. Align, Disambiguate and Walk: A Unified Approach for Measuring Semantic Similarity. Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (ACL 2013), Sofia, Bulgaria, August 4-9, 2013, pp. 1341-1351.