Experiment code for the AAAI'15 paper:

A Neural Probabilistic Model for Context Based Citation Recommendation

Please note that the code is experimental, so it contains two main part:

learning paper embeddings and calculate score (indexing)

The unprocessed data (SQL data) about the citation context and the cited papers are in: https://psu.box.com/v/refseer

• You are welcome to use the code under the terms of the license, however please acknowledge its use by citation: W. Huang, Z. Wu, C. Liang, P. Mitra, and C. Lee Giles. A Neural Probabilistic Model for Context Based Citation Recommendation. In the Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI'15), 2015.

• Instruction: The shared data is a SQL dump of citeseerx database with 3 tables: citations, citationContexts, and papers.

  • Important fields of table papers:
    1. id: each pdf will have a different id, this id is referred to as paperid in table citations;
    2. cluster: same paper (may be have more pdfs in our databases) will have a unique cluster number.
  • Important fields of table citations:
    1. id: this id is referred to as citationid in table citationContexts;
    2. cluster: the cluster number of the cited document;
    3. paperid: the id of citing document.
  • Important fields of table citationContexts:
    1. citationid: link to the citations table.
    2. context: citation contexts, citations are surrounded by =-= and -=-.

• Please use MySQL to import the data, I was told that there were some problems when importing 'citationContexts.sql' to Postgres.

• After the database is imported: These are the steps that may help you:

  • create new data format, remove citations (surrounded by -=- and =-=) :

  CitationContext      Cluster  (cited paper) 
  

  • learn word embedding from citation context

  • learn paper embedding from citation context (initial paper embedding)

  • learn word embedding and paper embedding simultaneously.

    • when learning paper embedding only use adj. and noun. words in citation context

    • when learning paper embeddings, I assigned a normalized weight for each noun and adj word in an context

      For example, For one pair of citation and citation context:

      w_1, w_2, ... , w_{n-1}, w_{n}              p_i
      

      when learning embedding of paper p_i , word w_1 ,w_2... w_{n} has different learning weight. I use the co-occurrence of word and paper in the whole corpus as weight.

Should you have more questions, please email me at gmail start with harrywy

All codes are under Penn State ownership and is licensed under a reative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.