Types

• Paradigmatic: A & B have paradigmatic relation if they can be substituted for each other
• Syntagmatic: A & B have syntagmatic relation if they can be combined with each other

Applications

• Text retrieval (e.g., use word associations to suggest a variation of a query)
• Automatic construction of topic map for browsing: words as nodes and associations as edges
• Compare and summarize opinions (e.g., what words are most strongly associated with “battery” in positive and negative reviews about iPhone 6, respectively?)

Algorithm: Paradigmatic Relation Mining

• Represent each word by its context (e.g. concat words in left_1, right_1, neighbors[4]... to generate psudo documents)
• Compute context similarity (e.g. sum of similarity of corresponding psudo documents)
• Words with high context similarity are likely to have paradigmatic relations

Algorithm: Syntagmatic Relation Mining

• by computing mutual information

Additional Readings

• Identify lexical atoms like "hot dog":
  Chengxiang Zhai, Exploiting context to identify lexical atoms: A statistical view of linguistic context. Proceedings of the International and Interdisciplinary Conference on Modelling and Using Context (CONTEXT-97), Rio de Janeiro, Brzil, Feb. 4-6, 1997. pp. 119-129.

• Use word graphs to find paradigmatic and syntagmatic relations
  Shan Jiang and ChengXiang Zhai, Random walks on adjacency graphs for mining lexical relations from big text data. Proceedings of IEEE BigData Conference 2014, pp. 549-554.

Reference

Text Mining: https://www.coursera.org/learn/text-mining