QUEST: Answering Complex Questions by joining Multi-Document Evidence with Quasi Knowledge Graphs


Description

QUEST is a method that can answer complex questions directly from textual sources on-the-fly, by computing similarity joins over partial results from different documents. It is completely unsupervised, avoiding training-data bottlenecks and being able to cope with rapidly evolving adhoc topics and formulation style in user questions. QUEST builds a noisy quasi KG with node and edge weights, consisting of dynamically retrieved entity names and relational phrases. It augments this graph with types and semantic alignments, and computes the best answers by an algorithm for Group Steiner Trees.

Our code and data are available here.

To cite QUEST:

For feedback and clarifications, please contact:

Sample Questions


Advanced Options
Number of GSTs

10

100


Entity Alignment

0.2

0.3


Predicate Alignment

0.7

0.8


Ranking Options





Number of Documents

Type Filtering

Note: If you change any of the advanced options, please re-enter the question and hit the "Answer" button again.

-->