IEEE SYP 2018 Demo

ieee-syp-2018

This version will be demonstrated in IEEE SYP 2018, on July 26, 2018.

Hypergraph-of-entity weighing and pruning

• Improved the Hypergraph-of-Entity by switching Document hyperedges to undirected, adding node and hyperedge weights and introducing a new pruning approach.

  • Pruning required the deletion of directed hyperedges, which was not supported by the Grph library. This was forked and implemented. We now use our own custom version of Grph.

• Implemented a Biased Random Walk Score, using node and hyperedge weights to randomly traverse the hypergraph.

  • Also improved random sampling efficiency and implemented a new non-uniform random sampler.

• Introduced the analysis module, with a new Random Walk stability test based on Kendall's coefficient of concordance W.

• Created an Hypergraph-of-Entity inspection method to export node and hyperedge weights to CSV for external analysis.

• Improved the R graph analysis utility for studying the discriminative power of node and hyperedge weights, based on the exported CSVs from inspect.

  • Also added a script to explore functions in order to build node and hyperedge weighting functions. This will also be helpful to build ranking functions later on.

• The reachability index has been disabled.

  • The entityWeight has mostly been deprecated (it does not scale) and doing this will save memory.

• Created a partial port of the Hypergraph-of-Entity in C++ and integrated it with the Python tool using Boost Python to create a C++ Python library.

  • The C++ implementation has already been deprecated and serves as an integration example.

• Added overall configurations for the selection of ranking function.

• Fixed several issues with the Dockerfile and automated Docker Hub builds.

• Fixed MongoDB issues with the storage of keys with a period.

0.3

This version includes:

• A more flexible configuration file in YAML, now considered global;
• Support for ranking function configuration for a single index, along with supported and default parameter configurations;
• Evaluation based on a selection of specific parameters;
• Dynamic description of models based on selected parameters;
• Query time bar chart visualization, shown when there are multiple tested parameters;
• Bar chart visualization for effectiveness metrics;
• Docker demo instance.