Entity Set Expansion with Meta Path in Knowledge Graph

This codebase contains implementation of the paper:

Zheng, Yuyan, et al. "Entity set expansion with meta path in knowledge graph." Pacific-Asia conference on knowledge discovery and data mining. Springer, Cham, 2017. [Paper]

Test cases

Running a model

To run the model, execute the following command: $ python main.py --seed_dir actor --seed_num 2

Available datasets are:

actor
software
movie
scientist

Requirements

The codebase is implemented in Python 3.6. Required packages are:

numpy    1.19.5
tqdm     4.62.3  

The Proposed Method

Candidate Entities Extraction

This method show in get_candidate.py. It is used to obtain the candidates based on the seeds.

• Step 1 obtains entity types of each seed.
• Step 2 generates the initial common candidates types by the intersection operation.
• Step 3 filters the initial candidates types with the concept hierarchy structure.
• Step 4 extracts candidate entities satisfying the ultimate candidates types.

Seed-Based Meta Path Generation

The goal is to automatically discover meta paths between seeds, shown in get_path.py

• Treenode edge is link(get_tree.py)
• Get seed pair
• judges whether the link is in the set of the given link type, whether the neighbor node isn’t visited before.
• Choose the tree node with max number of source set
• Choose the tree node with min number of tuples

Combination of Meta Path

• Get the weight of meta path using heuristic weight learning method, shown in get_path.py.
• combine meta paths to get the following ranking model, shown in order.py.