unKR: A Python Library for Uncertain Knowledge Graph Reasoning by Representation Learning

<p align="center"> <a href="https://pypi.org/project/unKR/"> <img alt="Pypi" src="https://img.shields.io/pypi/v/unKR"> </a> <a href="https://github.com/seucoin/unKR/blob/main/LICENSE"> <img alt="Pypi" src="https://img.shields.io/badge/license-Apache--2.0-yellowgreen"> </a> <!-- <a href=""> <img alt="LICENSE" src="https://img.shields.io/badge/license-MIT-brightgreen"> </a> --> <a href="https://seucoin.github.io/unKR/"> <img alt="Documentation" src="https://img.shields.io/badge/Doc-online-blue"> </a> </p>

unKR is an python library for uncertain Knowledge graph (UKG) Reasoning based on the PyTorch Lightning. It provides a unifying workflow to implement a variety of uncertain knowledge graph representation learning models to complete UKG reasoning. unKR consists of five modules: 1) Data Processor handles low-level dataset parsing and negative sampling, then generates mini-batches of data; 2) Model Hub implements the model algorithms, containing the scoring function and loss function; 3) Trainer conducts iterative training and validation; 4) Evaluator provides confidence prediction and link prediction tasks to evaluate models' performance; 5) Controller controls the training worklow, allowing for early stopping and model saving. These modules are decoupled and independent, making unKR highly modularized and extensible. Detailed documentation of the unKR is available at here.

unKR core development team will provide long-term technical support, and developers are welcome to discuss the work and initiate questions using issue.

<h3 align="center"> <img src="pics/unKR.svg", width="1000"> </h3> <!-- <p align="center"> <a href=""> <img src="pics/unKR.svg" width="1000"/></a> <p> -->

What's New

• [May, 2024] We perform instruction tuning with LLaMA-2-7B for UKG reasoning and achieve stronger performance in several UKG resoning tasks. We have relesed the implementation source code and tuned checkpoints.
• [Mar, 2024] Our paper "unKR: A Python Library for Uncertain Knowledge Graph Reasoning by Representation Learning" is accepted by SIGIR2024 Demonstration Track.
• [Feb, 2024] We have relesed this repo that includes the source code and documentation of unKR.

Demo

This is a demo shows the training and testing process of PASSLEAF model with unKR.

<!--  --> <img src="pics/demo.gif"> <!-- <img src="pics/demo.gif" width="900" height="500" align=center> -->

Datasets

unKR provides three public UKG datasets including CN15K, NL27K, and PPI5K. The following table shows the source, the number of entities, relations, and facts of each dataset.

| Dataset | Source | #Entity | #Relation | #Fact | |:-------:|:----------:|:---------:|:---------:|:-----------:| | CN15K | ConceptNet | 15000 | 36 | 241158 | | NL27K | NELL | 27221 | 404 | 175412 | | PPI5K | STRING | 4999 | 7 | 271666 |

Models

Now, nine uncertain knowledge graph representation learning models are available and they can be divided to two types: normal and few-shot models.

| Type | Model | |:--------:|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------:| | Normal | BEURrE, FocusE, GTransE, PASSLEAF, UKGE, UKGsE, UPGAT | | Few-shot | GMUC, GMUC+ |

Reproduced Results

unKR determines two tasks, confidence prediction and link prediction, to evaluate models' ability of UKG reasoning. For confidence prediction task, MSE (Mean Squared Error) and MAE (Mean Absolute Error) are reported. For link prediction task, Hits@k(k=1,3,10), MRR (Mean Reciprocal Rank), MR (Mean Rank) under both raw and filterd settings are reported. In addition, we choose high-confidence (>0.7) triples as the test data for link prediction.

Here are the reproduce results of nine models on NL27K dataset with unKR. See more results at here.

Confidence prediction

<table> <thead> <tr> <th>Type</th> <th>Model</th> <th>MSE</th> <th>MAE </th> </tr> </thead> <tbody align="center" valign="center"> <tr> <td rowspan="8">Normal</td> <td>BEUrRE</td> <td>0.08920 </td> <td>0.22194 </td> </tr> <tr> <td>PASSLEAF(ComplEx)</td> <td>0.02434 </td> <td>0.05176 </td> </tr> <tr> <td>PASSLEAF(DistMult)</td> <td>0.02309 </td> <td>0.05107 </td> </tr> <tr> <td>PASSLEAF(RotatE)</td> <td>0.01949 </td> <td>0.06253 </td> </tr> <tr> <td>UKGE_logi</td> <td>0.02868 </td> <td>0.05966 </td> </tr> <tr> <td>UKGE_rect</td> <td>0.03326 </td> <td>0.07015 </td> </tr> <tr> <td>UKGsE</td> <td>0.12202 </td> <td>0.27065 </td> </tr> <tr> <td>UPGAT</td> <td>0.02922 </td> <td>0.10107 </td> </tr> <tr> <td rowspan="2">Few-shot</td> <td>GMUC</td> <td>0.01300 </td> <td>0.08200 </td> </tr> <tr> <td>GMUC+</td> <td>0.01300 </td> <td>0.08600 </td> </tr> </tbody> </table>

Link prediction

<table> <thead> <tr> <th>Type</th> <th>Model</th> <th>Hits@1</th> <th>Hits@3</th> <th>Hits@10</th> <th>MRR</th> <th>MR</th> </tr> </thead> <tbody align="center" valign="center"> <tr> <td rowspan="10">Normal</td> <td>BEUrRE</td> <td>0.156 </td> <td>0.385 </td> <td>0.543 </td> <td>0.299 </td> <td>488.051 </td> </tr> <tr> <td>FocusE</td> <td>0.814 </td> <td>0.918 </td> <td>0.957 </td> <td>0.870 </td> <td>384.471 </td> </tr> <tr> <td>GTransE</td> <td>0.222 </td> <td>0.366 </td> <td>0.493 </td> <td>0.316 </td> <td>1377.564 </td> </tr> <tr> <td>PASSLEAF(ComplEx)</td> <td>0.669 </td> <td>0.786 </td> <td>0.876 </td> <td>0.741 </td> <td>138.808 </td> </tr> <tr> <td>PASSLEAF(DistMult)</td> <td>0.627 </td> <td>0.754 </td> <td>0.856 </td> <td>0.707 </td> <td>138.781 </td> </tr> <tr> <td>PASSLEAF(RotatE)</td> <td>0.687 </td> <td>0.816 </td> <td>0.884 </td> <td>0.762 </td> <td>50.776 </td> </tr> <tr> <td>UKGE_logi</td> <td>0.525 </td> <td>0.673 </td> <td>0.812 </td> <td>0.623 </td> <td>168.029 </td> </tr> <tr> <td>UKGE_rect</td> <td>0.500 </td> <td>0.647 </td> <td>0.800 </td> <td>0.599 </td> <td>125.233 </td> </tr> <tr> <td>UKGsE</td> <td>0.038 </td> <td>0.073 </td> <td>0.130 </td> <td>0.069 </td> <td>2329.501 </td> </tr> <tr> <td>UPGAT</td> <td>0.618 </td> <td>0.751 </td> <td>0.862 </td> <td>0.701 </td> <td>69.120 </td> </tr> <tr> <td rowspan="2">Few-shot</td> <td>GMUC</td> <td>0.3