Source code from the NEFUSI project
Please note that this is an intermediate version to work with sentences.
It is recommended to import the project with an IDE such as Eclipse. When this is not possible, you can proceed manually as follows:
...\nefusi-sentences\src>javac -cp combined-sentences.jar nefusi/*.java
...\nefusi-sentences\src>java -cp .;combined-sentences.jar nefusi.nefusi_Sentences
Then, you will see something like this:
Execution time in milliseconds : 377719
Training Cosine Test
0.8611 0.911 0.7805
What means that the neurofuzzy model completed the training phase with 86.11% accuracy, a cosine similarity (between ground truth and generated solution) with 91.1% accuracy and the test phase with 78.05% accuracy. By default, We use the lawSentence200 dataset that compares 200 sentences in the legal field. The inputs of the neural part have been generated with BERT. Please note that this accuracy is closed to state-of-the-art for the dataset lawSentence200. For more information, please refer to the publication mentioned below.
(Please be aware that the running the software under current configuration takes more than 5 minutes (11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz 1.80 GHz))
@inproceedings{martinez2021,
author = {Jorge Martinez-Gil and
Riad Mokadem and
Josef K{\"{u}}ng and
Abdelkader Hameurlain},
editor = {Matteo Golfarelli and
Robert Wrembel and
Gabriele Kotsis and
A Min Tjoa and
Ismail Khalil},
title = {A Novel Neurofuzzy Approach for Semantic Similarity Measurement},
booktitle = {Big Data Analytics and Knowledge Discovery - 23rd International Conference,
DaWaK 2021, Virtual Event, September 27-30, 2021, Proceedings},
series = {Lecture Notes in Computer Science},
volume = {12925},
pages = {192--203},
publisher = {Springer},
year = {2021},
url = {https://doi.org/10.1007/978-3-030-86534-4\_18},
doi = {10.1007/978-3-030-86534-4\_18},
timestamp = {Thu, 14 Oct 2021 10:06:19 +0200}
}
The development of NEFUSI is funded in the project NGI Zero Discovery by the NLnet Foundation and the European Commission. Project number: 2021-04-069
MIT