Artificial Intelligence News

New framework processing of knowledge graphs for AI applications


A research team led by professor Xindong Wu in Hefei, China has developed an unsupervised entity alignment framework to enhance the search process for related information in multiple knowledge graphs for artificial intelligence applications. The framework combines the advantages of various approaches and avoids relying on human labor to initiate the alignment process. They tested the framework on several datasets across languages ​​and measured the results, comparing them with those of fourteen other machine learning algorithms. Their model outperforms most of its competitors on two different metrics, and scores better than all of them when the metrics are combined into an overall score.

Credits: Tingting Jiang et al.

A research team led by professor Xindong Wu in Hefei, China has developed an unsupervised entity alignment framework to enhance the search process for related information in multiple knowledge graphs for artificial intelligence applications. The framework combines the advantages of various approaches and avoids relying on human labor to initiate the alignment process. They tested the framework on several datasets across languages ​​and measured the results, comparing them with those of fourteen other machine learning algorithms. Their model outperforms most of its competitors on two different metrics, and scores better than all of them when the metrics are combined into an overall score.

The group’s research was published on May 5 Intelligent ComputingJournal of Science Partners.

The new framework, called SE-UAE, scores higher on precision and recall than 12 of 14 competing algorithms, some supervised and some unsupervised. It scored higher overall for all three data sets. Experiments testing the robustness and scalability of the framework also achieved encouraging results.

The main advantage of the new framework is that it does not require complex data sets that are painstakingly annotated by humans. It can automatically handle data sets with missing information and combine data sets that have different internal structures. The results of the quantitative research thus show that it is not only convenient but also effective to use a combination of relatively easy automated methods of processing knowledge graphs for more sophisticated bootstrap.

Future research could further improve the efficiency and accuracy of the framework by tweaking one of the two modules of the framework.

The two modules of the framework are one that looks for surface similarities and another that looks for commonalities in relationships between entities. Both make use of a pair of knowledge graphs. In this case, the pair consists of knowledge graphs for the same content in two different languages, English and Japanese, French or Chinese. The data set was built by DBpedia from Wikipedia content.

The first module looks for not one but three types of surface similarities: the same name, the same meaning, and the same location on the two knowledge graphs. Importantly, the output of this module is used as input for the second module, which uses a type of neural network called a graph convolution network to automatically examine the internal structure of two knowledge graphs to find pairs of identical entities.

Once the framework has analyzed each pair of knowledge graphs and generated pairs of identical entities, researchers can check their work against the correct answers provided as part of the DBpedia dataset and assign scores according to their chosen evaluation metric.

Although knowledge graphs are essential for artificial intelligence applications such as recommendation systems, any structured representation of knowledge is generally incomplete. It is therefore desirable to combine information from multiple knowledge graphs through a process called entity alignment.

The easiest matching method is to compare surface attributes such as entity names. More sophisticated methods achieve better results, but usually require complex input data that must be generated manually first.

Wu’s co-authors on this paper are Tingting Jiang (who was Wu’s PhD student), Chenyang Bu and Yi Zhu.




Source link

Related Articles

Back to top button