What is Semantic Data Science?

Semantic Data Science uses artificial intelligence (AI) techniques to help you perform semantically meaningful data science. For example, this can include helping you understand tabular data by mapping tables to domains (such as, this table is about automobiles), mapping columns to concepts (such as, this column reflects the price of a car in dollars), or extracting meaningful features for a given table based on domain concepts (such as, most people compute body mass index (BMI) when they see the weight and height of a person in columns).

Semantic Data Science use cases

Today, automated data science attempts to help you build effective AI models by performing automated feature engineering, but they have no knowledge of the characteristics of a domain. However, the best data scientists use domain knowledge to build models that are ultimately more interpretable and meaningful. The goal of Semantic Data Science is to help data scientists build better models by leveraging extracted domain knowledge that captures the wisdom of a group of data scientists.

What is concept mapping?

Concept mapping is when columns in a table are semantically mapped to meaningful vocabularies or concepts. The vocabulary can either be a specialized glossary or generalized concepts from popular knowledge graphs such as Wikipedia.

What is feature extraction?

In feature extraction, useful features for a given domain are extracted from multiple sources of information to inform what might be useful features for data science. As an example, body mass index (BMI) and its relation to a person's height and weight might be extracted from Wikipedia, from text cells of notebooks in open source that manipulate weight and height, or from analysis of programs in open source that manipulate similar columns in other data sets. Column mapping is extremely useful in identifying how you might generalize these extracted features to apply it to a new problem.

Advanced reading

• Kavitha Srinivas, Takaaki Tateishi, Daniel Karl, Udayan Khurana, et al., Semantic Feature Discovery with Code Mining and Semantic Type Detection. In the proceedings of the 36th AAAI Conference on Artificial Intelligence (AAAI), 2022.
• Udayan Khurana, Sainyam Galhotra, Semantic Concept Annotation for Tabular Data. In the 30th ACM International Conference on Information & Knowledge Management, 2021.
• Sainyam Galhotra, and Udayan Khurana, Semantic Search over Structured Data. In the proceedings of the 29th ACM International Conference on Information & Knowledge Management (CIKM), 2020.
• Sainyam Galhotra, Udayan Khurana, Oktie Hassanzadeh, Kavitha Srinivas, Horst Samulowitz,Miao Qi. Automated Feature Enhancement for Predictive Modeling using External Knowledge. In the proceedings of the International Conference on Data Mining (ICDM), 2019.
• Udayan Khurana, Takaaki Tateishi, Daniel Karl, Kavitha Srinivas. Leveraging Semantics to Build Better AI Solutions.
• Jiménez-Ruiz, Ernesto, Oktie Hassanzadeh, Vasilis Efthymiou, Jiaoyan Chen, Kavitha Srinivas, and Vincenzo Cutrona. Results of Semtab 2020. In CEUR Workshop proceedings, vol. 2775, pp. 1-8. 2020.

Summary

This article provided an overview of Semantic Data Science, the challenges associated with it, and use cases to help identify how Semantic Data Science can be used in custom projects. It also included a list of conference publications for advanced reading.