Thesis of Michael Currant

Latent Knowledge Representation Learning for augmentation of structure and functionality of online services


The number of web APIs is always increasing, with many new web services and connected objects every day. In order to facilitate the use of these APIs by users, it has become common to employ a virtual assistant who can understand the user's intent and invoke the necessary APIs in his place.

Such a system requires a complex conversational model that can converse naturally with the user, correctly identify his intent(s), understand the structure and functionality of the various APIs available to the system, and invoke the required APIs to satisfy the user’s intent.

Our project focuses on the model’s comprehension of various APIs, their structure, semantic functionality, and their similarity, categorization and hierarchy relationships. Identifying techniques to enrich the metadata associated with APIs and find ways to represent their deterministic nature in a probabilistic domain.

Advisor: Boualem Benatallah
Coadvisor: Emmanuel Coquery