One Computer Science MSc thesis to be presented on 27 May

Tuesday, 27 May there will be a master thesis presentation in Computer Science at Lund University, Faculty of Engineering.

The presentation will take place in E:4130 (Lucas).

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname.lastname@cs.lth.se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions are found on this page.

10:15-11:00 in E:4130 (Lucas) N.B No more opponents for this presentation

Presenters: Pernilla Åström, Ludvig Svedberg
Title: Optimised Retrieval of Properties in a Graph Database with Machine Learning
Examiner: Jacek Malec
Supervisors: Mathias Haage (LTH), Lukas Gustavsson (Neo4j)

This study investigates how machine learning can be used to optimise the Neo4j graph database at the data storage layer. By predicting which properties will be more popular than others, they can be cached to reduce overall query latency. This is done by placing the properties in a fast and a slow storage. Due to Neo4j's unstructured nature, query statistics are not necessarily applicable between different nodes or databases. For this reason, machine learning is an interesting method.

First, this study measured the potential performance increase of caching the most popular properties for each node. This showed a potential latency and page faults reduction when utilising a fast memory storage.

Secondly, a general trend between properties in different databases was explored. A manual analysis found no clear trend, but some machine learning models showed potential for being able to be generalised in some datasets.

Finally, a machine learning model was integrated into the Neo4j system, and showed that a general model is not suitable for all datasets, but might work in some. A model trained for an individual dataset showed better results than a general one. Machine learning works best in systems with a lot of reads and few writes due to the expensive overhead of running the model. This study concludes that without further work, machine learning is not a viable solution.

Link to popular science summary: To be uploaded