Giuseppe M. Mazzeo's Homepage

About Me

I am a computer scientist and software engineer with experience in both research and industry. Since October 2016 I have been a Research Scientist at Meta (formerly Facebook), working on ads delivery systems and privacy infrastructure. My research background is in databases and artificial intelligence, including question answering over RDF data and clustering algorithms from my postdoc at UCLA. Earlier in my career I designed and developed enterprise information systems, mainly based on Java and open-source DBMSs. I like working on challenging and practically relevant problems, and I am particularly good at quickly turning ideas into prototypes.
Download my résumé

Latest Projects

CLUBS-POpen Source

A parameter-free parallel algorithm for clustering data around centroids

CLUBS-P is the parallel version of CLUBS⁺, a centroid-based clustering algorithm (like k-means). The algorithm does not require to exchange data distributed across the network, but only summary information. Two java implementations are available: one based on message passing , using an ad-hoc protocol for coordinating master and workers, and one based on Apache Spark .

CANaLI

Answering Controlled Natural Language Questions on RDF Knowledge Bases

Nowadays, many knowledge bases are publicly available, and they can be queried through their SPARQL endpoint. However, a formal query language like SPARQL is not suitable for non-technical users. We developed a system which accept questions posed in a controlled natural language, which follows a grammar defined by means of few syntactical rules, thast are specialized according to the underlying knowledge base. The system guides the users into typing correct questions, both syntactically and semantically.
CANaLI is the winner of the task 1 of QALD-6 challenge.

CLUBS⁺

CLUBS⁺ is a parameter-free clustering algorithm. It is a centroid-based algorithm, i.e., like k-means, the produced clusters are groups of data each concentrated around a centroid. The algorithm combines a divisive phase with an agglomerative one, and after each phase clusters are refined by isolating outliers. Despite being parameter-free, it outperforms classical algorithm such as K-means++ and Birch both in efficiency and accuracy.

Find out more

Condomani

Condomani is the most used SaaS for property management in Italy. Besides the complete account management of properties, Condomani provides a social platform that enables the property manager and tenants to easily communicate and to handle maintenance tickets.

Find out more

Affibox 2010

Affibox 2010 is the management software used by Pubbliemme Srl, one of the main Italian outdoor advertising companies. It supports the overall workflows across the functional business units of the company, located in different branch offices, from the first contact with the customer to the final billposting.

GestOro

GestOro is the management software used by Banco Metalli Italiano SpA, the main Italian precious medal trading company. It supports all the business processess of the company, including the automatic generation of the invoices, based on a complex mechanism depending on the continuously changing quotation of precious metals.

Other Projects

eJRM

The eJRM (electronic Justice Relationship Management) project is an Italian national project involving several universities and companies. It aims at creating a novel model for handling relationships between citizens and the justice system. In this project, my main task was building a system for predicting the outcome of lawsuits from the text describing the case.

Find out more

TETRIs

The TETRis (Innovative Services over TETRA) project is an Italian national project involving several universities and companies. It aims at creating innovative services in the Smart Countries and Smart Cities context using the TETRA standard. In this project, my main task was building an OLAP system for analyzing the logs of communications over the TETRA network.

Find out more

GHBH

GHBH (Grid Hierarchical Binary Histograms) is a technique for summarizing multi-dimensional data. Summarized data enables a fast approximate evaluation of range queries, which can effectively support OLAP analysis (an approximate result is often enough, and long conputation on detailed data are saved) and selectivity estimation for optimal query plan evaluation. GHBH uses a space efficient-approach for construction histograms that approximate original data with an excellent accuracy.

Find out more

JSON-RPC Servlets Open Source

Libraries for easily creating JSON-RPC web services through Java Servlets.

View on GitHub

CRUD Services Open Source

Libraries supporting the creation of CRUD services.

View on GitHub

AssiCloud

AssiCloud is a SaaS platform for insurance brokers. THe broker can manage all the insurance life-cycle.

OroSoft

OroSoft is a SaaS platform for managing multiple cash-for-gold stores. The system is compliant with Italian laws regarding precious metal trading.

Wellcard

Wellcard is a multi-store reward system based in Southern Italy. The customers can collect points in the participating stores by using a mobile app.

Work Experience

Research Scientist - Meta (2016/10 - Present)

I joined Facebook (now Meta) in October 2016.

I was on the initial team that built Campaign Budget Optimization (CBO, now Advantage+ campaign budget), and later led the backend development that brought it to over 60% of Ads revenue.
In Privacy Infra, I rebuilt the classification infrastructure that processes all Meta data to meet privacy requirements, making the system modular and easily extensible. A presentation is available at PEPR '25.

Postdoctoral Scholar - University of California, Los Angeles (2014/01 - 2016/10)

I was a postdoc in the Scalable Analytics Institute at the UCLA computer science department, working with Prof. Carlo Zaniolo on clustering algorithms and question answering over RDF data. Main projects: CANaLI, CLUBS-P, CLUBS⁺.

Research associate - University of Milano-Bicocca (2012/05 - 2013/12)

Worked with Prof. Enza Messina and Prof. Sergio Flesca at the development of a system for system for predicting the outcome of lawsuits from the text describing the case. Main project: eJRM

Co-founder and CTO - Open Knowledge Technologies Srl (2010/11 - 2012/04)

I was a co-founder of this start-up, now spin-off of University of Calabria, where I initially served as CTO. Main projects: Condomani, TETRIs.

Co-founder and Software Engineer - Zenit Srl (2008/11 - 2012/04)

I was co-founder of this start-up and served as Software Engineer. Main projects: Affibox 2010, GestOro.

Research associate - ICAR-CNR (2008/01 - 2010/02)

Worked with Prof. Domenico Saccà, Prof. Filippo Furfaro, and Dr. Elio Masciari on devising technique for multi-dimensional data compression. As main results, a technique for building privacy-preserving histograms and a technique for distributing and querying histograms on a P2P network were devised.

Contract professor - University of Calabria (2007/01 - 2010/12)

I was instructor for the upper-division course of Information Systems (around 100 students per academic year) at the department of CS, where I also served as first-level thesis advisor for 30+ students.

Selected publications

M. Atzori, S. Gao, Giuseppe M. Mazzeo, Carlo Zaniolo: Answering End-User Questions, Queries and Searches on Wikipedia and its History. IEEE Data Eng. Bull. 39(3) 2016

B. Fazzinga, Filippo Furfaro, Elio Masciari, Giuseppe M. Mazzeo: Privacy or Security?: Take A Look And Then Decide. SSDBM 2016

Giuseppe M. Mazzeo, Carlo Zaniolo: Answering Controlled Natural Language Questions on RDF Knowledge Bases. EDBT 2016: 608-611

Elio Masciari, Giuseppe M. Mazzeo, Carlo Zaniolo: A New, Fast and Accurate Algorithm for Hierarchical Clustering on Euclidean Distances. PAKDD 2013: 111-122

Francesco Buccafurri, Filippo Furfaro, Giuseppe M. Mazzeo, Domenico Saccà: A quad-tree based multiresolution approach for two-dimensional summary data. Information Systems 36(7): 1082-1103 (2011)

Filippo Furfaro, Giuseppe M. Mazzeo, Andrea Pugliese: Managing Multidimensional Historical Aggregate Data in Unstructured P2P Networks. IEEE TKDE 22(9): 1313-1330 (2010)

Filippo Furfaro, Giuseppe M. Mazzeo, Domenico Saccà, Cristina Sirangelo: Compressed hierarchical binary histograms for summarizing multi-dimensional data. KAIS 15(3): 335-380 (2008)

Alfredo Cuzzocrea, Filippo Furfaro, Giuseppe M. Mazzeo: A Probabilistic Approach for Computing Approximate Iceberg Cubes. DEXA 2008: 348-361

Filippo Furfaro, Giuseppe M. Mazzeo, Domenico Saccà: A Probabilistic Framework for Building Privacy-Preserving Synopses of Multi-dimensional Data. SSDBM 2008: 114-130

Filippo Furfaro, Giuseppe M. Mazzeo, Cristina Sirangelo: Exploiting Cluster Analysis for Constructing Multi-dimensional Histograms on Both Static and Evolving Data. EDBT 2006: 442-459

My GitHub

Skills

I have successfully dealt with several technologies in my life. My good point is that I can learn new technologies quicky, the bad one is that I forget them just as quickly when I stop using them (not sure it is actually a bad point, maybe just my memory management daemon doing its job). However, you need to have a precise idea of what to do, for the rest there is StackExchange, right?