Guillaume Plique

Jupyter notebooks are a great tool for exploratory data analysis as it is very easy to process and visualize data using traditional charts. However, they lack utilities to properly explore networks interactively. The ipysigma library therefore proposes a Jupyter widget enabling its users to perform visual network analysis from the comfort of a notebook. ipysigma makes it simple to tweak a network's visual variables to display it exactly as you intend. This way, you can perform a work at the crossroad between Python processing of graph data and visual exploration like you would do for example with Gephi. It supports networkx and igraph seamlessly and can be easily used by numpy and pandas users all the same. Different usecases of ipysigma will be showcased during the talk through a dataset about FOSDEM history. We will also demonstrate how ipysigma is able to render synchronized & interactive "small multiples" of a same network so that one can easily compare different features. ipysigma is developed at the médialab of SciencesPo and uses graphology and sigma.js (JavaScript libraries able to render interactive graphs in web browsers).

Web mining, as represented mostly by the scraping & crawling practices, is not a straightforward task and requires a variety of skills related to web technologies. However, web mining can be incredibly useful to social sciences since it enables researchers to tap into a formidable source of information about society.

But researchers may not have the possibility to invest copious amount of times into learning web technologies in and out. They usually rely on engineers to collect data from the web. The object of this talk is to explain how Sciences Po's médialab designed & developed tools to empower researchers and enable them to perform web mining tasks to answer their research questions. Here is an example of issues we will tackle during this talk:

• How a social sciences laboratory life can be a very fruitful context for tool R&D regarding webmining
• How to create performant & effective webmining tools that anyone can use (multithreading, parallelism, JS execution, complex spiders etc.)
• How to re-localize data collection: researchers should be able to conduct their own collections without being dependent on external servers or resources
• How to teach researchers the necessary skills: HTML, the DOM, CSS selection etc.

Examples will be taken mainly from the minet CLI tool and the artoo.js bookmarklet.

Speaker

Guillaume Plique is a research engineer working for SciencesPo's médialab. He assists social sciences researchers daily with their methods and maintain a variety of FOSS tools geared toward the social sciences community and also developers.

There is a tenacious misconception nowadays that people working with JavaScript do not need to know much about data structures because developing for the web is still often deemed to be, and this cannot be more false, an easier task than "real" programming.

Web applications have become complex beasts and node.js allows to build programs that used to be other backend languages' turf. JavaScript developers now need to develop, as with any other programming language, custom data structures to solve particular problems when arrays and objects are simply not enough. The intention of this presentation is therefore to compile series of tips, tricks and use cases regarding the implementation of data structures in JavaScript and the challenges it poses. It is indeed tricky to ensure consistent performance in a high-level language with JIT and where engines like v8 apply a lot of optimization magic. One has to be able to evolve around those constraints. Examples will be taken from libraries such as graphology, sigma.js' newest version and finally mnemonist to demonstrate that 1. performant data structures can be designed for JavaScript using the language's capabilities and that 2. everyone can use them to solve problems more efficiently.

Hyphe, a web crawler for social scientists developed by the SciencesPo médialab, introduced the novel concept of web entities to provide a flexible and evolutive way of grouping web pages in situations where the notion of website is not relevant enough (either too large, for instance with Twitter accounts, newspaper articles or Wikipedia pages, or too constrained to group together multiple domains or TLDs...). This comes with technical challenges since indexing a graph of linked web entities as a dynamic layer based on a large number of URLs is not as straightforward as it may seem.

We aim at providing the graph community with some feedback about the design of an on-file index - part Graph, part Trie - named the "Traph", to solve this peculiar use-case. Additionally we propose to retrace the path we followed, from an old Lucene index, to our experiments with Neo4j, and lastly to our conclusion that we needed to develop our own data structure in order to be able to scale up.

Speakers

• Paul Girard
• Mathieu Jacomy
• Benjamin Ooghe-Tabanou
• Guillaume Plique

While a lot of languages already have comprehensive libraries to handle graph data (like networkx in python), the same cannot be said for JavaScript.

This talk discusses how we were driven to create a multipurpose graph object specification for JavaScript and what were the design issues we soon had to tackle from implementation choices to API naming.

The result of this work is the graphology specification & reference implementation which we present to open it to feedbacks & contributions.

We will first review the state of the art of graph libraries for JavaScript and discuss the issues which brought us to create a new projet from scratch.

Then we will explain why we chose to design an open specification for a standard API rather than only providing a library.

Finally we will present the reference implementation & the attached library of algorithms and justify the technical choices made, before presenting our future roadmap and related projects (a new version of the graph rendering library, sigma.js, notably).

Alexis Jacomy

Alexis is CTO of Matlo, a data discovery web app. He spent last years working on tools to help people explore data. He also develops multiple JavaScript tools with Guillaume Plique, notably sigma.js to visualize networks in web pages.

https://github.com/jacomyal

Guillaume Plique

Guillaume Plique is a developer at SciencesPo's médialab working on a variety of Open Source projects in JavaScript, Clojure & Python. He specializes in developing software for social sciences researchers and often works with graphs, natural language processing and UIs.

https://github.com/Yomguithereal