Gaël POUX

[Gaël POUX @GaelPoux]

As an example, we coupled the simplest iteration of SIMSBM to our method to retrieve geographical clusters of roman inscriptions over time. We retrieve some simple historical facts, linked to the expansion of the Roman empire over time.

[Gaël POUX @GaelPoux]

Our latest work "Dynamic MMSBM for Weighted Labeled Networks" with @jvelcin and S. Loudcher @ERIC_EA3083 got accepted for publication at #sigir2023!

Find more below or at https://t.co/B3SWj31Wvg

[Gaël POUX @GaelPoux]

Each data slice needs very few observations to obtain very good results, because slices "help" each other during the optimization procedure. This makes our approach fit for situations where data is scarce --typically in social sciences.

[Gaël POUX @GaelPoux]

In our paper, we test the boundaries of this extension, and show it is fit for a number of problems. We also solve computational problems that arise and manage to retain a linear computation time.

[Gaël POUX @GaelPoux]

Our (other!) latest work with @jvelcin and S. Loudcher @ERIC_EA3083 on Multivariate Dirichlet-Hawkes Processes got accepted for publication at @ecir2023 ; study topical interactions at scale, using this new nonparameteric approach!

Find more below or at https://t.co/MFKLYAAg8E

[Gaël POUX @GaelPoux]

However, there was no interplay between topics; politics could not trigger sports or business related tweets.

We propose an extension to this approach, that alleviates this assumption. Topics can now influence each other.

[Gaël POUX @GaelPoux]

We propose to infer not only clusters of memes, but also the dynamic networks they spread on ; when who spreads what to whom.

[Gaël POUX @GaelPoux]

Our latest work with @jvelcin and S. Loudcher @ERIC_EA3083 on Dirichlet-Point processes just got accepted for publication at @ecir2023 ; it shows how to infer topic-dependent diffusion networks at scale from data streams!

Find more below or at https://t.co/gjZ3Dz6iue

[Gaël POUX @GaelPoux]

Consider a stream of timestamped diffusion events, represented as triplets (user, date, content), where content can be a text, a picture, a song, or anything else you can cluster using Bayesian modelling.

[Unité de Recherche ERIC @ERIC_EA3083]

13/09/22 – Soutenance de thèse de Gaël Poux-Médard : Interactions entre informations dans les processus de diffusion / Interactions in #InformationSpread. La soutenance aura lieu le ma https://t.co/ijyKUu40Wz

[Gaël POUX @GaelPoux]

SIMSBM is a framework, for it allows to easily recover recently developed MMSBMs variations as special cases, such as MMSBM, Bi-MMSBM, T-MBM, IMMSBM. Try it yourself at https://t.co/JuvVDgWdqU

[Gaël POUX @GaelPoux]

Our latest work with @jvelcin and Sabine Loudcher on Mixed Membership SBMs just got accepted for publication at @icdm2022 !

Find more below or check it out at https://t.co/zCkksh4K6l

[Gaël POUX @GaelPoux]

Diagnose a disease given a combination of 1, or 2, or 3 symptoms; guess a movie rating given its director and lead actor, or also given a second actor; guess a player's move given an opponent's move, and maybe the player's last move; ...
Modeling choices are infinite!

[Lena Mangold @lenafmangold]

Excited for today's launch of our new Computational Social Science seminar series at @CentreMarcBloch, and to hear @jvelcin and @GaelPoux speak about “Different ways for modeling time with textual data”.