Analyse comparative de modèles statistique et d’apprentissage profond pour l’étude du parcours utilisateur sur un site internet

Abstract : A website’s popularity can be evaluated using precise indicators such as the number of visits per day and the number of viewed webpages per visit. However, sytems that are used nowadays are mostly based on cookies, which can be blocked by users. We asked ourselves whether or not these analyses gave an accurate overview of a website’s actual traffic. A study of data which is automatically saved onto a server (log apache data) was carried out. During the study, other challenges arose and were dealt with. The work presented here consists of three keys steps. Firstly, a filter was applied to the data so as to store clean data in a database. Then, a dashboard was put in place to give marketing teams and management access to information regarding user habits. A comparison between this dashboard and a classic cookie based dashboard was done. Finally, another comparison was done, this time, between a classic statistics model and a deep-learning LSTM model, to see if they could similarly predict how a page can interest users based on their navigation data. It turned out that both models had a very similar accuracy, roughly 30 %. The study can be reproduced by any website team with server data.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [21 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-01961614
Contributor : Agrocampus Ouest <>
Submitted on : Thursday, December 20, 2018 - 9:32:12 AM
Last modification on : Tuesday, February 5, 2019 - 1:18:44 AM
Long-term archiving on : Friday, March 22, 2019 - 12:26:49 PM

File

2018_GORIEU_Mathilde_Data_Scie...
Files produced by the author(s)

Identifiers

  • HAL Id : dumas-01961614, version 1

Collections

Citation

Mathilde Gorieu. Analyse comparative de modèles statistique et d’apprentissage profond pour l’étude du parcours utilisateur sur un site internet. Sciences du Vivant [q-bio]. 2018. ⟨dumas-01961614⟩

Share

Metrics

Record views

44

Files downloads

60