Josiane Mothe

Doctoral supervision: Supervision of 21 theses, of which 20 defended (1 per year on average).

In information retrieval (IR), I am particularly interested in texts. The aim of Information Retrieval systems is to retrieve documents that answer to a user's query. Three PhD students I supervised tackled the research challenge of studying the specific challenge of the meaning carried out by texts (documents) and queries. B. Ralalason (2006-2010, now a university research engineer) and N. Hernandez (2002-2005, senior lecturer, University of Toulouse) have focused on the use of ontologies in IR. L. Ermakova (2012-2016, now lecturer at Univ. Brest) deals with contextualisation of short texts which aims to help the understanding of texts such as tweets for instance. The HDR of G. Hubert (2010, Associate Professor, University of Toulouse) deals with contextual IR.
Recommendation systems are another facet of IR: they recommend documents (texts, items) to users while they browse information. In this context, D. Dudognon proposed in his thesis (2010-2014, research engineer at OverBlog) a model allowing the diversification of recommendations. This model has been implemented in the engine used by the European leader in blogging platforms (OverBlog). J. Louëdec's thesis (2013-2016, now engineer at CDiscount) focuses on incremental learning models in recommendation systems in order to better take into account the dynamics (new documents, new users).
Queries are varied and search techniques are not adapted to handle all queries effectively. I am working on models to adapt the query processing, in particular using predictive models and developping query performance predictors. A. Benammar (1999-2003, associate professor at the University of Sfax, Tunisia) proposed a query reformulation model adapted to the user's profile. Y. Champclaux (2005-2009) proposed a search model based on graphs. D. Kompaoré (2004-2008, independent consultant), J. Randriamparany (2008-2012, university assistant, Madagascar) proposed adaptive models based on the analysis of the influence of search engine parameters. A. Chifu (2011-2015, assistant professor at the University of Marseille, France) defined predictors of query difficulty, which can then be used to guide the specific treatments to be applied. This work has been continued as part of the post-doc of Md Zia Ullah (2018-2022, lecturer at the Univeristy of Edinburgh, Scotland).
Exploration and machine learning methods must be able to adapt to the complexity of the data. The PhD theses of Léa Laporte (2010-2013, Senior Lecturer, INSA Lyon) and A. Bigot (2011-2014) benefited from a math-info co-supervision to propose parsimonious classification methods and analysis scenarios for the adaptation of IR methods. Machine learning is also a core aspect of the PhD theses in collaboration with Airbus to detect anomalies either in the analysis connection logs (O. Coustié, 2018-2021, now engineer at Airbus), or in time series from sensor collections (C. Lejeune, 2018-2021). The detection of abnormal behaviour in blogs can allow the detection of signs of depression (F. Ramiandrisoa, 2017-2020) or the analysis of sensor data can help to predict turbulences during flights (T. Li, 2020-in progress).
Image analysis for cancer diagnosis is based on the use of deep neural networks and their optimal configuration in the PhD theses of S. Mejbri (2016-2019) and Ismat Ara Reshma (2018-2021) in collaboration with the Oncopole de Toulouse. Combining texts and images is used to automatically detect deforestation in N. Neptune's PhD thesis (2017-2022, ATER at the Univesity of Toulouse, France). But also for the detection and prevention of collisions in flight (M. S. Rahman, 2019-2022, engineer at Sogeti, Toulouse, France).

Master's theses: Supervision of 2 students per year on average.

Among the many Master students I have supervised, I would like to underline the attractiveness of the research I carry out by the application of scholarship students:

1 Italian student under the Erasmus Mundus scholarships (5 months, 2017).
1 Indian student in the framework of the Chapka scholarships (2 months, 2014-2015) who developed a tool to compare the dissemination of information on online newspapers and social networks.
1 Italian student under the Erasmus Mundus scholarships (7 months, 2014-2015).
1 Indian student under the Chapka scholarships (2 months, 2013-2014) who developed a graphical NoSQL database for connection log analysis.

IR/IC: International peer Reviewed/Conference ; NR/NC: National peer Reviewed/Conference; O: Other

Md Siddiqur Rahman (2019-2022):
Deep learning for conflict resolution between aircraft en route - two complementary approaches.
Supervision ratio: 70%.
Publications: 4 IC, 2 IR.
=> Engineer at Sogeti Toulouse.

Oihana Coustié (2018-2021):
Anomaly detection model in application connection logs.
Supervision ratio: 50%.
Publications: 2 IC, 2 NC.
=> Engineer at Airbus.

Clément Lejeune (2018-2021):
Functional anomaly detection model.
Supervision ratio: 50%.
Publications: 1 IR, 1 NC.

Ismat Ara Reshma (2018-2021):
CNN image segmentation model for cancer detection.
Supervision ratio: 50%.
Publications: 2 IC, 3 IR, 1 NC.

Nathalie Neptune (2017-2022):
Monitoring deforestation from satellite images.
Supervision ratio: 70%.
Publications: 1 NR, 4 NC.
=> ATER at University of Toulouse.

Faneva Ramiandrisoa (2017-2020):
Detection of depressed people from social networks.
Supervision ratio: 100%.
Publications: 3 IC, 7 NC.

Sonia Mejbri (2016-2019):
Deep neural networks for image-based cancer detection.
Supervision ratio: 50%.
Publications: 2 IC.

Thi Bich Ngoc Hoang (2015-2018):
Information dissemination, information and knowledge extraction in social networks.
Supervision ratio: 100%.
Publications: 3 IR, 3 IC, 3 NC.

Jonathan Louëdec (2013-2016):
Recommendation systems and incremental learning.
Supervision ratio: 33%.
Publications: 2 NR, 1 IC, 4 NC.
=> Engineer at CDiscount.

Liana Ermakova (2012-2016):
Contextualisation of short texts.
Supervision ratio: 100%.
Publications: 1 IR, 1 NR, 14 IC, 2 NC.
=> Lecturer at University of Brest.

Adrian Chifu (2011-2015):
Disambiguation of queries in an information retrieval system.
Supervision ratio: 100%.
Publications: 3 IR, 4 IC, 7 NC.
=> Assistant professor at University of Marseille.

Anthony Bigot (2011-2014):
Adapting search engines to queries.
Supervision ratio: 50%.
Publications: 2 IR, 1 NR, 2 NC, 3 O.

Damien Dudognon (2010-2014):
Recommendation systems.
Supervision ratio: 50%.
Publications: 1 IR, 4 IC, 2 NC, 3 O.
=> Research engineer at OverBlog.

Léa Laporte (2010-2013):
Document Scheduling in IR and Machine Learning.
Supervision ratio: 50%.
Publications: 2 IR, 2 NR, 3 IC, 7 NC, 3 O.
=> Senior lecturer at INSA Lyon.

Joelson Randriamparany (2008-2012):
Contextual information retrieval: adapting search engines.
Supervision ratio: 100%.
Publications: 1 IC, 2 NC.
=> University assisant in Madagascar.

Bachelin Ralalason (2006-2010):
Using ontologies for context representation in information retrieval.
Supervision ratio: 50 %.
Publications: 2 IR, 1 NR, 4 IC, 2 IN.
=> University research engineer.

Yael Champclaux (2005-2009):
Definition of a graph-based model in information retrieval.
Supervision ratio: 30%.
Publications: 2 IC, 2 NC.

Désiré Kompaoré (2004-2008):
Contextual information retrieval: difficult queries and system fusion.
Supervision ratio: 100%.
Publications: 1 NR, 4 IC, 5 NC, 1 O.
=> Independant consultant.

Nathalie Hernandez (2002-2005):
Using ontology-based domain representation for information retrieval.
Publications: 1 IR, 3 NR, 2 O, 4 IC, 3 NC.
=> Senior lecturer at University of Toulouse.

Anis Ben Ammar (1999-2003):
Combining information mining and information retrieval techniques for query reformulation and user profiling.
Supervision ratio: 80%.
Publications: 1 IR,3 IC, 2 NC.
=> Associate professor at University of Sfax.