Table of Contents
Franck Michel
Université Côte d’Azur, CNRS, Inria.
Member of I3S laboratory, and Inria's Wimmics team.
#KnowledgeGraph #RDF #GraphRAG #OpenScience #OpenData #FAIR #AttentionEconomy #Biodiversity
Research activities
Knowledge Graphs, Integration of Heterogeneous Data
My research deals with the integration of heterogeneous data based on a knowledge engineering approach and the construction of Knowledge Graphs. The goal is to overcome data structural and semantic heterogeneity in order to reconcile large, distributed data sets, and to publish them on the web.
With the arisal of Large Language Models (LLM), I’m involved in new questions such as: How to “speak to the data” by translating natural language questions into SPARQL queries (text-to-SPARQL)? How to enrich existing knowledge bases by extracting knowledge graphs from text?
Open Science
Relatedly, I actively participate in promoting the adoption of the Open Science practices, meant to spread the methods, results and products of scientific research, through the open access publishing, open data and the FAIR principles, and open source software.
Regulation of the Attention Economy
I’m conducting a research about regulating the Attention Economy, that is, the generalization of attention-capturing techniques by online platforms. Leveraging addictive design, cognitive biases and emotions, these platforms yield multiple detrimental side effects on human societies, such as causing public health issues, polarizing opinions, spreading false information, and threatening economies and democracies.
Research projects and communities
Here are some projects I am or was involved in:
- MetaboLinkAI (ANR): transforming metabolomics data into actionable insights through AI-powered, knowledge graph-driven solutions
- ISSA (Collex-Persée): Semantic Indexing of a Scientific archive and Associated Services
- D2KAB (ANR): From Data to Knowledge in Agronomy, Agriculture and Biodiversity
- DeKaloG (ANR): Decentralized Knowledge Graphs
- TAXREF-LD: the French Linked Data Taxonomic Registry
Here are the community projects I'm currently involved in:
- COST GOBLIN: COST action Global Open Knowledge Base Linking and Integration
- COST DKG: COST action Distributed Knowledge Graphs
- Bioschemas: Schema.org for Life Sciences
Publications
Contact
Address:
Université Côte d’Azur, CNRS, Inria - I3S, UMR 7271
930 route des Colles - Bât. Les Templiers
BP 145 - 06903 Sophia Antipolis CEDEX - France
Email: franck [dot] michel [at] inria [dot] fr
Find me on: ResearchGate, Github, LinkedIn, Twitter, SlideShare, Flickr, Instagram
Selected talks
Pay Attention: A Call to Regulate the Attention Market & Prevent Algorithmic Emotional Governance
Recherche, exploration et bibliométrie dans une archive scientifique ouverte
Given 2024-09-10.
Open Science, reproducible research, and the citation of articles, code and data alike
Given 2024-04-04.
ISSA: Generic Knowledge Model and Visualization tools to Help Scientists Make Sense of Archive
Wimmics Monthly Seminar 2022-12-15 / ISWC 2022 resource track replay
Covid-on-the-Web: Knowledge Graph and Services to Advance COVID-19 Research
Presented at the ISWC 2020 conference, resource track.
Bioschemas: Marking up biodiversity websites for data discovery & integration
TDWG webinar series, 2021-03.
Integration of biodiversity data from web pages to knowledge graphs, a computer scientist view point
DIADE research unit seminars (http://diade.ird.fr), 2021-04-13.
Open Datasets Creation and publication
Open data is one of the three pillars of Open Science. It enables transparency, collaboration and innovation by applying the FAIR principles, ensuring data is Findable, Accessible, Interoperable, and Reusable.
Below are some of the open datasets I have participated in creating and/or publishing.
ISSA Agritrop Dataset, Semantic index of the Agritrop open scientific archive. github DOI
TAXREF-LD, Linked Data knowledge graph of the French taxonomic register. 2017-2024. github sparql article DOI
Covid-on-the-Web. Knowledge graph produced by processing the scholarly articles of the COVID-19 Open Research Dataset (CORD-19). 2020. github sparql article DOI.
WASABI RDF Knowledge Graph. An RDF representation of the WASABI corpus of songs enriched with metadata extracted from music databases on the Web, and resulting from the processing of song lyrics and from audio analysis. 2020. github sparql article DOI
WeKG-MF, Weather Knowledge Graph of Météo France Meteorological Observations. 2022. github sparql article DOI
WheatGenomicsSLKG, Wheat Genomics Scientific Literature Knowledge Graph. 2023. github sparql DOI
Wheat Observations Knowledge Graph. Soft wheat phenotype observations data including the result of observation campaigns carried out on micro-parcels located in France between 1999 and 2015. Relies on the Plant Phenotype Experiment Ontology (PPEO) and the CO_321 Wheat Crop Ontology. 2024. DOI
Number of public photos uploaded to Flickr from 2004 to 2021. This dataset reports the number of photos uploaded to Flickr every day, hour by hour (CET) from 2004 to 2021. Only public photos are considered, private photos as well as other type of material (e.g. videos) are not accounted for. DOI github
Open source Software Development
Along with Open Data, Open Source software is a key pillar of Open Science, promoting transparency, reproducibility, and collaboration by making code freely available, shareable, and adaptable to support robust and verifiable research.
Gen²KGBot – Generic Generative Knowledge Graph Robot: Yousouf Taghzouti, Franck Michel, Tao Jiang, Louis-Felix Nothias, Fabien Gandon (2025). Github
Q²Forge – End-to-end pipeline to generate (question, SPARQL query) pairs for a given Knowledge Graph. Yousouf Taghzouti, Franck Michel, Tao Jiang, Louis-Felix Nothias, Fabien Gandon (2025). Github
ISSA visualization and search web application: Franck MICHEL, Youssef Mekouar (2022). Visualization: github DOI. Backend: github DOI
ISSA Processing Pipeline: Anna Bobasheva, Franck MICHEL (2022). github DOI
WheatGenomicsSLKG visualization and search web application: Franck MICHEL, Youssef Mekouar (2022). Visualization: github DOI. Backend: github DOI
SPARQL Micro-Services: Querying Web APIs with SPARQL. Franck Michel. 2018. github
Morph-xR2RML: MongoDB-to-RDF translation and SPARQL rewriting: Franck Michel, Freddy Pryiatna. Implementation of the xR2RML mapping language for MongoDB databases. 2017. github DOI
The VO Administration and operations PORtal (VAPOR). Franck Michel, Flavien Forestier. 2014. web DOI
EGI Virtual Organisations Support Tools. Franck Michel. 2013. web DOI
NeuroLOG platform. Alban Gaignard, Franck Michel, Johan Montagnat, Javier Rojas Balderrama, Farooq Ahmad, Bacem Wali. 2008. web
Background and Position
- CNRS Research engineer (IR), Université Côte d'Azur, CNRS, Inria, I3S laboratory. Jan. 2011 until now.
- PhD in Computer Sciences at Université Côte d'Azur, March 2017. Manuscript
- Expert software engineer in IRISA team VisAGeS, May 2008 to Dec. 2010
- Expert telecom engineer, company Capgemini Telecom, Media Networks, 1999 à 2008
- Development engineer, company Nortel Networks France, 1995 à 1999
- Engineering degree in Computer Sciences, INSA de Rennes, 1995
Organizing and Program Committees
I was/am a member of the program committees for the following conferences and/or workshops:
- SEMANTiCS 2024, 20th International Conference on Semantic Systems
- ECAI 2024, European Conference on AI
- ESWC 2024, The Extended Semantic Web Conference
- The Web Conference 2024, The Extended Semantic Web Conference
- SEMANTiCS 2023, 19th International Conference on Semantic Systems
- ESWC 2022, The Extended Semantic Web Conference
- KGCW 2022, Third International Workshop on Knowledge Graph Construction
- SCG 2021, First workshop on Squaring the circle on graphs
- KGCW 2021, Second International Workshop on Knowledge Graph Construction
- ESWC 2021, The Extended Semantic Web Conference
- ICCS 2020, The International Conference on Computational Science
- IJCAI 2020, 29th International Joint Conference on Artificial Intelligence
- SEMANTiCS 2019, 15th International Conference on Semantic Systems
- Knowledge Graph Building (KBD), workshop of the Extended Semantic Web Conference 2019 (ESWC)
- Hypermedia Multi-Agent Systems (HyperAgents 2019), workshop of the Web Conference 2019
- EKAW 2018, 21th International Conference on Knowledge Engineering and Knowledge Management
- SEMANTiCS 2018, 14th International Conference on Semantic Systems
- ISWC 2018, 17th International Semantic Web Conference
- ICCS 2018, 23rd International Conference on Conceptual Structures
- WWW 2018, The Web Conference 2018
- ISWC 2017, 16th International Semantic Web Conference
- SEMANTiCS 2017, 13th International Conference on Semantic Systems
- ICCS 2016, 22nd International Conferences on Conceptual Structures
- SI&IA 2015, Systèmes d'Information et Intelligence Artificielle 2015
I was/am a member of the organizing committees for the following conferences and/or workshops:
- Ontology Alignment Evaluation Initiative 2022: ontology provider for the complex alignment and biodiversity tracks
- Ontology Alignment Evaluation Initiative 2021: ontology provider for the complex alignment and biodiversity tracks
Wild ideas
Large Language Models as the components of a conscious AI?
AIs, and LLMs in particular, are not conscious. They are reactive systems, they respond to an input by producing an output. By contrast, consciousness can be defined as the ability to form thoughts for oneself, without the need for external stimulus.
What if we fine-tuned several LLMs to collaborate together, following the model of human psyche.
- The “conscious” model would be fine-tuned to remain in the realm of values, morality, norms, and logical thinking. This is the one that would interact with the “outside” world and provide material to the “unconscious” model.
- The “unconscious” model (or “subconscious” depending on the definitions) would be fine-tuned to phrase drive, desires, regardless of any norms nor value system.
- The “preconscious” model would be fine-tuned to filter/rewrite outputs of the “unconscious” to let only acceptable outputs make their way to the “conscious”, while also providing it with material in a feed-back loop.
This way, we could imagine being able to design some sort of a conscious AI system. But this raises multiple questions: How would it be bootstrapped? Individually, each of the 3 LLMs remains a question-answering system, it does not take the initiative of producing an output. So how to start this, and once this starts, how to control the flow? If we skip the conscious model (to try and simulate sleep) and leave the unconscious model talk with itself, could it come up with dreams?
Frugality by design: less is good
Multiple tools of the daily life consume energy and/or resource even though that's not the intend of the user. These tools must be redesigned with a specific bias towards frugality.
This can be implemented as a default behavior, as a nudge etc. Examples:
- Mixer tap delivers heated water by default: In the middle position, most mixer taps mix half ambient temperature water and half heated water. Although users may not need heated water. These should be redesigned with a middle position that only delivers ambient temperature water, so that getting heated water will require a deliberate action from the user.
- Public space fountain water delivers cooled by default: Fountains in public spaces usually deliver cooled water by default although users may not want that. These should be redesigned with a default ambient temperature water, so that getting cooled water will require a deliberate action from the user.