Exploring the Invariance of Analogical Reasoning Across Different Knowledge Graph Embedding Models

Analogical reasoning, expressed with analogical quadruples of the form “a is to b as c is to d” (e.g. Paris is to France as Berlin is to Germany), is a natural way for human beings to reason about new situations based on the knowledge gained from experiencing similar situations. Its insights have been proven in various human cognitive tasks, such as natural language learning or problem-solving, as well as recently in Machine learning through analogy-based classifiers (Lim et al., 2019) and retrievers (Marquer et al., 2025).

The past work of (Jarnac and al., 2023) has demonstrated that analogy-based classifiers can be applied to knowledge graph (KG) management tasks, showing great results for domain-specific KG bootstrapping, and paving the road for testing this analogy-based classifier on other KG management tasks.

However, to make the use of an analogy-based classifier on a KG, it is necessary to first compute numerical representations of its components (entities and relations). This necessity has led to the use of knowledge graph embeddings, which model graph elements into continuous vector spaces. Given the diversity of embedding approaches belonging to different families (Translational, Neural Networks, Tensor Decomposition …) and for each specific assumption and representation properties (Ji et al., 2020, Ali et al., 2021), it is interesting to address the impact of the choice of an embedding model on the performance of an analogy-based model.

This internship aims to address some of the following questions:

• What are the consequences of using different KG embedding models as input to an analogy-based classifier?
• Do analogical proportions built with KG entities remain invariant under different representation spaces of this KG?
• Is the choice of the methodology used to select analogy left pairs invariant through different KG representations?

Internship plan:

1. Understanding key concepts of KG and analogical reasoning through a literature review.
2. Training several embedding models on a dedicated knowledge graph.
3. Applying and evaluating the proposed analogy-based classifier using different embeddings.
4. Exploring selection strategies for analogy left pairs at inference and comparing their robustness with different representations.

This work will give a better understanding on the variability introduced by the choice of a specific KG embedding model, especially when applying analogy-based classifiers.

You are studying in Master Year 2 / final year of engineering school, with a specialty in computer science or applied mathematics. You are proficient in:

• Python programming
• Machine Learning / Deep Learning, especially with frameworks like PyTorch or Tensorflow
• Knowledge of the Semantic Web (RDF, RDFS, OWL, SPARQL, knowledge graphs and ontologies) would be appreciated.
• Ability to read and write in English

You are curious, eager to learn, face challenges, experiment and discover by yourself.

• Subsidized meals
• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage