pith. sign in

arxiv: 2605.26819 · v1 · pith:Z25CMOF2new · submitted 2026-05-26 · 💻 cs.IR · cs.AI

RAGEAR: Retrieval-Augmented Graph-Enhanced Academic Recommender

Francesco Granata , Lorenzo Lamazzi , Misael Mongiov\`i , Francesco Poggi , Valeria Secchini This is my paper

Pith reviewed 2026-06-29 15:57 UTC · model grok-4.3

classification 💻 cs.IR cs.AI
keywords course recommendationknowledge graphdense retrievalacademic recommendertranscript chunksranking aggregationcurricular constraintsneurosymbolic system
0
0 comments X

The pith

RAGEAR aggregates transcript chunks via a knowledge graph to improve course recommendation rankings over simple sum baselines.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents RAGEAR as a system that retrieves semantically relevant chunks from lecture transcripts and feeds them into a knowledge graph that models courses, lessons, credits, study plans, and prerequisites. The core step is a graph-aware aggregation function that turns chunk-level similarity scores into course-level rankings by factoring in the share of evidence per course, the strength of chunk ranks, and how evidence spreads across lessons. If this holds, recommendations would better respect both what is actually taught in the transcripts and the formal curricular rules that metadata-only systems often miss. A sympathetic reader would care because mismatched course suggestions waste student time and risk violating degree requirements.

Core claim

RAGEAR's graph-aware aggregation function propagates chunk-level evidence to course-level recommendations by combining the share of retrieved similarity associated with a course, the rank-based strength of its relevant chunks, and the distribution of evidence across lessons; this produces higher ranking quality than a transcript-based normalized SumP baseline, especially for top-ranked items, as measured on 152 student-like queries via human and LLM assessments.

What carries the argument

Graph-aware aggregation function that scores courses from transcript chunks by combining similarity share, rank strength, and lesson distribution.

If this is right

  • Lecture transcripts yield stronger retrieval signals than metadata alone.
  • The aggregation step improves top-ranked precision compared with normalized sum-of-similarities.
  • Symbolic filtering on the graph can be applied without discarding content-based evidence.
  • Recommendations become sensitive to study-plan and credit constraints while still using fine-grained lesson content.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same chunk-to-entity aggregation pattern could be tested in other constrained recommendation settings such as treatment planning.
  • Adding longitudinal student performance data to the aggregation might further sharpen the evidence distribution term.
  • The hybrid retrieval-plus-graph design suggests that purely neural or purely symbolic academic recommenders are each missing part of the signal.

Load-bearing premise

The human and LLM evaluations on 152 queries serve as valid proxies for real student satisfaction, and the knowledge graph captures all relevant curricular constraints and prerequisites.

What would settle it

A controlled study in which actual students receive either RAGEAR or baseline recommendations, then report satisfaction or enrollment follow-through rates after one semester.

Figures

Figures reproduced from arXiv: 2605.26819 by Francesco Granata, Francesco Poggi, Lorenzo Lamazzi, Misael Mongiov\`i, Valeria Secchini.

Figure 1
Figure 1. Figure 1: High-level architecture integrating User Interface, Knowledge Graph, Dense Retrieval Module, and Recommendation Algorithm. to capture fine-grained conceptual evidence that may appear in specific lectures even when it is not mentioned in the course title, abstract, or syllabus metadata. Each lecture transcript is segmented into semantically coherent chunks, as described in Section 3. Each chunk is associate… view at source ↗
Figure 2
Figure 2. Figure 2: Graffoo diagram depicting the ontology module with information about content, courses and lectures At the instructional-content level, the ontology represents courses as struc￾tured collections of lessons and transcript chunks. As showed in the Graffoo [7] diagram in Fig.2, a Course is linked to its instructional units and transcripts, and each transcript is decomposed into Chunk instances. Chunks are asso… view at source ↗
Figure 3
Figure 3. Figure 3: Example query and user interface output. The interface displays three recom￾mended courses with metadata and supporting transcript chunks. Some information has been modified for copyright and privacy reasons. a text description of their academic interests. Once submitted, the query is trans￾mitted via an HTTP POST request to the other components. The interface then dynamically displays three recommended co… view at source ↗
read the original abstract

We present RAGEAR (Retrieval-Augmented Graph-Enhanced Academic Recommender), a neurosymbolic recommender system for academic course recommendation. RAGEAR combines dense retrieval over full lecture transcripts with a symbolic Knowledge Graph modelling courses, lessons, transcript chunks, credits, study plans, and curricular information. The Knowledge Graph supports symbolic filtering and contextualisation based on structured constraints, such as credits, academic disciplines, study plans, and prerequisites. Unlike metadata-based approaches, it exploits fine-grained instructional content by retrieving transcript chunks semantically aligned with a student's query. The main contribution is a graph-aware aggregation function that propagates chunk-level evidence to course-level recommendations. The score combines three factors: the share of retrieved similarity associated with a course, the rank-based strength of its relevant chunks, and the distribution of evidence across lessons. We evaluate RAGEAR on 152 student-like queries through a human evaluation sample and a large-scale LLM-based relevance assessment. Results show that lecture transcripts improve over metadata-only retrieval, and that RAGEAR further improves ranking quality over a transcript-based normalized SumP baseline, especially for top-ranked recommendations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The paper claims to introduce RAGEAR, a neurosymbolic academic course recommender combining dense retrieval over lecture transcripts with a symbolic knowledge graph (modeling courses, lessons, chunks, credits, study plans, and prerequisites) for filtering and contextualization. Its main technical contribution is a graph-aware aggregation function that propagates chunk-level evidence to course recommendations via three factors: share of retrieved similarity, rank-based chunk strength, and evidence distribution across lessons. Evaluation on 152 student-like queries via human sample and LLM-based relevance assessment reports that transcripts outperform metadata-only retrieval and that RAGEAR further improves ranking quality over a transcript-based normalized SumP baseline, especially at top ranks.

Significance. If the empirical improvements hold under rigorous validation, the work offers a concrete example of neurosymbolic integration in educational IR, where the KG enables structured constraints while the aggregation function provides an explicit mechanism for evidence propagation from fine-grained content. The explicit evaluation protocol on 152 queries and the parameter-free character of the aggregation (once defined) are strengths that support reproducibility and falsifiability in the domain.

minor comments (3)
  1. [Abstract] Abstract: the directional claims of improvement would be more informative if accompanied by the key numerical metrics, confidence intervals, or statistical test results that appear in the evaluation section.
  2. [Evaluation] Evaluation section: clarify the exact construction of the 152 queries and the normalization procedure applied to the SumP baseline so that the reported gains can be directly reproduced.
  3. [Results] Figure or table presenting the ranking results: ensure error bars or significance markers are included to allow readers to assess whether the top-rank improvements are statistically distinguishable from the baseline.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. No specific major comments appear in the provided report, so we have no individual points requiring detailed rebuttal or revision at this stage. We remain available to address any minor suggestions that may be supplied separately.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper describes an empirical neurosymbolic recommender with a graph-aware aggregation function evaluated on 152 queries against a normalized SumP baseline. No equations, derivations, or first-principles predictions are presented that could reduce to inputs by construction. The central claims rest on explicit experimental comparisons rather than self-referential fitting or self-citation chains. This is the expected non-finding for an applied systems paper whose contributions are architectural and evaluative.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review prevents enumeration of free parameters, axioms, or invented entities; the central claim rests on unstated assumptions about query representativeness, graph completeness, and evaluation validity.

pith-pipeline@v0.9.1-grok · 5736 in / 1080 out tokens · 30411 ms · 2026-06-29T15:57:02.262044+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 16 canonical work pages · 1 internal anchor

  1. [1]

    Machine Learning and Knowledge Extraction5(2), 560–596 (2023)

    Algarni, S., Sheldon, F.: Systematic review of recommendation systems for course selection. Machine Learning and Knowledge Extraction5(2), 560–596 (2023). https://doi.org/10.3390/make5020033

    work page doi:10.3390/make5020033 2023

  2. [2]

    In: Hitzler, P., Gangemi, A., Janowicz, K., Krisnadhi, A., Presutti, V

    Blomqvist, E., Hammar, K., Presutti, V.: Engineering Ontologies with Patterns - The eXtreme Design Methodology. In: Hitzler, P., Gangemi, A., Janowicz, K., Krisnadhi, A., Presutti, V. (eds.) Ontology Engineering with Ontology Design Patterns, Studies on the Semantic Web, vol. 25, pp. 23–50. IOS Press (2016). https://doi.org/10.3233/978-1-61499-676-7-23

    work page doi:10.3233/978-1-61499-676-7-23 2016

  3. [3]

    In: Journal of Web Semantics (2005).https://doi.org/10.1016/j.websem.2005.09.001

    Carroll, J.J., Bizer, C., Hayes, P., Stickler, P.: Named graphs. In: Journal of Web Semantics (2005).https://doi.org/10.1016/j.websem.2005.09.001

    work page doi:10.1016/j.websem.2005.09.001 2005

  4. [4]

    Dai, Z., Callan, J.: Deeper text understanding for ir with contextual neural lan- guagemodeling.p.985–988(2019).https://doi.org/10.1145/3331184.3331303

    work page doi:10.1145/3331184.3331303 2019

  5. [5]

    Knowledge-Based Systems194, 105385 (2020).https://doi.org/10

    Esteban, A., Zafra, A., Romero, C.: Helping university students to choose elective courses by using a hybrid multi-criteria recommendation system with genetic op- timization. Knowledge-Based Systems194, 105385 (2020).https://doi.org/10. 1016/j.knosys.2019.105385

    work page arXiv 2020

  6. [6]

    Explosion AI: spaCy: Industrial-Strength Natural Language Processing in Python (2023),https://spacy.io, version 3.7.2

    2023

  7. [7]

    In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

    Falco, R., Gangemi, A., Peroni, S., Shotton, D., Vitali, F.: Modelling owl ontologies with graffoo. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 8798, p. 320 – 325 (2014).https://doi.org/10.1007/978-3-319-11955-7_42

    work page doi:10.1007/978-3-319-11955-7_42 2014

  8. [8]

    IEEE Access8, 189069– 189088 (2020).https://doi.org/10.1109/ACCESS.2020.3031572

    Fernández-García, A.J., Rodríguez-Echeverría, R., Preciado, J.C., Manzano, J.M.C., Sánchez-Figueroa, F.: Creating a recommender system to support higher education students in the subject enrollment decision. IEEE Access8, 189069– 189088 (2020).https://doi.org/10.1109/ACCESS.2020.3031572

    work page doi:10.1109/access.2020.3031572 2020

  9. [9]

    In: The Second Text REtrieval Conference (TREC-2)

    Fox, E.A., Shaw, J.A.: Combination of multiple searches. In: The Second Text REtrieval Conference (TREC-2). pp. 243–252. NIST (1994)

    1994

  10. [10]

    Knowledge and Information Sys- tems68(1) (2026).https://doi.org/10.1007/s10115-025-02631-y

    Gangemi, A., Graciotti, A., Meloni, A., Nuzzolese, A.G., Presutti, V., Refor- giato Recupero, D., Russo, A.: Text2amr2fred, converting text into rdf/owl knowl- edge graphs via abstract meaning representation. Knowledge and Information Sys- tems68(1) (2026).https://doi.org/10.1007/s10115-025-02631-y

    work page doi:10.1007/s10115-025-02631-y 2026

  11. [11]

    In: 2013 IEEE 13th International Conference on Ad- vanced Learning Technologies

    Hu, L., Du, Z., Tong, Q., Liu, Y.: Context-aware recommendation of learning resources using rules engine. In: 2013 IEEE 13th International Conference on Ad- vanced Learning Technologies. pp. 181–183 (2013).https://doi.org/10.1109/ ICALT.2013.56

    2013

  12. [12]

    https://doi.org/10.1109/ACCESS.2018.2889635

    Ibrahim, M.E., Yang, Y., Ndzi, D.L., Yang, G., Al-Maliki, M.: Ontology-based personalizedcourserecommendationframework.IEEEAccess7,5180–5199(2019). https://doi.org/10.1109/ACCESS.2018.2889635

    work page doi:10.1109/access.2018.2889635 2019

  13. [13]

    Journal of Web Semantics85, 100857 (2025).https://doi.org/ 10.1016/j.websem.2024.100857

    Jaldi, C.D., Ilkou, E., Schroeder, N., Shimizu, C.: Education in the era of neu- rosymbolic ai. Journal of Web Semantics85, 100857 (2025).https://doi.org/ 10.1016/j.websem.2024.100857

    work page doi:10.1016/j.websem.2024.100857 2025

  14. [14]

    In: Proceedings of the International Conference on Educational Data Mining

    Li,Z.,Wang,M.,Zhao,L.:Coursekg:Aknowledge-graphapproachforpersonalised course and learning-path recommendation. In: Proceedings of the International Conference on Educational Data Mining. pp. 512–524. EDM (2024).https:// doi.org/10.3390/app14072710 RAGEAR: Retrieval-Augmented Graph-Enhanced Academic Recommender 17

    work page doi:10.3390/app14072710 2024

  15. [15]

    In: 2022 9th International Con- ference on Computing for Sustainable Global Development (INDIACom)

    Malhotra, I., Chandra, P., Lavanya, R.: Course recommendation using domain- based cluster knowledge and matrix factorization. In: 2022 9th International Con- ference on Computing for Sustainable Global Development (INDIACom). pp. 12–

    2022

  16. [16]

    IEEE (2022).https://doi.org/10.23919/INDIACom54597.2022.9763281

    work page doi:10.23919/indiacom54597.2022.9763281 2022

  17. [17]

    In: Nguyen, N.T., Iliadis, L., Ma- glogiannis, I., Trawiński, B

    Nguyen, T.T.M., Tran, T.P.Q.: A knowledge graph embedding based approach for learning path recommendation for career goals. In: Nguyen, N.T., Iliadis, L., Ma- glogiannis, I., Trawiński, B. (eds.) Computational Collective Intelligence. pp. 66–

  18. [18]

    Springer International Publishing, Cham (2021).https://doi.org/10.1007/ 978-3-030-88081-1_6

    2021

  19. [19]

    In: Proceedings of the International Conference on Ad- vanced Learning Technologies

    Obeid, N., Lahoud, C., El Khoury, R.: An ontology-based recommender system for higher education. In: Proceedings of the International Conference on Ad- vanced Learning Technologies. pp. 45–49. IEEE (2018).https://doi.org/10. 1145/3184558.3191533

    work page arXiv 2018

  20. [20]

    In: International conference on machine learning

    Radford,A.,Kim,J.W.,Xu,T.,Brockman,G.,McLeavey,C.,Sutskever,I.:Robust speech recognition via large-scale weak supervision. In: International conference on machine learning. pp. 28492–28518. PMLR (2023).https://doi.org/10.5555/ 3618408.3619590

    work page arXiv 2023

  21. [21]

    In: 2020 IEEE International Conference on Computing, Power and Communication Technologies (GUCON)

    Sankhe, V., Shah, J., Paranjape, T., Shankarmani, R.: Skill based course rec- ommendation system. In: 2020 IEEE International Conference on Computing, Power and Communication Technologies (GUCON). pp. 573–576 (2020).https: //doi.org/10.1109/GUCON48875.2020.9231074

    work page doi:10.1109/gucon48875.2020.9231074 2020

  22. [22]

    Multilingual E5 Text Embeddings: A Technical Report

    Wang, L., Yang, N., Huang, X., Yang, L., Majumder, R., Wei, F.: Multilingual e5 text embeddings: A technical report. arXiv preprint arXiv:2402.05672 (2024). https://doi.org/10.48550/arXiv.2402.05672

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2402.05672 2024

  23. [23]

    Lecture Notes in Com- puter Science12657 LNCS, 150 – 163 (2021).https://doi.org/10.1007/ 978-3-030-72240-1_11

    Zhang, X., Yates, A., Lin, J.: Comparing score aggregation approaches for document retrieval with pretrained transformers. Lecture Notes in Com- puter Science12657 LNCS, 150 – 163 (2021).https://doi.org/10.1007/ 978-3-030-72240-1_11

    2021