Pith. sign in

REVIEW 3 major objections 5 minor 5 references

A Simple Recommender Engine for Matching Final-Year Project Student with Supervisor

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read A minimal recommender engine built on Euclidean distance can rank final-year project supervisors from a student's ratings of five interest areas.

desk verdict A textbook recommender applied to a small admin task, undercut by a mislabeled example table that inverts the ranking. read the letter →

arxiv 1908.03475 v1 pith:I5KFYCP2 submitted 2019-08-08 cs.IR

classification cs.IR
keywords recommenderenginesystemEuclideandistancefinalyearprojectsupervisormatchinginterestratingsimilarityscoreproofofconcept
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a very simple recommender engine, built on Euclidean distance, can help a final-year project student find a suitable supervisor. The student enters ratings for five interest areas, and the engine compares that query with a stored five-dimensional profile for each lecturer, returning a ranked list of the best matches. The authors implemented the engine in Java, populated lecturer profiles from previous project titles and the supervisors students actually chose, and tested it on 51 returned questionnaires from IT and computer science students. Their conclusion is deliberately modest: Euclidean distance 'seems to achieve the goal' of recommending lecturers, and the work is a feasible proof-of-concept rather than a finished system. A sympathetic reader would take the contribution to be the demonstration that even the simplest distance-based ranking can make the supervisor-selection problem tractable.

What carries the argument

The machinery is the Euclidean distance score and its conversion to a bounded similarity value. Each Lecturer object is a vector $(p_1,\dots,p_5)$ over the five areas—Multimedia, Web Application, Network, Artificial Intelligence, and Mobile Application—and the student's questionnaire ratings form the query vector $(q_1,\dots,q_5)$. The engine computes $d(\mathbf{p},\mathbf{q}) = \sqrt{\sum_{i=1}^5 (p_i - q_i)^2}$, applies the Figure 2 conversion to map the distance into a similarity score between 0.0 and 1.0, and sorts descending to produce the recommendation. The work this machinery does is to convert an open-ended human decision—'who should supervise my project?'—into a fixed geometric ranking whose correctness depends entirely on how faithfully the five stored numbers capture each lecturer's actual expertise.

What would settle it

Have two independent coders build the five-area rating vector for each lecturer from the same project-title list, then run one fixed student query; if the top three recommended supervisors differ between the two codings, the ranking is an artifact of the CSV encoding rather than a stable measure of lecturer expertise.

Watch

Extended reading notes

Core claim

The paper's central claim is that a supervisor can be represented by a five-dimensional vector of interest-area ratings and a student by a matching query vector, and that ranking lecturers by Euclidean distance, converted into a 0-to-1 similarity score, returns a list of the lecturers who best fit the student's stated interests. The authors demonstrate this with sample output in which a student with strong Multimedia and Web Application interests receives a top-five list led by lecturers whose profiles sit closest in that five-dimensional space. They further claim that this result is enough to show the approach is feasible for the final-year-project matching problem, while acknowledging that the engine is far from complete and cannot reliably tell whether a group of lecturers has consistently similar interests.

Load-bearing premise

The load-bearing premise is that a lecturer's expertise can be captured accurately and stably by five fixed ratings derived from past project titles and students' supervisor choices; if that profile is wrong or unstable, the distance-based ranking is meaningless even though the software runs.

Editorial extensions

If this is right

  • Other institutions with a similar project-allocation workflow could deploy the same engine with only a CSV of supervisor profiles and no training data.
  • A student who spreads ratings evenly near the middle will receive a less discriminating list than one with strong preferences, because Euclidean distance compresses small differences across all dimensions into a short total distance.
  • Because similarity is computed from raw distance, the engine cannot by itself tell whether two lecturers' interest patterns are consistently similar, which is why the authors flag Pearson correlation as future work.
  • The five fixed categories define the ceiling of what the engine can express; adding more areas, as the authors plan, would change the geometry and therefore the rankings.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An unstated consequence of choosing Euclidean distance is that two lecturers with the same average profile but opposite preference shapes—one strong in Multimedia and weak in Network, the other the reverse—are treated as very different, so the ranking rewards matching a student's overall interest level rather than their preference pattern.
  • A natural extension the paper does not explore is to run the same engine in reverse: using lecturer profiles as queries would cluster students by interest, which could help supervisors form project groups or spot emerging interest areas before project allocation.
  • The engine's transferability is untested: the five areas were induced from one faculty's project titles, so applying the same code elsewhere without re-deriving the area set would silently compress any missing specialty into the nearest existing category and could produce confidently wrong recommendations.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This manuscript describes a Java-based recommender engine intended to match final-year project students with potential supervisors at UiTM Perlis. The engine takes a student's ratings in five interest areas (Multimedia, Web Application, Network, Artificial Intelligence, Mobile Application) and computes Euclidean distances to lecturer profiles stored in CSV format, ranking supervisors by a similarity score derived from the distance. The authors report a single illustrative test with five supervisors and conclude that the approach 'seems to achieve the goal' while acknowledging the engine is only a proof-of-concept and far from complete.

Significance. If the demonstration were valid, the paper would provide a simple, transparent proof-of-concept for a content-based recommender in an academic advising context. The work has some praiseworthy elements: the algorithm is clearly stated (Euclidean distance with 1/(1+d) similarity conversion), the implementation is described as a reusable Java class, and the authors explicitly temper their claims by calling the system a proof-of-concept. However, the sole reported result is internally inconsistent with the stated formula, and the paper lacks the data and procedural detail needed to reproduce or evaluate the ranking. As it stands, the central feasibility claim is unsupported, and the manuscript is not publishable in its present form.

major comments (3)
  1. [Recommender Engine, Table 1] Table 1 reports 'Similarity' values of 44.95, 37.62, 34.83, 32.04, and 29.21, but the conversion formula in Figure 2 (similarity = 1/(1+distance)) is bounded between 0.0 and 1.0. The reported values are therefore impossible as similarity scores. If they are actually raw Euclidean distances, the column is mislabeled and the ranking is inverted, since larger distance means less similar; the top-listed lecturer would then be the least suitable match. This inconsistency invalidates the only demonstration of the paper's central claim and makes the 'best match' ranking uninterpretable.
  2. [Data Modeling, Figure 4] The Data Modeling section states that lecturer profiles are transformed from previous project titles and chosen-supervisor data into a CSV of interest values, but it never specifies how the numeric value for each of the five interest dimensions is computed from those raw data. Without this construction rule, the Euclidean distances and the resulting recommendation list cannot be reproduced or checked, so the feasibility of the approach is not established.
  3. [Conclusion] The conclusion asserts that 'Euclidean distance seems to achieve the goal for recommending lecturers,' but the only supporting evidence is a single unvalidated output (Table 1/Figure 5). There is no ground-truth comparison, no baseline, no error analysis, and no measure of whether the recommended supervisors are actually suitable. The authors' own caveat that the engine is 'far from complete' is appropriate, but it does not compensate for the absence of any valid test of the ranking; the assertion therefore overstates what the experiments show.
minor comments (5)
  1. [Introduction] The Introduction says the project titles 'falls within four general areas' and then lists five: Multimedia, Web Application, Network, Artificial Intelligence, and Mobile Application. Please correct the count.
  2. [Figure 1 and throughout] The spelling 'Euclidian' appears in the Figure 1 caption; elsewhere the manuscript uses 'Euclidean'. Please unify.
  3. [References] Reference [5] contains a typo: 'Eidenburgh' should be 'Edinburgh'.
  4. [Figure 5] Figure 5 is described as sample raw output, but the typeset image is not legible; please provide a readable screenshot or reproduce the output in a table.
  5. [Data Modeling] Please clarify how missing or partial lecturer data (e.g., a new lecturer with no previous project titles) is handled in the vector construction.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the recommender ranking is a direct Euclidean-distance computation with no fitted parameters or load-bearing self-citations.

full rationale

The derivation chain is explicit and self-contained: the student supplies interest ratings, each lecturer is represented by a vector built from questionnaire/CSV data, the engine computes Euclidean distance and converts it to a similarity score via the stated 1/(1+d) formula, and the list is sorted by that score. No parameter is fitted to the predicted output, no lecturer profile is defined in terms of the recommended ranking, and no prior work by the same authors is cited to justify the algorithm or force the choice of formula. The reported Table 1 values (44.95, 37.62, etc.) are inconsistent with the paper's own bounded similarity formula, and the paper does not disclose exactly how the lecturer vector numeric values are derived from the questionnaire responses; however, these are correctness and reproducibility concerns, not circularity. The central claim is only that a proof-of-concept Euclidean-distance recommender is feasible, which is exactly what the stated computation would show if the data were internally consistent. Because nothing in the derivation reduces by construction to its own inputs, the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the suitability of five hand-picked interest categories, the reliability of self-reported ratings, the assumption that lecturer profiles derived from past project titles and student choices capture expertise, and the choice of Euclidean distance as the matching metric. None of these are validated against independent data. There are no numeric free parameters fitted to data, but the modeling choices are assumptions rather than derived facts.

assumptions (4)
  • standard math The Euclidean distance formula (Figure 1) and the distance-to-similarity conversion (Figure 2) correctly compute dissimilarity and bounded similarity for numeric vectors.
    This is a standard mathematical fact; the paper invokes it in Background and does not need to prove it.
  • domain assumption Lecturer expertise can be represented as a fixed vector over the five chosen interest areas, built from previous project titles and student choices.
    Data Modeling describes the Lecturer object and CSV rows, but provides no evidence that this representation is complete or consistent enough for meaningful matching.
  • domain assumption Students' 1-5 self-ratings of interest are reliable proxies for their actual project interests.
    The questionnaire is described in Methodology, but no reliability or validity test is reported.
  • domain assumption Euclidean distance is an appropriate similarity measure for this particular matching task.
    The paper selects the algorithm from the recommender systems literature without comparing to alternatives such as Pearson correlation, which the conclusion itself mentions as future work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Simple Recommender Engine for Matching Final-Year Project Student with Supervisor." pith.science (2026). https://pith.science/paper/I5KFYCP2

@misc{pith2026190803475,
  author       = {Pith},
  title        = {Pith review of: A Simple Recommender Engine for Matching Final-Year Project Student with Supervisor},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I5KFYCP2}},
  note         = {Machine review of arXiv:1908.03475}
}
read the original abstract

This paper discusses a simple recommender engine, which can match final year project student based on their interests with potential supervisors. The recommender engine is constructed based on Euclidean distance algorithm. The initial input data for the recommender system is obtained by distributing questionnaire to final year students and recording their response in CSV format. The recommender engine is implemented using Java class and application, and result of the initial tests has shown promises that the project is feasible to be pursued as it has the potential of solving the problem of final year students in finding their potential supervisors.

Figures

Figures reproduced from arXiv: 1908.03475 by the authors.

Figure 1
Figure 1. , where p is data points, d is the distance function, Sp is the component value of the data points [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Euclidean distance to similarity score conversion formula [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Recommender engine for matching supervisor methodology [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Sample CSV data Recommender Engine The Java class is then implemented as a Java application to create a rudimentary recommender engine, which accepts rating of each interest area (Multimedia, Web application, Network, Artificial Intelligence and Mobile Application) and…
Figure 5
Figure 5. Figure 5: Sample Raw output from Java application [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages

  1. [1]

    M. L. Yap, N. Abdul Manaf, W. N. A. Wan Hariri, A. Ahmad, N. Janom, S. Abdul Rahman, A. S. Abdullah, and N. S. Kamal Bashah, Essentials of Computing Sciences - Project Administration, 1st ed. Shah Alam, Selangor: Faculty of Computer and Mathematical Sciences

  2. [2]

    Ricci, L

    F. Ricci, L. Rokach, B. Shapira, and P. B. Kantor, Recommender Systems Handbook. Springer, 2010

  3. [3]

    A new distance between two bodies of evidence,

    A.-L. Jousselme, D. Grenier, and É. Bossé, “A new distance between two bodies of evidence,” Inf. Fusion, vol. 2, no. 2, pp. 91–101, 2001

  4. [4]

    An efficient euclidean distance transform,

    D. G. Bailey, “An efficient euclidean distance transform,” in Combinatorial Image Analysis, Springer, 2005, pp. 394–408

  5. [5]

    Similarity and recommender systems

    H. Shimodaira, “Similarity and recommender systems.” School of Informatics, The University of Eidenburgh, 21-Jan-2014

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.