{"id":"301c74c1-c683-40df-9e29-b23aa8c82570","arxiv_id":"1908.03475","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A proof-of-concept recommender using Euclidean distance ranks supervisors for final-year students based on a five-question interest questionnaire, but it is not evaluated with data or code.","lead":"This paper describes a small Java program that ranks potential final-year project supervisors for students by comparing interest ratings using Euclidean distance. It is a proof-of-concept from one university, without released data, code, or evaluation.","discovery_kind":"incremental","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported similarity scores in Table 1 exceed the theoretical maximum of the stated 1/(1+d) formula, so the sole demonstration is internally inconsistent and cannot support the central claim.","rationale":"I read the paper as a proof-of-concept claim that a Euclidean-distance recommender can rank supervisors for final-year students. The math of Euclidean distance and the 1/(1+d) similarity transform are standard, and the Java implementation is plausibly straightforward. The load-bearing evidence is the sample result in Table 1, which is the only demonstration that the algorithm 'works.' My concern is more specific than the reader's: even before questioning whether the lecturer profile vectors are accurate, the paper's own reported outputs violate the bounds of its similarity formula. This is an internal inconsistency, not merely a matter of missing evaluation. If the values in Table 1 are raw distances, the ranking is backwards; if they are some other transform, the paper never says what it is. Either way, the central claim rests on an uninterpretable artifact. The reader's weakest assumption about the validity of lecturer profiles is also valid and remains a real problem, but the Table 1 inconsistency is the most direct falsification of the presented evidence. Since the paper already lacks code, data, and evaluation, this additional inconsistency reinforces the REJECT verdict rather than changing it.","tokens_in":2975,"tokens_out":2427,"duration_ms":27021,"concrete_test":"Reconstruct the recommender from the paper's description: take a lecturer profile vector in the CSV format shown in Figure 4, compute the Euclidean distance from Student X's input (Multimedia=5.0, Web Application=4.5, Network=1.0, AI=2.5, Mobile Application=3.0), then apply the stated similarity conversion 1/(1+distance) and compare with Table 1. If the resulting values do not match either the reported numbers or the raw distances, Table 1 is not reproducible from the described algorithm. If they match raw distances, then the ranking order in Table 1 is opposite to a similarity-based recommendation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that Euclidean distance 'seems to achieve the goal' of recommending supervisors. The only evidence is Table 1, which lists five supervisors with 'similarity' values of 44.95, 37.62, 34.83, 32.04, and 29.21. Yet the paper defines similarity as 1/(1+distance), which is bounded between 0.0 and 1.0. The reported values are far outside this range, so they cannot be the stated similarity scores. If they are actually raw Euclidean distances, they are mislabeled, and the ranking is inverted: a larger distance means less similar, so the top-listed lecturer would be the least suitable match. The paper provides no code, no CSV data, and no explicit formula for constructing a lecturer's five-dimensional vector from project titles and student choices, making it impossible to reproduce or verify the table. The Data Modeling section describes the CSV format but not how the numeric interest values for each lecturer are computed. Without resolving this inconsistency, the demonstration is uninterpretable: it neither confirms nor tests the proposed recommender, and the conclusion overstates what is shown.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":3101,"tokens_out":4139,"duration_ms":37285,"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":[{"comment":"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.","section":"Recommender Engine, Table 1"},{"comment":"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.","section":"Data Modeling, Figure 4"},{"comment":"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.","section":"Conclusion"}],"minor_comments":[{"comment":"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.","section":"Introduction"},{"comment":"The spelling 'Euclidian' appears in the Figure 1 caption; elsewhere the manuscript uses 'Euclidean'. Please unify.","section":"Figure 1 and throughout"},{"comment":"Reference [5] contains a typo: 'Eidenburgh' should be 'Edinburgh'.","section":"References"},{"comment":"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.","section":"Figure 5"},{"comment":"Please clarify how missing or partial lecturer data (e.g., a new lecturer with no previous project titles) is handled in the vector construction.","section":"Data Modeling"}],"recommendation":"reject","confidential_remarks":"This manuscript is a short conference paper whose central exhibit (Table 1) is numerically inconsistent with its own stated formula and whose data and code are not available. Even with a corrected table, the lack of any ground-truth evaluation would remain a serious gap; a publishable version would need a reproducible construction of the lecturer vectors, a meaningful evaluation against actual match outcomes or expert judgment, and more than one illustrative example. I would not invite a revision of this paper as is."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a minor proof-of-concept that applies textbook Euclidean distance to a small administrative matching problem, and the only demonstration they show has a load-bearing mistake. Not worth serious referee time.\n\nWhat's actually new is close to nothing: they apply a standard recommender method (Euclidean distance plus the usual 1/(1+d) similarity conversion) to a five-category student/supervisor matching task. The citations for the algorithm are fine. The paper is clearly written for a practitioner audience, and the motivation is real — students at UiTM Perlis don't know faculty research areas. The questionnaire response rate (51/53) is decent, and the Java implementation sounds straightforward.\n\nThe problems dominate. The stress-test is right: Table 1 lists 'similarity' values between 44.95 and 29.21, but the paper defines similarity as 1/(1+distance), which is bounded between 0 and 1. Those numbers cannot be similarity scores. If they are raw Euclidean distances, they're mislabeled and the ranking is backwards — the first row is the least similar, not the most. So the single demonstration in the paper is uninterpretable, and the conclusion that Euclidean distance 'seems to achieve the goal' is unsupported. There is also no released code or data, no description of how lecturer vectors are built from project titles, no baseline, and no evaluation against the current manual process. The paper admits it is 'far from complete,' which the abstract's 'potential of solving the problem' overstates. Minor issues: the text says 'four general areas' then lists five, and the CSV sample in Figure 4 lacks an explanation of what the numbers are.\n\nThe math as stated is standard and the reference list is appropriate. The weakness is entirely in the evidence and the interpretation of the output. This is a classroom-level project write-up, not a research contribution. A desk reject is appropriate; I wouldn't send it to reviewers unless the authors can add code, data, and a corrected, evaluated example.","headline":"A textbook recommender applied to a small admin task, undercut by a mislabeled example table that inverts the ranking.","tokens_in":3679,"tokens_out":4431,"would_cite":false,"duration_ms":43568,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A minimal recommender engine built on Euclidean distance can rank final-year project supervisors from a student's ratings of five interest areas.","keywords":["recommender engine","recommender system","Euclidean distance","final year project","supervisor matching","interest rating","similarity score","proof of concept"],"falsifier":"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.","tokens_in":2727,"feed_emoji":"🎓","tokens_out":8738,"duration_ms":88629,"temperature":0.7,"pith_summary":"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.","feed_headline":"A simple distance rule ranks project supervisors for students","feed_subtitle":"Proof-of-concept engine turns five area-of-interest ratings into a ranked list of potential supervisors.","key_machinery":"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.","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"Defines the local final-year project requirement with proposal and construction phases, motivating the supervisor-matching problem.","marker":"[1]"},{"why":"Supplies the standard definition of recommender systems as software tools that provide suggestions, framing the engine as a recommender.","marker":"[2]"},{"why":"Cited for the Euclidean distance definition between multidimensional data points, the core algorithm used for ranking.","marker":"[3]"},{"why":"Supports the claim that Euclidean distance is useful because it parallels real-world object distance, justifying the algorithm choice.","marker":"[4]"},{"why":"Provides the distance-to-similarity conversion formula that turns unbounded distance into the 0.0-to-1.0 score used for the ranked list.","marker":"[5]"}],"fun_headline_variants":["Euclidean distance matches students to project supervisors","Simple vector scoring pairs final-year students with supervisors","Distance-based engine ranks supervisors by student interests","Five interest ratings drive supervisor ranking for students"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Euclidean distance matches students to project supervisors","Simple vector scoring pairs final-year students with supervisors","Distance-based engine ranks supervisors by student interests","Five interest ratings drive supervisor ranking for students"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000135,"raw_usage":{"total_tokens":1050,"prompt_tokens":759,"completion_tokens":291,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":375,"completion_tokens_details":{"reasoning_tokens":234}},"tokens_in":375,"tokens_out":291,"duration_ms":3528,"temperature":1.0,"reasoning_tokens":234,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:27:51.930352+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the local final-year project requirement with proposal and construction phases, motivating the supervisor-matching problem."},{"cited_title":"Ricci, L","cited_arxiv_id":null,"evidence_quote":"Supplies the standard definition of recommender systems as software tools that provide suggestions, framing the engine as a recommender."},{"cited_title":"A new distance between two bodies of evidence,","cited_arxiv_id":null,"evidence_quote":"Cited for the Euclidean distance definition between multidimensional data points, the core algorithm used for ranking."},{"cited_title":"An efficient euclidean distance transform,","cited_arxiv_id":null,"evidence_quote":"Supports the claim that Euclidean distance is useful because it parallels real-world object distance, justifying the algorithm choice."},{"cited_title":"Similarity and recommender systems","cited_arxiv_id":null,"evidence_quote":"Provides the distance-to-similarity conversion formula that turns unbounded distance into the 0.0-to-1.0 score used for the ranked list."}],"review_version":1}