Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation

Matt Sandler

cs.IRSep 24, 2026

Nearest but Not Dearest: Shared Curator-Feedback Infrastructure for Content-Only Search and Recommendation

Authors: Matt Sandler

Organizations: Feed.fm San Francisco, California, USA

Abstract

A deployed B2B music-discovery platform serves both query-driven search (text prompts, vibe tags) and seed-driven recommendation (seed-track and artist stations) over one licensed catalog, one LAION-CLAP joint audio-text embedding space, one candidate-generation filter, and one ranking head -- and neither path consumes end-listener behavioral signal. In this content-only regime, curator judgment is the principal feedback signal available, and offline cosine similarity predicts it poorly: 38% of cosine-nearest neighbors are rejected by curators. The rejections reveal a clean partition: a majority (55%) are sound failures the encoder could address (style, tempo, mood mismatch), and a substantial minority (37%) are context failures orthogonal to the waveform (wrong language, holiday content, devotional content, rights and lyric flags). We deploy this sound-vs-context decomposition as feedback infrastructure, routing each failure mode to the layer that can absorb it: context failures to a constraint filter at candidate generation, sound failures to an embedding reweighting head at the representation layer -- both below the search/recommendation split, so a single curator loop maintains both experiences. On 1,200 curator judgments collected over two production rounds one month apart, the combined intervention reduces rejection rate from 38.17% to 28.83% (-24.5% relative, McNemar chi-squared = 22.4, p = 2.2e-6). An accounting decomposition attributes 4.08 pp of the drop to filter-eligible categories and 5.25 pp to the rest; the deployment was unblinded and compound, so this is a production accounting bound, not a causal estimate. We present this as an industrial case study rather than a validated general method, and close with lessons for content-only discovery: the failure partition is orthogonal to the paradigm partition, and corrections land in layers shared by both paradigms.

Figures & tables

Panel of Figure 1 — Figure 1. The curator-feedback loop. A rejection categorized as $\mathcal{C}_{\text{ctx}}$ is routed to the constraint filter (catalog-side intervention); a rejection in $\mathcal{C}_{\text{sound}}\cup\mathcal{C}_{\text{other}}$ is routed to the reweighting head along the non-filter path (representation-side intervention). The two mechanisms address disjoint failure modes and update on different cadences. Block diagram of the production retrieval pipeline. A seed track flows into a LAION-CLAP embedding, then a constraint filter, then a reweighting head, then top-k recommendations, then a curator judgment. A decision diamond below routes context failures back to the constraint filter and sound or uncategorized failures back to the reweighting head.

Failure mode	Count	% rej.
$\mathcal{C}_{\text{sound}}$	253	55.2%
style	168	36.7%
tempo	52	11.4%
mood	33	7.2%
$\mathcal{C}_{\text{ctx}}$	169	36.9%
holiday	68	14.8%

Table 1. Round-1 (baseline) rejection breakdown over

N=1{,}200

recommendations,

n=458

thumbs-downs with a recorded failure-mode code. Zero-count

\mathcal{C}_{\text{ctx}}

categories (e.g., bad_quality ) are omitted from the table. other is reported as a separate row because the curator interface does not disaggregate sound-related from contextual “other” selections; the accounting analysis in § 4.3 conservatively allocates

\mathcal{C}_{\text{other}}

to the non-filter group alongside

\mathcal{C}_{\text{sound}}

Figure 2. Cosine similarity distribution by curator judgment on Round-1 (

N=1{,}149

judgments with a clear binary outcome and recorded failure-mode code). Accepted (thumbs-up,

n=691

) and rejected (thumbs-down,

n=458

) distributions overlap heavily; the small mean shift (0.897 vs. 0.881, dashed lines) yields AUC

=0.644

, the empirical basis for the partition-as-routing-criterion argument in § 3.3 . Overlaid histograms of cosine similarity between seed and retrieved neighbor, separated by curator judgment. The accepted (thumbs-up) distribution and the rejected (thumbs-down) distribution overlap substantially across the 0.80–0.95 range, with the accepted distribution shifted slightly higher.

Metric	R1	R2	$\Delta$	Rel.
Seeds	400	400	—	—
Judged recs	1,200	1,200	—	—
Thumbs-downs	458	346	$-112$	—
Rejection rate $\rho$	38.17%	28.83%	$-9.33$ pp	$\mathbf{-24.5\%}$

Table 2. Round-1 baseline vs. Round-2 (constraint filter + reweighting head) on

N=1{,}200

judgments per round. Two-proportion pooled

z=4.84

p=1.3\times 10^{-6}

Category group	R1 TD	R2 TD	$\Delta$	pp contrib.
Filter-eligible ( $\mathcal{C}_{\text{ctx}}$ )	169	120	$-49$	$-4.08$ pp
Non-filter ( $\mathcal{C}_{\text{sound}}\cup\mathcal{C}_{\text{other}}$ )	289	226	$-63$	$-5.25$ pp
Total	458	346	$-112$	$-9.33$ pp

Table 3. Round-over-round accounting decomposition by category group. “Filter-eligible” contains

\mathcal{C}_{\text{ctx}}

; “Non-filter” contains

\mathcal{C}_{\text{sound}}

and other . The deployment was unblinded and compound; this attribution is arithmetic across categories, not a controlled per-mechanism causal estimate.

Figure 3. Rejection-rate waterfall, R1 baseline

\to

R2 result. Filter-eligible categories (

\mathcal{C}_{\text{ctx}}

) account for

-4.08

pp of the drop; non-filter categories (

\mathcal{C}_{\text{sound}}\cup\mathcal{C}_{\text{other}}

) account for

-5.25

pp. The two mechanisms shipped together (unblinded compound deployment); this is an arithmetic accounting decomposition, not a controlled per-mechanism ablation. Waterfall chart of curator rejection rate. Round 1 baseline starts at 38.17 percent. A first step labeled "filter-eligible categories" lowers the rate by 4.08 percentage points to 34.09 percent. A second step labeled "non-filter categories" lowers the rate by a further 5.25 percentage points to 28.84 percent. The Round 2 result is shown as 28.83 percent.

Recommender Systems

Recommender systems predict and rank items for people using interaction histories, content, context, and feedback.

See all 508 papers

Aug 25, 2026

Keeping the Index Open: The Recommendation-Side Cost of Shared Search and Recommendation

A shared search-and-recommendation index must score new items from features alone because search has no exploration slot. In a public log covering both surfaces over one catalog,

38.6\%

of held-out query-search impressions show an item never previously shown or visited. For user-cold engagements, the feature-based tower serves this demand without measurable loss against

99

sampled negatives (

0.9595

Recall@20 versus

0.9510

warm). A lexical baseline reaches similar parity, while a full-catalog check remains statistically undecided. Dual-encoder retrieval therefore keeps the index open to new items, unlike an ID-softmax recommender that requires retraining. We price this openness on recommendation against six sequential baselines, each retrained and tuned through five rounds on corrected targets. A float32 timestamp bug had reordered leave-one-out targets for

19.7\%

of users. On MovieLens-1M, warm accuracy trails the strongest retrained baseline by

5.2\%

Recall@20 and

11.4\%

NDCG@20. On MIND, the gap narrows to

0.8

3.6\%

relative to the five strongest baselines, though the model ranks sixth of seven. Under strict zero-leakage cold-start evaluation, the content tower achieves

0.172 \pm 0.006

Recall@20,

1.4\times

the strongest retrained dedicated method (

0.124 \pm 0.007

) and

3\times

a training-free floor, without cold-specific training. Exact full-softmax training raises Recall@20 by

54\%

on MIND-small and

6.9\%

on MovieLens-1M over sampled InfoNCE, but recomputes the full catalog each step and exhausts accelerator memory at

240

K items. Approximate nearest-neighbor search explains none of the remaining gap, serving cost does not regress against ID-softmax retrieval, and a history-window sweep explains half the post-recipe remainder. Exact-quality training at catalog scale remains the open problem.

Theodore Rogers, Joe Standerfer, Dmitrii Timoshenko +3

Amazon Web Services United States

Recommender Systems Cold-Start Recommendation Dense Retrieval Sequential Recommendation

Ask PDF

Aug 27, 2026

Recovering Expert Critic-Sourced Network Adjacency between Musical Artists from Acoustic Distributions: A Construct-Validity Approach

Music recommendation relies primarily on two signals: user-item interactions, which fail in the cold-start regime, and intrinsic musical content, available for any recording. We argue that a third, largely untapped signal is both richer and more principled: critical adjacency, the pairwise relation established when an expert critic explicitly links two artists in long-form prose. It encodes deliberate judgments about which artists belong together. Prior work established its internal validity, showing it recovers coherent, interpretable communities and can match collaborative filtering in user-satisfaction simulations, with no user data. What has been missing is external validation: whether this critic-sourced relation is grounded in the music itself versus sociological context. We test it against acoustic content, reframing the question as one of construct validity. Representing artists as empirical distributions over 80 low-level Essentia acoustic descriptors and modeling pairwise proximity via marginal optimal-transport (Wasserstein) distances, we evaluate how far critical adjacency is sonically recoverable under a cold-start, artist-disjoint split. Our ensemble recovers these edges at out-of-sample AUC of 0.767 (95% CI 0.761-0.775). Recoverability rises monotonically with critical consensus, reaching 0.865 on multi-source attested edges. Stratified evaluations align with sociological models of genre: tightly bounded, scene-based genres show higher recoverability than broad industry umbrella terms. Critical discourse is thus a rich source of information for recommendation, decomposing into a reproducible "sonic core" and a "sociological remainder" driven by narrative positioning, subcultural context, and canonical placement. The work offers both a scalable cold-start discovery mechanism and a sociologically grounded approach to MIR and MRS research.

Elena Badillo-Goicoechea, Fengfeng He

University of Chicago Chicago, Illinois, USA

Cold-Start Recommendation Link Prediction Recommender Systems Music Information Retrieval Music Recommendation

Ask PDF

Sep 20, 2026

Explainable Recommendations at Scale: LLM Rationales for YouTube Music Artist Discovery

Modern music streaming platforms face a persistent tradeoff: exploiting familiar content versus driving the exploration of novel items. While users frequently desire discovery, they hesitate to select unknown artists over proven favorites. Providing transparent, natural language rationales that explain why an unexplored item is recommended lowers this barrier. However, while Large Language Models (LLMs) excel at this nuanced explainability, their real-time deployment is severely bottlenecked by prohibitive inference costs and computational overhead. In this paper, we present an industry case study of a decoupled recommendation architecture that successfully scales exploration without compromising latency. Our system isolates LLM inference asynchronously offline, pre-computing personalized candidate pools of undiscovered artists alongside tailored rationales. Large-scale online A/B experiments validate our design. We demonstrate that combining LLM-backed recommendations with these explanatory rationales significantly reduces the trust barrier for new content, yielding statistically significant improvements in both user exploration and overall engagement on the discovery surfaces.

Xiao Liu, Yanwei Song, Srivaths Ranganathan +7

Google LLC USA

Music Recommendation Recommender Systems Explainable Recommendation Large Language Model-Based Recommendation LLM Serving

Ask PDF

Abstract

Recommender Systems

Explore similar work

Keeping the Index Open: The Recommendation-Side Cost of Shared Search and Recommendation

Recovering Expert Critic-Sourced Network Adjacency between Musical Artists from Acoustic Distributions: A Construct-Validity Approach

Explainable Recommendations at Scale: LLM Rationales for YouTube Music Artist Discovery