Papers › ELECTRA and GPT-4o: Cost-Effective Partners for Sentiment Analysis

ELECTRA and GPT-4o: Cost-Effective Partners for Sentiment Analysis

29 Dec 2024arXiv:2501.00062archive 2025-07-28

James P. Beno

Bidirectional transformers excel at sentiment analysis, and Large Language Models (LLM) are effective zero-shot learners. Might they perform better as a team? This paper explores collaborative approaches between ELECTRA and GPT-4o for three-way sentiment classification. We fine-tuned (FT) four models (ELECTRA Base/Large, GPT-4o/4o-mini) using a mix of reviews from Stanford Sentiment Treebank (SST) and DynaSent. We provided input from ELECTRA to GPT as: predicted label, probabilities, and retrieved examples. Sharing ELECTRA Base FT predictions with GPT-4o-mini significantly improved performance over either model alone (82.50 macro F1 vs. 79.14 ELECTRA Base FT, 79.41 GPT-4o-mini) and yielded the lowest cost/performance ratio ($0.12/F1 point). However, when GPT models were fine-tuned, including predictions decreased performance. GPT-4o FT-M was the top performer (86.99), with GPT-4o-mini FT close behind (86.70) at much less cost ($0.38 vs. $1.59/F1 point). Our results show that augmenting prompts with predictions from fine-tuned encoders is an efficient way to boost performance, and a fine-tuned GPT-4o-mini is nearly as good as GPT-4o FT at 76% less cost. Both are affordable options for projects with limited resources.

PaperPDFCode

Syntology This paper's arXiv id is not in Syntology's Atlas id list (549,449 ids, as this build read them; the read time is in the build record), so this page links to no Syntology page for it. The list is not the whole graph; Syntology's feed shows papers as they enter the graph.

Code

jbeno/sentiment officialmentioned in papermentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators (arXiv:2003.10555; that paper's own run record: 31 ran (of which 7 constructed an object rather than computing a result; 18 with no instrument failure: 2 honoured, 2 violated, 14 with no contract checked; 13 where Syntology's instrument failed) · 9 unverified); nothing checks that they implement this paper's method. report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

A paper named beside a repository with no sample linked to this paper is shown with that paper's own run record, not this paper's: “ran” means executed on a synthesized input, not that the code is correct, and “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Papers are listed in arXiv-id order, at most three per repository.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Sentiment AnalysisSentiment Classification

Datasets

Introduced by this paper, per the archive.

Sentiment Merged

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Sentiment Analysis DynaSent GPT-4o Fine-Tuned (Minimal) Macro F1 89 #1 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o-mini Fine-Tuned Macro F1 86.9 #2 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o + ELECTRA Large FT (Prompt, Label, Examples) Macro F1 81.53 #3 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o (Prompt) Macro F1 80.22 #4 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o-mini + ELECTRA Large FT (Prompt, Label, Probabilities) Macro F1 79.72 #5 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o-mini + ELECTRA Large FT (Prompt, Label) Macro F1 77.94 #6 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o + ELECTRA Large FT Macro F1 77.69 #7 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o-mini (Prompt) Macro F1 77.35 #8 of 12 Archive leaderboard report
Sentiment Analysis DynaSent ELECTRA Large Fine-Tuned Macro F1 76.29 #9 of 12 Archive leaderboard report
Sentiment Analysis DynaSent GPT-4o-mini + ELECTRA Base FT Macro F1 76.19 #10 of 12 Archive leaderboard report
Sentiment Analysis DynaSent ELECTRA Base Fine-Tuned Macro F1 71.83 #11 of 12 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o-mini Fine-Tuned Macro F1 75.68 #1 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o Fine-Tuned (Minimal) Macro F1 73.99 #2 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o + ELECTRA Large FT Macro F1 72.94 #3 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o (Prompt) Macro F1 72.2 #4 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o + ELECTRA Large FT (Prompt, Label, Examples) Macro F1 72.06 #5 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o-mini + ELECTRA Large FT (Prompt, Label, Examples) Macro F1 71.98 #6 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o-mini + ELECTRA Base FT Macro F1 71.72 #7 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o-mini + ELECTRA Large FT (Prompt, Label) Macro F1 70.99 #8 of 11 Archive leaderboard report
Sentiment Analysis SST-3 ELECTRA Large Fine-Tuned Macro F1 70.90 #9 of 11 Archive leaderboard report
Sentiment Analysis SST-3 GPT-4o-mini (Prompt) Macro F1 70.67 #10 of 11 Archive leaderboard report
Sentiment Analysis SST-3 ELECTRA Base Fine-Tuned Macro F1 69.95 #11 of 11 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o Fine-Tuned (Minimal) Macro F1 86.99 #1 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o-mini Fine-Tuned Macro F1 86.77 #2 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o-mini + ELECTRA Large FT (Prompt, Label) Macro F1 83.49 #3 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o + ELECTRA Large FT (Prompt, Label, Examples) Macro F1 83.09 #4 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o-mini + ELECTRA Base FT (Prompt, Label) Macro F1 82.74 #5 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged ELECTRA Large Fine-Tuned Macro F1 82.36 #6 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o + ELECTRA Large FT (Prompt, Label) Macro F1 81.57 #7 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o (Prompt) Macro F1 80.14 #8 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged GPT-4o-mini (Prompt) Macro F1 79.52 #9 of 10 Archive leaderboard report
Sentiment Analysis Sentiment Merged ELECTRA Base Fine-Tuned Macro F1 79.29 #10 of 10 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamAttentionAttention DropoutBASEBPECosine AnnealingDense ConnectionsDiscriminative Fine-TuningDropoutELECTRAFAVOR+GPTLayer NormalizationLinear LayerLinear Warmup With Cosine AnnealingLinear Warmup With Linear DecayMulti-Head AttentionPerformerResidual ConnectionSoftmaxWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections