{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/electra-and-gpt-4o-cost-effective-partners","title":"ELECTRA and GPT-4o: Cost-Effective Partners for Sentiment Analysis","arxiv_id":"2501.00062","date":"2024-12-29","proceeding":null,"authors":["James P. Beno"],"abstract":"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.","url_abs":"https://arxiv.org/abs/2501.00062v2","url_pdf":"https://arxiv.org/pdf/2501.00062v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"electra-and-gpt-4o-cost-effective-partners","repo_url":"https://github.com/jbeno/sentiment","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"},{"task_slug":"sentiment-classification","task_name":"Sentiment Classification"}],"methods":[{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"attention-dropout","method_name":"Attention Dropout"},{"method_slug":"base","method_name":"BASE"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"cosine-annealing","method_name":"Cosine Annealing"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"discriminative-fine-tuning","method_name":"Discriminative Fine-Tuning"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"electra","method_name":"ELECTRA"},{"method_slug":"favor","method_name":"FAVOR+"},{"method_slug":"gpt","method_name":"GPT"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"linear-warmup-with-cosine-annealing","method_name":"Linear Warmup With Cosine Annealing"},{"method_slug":"linear-warmup-with-linear-decay","method_name":"Linear Warmup With Linear Decay"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"performer","method_name":"Performer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"weight-decay","method_name":"Weight Decay"},{"method_slug":"wordpiece","method_name":"WordPiece"}],"datasets_introduced":[{"slug":"sentiment-merged","name":"Sentiment Merged","full_name":"SST-3, DynaSent R1/R2"}],"methods_introduced":[],"results":[{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o Fine-Tuned (Minimal)","rank_in_archive_order":1,"of":12,"metrics":{"Macro F1":"89"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o-mini Fine-Tuned","rank_in_archive_order":2,"of":12,"metrics":{"Macro F1":"86.9"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o + ELECTRA Large FT (Prompt, Label, Examples)","rank_in_archive_order":3,"of":12,"metrics":{"Macro F1":"81.53"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o (Prompt)","rank_in_archive_order":4,"of":12,"metrics":{"Macro F1":"80.22"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o-mini + ELECTRA Large FT (Prompt, Label, Probabilities)","rank_in_archive_order":5,"of":12,"metrics":{"Macro F1":"79.72"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o-mini + ELECTRA Large FT (Prompt, Label)","rank_in_archive_order":6,"of":12,"metrics":{"Macro F1":"77.94"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o + ELECTRA Large FT","rank_in_archive_order":7,"of":12,"metrics":{"Macro F1":"77.69"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o-mini (Prompt)","rank_in_archive_order":8,"of":12,"metrics":{"Macro F1":"77.35"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"ELECTRA Large Fine-Tuned","rank_in_archive_order":9,"of":12,"metrics":{"Macro F1":"76.29"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"GPT-4o-mini + ELECTRA Base FT","rank_in_archive_order":10,"of":12,"metrics":{"Macro F1":"76.19"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-dynasent","task":"Sentiment Analysis","dataset":"DynaSent","model":"ELECTRA Base Fine-Tuned","rank_in_archive_order":11,"of":12,"metrics":{"Macro F1":"71.83"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o-mini Fine-Tuned","rank_in_archive_order":1,"of":11,"metrics":{"Macro F1":"75.68"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o Fine-Tuned (Minimal)","rank_in_archive_order":2,"of":11,"metrics":{"Macro F1":"73.99"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o + ELECTRA Large FT","rank_in_archive_order":3,"of":11,"metrics":{"Macro F1":"72.94"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o (Prompt)","rank_in_archive_order":4,"of":11,"metrics":{"Macro F1":"72.2"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o + ELECTRA Large FT (Prompt, Label, Examples)","rank_in_archive_order":5,"of":11,"metrics":{"Macro F1":"72.06"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o-mini + ELECTRA Large FT (Prompt, Label, Examples)","rank_in_archive_order":6,"of":11,"metrics":{"Macro F1":"71.98"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o-mini + ELECTRA Base FT","rank_in_archive_order":7,"of":11,"metrics":{"Macro F1":"71.72"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o-mini + ELECTRA Large FT (Prompt, Label)","rank_in_archive_order":8,"of":11,"metrics":{"Macro F1":"70.99"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"ELECTRA Large Fine-Tuned","rank_in_archive_order":9,"of":11,"metrics":{"Macro F1":"70.90"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"GPT-4o-mini (Prompt)","rank_in_archive_order":10,"of":11,"metrics":{"Macro F1":"70.67"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sst-3","task":"Sentiment Analysis","dataset":"SST-3","model":"ELECTRA Base Fine-Tuned","rank_in_archive_order":11,"of":11,"metrics":{"Macro F1":"69.95"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o Fine-Tuned (Minimal)","rank_in_archive_order":1,"of":10,"metrics":{"Macro F1":"86.99"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o-mini Fine-Tuned","rank_in_archive_order":2,"of":10,"metrics":{"Macro F1":"86.77"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o-mini + ELECTRA Large FT (Prompt, Label)","rank_in_archive_order":3,"of":10,"metrics":{"Macro F1":"83.49"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o + ELECTRA Large FT (Prompt, Label, Examples)","rank_in_archive_order":4,"of":10,"metrics":{"Macro F1":"83.09"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o-mini + ELECTRA Base FT (Prompt, Label)","rank_in_archive_order":5,"of":10,"metrics":{"Macro F1":"82.74"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"ELECTRA Large Fine-Tuned","rank_in_archive_order":6,"of":10,"metrics":{"Macro F1":"82.36"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o + ELECTRA Large FT (Prompt, Label)","rank_in_archive_order":7,"of":10,"metrics":{"Macro F1":"81.57"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o (Prompt)","rank_in_archive_order":8,"of":10,"metrics":{"Macro F1":"80.14"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"GPT-4o-mini (Prompt)","rank_in_archive_order":9,"of":10,"metrics":{"Macro F1":"79.52"},"uses_additional_data":false},{"leaderboard":"/sota/sentiment-analysis-on-sentiment-merged","task":"Sentiment Analysis","dataset":"Sentiment Merged","model":"ELECTRA Base Fine-Tuned","rank_in_archive_order":10,"of":10,"metrics":{"Macro F1":"79.29"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}