Task Complexity Matters: An Empirical Study of Reasoning in LLMs for Sentiment Analysis

Donghao Huang; Zhaoxia Wang

REVIEW 4 major objections 5 minor 17 references

Reasoning in LLMs is task-dependent: it hurts simple sentiment classification and only helps fine-grained 27-class emotion recognition.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 20:02 UTC pith:MPGBVWKS

load-bearing objection A useful large-scale benchmark-style study whose headline claim about reasoning hurting simple tasks is only half supported once you separate distillation from thinking-mode effects. the 4 major comments →

arxiv 2602.24060 v1 pith:MPGBVWKS submitted 2026-02-27 cs.CL cs.AI

Task Complexity Matters: An Empirical Study of Reasoning in LLMs for Sentiment Analysis

Donghao Huang , Zhaoxia Wang This is my paper

classification cs.CL cs.AI

keywords reasoning LLMssentiment analysistask complexityover-deliberationfew-shot learningPareto frontieremotion recognitiondistilled reasoning models

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper sets out to test the narrative that adding reasoning capabilities to large language models improves performance across language tasks. It compares base/non-thinking models with distilled/thinking variants over 504 configurations spanning seven model families and three sentiment datasets of increasing granularity: binary (IMDB), five-class (Amazon), and 27-class emotion (GoEmotions). The central finding is a monotonic reversal: non-reasoning models win by 4.8 and 3.6 F1 points on the simpler tasks, while reasoning models win by 2.0 points on the most complex one. The authors argue the number of target classes serves as a proxy for task complexity and that reasoning degrades simple tasks through systematic over-deliberation. If correct, the practical consequence is that reasoning should be switched on only for complex discrimination, not uniformly.

Core claim

Reasoning effectiveness in LLMs is task-dependent: on binary sentiment, reasoning variants lose up to 19.9 F1 points; on five-class, up to 18.4; on 27-class emotion recognition, they gain up to 16.0. Across 504 configurations, base models lead by 4.8±6.3 pp (IMDB) and 3.6±2.2 pp (Amazon), while reasoning models lead by 2.0±1.0 pp on GoEmotions. The authors attribute this to over-deliberation, where reasoning chains introduce spurious hedges on simple tasks but enable fine-grained distinctions on complex ones. Pareto analysis shows base models dominate except on the 27-class task, where the 2.1×–54× overhead is justified.

What carries the argument

The operating machinery is a three-point complexity ladder built from the number of target classes—IMDB (2), Amazon (5), and GoEmotions (27)—treated as a proxy for task complexity. Reasoning effects are isolated by paired comparisons: distilled reasoning models against their base counterparts, and same-model thinking (T) versus non-thinking (N) modes. The outcome measures are F1 difference, failure rate (share of comparisons where reasoning loses), and per-sample latency for Pareto frontier analysis. The explanatory mechanism is over-deliberation: on simple tasks, reasoning chains introduce spurious considerations and hedge, flipping otherwise correct classifications.

Load-bearing premise

The argument stands on treating the number of target classes (2, 5, 27) as a proxy for task complexity; if the systematic pattern instead comes from the datasets' differences in domain, text length, label noise, or evaluation metric, the central conclusion collapses.

What would settle it

A single-domain controlled experiment would settle this: take one sentiment corpus and create 2-, 5-, and 27-class labelings (or collapse GoEmotions and expand IMDB), then run the same model pairs. If reasoning gains do not increase monotonically with class count within that fixed domain, the paper's central claim fails. Alternatively, finding a 27-class dataset whose classes are trivially separable, and showing reasoning still helps, would also falsify the complexity-as-granularity interpretation.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

For binary and five-class sentiment, base/non-thinking models deliver higher F1 with lower latency, so deployment should not default to enabling reasoning.
Few-shot prompting improves over zero-shot in most configurations regardless of model type, and narrows the reasoning gap on simple tasks while preserving reasoning gains on complex ones.
Reasoning is justifiable only for fine-grained 27-class emotion recognition, where accuracy gains offset the 2.1×–54× computational overhead.
Distilled reasoning variants underperform their base models on simpler tasks but can recover and even surpass them with few-shot examples on complex tasks.
Model selection for sentiment systems should be guided by task complexity rather than an assumption that reasoning universally helps.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

A direct test the paper leaves implicit would vary class count within a single domain (e.g., collapsing GoEmotions to 5 and 2 classes) to separate granularity effects from domain effects; if the gradient does not track class count, the central conclusion would need revision.
The over-deliberation mechanism suggests a hybrid architecture that gates reasoning on input ambiguity: simple reviews skip deliberation while ambiguous emotion texts invoke it, potentially capturing the complex-task gains without the simple-task losses.
If the task-complexity pattern extends beyond sentiment, the widespread practice of keeping thinking modes always on would be questionable; a natural next test is another fine-grained classification task with many classes, such as natural language inference or intent detection.
The finding that few-shot examples help more reliably than reasoning modes implies that in-context demonstrations may be a cheaper route to fine-grained classification than expensive reasoning compute, which could be tested on larger class-count datasets.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

A useful large-scale benchmark-style study whose headline claim about reasoning hurting simple tasks is only half supported once you separate distillation from thinking-mode effects.

read the letter

You should know this paper before you cite its headline. It's a broad, systematic evaluation of 504 configs across seven model families on IMDB, Amazon, and GoEmotions, and it does a lot of things right: direct comparisons of distilled reasoning models to their base checkpoints, thinking vs. non-thinking modes on the same models, a Pareto cost-performance analysis, and a released repo. That scale and the explicit cross-architecture framing are genuinely new relative to the financial-overthinking and DeepSeek-R1 work. The finding that few-shot prompting often recovers the distilled gap is also useful and under-appreciated in the field.

But the central claim — reasoning degrades simple tasks and helps complex ones — is only partially supported by the paper's own numbers. Table 3 aggregates two incommensurable comparisons: distillation (Table 1) and runtime thinking modes (Table 2). On IMDB, Table 2 shows thinking modes are neutral or positive in most cases (9/14 positive, FR=36%), so the big aggregate drop is driven almost entirely by distillation, especially the DSR1 vs. DSV3 zero-shot gap (-19.9 pp). The stress-test note is right: the paper's own FR numbers contradict a monotonic degradation story for simpler tasks. The real pattern is more like "distilled reasoning models underperform on binary tasks" plus "thinking modes help on 27-class emotion recognition," not a single task-complexity gradient that applies to all reasoning mechanisms.

The class-count-as-complexity proxy has a real confound: the three datasets differ in domain, text length, and label noise. The authors say the systematic gradient would be "unlikely" if domain differences dominated, but they don't test that. That's a testable claim and they didn't run the experiment. Also, the quantitative magnitudes come from a single run each: no repeated seeds, no confidence intervals, no significance tests. The "best-shot" numbers are selected post hoc from seven shot levels, which inflates apparent differences. The repo is promised but without a commit hash, so exact reproducibility is uncertain.

Qualitative error analysis on over-deliberation is plausible and nicely complements the numbers, but it's anecdotal rather than a systematic coding of errors.

All that said, the paper is honestly written and the limitations section acknowledges several of these gaps. It's not a wasted effort — it's a useful resource for anyone doing model selection on sentiment/emotion tasks, and the Pareto analysis is a good practical contribution. But the abstract and Discussion overstate the uniformity of the task-complexity effect.

My take: it deserves a serious referee, but it needs major revision — reanalyze Table 3 separating distillation from thinking modes, add uncertainty estimates, and ideally disentangle class count from dataset domain. I'd bring it to a reading group if the group cares about LLM evaluation methodology. I'd cite it cautiously, mainly for the Pareto analysis and the raw comparative numbers. Serious thinker: yes, the work is coherent on its own terms, even though I disagree with the aggregation.

Referee Report

4 major / 5 minor

Summary. The paper tests whether reasoning-augmented LLMs uniformly improve sentiment analysis, using 504 configurations spanning seven model families, three datasets of increasing class granularity (binary, 5-class, 27-class), and seven shot levels. It compares distilled/thinking models with base/non-thinking counterparts and reports that reasoning effectiveness is task-dependent: base/non-thinking models outperform reasoning variants by 4.8 pp on IMDB and 3.6 pp on Amazon, while reasoning variants gain 2.0 pp on GoEmotions. It also reports few-shot recovery, a Pareto efficiency-performance analysis, and a qualitative over-deliberation account of why reasoning harms simple tasks.

Significance. If its central claim were established, the paper would be a useful counterweight to the narrative that reasoning universally helps, with concrete deployment guidance for sentiment classification. Its strengths include the release of data, prompts, code, and results; broad architectural coverage; explicit base-model comparisons for distilled reasoning variants; latency measurements; and a clearly falsifiable empirical design. The claim is currently weakened by the way two different reasoning operationalizations are pooled and by the untreated correlation between class count and dataset identity.

major comments (4)

[Aggregate Performance by Task Complexity, Tables 2 and 3] Table 3's headline aggregation pools Table 1's distilled-vs-base comparisons with Table 2's same-model thinking-vs-non-thinking comparisons. These are different mechanisms, different model families, and different model sizes. The paper's own Table 2 shows that on IMDB, thinking underperforms non-thinking in only 5 of 14 comparisons (FR=36%), with several zero-shot differences positive (+0.5 for Qwen3-4B, +1.6 for Qwen3-14B, +2.1 for Magistral; mean zero-shot thinking difference ≈ +0.5 pp). The aggregate IMDB penalty of −4.8±6.3 pp is therefore driven by Table 1, especially DSR1 vs DSV3 at zero shot (−19.9 pp). The central conclusion that 'reasoning degrades simpler tasks' is not supported for runtime-activated thinking on IMDB; it is a distillation-specific effect. The aggregate should be decomposed by reasoning type, or the claim should be narrowed accordingly.
[Methodology, Datasets] The paper treats the number of target classes (2, 5, 27) as a proxy for task complexity, but the three datasets differ in domain, text length, label distribution, label noise, and evaluation metric (binary F1 vs weighted F1). The monotonic gradient in Tables 1–3 could be explained by any of these confounds. The assertion that the gradient would be 'unlikely' if domain differences dominated is not a substitute for a test. A control is needed: for example, binarized or coarse-grained versions of GoEmotions/Amazon, multiple datasets at the same class granularity, or a within-domain comparison. Without this, the causal attribution to task complexity is underdetermined.
[Metrics and Evaluation, Tables 1–3] The headline quantitative claims rely on small differences (e.g., +2.0±1.0 pp on GoEmotions, −3.6±2.2 pp on Amazon) but no repeated runs, confidence intervals, or significance tests are reported. Exemplar sampling uses a single seed (seed=42), and the reported standard deviations are across model pairs and shot levels, not sampling variability. At least bootstrap confidence intervals over test examples, or multiple seeds for few-shot exemplars, should be provided for the main aggregated differences before the task-complexity gradient is stated as robust.
[Discussion, 'Why Reasoning Fails'] The qualitative claim of systematic over-deliberation is presented as a mechanistic finding, but no methodology is described: no sample size, sampling procedure, coding scheme, or inter-annotator agreement. The evidence is anecdotal (e.g., the 'subplot weakness' example). If this analysis is a contribution, it needs a systematic protocol; otherwise it should be framed as an illustrative hypothesis rather than evidence.

minor comments (5)

[Throughout] Several typos and formatting artifacts: 'T able' in table captions, 'F ew-shot' in section headings, and a corrupted URL in the footnote ('inﬂaton').
[Tables 1 and 2] The 'best-shot' column selects the highest F1 across shot levels post hoc. This should be stated explicitly wherever best-shot numbers are used, and per-shot curves or a fixed-shot protocol would aid interpretation of the 'few-shot recovery' claims.
[Table 3] The table says '12 model pairs', but these are not independent: five are distillation pairs and seven are thinking-mode pairs, with differing base architectures. Clarify that the weighted/mean aggregation is over heterogeneous comparisons.
[Figure 1] The caption states 'DSR1/V3 excluded', while the text says the Pareto analysis covers all 504 configurations. Make the exclusion explicit in the main text and state how many configurations remain.
[Model families] The 'seven model families' includes both base and reasoning variants of the same underlying models; the wording could be clarified to distinguish model families from model variants.

Circularity Check

0 steps flagged

No circularity: the study is an empirical benchmark with externally defined task complexity and observed performance differences.

full rationale

The paper's central claim—that reasoning effectiveness varies with task complexity—is an empirical generalization from 504 measured configurations. Task complexity is defined independently of the outcome as class count (2, 5, 27), and reasoning effectiveness is operationalized as observed F1 differences between model variants (distilled vs. base, thinking vs. non-thinking); neither quantity is defined in terms of the conclusion. No parameter is fitted and then renamed as a prediction; Table 3 is an aggregation of Tables 1 and 2, not a derivation from the conclusion. The only self-citations to Huang & Wang [4] are background/motivation (prior DeepSeek-R1 evaluation) and dataset selection; they are not load-bearing and no uniqueness theorem or ansatz is imported from them. The over-deliberation explanation is a post-hoc qualitative interpretation, not part of the derivation chain. Concerns about confounds (domain, length, metric) or about aggregating distillation and runtime-thinking comparisons are validity/correctness issues, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

The central claim rests on experimental-design assumptions rather than fitted parameters or invented entities. The most important is the class-count-as-complexity proxy; also the isolation of reasoning via mode comparisons and the single-label GoEmotions subset. No new physical or conceptual entities are introduced; 'over-deliberation' is a descriptive label, not an independent entity.

axioms (4)

domain assumption Number of target classes is a valid proxy for task complexity.
Methodology, Datasets: 'the number of target classes (2, 5, 27) serving as a proxy for task complexity.' This is load-bearing for the central claim and untested against confounds like domain and text length.
domain assumption Thinking/non-thinking mode comparisons isolate reasoning effects.
Methodology, Models: the paper compares thinking (T) and non-thinking (N) modes within Qwen3, Granite3.3, and Magistral to isolate reasoning. This assumes the modes differ only in reasoning activation and not in other training or decoding properties.
domain assumption Single-label GoEmotions subset preserves the complexity ordering.
Methodology, Datasets: GoEmotions is originally multi-label, but only single-label samples are used, 'which may underestimate the original multi-label difficulty' (Limitations). The ordering of task complexity relies on this subset.
domain assumption Default hyperparameters are representative and allow fair comparison.
Infrastructure section states 'All models evaluated using recommended default hyperparameters'. Exact decoding parameters are not reported, so the comparability of latency and F1 across models rests on an unstated assumption.

pith-pipeline@v1.3.0-alltime-deepseek · 8941 in / 9524 out tokens · 82114 ms · 2026-08-02T20:02:10.483862+00:00 · methodology

0 comments

read the original abstract

Large language models (LLMs) with reasoning capabilities have fueled a compelling narrative that reasoning universally improves performance across language tasks. We test this claim through a comprehensive evaluation of 504 configurations across seven model families--including adaptive, conditional, and reinforcement learning-based reasoning architectures--on sentiment analysis datasets of varying granularity (binary, five-class, and 27-class emotion). Our findings reveal that reasoning effectiveness is strongly task-dependent, challenging prevailing assumptions: (1) Reasoning shows task-complexity dependence--binary classification degrades up to -19.9 F1 percentage points (pp), while 27-class emotion recognition gains up to +16.0pp; (2) Distilled reasoning variants underperform base models by 3-18 pp on simpler tasks, though few-shot prompting enables partial recovery; (3) Few-shot learning improves over zero-shot in most cases regardless of model type, with gains varying by architecture and task complexity; (4) Pareto frontier analysis shows base models dominate efficiency-performance trade-offs, with reasoning justified only for complex emotion recognition despite 2.1x-54x computational overhead. We complement these quantitative findings with qualitative error analysis revealing that reasoning degrades simpler tasks through systematic over-deliberation, offering mechanistic insight beyond the high-level overthinking hypothesis.

Figures

Figures reproduced from arXiv: 2602.24060 by Donghao Huang, Zhaoxia Wang.

**Figure 1.** Figure 1: Performance vs. computational cost. Circles: base/non-thinking; triangles: reasoning/thinking. Colors: IMDB (blue), Amazon (green), GoEmotions (red). Dashed lines: Pareto frontiers. Evaluated on H100 via Ollama; DSR1/V3 excluded. Performance variability and model selection. As shown in [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗

discussion (0)

Reference graph

Works this paper leans on

17 extracted references · 6 linked inside Pith

[1]

arXiv preprint arXiv:2501.12948 (2025)

DeepSeek-AI: Deepseek-r1: Incentivizing reasoning capability in llms via reinforce- ment learning. arXiv preprint arXiv:2501.12948 (2025)

Pith/arXiv arXiv 2025
[2]

In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

Demszky, D., Movshovitz-Attias, D., Ko, J., Cowen, A., Nemade, G., Ravi, S.: Goemotions: A dataset of ﬁne-grained emotions. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 4040–4054 (2020)

2020
[3]

In: Findings of the Association for Computational Linguistics: ACL

Hsieh, C.Y., Li, C.L., Yeh, C.K., Nakhost, H., Fujii, Y., Ratner, A.: Distilling step- by-step! outperforming larger language models with less training data and smaller model sizes. In: Findings of the Association for Computational Linguistics: ACL
[4]

IEEE Intelligent Systems 40(6), 52–63 (2025)

Huang, D., Wang, Z.: Explainable sentiment analysis with deepseek-r1: Perfor- mance, eﬃciency, and few-shot learning. IEEE Intelligent Systems 40(6), 52–63 (2025). https://doi.org/10.1109/MIS.2025.3614967

arXiv 2025
[5]

https://www.ibm.com/granite/docs/models/granite/ (2025), accessed on July 20, 2025 12 D

IBM: Granite 3.3 model documentation. https://www.ibm.com/granite/docs/models/granite/ (2025), accessed on July 20, 2025 12 D. Huang et al

2025
[6]

arXiv preprint arXiv:2506.04574 (2025)

Li, J., Chen, M., Wang, Q., Zhang, Y.: Reasoning or overthinking: Evaluating large language models on ﬁnancial sentiment analysis. arXiv preprint arXiv:2506.04574 (2025)

Pith/arXiv arXiv 2025
[7]

In: IEEE 10th International Conference on Awareness Science and Technology (iCAST)

Mukherjee, A., Mukhopadhyay, S., Panigrahi, P., Goswami, S.: Utilization of over- sampling for multi-class sentiment analysis on amazon review dataset. In: IEEE 10th International Conference on Awareness Science and Technology (iCAST). pp. 1–6. IEEE (2019)

2019
[8]

arXiv preprint arXiv:2407.01687 (2024)

Prabhakar, A., Khandelwal, U., Kembhavi, A.: Deciphering the factors inﬂuencing the eﬃcacy of chain-of-thought: Probability, memorization, and noisy reasoning. arXiv preprint arXiv:2407.01687 (2024)

Pith/arXiv arXiv 2024
[9]

arXiv preprint arXiv:2505.09388 (2025)

Qwen Team: Qwen3 technical report. arXiv preprint arXiv:2505.09388 (2025)

Pith/arXiv arXiv 2025
[10]

arXiv preprint arXiv:2506.10910 (2025)

Rastogi, A., Jiang, A.Q., Lo, A., Berrada, G., Lample, G., Rute, J., Barmentlo, J., Yadav, K., Khandelwal, K., Chandu, K.R., Blier, L., Saulnier, L., Dinot, M., Darrin, M., Gupta, N., Soletskyi, R., Vaze, S., Le Scao, T., Wang, Y.: Magistral: Transparent reasoning via reinforcement learning. arXiv preprint arXiv:2506.10910 (2025)

Pith/arXiv arXiv 2025
[11]

International Journal of Advanced Computer Science and Applications 13(9), 618–624 (2022)

Steinke, I., Wier, J., Simon, L., Seetan, R.: Sentiment analysis of online movie re- views using machine learning. International Journal of Advanced Computer Science and Applications 13(9), 618–624 (2022)

2022
[12]

Artiﬁcial Intelligence Review 58(3), 75 (2025)

Wang, Z., Huang, D., Cui, J., Zhang, X., Ho, S.B., Cambria, E.: A review of chinese sentiment analysis: subjects, methods, and trends. Artiﬁcial Intelligence Review 58(3), 75 (2025)

2025
[13]

In: Advances in neural information processing systems

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D.: Chain-of-thought prompting elicits reasoning in large language models. In: Advances in neural information processing systems. vol. 35, pp. 24824–24837 (2022)

2022
[14]

arXiv preprint arXiv:2402.13116 (2024)

Xu, X., Li, M., Wang, C., Luo, T., Wang, X., Wang, J.: A survey on knowledge distillation of large language models. arXiv preprint arXiv:2402.13116 (2024)

Pith/arXiv arXiv 2024
[15]

In: Proceedings of NAACL 2024 (2024)

Zhang, A., Li, B., Wang, C., Liu, D., Chen, E.: Sentiment analysis in the era of large language models: A reality check. In: Proceedings of NAACL 2024 (2024)

2024
[16]

arXiv preprint arXiv:2503.03225 (2025)

Zhang, L., Liu, P., Wang, H., Chen, S.: Targeted distillation for sentiment analysis. arXiv preprint arXiv:2503.03225 (2025)

arXiv 2025
[2023]

8003–8017 (2023)

pp. 8003–8017 (2023)

2023

[1] [1]

arXiv preprint arXiv:2501.12948 (2025)

DeepSeek-AI: Deepseek-r1: Incentivizing reasoning capability in llms via reinforce- ment learning. arXiv preprint arXiv:2501.12948 (2025)

Pith/arXiv arXiv 2025

[2] [2]

In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

Demszky, D., Movshovitz-Attias, D., Ko, J., Cowen, A., Nemade, G., Ravi, S.: Goemotions: A dataset of ﬁne-grained emotions. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 4040–4054 (2020)

2020

[3] [3]

In: Findings of the Association for Computational Linguistics: ACL

Hsieh, C.Y., Li, C.L., Yeh, C.K., Nakhost, H., Fujii, Y., Ratner, A.: Distilling step- by-step! outperforming larger language models with less training data and smaller model sizes. In: Findings of the Association for Computational Linguistics: ACL

[4] [4]

IEEE Intelligent Systems 40(6), 52–63 (2025)

Huang, D., Wang, Z.: Explainable sentiment analysis with deepseek-r1: Perfor- mance, eﬃciency, and few-shot learning. IEEE Intelligent Systems 40(6), 52–63 (2025). https://doi.org/10.1109/MIS.2025.3614967

arXiv 2025

[5] [5]

https://www.ibm.com/granite/docs/models/granite/ (2025), accessed on July 20, 2025 12 D

IBM: Granite 3.3 model documentation. https://www.ibm.com/granite/docs/models/granite/ (2025), accessed on July 20, 2025 12 D. Huang et al

2025

[6] [6]

arXiv preprint arXiv:2506.04574 (2025)

Li, J., Chen, M., Wang, Q., Zhang, Y.: Reasoning or overthinking: Evaluating large language models on ﬁnancial sentiment analysis. arXiv preprint arXiv:2506.04574 (2025)

Pith/arXiv arXiv 2025

[7] [7]

In: IEEE 10th International Conference on Awareness Science and Technology (iCAST)

Mukherjee, A., Mukhopadhyay, S., Panigrahi, P., Goswami, S.: Utilization of over- sampling for multi-class sentiment analysis on amazon review dataset. In: IEEE 10th International Conference on Awareness Science and Technology (iCAST). pp. 1–6. IEEE (2019)

2019

[8] [8]

arXiv preprint arXiv:2407.01687 (2024)

Prabhakar, A., Khandelwal, U., Kembhavi, A.: Deciphering the factors inﬂuencing the eﬃcacy of chain-of-thought: Probability, memorization, and noisy reasoning. arXiv preprint arXiv:2407.01687 (2024)

Pith/arXiv arXiv 2024

[9] [9]

arXiv preprint arXiv:2505.09388 (2025)

Qwen Team: Qwen3 technical report. arXiv preprint arXiv:2505.09388 (2025)

Pith/arXiv arXiv 2025

[10] [10]

arXiv preprint arXiv:2506.10910 (2025)

Rastogi, A., Jiang, A.Q., Lo, A., Berrada, G., Lample, G., Rute, J., Barmentlo, J., Yadav, K., Khandelwal, K., Chandu, K.R., Blier, L., Saulnier, L., Dinot, M., Darrin, M., Gupta, N., Soletskyi, R., Vaze, S., Le Scao, T., Wang, Y.: Magistral: Transparent reasoning via reinforcement learning. arXiv preprint arXiv:2506.10910 (2025)

Pith/arXiv arXiv 2025

[11] [11]

International Journal of Advanced Computer Science and Applications 13(9), 618–624 (2022)

Steinke, I., Wier, J., Simon, L., Seetan, R.: Sentiment analysis of online movie re- views using machine learning. International Journal of Advanced Computer Science and Applications 13(9), 618–624 (2022)

2022

[12] [12]

Artiﬁcial Intelligence Review 58(3), 75 (2025)

Wang, Z., Huang, D., Cui, J., Zhang, X., Ho, S.B., Cambria, E.: A review of chinese sentiment analysis: subjects, methods, and trends. Artiﬁcial Intelligence Review 58(3), 75 (2025)

2025

[13] [13]

In: Advances in neural information processing systems

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D.: Chain-of-thought prompting elicits reasoning in large language models. In: Advances in neural information processing systems. vol. 35, pp. 24824–24837 (2022)

2022

[14] [14]

arXiv preprint arXiv:2402.13116 (2024)

Xu, X., Li, M., Wang, C., Luo, T., Wang, X., Wang, J.: A survey on knowledge distillation of large language models. arXiv preprint arXiv:2402.13116 (2024)

Pith/arXiv arXiv 2024

[15] [15]

In: Proceedings of NAACL 2024 (2024)

Zhang, A., Li, B., Wang, C., Liu, D., Chen, E.: Sentiment analysis in the era of large language models: A reality check. In: Proceedings of NAACL 2024 (2024)

2024

[16] [16]

arXiv preprint arXiv:2503.03225 (2025)

Zhang, L., Liu, P., Wang, H., Chen, S.: Targeted distillation for sentiment analysis. arXiv preprint arXiv:2503.03225 (2025)

arXiv 2025

[17] [2023]

8003–8017 (2023)

pp. 8003–8017 (2023)

2023