Recent Large Reasoning Models (LRMs) likeDeepSeek-R1 have demonstrated remarkable success in complex reasoning tasks, exhibiting human-like patterns in exploring multiple alternative solutions.Upon closer inspection, however, we uncover a surprising phenomenon: The First is The Best, where alternative solutions are not merely suboptimal but potentially detrimental.This observation challenges widely accepted test-time scaling laws, leading us to hypothesize that errors within the reasoning path scale concurrently with test time.Through comprehensive empirical analysis, we characterize errors as a forest-structured Forest of Errors (FoE) and conclude that FoE makes the First the Best, which is underpinned by rigorous theoretical analysis.Leveraging these insights, we propose RED, a self-guided efficient reasoning framework comprising two components: I) Refining First, which suppresses FoE growth in the first solution; and II) Discarding Subs, which prunes subsequent FoE via dualconsistency.Extensive experiments across five benchmarks and six backbone models demonstrate that RED outperforms eight competitive baselines, achieving performance gains of up to 19.0% while reducing token consumption by 37.7% ∼ 70.4%.Moreover, comparative experiments on FoE metrics shed light on how RED achieves effectiveness.... ... ...

Large Language Models (LLMs) have recently achieved remarkable progress in mathematical reasoning. To enable such capabilities, many existing works distill strong reasoning models into long chains of thought or design algorithms to construct high-quality math question-answer (QA) data for training. However, these efforts primarily focus on generating correct reasoning paths and answers, while largely overlooking the correctness of the questions themselves. In this work, we present ValiMath, a benchmark consisting of 2147 human-verified mathematical questions covering a wide range of domains such as arithmetic, algebra, and geometry, which are synthesized and curated from the NuminaMath dataset. Each question is annotated with its logical structure, domain coverage, and question correctness, enabling fine-grained evaluation of question quality. ValiMath serves as a high-quality gold-standard test set for validating mathematical questions in LLM training corpora. Building upon this benchmark, we further propose MathQ-Verify, a pipeline that performs fine-grained parsing of mathematical questions into atomic assumptions and conclusions, and evaluates their semantic soundness through consistency checks. This pipeline achieves high precision in detecting flawed questions and provides a reliable foundation for cleaning noisy mathematical datasets. Experiments show that MathQ-Verify achieves state-of-the-art performance across multiple benchmarks, improving the F1 score by up to 25 percentage points over the direct verification baseline. MathQ-Verify offers a scalable and accurate solution for curating reliable mathematical datasets, reducing label noise and avoiding unnecessary computation on invalid questions. Our code and data are available at the repository https://github.com/OpenDCAI/MathQ-Verify.

Graph Neural Networks (GNNs) have shown remarkable capability in modeling complex relational data, yet their effectiveness is constrained by subtle but critical limitations in the message-passing paradigm. We identify a fundamental issue -- termed aggregation collapse -- arising from the aggregation step that combines representations of central and neighboring nodes with varying sizes. This bottleneck manifests in two forms of inherent information loss: (i) feature information loss, occurring when value distributions in learned representations become concentrated and indistinguishable after aggregation, and (ii) structure information loss, where variable node degrees are obscured by aggregation functions. To address these challenges, we propose complementary solutions: (i) nonlinear feature mapping and distribution re-scaler, which diversify input feature distributions prior to aggregation to preserve valuable feature-level details, and (ii) post-aggregation structural encoding, which retains essential structure information. Our methods are theoretically grounded, easy to implement, and broadly applicable to existing GNN models. Empirical evaluations demonstrate that our techniques successfully alleviate aggregation collapse, leading to improved performance and scalability in GNNs.

Pre-trained models have brought significant improvements to many NLP tasks and have been extensively analyzed. But little is known about the effect of fine-tuning on specific tasks. Intuitively, people may agree that a pre-trained model already learns semantic representations of words (e.g. synonyms are closer to each other) and fine-tuning further improves its capabilities which require more complicated reasoning (e.g. coreference resolution, entity boundary detection, etc). However, how to verify these arguments analytically and quantitatively is a challenging task and there are few works focus on this topic. In this paper, inspired by the observation that most probing tasks involve identifying matched pairs of phrases (e.g. coreference requires matching an entity and a pronoun), we propose a pairwise probe to understand BERT fine-tuning on the machine reading comprehension (MRC) task. Specifically, we identify five phenomena in MRC. According to pairwise probing tasks, we compare the performance of each layer's hidden representation of pre-trained and fine-tuned BERT. The proposed pairwise probe alleviates the problem of distraction from inaccurate model training and makes a robust and quantitative comparison. Our experimental analysis leads to highly confident conclusions: (1) Fine-tuning has little effect on the fundamental and low-level information and general semantic tasks. (2) For specific abilities required for downstream tasks, fine-tuned BERT is better than pre-trained BERT and such gaps are obvious after the fifth layer.

The E-learning services recommendation is essential in enabling precision instruction and personalized learning. In this paper, a new personalized E-learning services recommendation algorithm is proposed to solve the problem of low accuracy, recall and effectiveness. The algorithm builds user similarity matrix based on both user information data and user behavior data. In order to achieve the goal of bettering things, this paper creates an asymmetric similarity matrix based on the user learning ability and designs an E-learning services ranking strategy to make personalized E-learning service recommendation better. The application of the recommendation algorithm in the personalized E-learning platform of a software college shows that the new algorithm can improve the accuracy, recall and effectiveness compared with the traditional recommendation algorithm.

For each subject, the syllabus specifies what knowledge points students should master and how well they should master them. Whether students can acquire the ability required by the syllabus is an important evaluation criterion for a student's learning achievements. However, most of the exercises for k-12 students at present contain knowledge points in the examination syllabus, but the difficulty of the examination of knowledge points is inconsistent with the teaching syllabus. Solving and studying such exercises does not help students acquire the skills required by the syllabus quickly. Therefore, students should choose exercises that fit well with the syllabus when practicing. However, at present, there is no way to recommend exercises that fit well with the syllabus for students, so as to improve learning efficiency. In order to solve the above problems, this paper proposes an exercise recommendation method for k-12 students based on the syllabus. It is proved by experiment that this method can recommend exercises that fit well with the syllabus for students, so as to improve their learning efficiency. The use of this method can help students achieve better learning outcomes and achieve higher scores in the final examination than ordinary students under the condition of learning and mastering the same number of exercises with the same difficulty.

Students cheating in exams destroys the fair principle of evaluation and affects the normal teaching order of the school. Therefore, the examination cheating detection has the vital significance. The existing cheating detection methods have disadvantages such as insufficient modeling accuracy for students, lag in cognitive diagnosis, difficulty in detecting multi-source plagiarism and low accuracy. In order to solve the disadvantages of traditional methods, this paper proposes a method for detecting cheating in multi-index examinations based on feed-forward neural network. This paper first proposes RAE algorithm which combines linear regression and EM algorithm for students' cognitive diagnosis. We use RAE algorithm and LSTM neural network to obtain the knowledge point mastery degree of each student based on the history problem solving and the knowledge point mastery degree based on the exam problem solving. Then, according to the information of students' cognitive level, seat distribution in the examination room, students' habit of guessing answers at normal times, similarity of examination papers, etc., we get several indicators to judge whether students cheat. Finally, we take various indicators obtained through various methods as characteristics and use feed-forward neural network to classify whether students cheat or not. The experimental results show that the accuracy and recall of this method are significantly higher than those of several popular methods.