摘要: 从中文的角度, "摘要"就是把论文重要的内容摘录出来; 从英文的角度, Abstract 就是把重要的内容抽取出来. 它通常包括三个部分: 已有工作的评述, 本文工作的描述, 实验结果. 有些期刊明确要求按照 Background, contribution, experiments 三个方面描述.
逐句分析
我的规范包括 10 句:
- 问题及其重要性
本句可以说明问题所属的领域, 解释最重要的概念, 或者强调问题的重要性.
– 反例: Attribute reduction is an important issue in data mining.
分析: 只会说 important, 干巴巴的. 而且 important 没有任何特色, 任何论文都可以说自己的问题 important.
– 正例: Recommender systems guide their users in decisions related to personal opinions about items.
分析: 针对性强, 有营养. - 已有工作
本句可以描述该问题的流行解决方案. 如果说我们做研究是 “站在巨人的肩膀上”, 本句就描述巨人长什么样子. - 已有工作局限性
本句以 However 开头, 需要注意
– 指责不要太强烈, 要尊重别人的劳动;
– 本句干的事情是挖坑, 要保证自己的方法能把坑填上;
– 实在不行也可以不写本句, 但这样显得没有底气. - 本文工作
– 以 In this paper 开始
– 是题目的扩展
– 出现算法的缩写
– 可以超过 20 个单词, 这也是全文唯一可以超过 20 个单词的句子 - 本文方法的第 1 个技术/步骤/方面/优势/贡献
- 本文方法的第 2 个技术/步骤/方面/优势/贡献
- 本文方法的第 3 个技术/步骤/方面/优势/贡献
如果不能扯出 3 个方面, 论文的工作量就显得不够
这几句作用是填坑, 最好与前面挖的坑对应 - 实验设置
给出数据的领域、来源、数量. - 实验结果
提高了准确性、效率等等. - 提升
吹嘘一下论文的意义: 为该领域打开一扇门之类
例 1. Three-way active learning through clustering selection.
- In clustering-based active learning, the performance of the learner relies heavily on the quality of clustering results.
第 1 句: 问题的重要性 - Empirical studies have shown that different clustering techniques are applicable to different data.
第 2 句: 已有工作 - We propose the three-way active learning through clustering selection (TACS) algorithm to dynamically select the appropriate techniques in the learning process.
第 4 句: 本文内容 (注意并没有第 3 句) - The algorithm follows the coarse-to-fine scheme of granular computing coupled with three-way instance processing.
第 4’ 句: 主要内容的补充说明 - For label query, we select both representative instances with density peaks, and edge instances with the maximal total
distance.
第 5 句: 第 1 项技术 - For block partition, we revise six popular clustering techniques to accommodate the binary splitting.
第 6 句: 第 2 项技术 - For clustering evaluation, we define weighted entropy with 1-nearest-neighbor.
第 7 句: 第 3 项技术 - For insufficient labels, we design tree pruning techniques with the help of a block queue.
第 7’ 句: 第 4 项技术 - Experiments were undertaken on twelve UCI data sets.
第 8 句: 实验设置 - The results show that TACS is superior to single clustering technique based algorithms and other state-of-the-art active learning algorithms.
第 9 句: 实验结果
例 2. Test-cost-sensitive attribute reduction
- In many data mining and machine learning applications, there are two objectives in the task of classification: one is decreasing the test cost, the other is improving the classification accuracy.
理文编辑把两个句子合并, 导致比较长 - Most existing research work focuses on the latter, with attribute reduction serving as an optional pre-processing stage to remove redundant attributes.
没写 However, 但挖了坑 - In this paper, we point out that when tests must be undertaken in parallel, attribute reduction is mandatory in dealing with the former objective.
填坑 - With this in mind, we posit the minimal test cost reduct problem which constitutes a new, but more general, difficulty than the classical reduct problem.
本质上相当于 First - We also define three metrics to evaluate the performance of reduction algorithms from a statistical viewpoint.
本质上相当于 Second - A framework for a heuristic algorithm is proposed to deal with the new problem; specifically, an information gain-based
λ
\lambda
λ-weighted reduction algorithm is designed, where weights are decided by test costs and a non-positive exponent
λ
\lambda
λ, which is the only parameter set by the user.
本质上相当于 Third - The algorithm is tested with three representative test cost distributions on four UCI (University of California-Irvine) data sets.
实验设置 - Experimental results show that there is a trade-off while setting
λ
\lambda
λ, and a competition approach can improve the quality of the result significantly.
结果 - This study suggests potential application areas and new research trends concerning attribute reduction.
吹牛. 不过本文的引用次数确实达到了 300+
例 3. Three-way recommender systems based on random forest
- Recommender systems guide their users in decisions related to personal opinions about items.
- Most existing systems implicitly assume the recommender behavior as a binary classification.
- That is, the incoming item is either recommended or not.
虽然没有用 However, 但有那个意思. - In this paper, we propose a framework integrating three-way decision and random forest to build recommender systems.
- First, we consider both misclassification costs and teacher cost.
- Misclassification costs are paid for wrong recommendation behaviors, while teacher cost is paid to consult the user actively for her tendency.
- Second, with these costs, a three-way decision model
is built and rational settings of α ∗ \alpha^* α∗ and β ∗ \beta^* β∗ are computed. - Third, we build a random forest to compute the probability P P P that a user likes an item.
- Fourth,
α
∗
\alpha^*
α∗,
β
∗
\beta^*
β∗ and
P
P
P are employed for determining the recommender behavior to users.
总共 4 个方面 - The performance of the recommender is evaluated by the average cost.
- Experiments results on the well-known MovieLens dataset show that the threshold pair ( α ∗ , β ∗ ) (\alpha^*, \beta^*) (α∗,β∗) determined by three-way decision is optimal not only on the training set, but also on the testing set.
例 4. Multi-label active learning through serial–parallel neural networks
- Multi-label active learning is an extension of supervised learning with high-dimensional label spaces and interactive scenarios.
问题及其特点 - Its key issues include the exploitation of label correlations, handling of missing labels, and selection of query labels.
核心问题, 为自己的内容作准备 - Various techniques have been proposed for this purpose; however, there is still room for performance improvement.
However 的典型用法 - In this study, we propose a multi-label active learning through serial–parallel neural networks (MASP) algorithm with simple and effective mechanisms.
核心句, 同时给出算法的缩写 - For label correlations, the serial part of the network extracts features that are common to all the labels.
解决第一个问题的技术
5’ This mechanism is more effective than the explicit feature extraction or compressed
sensing methods.
吹嘘一下优点 - Regarding the missing labels, the network sets the corresponding losses to zero for
backpropagation.
解决第二个问题的技术. Regarding 起到和 For 同样的作用, 而且避免简单的重复
6’ Thus, it does not require label completions that may introduce additional errors.
吹嘘一下优点 - For label queries, the parallel part of the network provides independent pairwise predictions for each label.
解决第三个问题的技术
7’ Such pairwise predictions present appropriate information for computing label uncertainty.
吹嘘是一种习惯 - Three sets of experiments were conducted on 22 benchmark datasets using 14 popular algorithms for comparison.
实验设置 - The results show that our algorithm achieves state-of-the-art active learning performance.
实验结果
MASP 是至今 (2022年末) 为止我们包装得最八股的一篇论文.