Continuous machine learning pipelines are common in industrial settings where
models are periodically trained on data streams. Unfortunately, concept drifts
may occur in data streams where the joint distribution of the data X and label
y, P(X, y), changes over time and possibly degrade model accuracy. Existing
concept drift adaptation approaches mostly focus on updating the model to the
new data possibly using ensemble techniques of previous models and tend to
discard the drifted historical data. However, we contend that explicitly
utilizing the drifted data together leads to much better model accuracy and
propose Quilt, a data-centric framework for identifying and selecting data
segments that maximize model accuracy. To address the potential downside of
efficiency, Quilt extends existing data subset selection techniques, which can
be used to reduce the training data without compromising model accuracy. These
techniques cannot be used as is because they only assume virtual drifts where
the posterior probabilities P(y|X) are assumed not to change. In contrast, a
key challenge in our setup is to also discard undesirable data segments with
concept drifts. Quilt thus discards drifted data segments and selects data
segment subsets holistically for accurate and efficient model training. The two
operations use gradient-based scores, which have little computation overhead.
In our experiments, we show that Quilt outperforms state-of-the-art drift
adaptation and data selection baselines on synthetic and real datasets.

通过使用 Quilt，这是一个数据中心的框架来鉴定和选择能够最大化模型准确性的数据片段，从而提高模型准确性和效率。

Quilt：面对概念漂移的鲁棒数据段选择

Quilt: Robust Data Segment Selection against Concept Drifts

Due to the advent and increase in the popularity of the Internet, people have
been producing and disseminating textual data in several ways, such as reviews,
social media posts, and news articles. As a result, numerous researchers have
been working on discovering patterns in textual data, especially because social
media posts function as social sensors, indicating peoples' opinions,
interests, etc. However, most tasks regarding natural language processing are
addressed using traditional machine learning methods and static datasets. This
setting can lead to several problems, such as an outdated dataset, which may
not correspond to reality, and an outdated model, which has its performance
degrading over time. Concept drift is another aspect that emphasizes these
issues, which corresponds to data distribution and pattern changes. In a text
stream scenario, it is even more challenging due to its characteristics, such
as the high speed and data arriving sequentially. In addition, models for this
type of scenario must adhere to the constraints mentioned above while learning
from the stream by storing texts for a limited time and consuming low memory.
In this study, we performed a systematic literature review regarding concept
drift adaptation in text stream scenarios. Considering well-defined criteria,
we selected 40 papers to unravel aspects such as text drift categories, types
of text drift detection, model update mechanism, the addressed stream mining
tasks, types of text representations, and text representation update mechanism.
In addition, we discussed drift visualization and simulation and listed
real-world datasets used in the selected papers. Therefore, this paper
comprehensively reviews the concept drift adaptation in text stream mining
scenarios.

这篇论文对文本流场景中的概念漂移适应进行了系统的文献综述，包括文本漂移类别、文本漂移检测类型、模型更新机制、流式挖掘任务、文本表示类型以及文本表示更新机制，并讨论了漂移可视化和模拟，列举了所选论文中使用的真实世界数据集。