Refine
Year of publication
Institute
Document Type
- Conference Proceeding (34)
- Article (11)
- Book (1)
- Part of a Book (1)
- Course Material (1)
- Other (1)
- Report (1)
Keywords
- CAD (15)
- civil engineering (14)
- Bauingenieurwesen (13)
- Natural language processing (4)
- Information extraction (3)
- Architektur (2)
- Clustering (2)
- architecture (2)
- Active learning (1)
- Architectural design (1)
The progress in natural language processing (NLP) research over the last years, offers novel business opportunities for companies, as automated user interaction or improved data analysis. Building sophisticated NLP applications requires dealing with modern machine learning (ML) technologies, which impedes enterprises from establishing successful NLP projects. Our experience in applied NLP research projects shows that the continuous integration of research prototypes in production-like environments with quality assurance builds trust in the software and shows convenience and usefulness regarding the business goal. We introduce STAMP 4 NLP as an iterative and incremental process model for developing NLP applications. With STAMP 4 NLP, we merge software engineering principles with best practices from data science. Instantiating our process model allows efficiently creating prototypes by utilizing templates, conventions, and implementations, enabling developers and data scientists to focus on the business goals. Due to our iterative-incremental approach, businesses can deploy an enhanced version of the prototype to their software environment after every iteration, maximizing potential business value and trust early and avoiding the cost of successful yet never deployed experiments.
The integration of frequently changing, volatile product data from different manufacturers into a single catalog is a significant challenge for small and medium-sized e-commerce companies. They rely on timely integrating product data to present them aggregated in an online shop without knowing format specifications, concept understanding of manufacturers, and data quality. Furthermore, format, concepts, and data quality may change at any time. Consequently, integrating product catalogs into a single standardized catalog is often a laborious manual task. Current strategies to streamline or automate catalog integration use techniques based on machine learning, word vectorization, or semantic similarity. However, most approaches struggle with low-quality or real-world data. We propose Attribute Label Ranking (ALR) as a recommendation engine to simplify the integration process of previously unknown, proprietary tabular format into a standardized catalog for practitioners. We evaluate ALR by focusing on the impact of different neural network architectures, language features, and semantic similarity. Additionally, we consider metrics for industrial application and present the impact of ALR in production and its limitations.
Often, research results from collaboration projects are not transferred into productive environments even though approaches are proven to work in demonstration prototypes. These demonstration prototypes are usually too fragile and error-prone to be transferred
easily into productive environments. A lot of additional work is required.
Inspired by the idea of an incremental delivery process, we introduce an architecture pattern, which combines the approach of Metrics Driven Research Collaboration with microservices for the ease of integration. It enables keeping track of project goals over the course of the collaboration while every party may focus on their expert skills: researchers may focus on complex algorithms,
practitioners may focus on their business goals.
Through the simplified integration (intermediate) research results can be introduced into a productive environment which enables
getting an early user feedback and allows for the early evaluation of different approaches. The practitioners’ business model benefits throughout the full project duration.
Messenger apps like WhatsApp and Telegram are frequently used for everyday communication, but they can also be utilized as a platform for illegal activity. Telegram allows public groups with up to 200.000 participants. Criminals use these public groups for trading illegal commodities and services, which becomes a concern for law enforcement agencies, who manually monitor suspicious activity in these chat rooms. This research demonstrates how natural language processing (NLP) can assist in analyzing these chat rooms, providing an explorative overview of the domain and facilitating purposeful analyses of user behavior. We provide a publicly available corpus of annotated text messages with entities and relations from four self-proclaimed black market chat rooms. Our pipeline approach aggregates the extracted product attributes from user messages to profiles and uses these with their sold products as features for clustering. The extracted structured information is the foundation for further data exploration, such as identifying the top vendors or fine-granular price analyses. Our evaluation shows that pretrained word vectors perform better for unsupervised clustering than state-of-the-art transformer models, while the latter is still superior for sequence labeling.
Supervised machine learning and deep learning require a large amount of labeled data, which data scientists obtain in a manual, and time-consuming annotation process. To mitigate this challenge, Active Learning (AL) proposes promising data points to annotators they annotate next instead of a subsequent or random sample. This method is supposed to save annotation effort while maintaining model performance.
However, practitioners face many AL strategies for different tasks and need an empirical basis to choose between them. Surveys categorize AL strategies into taxonomies without performance indications. Presentations of novel AL strategies compare the performance to a small subset of strategies. Our contribution addresses the empirical basis by introducing a reproducible active learning evaluation (ALE) framework for the comparative evaluation of AL strategies in NLP.
The framework allows the implementation of AL strategies with low effort and a fair data-driven comparison through defining and tracking experiment parameters (e.g., initial dataset size, number of data points per query step, and the budget). ALE helps practitioners to make more informed decisions, and researchers can focus on developing new, effective AL strategies and deriving best practices for specific use cases. With best practices, practitioners can lower their annotation costs. We present a case study to illustrate how to use the framework.
In recent years, the development of large pretrained language models, such as BERT and GPT, significantly improved information extraction systems on various tasks, including relation classification. State-of-the-art systems are highly accurate on scientific benchmarks. A lack of explainability is currently a complicating factor in many real-world applications. Comprehensible systems are necessary to prevent biased, counterintuitive, or harmful decisions.
We introduce semantic extents, a concept to analyze decision patterns for the relation classification task. Semantic extents are the most influential parts of texts concerning classification decisions. Our definition allows similar procedures to determine semantic extents for humans and models. We provide an annotation tool and a software framework to determine semantic extents for humans and models conveniently and reproducibly. Comparing both reveals that models tend to learn shortcut patterns from data. These patterns are hard to detect with current interpretability methods, such as input reductions. Our approach can help detect and eliminate spurious decision patterns during model development. Semantic extents can increase the reliability and security of natural language processing systems. Semantic extents are an essential step in enabling applications in critical areas like healthcare or finance. Moreover, our work opens new research directions for developing methods to explain deep learning models.
In collaborative research projects, both researchers and practitioners work together solving business-critical challenges. These projects often deal with ETL processes, in which humans extract information from non-machine-readable documents by hand. AI-based machine learning models can help to solve this problem.
Since machine learning approaches are not deterministic, their quality of output may decrease over time. This fact leads to an overall quality loss of the application which embeds machine learning models. Hence, the software qualities in development and production may differ.
Machine learning models are black boxes. That makes practitioners skeptical and increases the inhibition threshold for early productive use of research prototypes. Continuous monitoring of software quality in production offers an early response capability on quality loss and encourages the use of machine learning approaches. Furthermore, experts have to ensure that they integrate possible new inputs into the model training as quickly as possible.
In this paper, we introduce an architecture pattern with a reference implementation that extends the concept of Metrics Driven Research Collaboration with an automated software quality monitoring in productive use and a possibility to auto-generate new test data coming from processed documents in production.
Through automated monitoring of the software quality and auto-generated test data, this approach ensures that the software quality meets and keeps requested thresholds in productive use, even during further continuous deployment and changing input data.
Companies often build their businesses based on product information and therefore try to automate the process of information extraction (IE). Since the information source is usually heterogeneous and non-standardized, classic extract, transform, load techniques reach their limits. Hence, companies must implement the newest findings from research to tackle the challenges of process automation. They require a flexible and robust system that is extendable and ensures the optimal processing of the different document types. This paper provides a distributed microservice architecture pattern that enables the automated generation of IE pipelines. Since their optimal design is individual for each input document, the system ensures the ad-hoc generation of pipelines depending on specific document characteristics at runtime. Furthermore, it introduces the automated quality determination of each available pipeline and controls the integration of new microservices based on their impact on the business value. The introduced system enables fast prototyping of the newest approaches from research and supports companies in automating their IE processes. Based on the automated quality determination, it ensures that the generated pipelines always meet defined business requirements when they come into productive use.
We conducted a scoping review for active learning in the domain of natural language processing (NLP), which we summarize in accordance with the PRISMA-ScR guidelines as follows:
Objective: Identify active learning strategies that were proposed for entity recognition and their evaluation environments (datasets, metrics, hardware, execution time).
Design: We used Scopus and ACM as our search engines. We compared the results with two literature surveys to assess the search quality. We included peer-reviewed English publications introducing or comparing active learning strategies for entity recognition.
Results: We analyzed 62 relevant papers and identified 106 active learning strategies. We grouped them into three categories: exploitation-based (60x), exploration-based (14x), and hybrid strategies (32x). We found that all studies used the F1-score as an evaluation metric. Information about hardware (6x) and execution time (13x) was only occasionally included. The 62 papers used 57 different datasets to evaluate their respective strategies. Most datasets contained newspaper articles or biomedical/medical data. Our analysis revealed that 26 out of 57 datasets are publicly accessible.
Conclusion: Numerous active learning strategies have been identified, along with significant open questions that still need to be addressed. Researchers and practitioners face difficulties when making data-driven decisions about which active learning strategy to adopt. Conducting comprehensive empirical comparisons using the evaluation environment proposed in this study could help establish best practices in the domain.