Refine
Year of publication
Institute
Document Type
- Conference Proceeding (34)
- Article (10)
- 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)
IASSE-2004 - 13th International Conference on Intelligent and Adaptive Systems and Software Engineering eds. W. Dosch, N. Debnath, pp. 245-250, ISCA, Cary, NC, 1-3 July 2004, Nice, France We introduce a UML-based model for conceptual design support in civil engineering. Therefore, we identify required extensions to standard UML. Class diagrams are used for elaborating building typespecific knowledge: Object diagrams, implicitly contained in the architect’s sketch, are validated against the defined knowledge. To enable the use of industrial, domain-specific tools, we provide an integrated conceptual design extension. The developed tool support is based on graph rewriting. With our approach architects are enabled to deal with semantic objects during early design phase, assisted by incremental consistency checks.
Der konstruktive Entwurf wird in derzeitigen CAD-Systemen gut unterstützt, nicht aber der konzeptuelle Gebäude-Entwurf. Dieser abstrahiert von konstruktiven Elementen wie Linie, Wand oder Decke, um auf die Konzepte, d.h. die eigentlichen Funktionen, heraus zu arbeiten. Diese abstraktere, funktionale Sichtweise auf ein Gebäude ist während der frühen Entwurfsphase essentiell, um Struktur und Organisation des gesamten Gebäudes zu erfassen. Bereits in dieser Phase muss Fachwissen (z. B. rechtliche, ökonomische und technische Bestimmungen) berücksichtigt werden. Im Rahmen des vorliegenden Projekts werden Software-Werkzeuge integriert in industrielle CAD-Systeme entwickelt, die den konzeptuellen Gebäude-Entwurf ermöglichen und diesen gegen Fachwissen prüfen. Das Projekt ist in zwei Teile gegliedert. Im Top-Down-Ansatz werden Datenstrukturen und Methoden zur Strukturierung, Repräsentation und Evaluation von gebäudespezifischem Fachwissen erarbeitet. Dieser Teil baut auf den graphbasierten Werkzeugen PROGRES und UPGRADE des Lehrstuhls auf. Der Bottom-Up-Ansatz ist industriell orientiert und hat zum Ziel, das kommerzielle CAD-System ArchiCAD zu erweitern. Hierbei soll der frühe, konzeptuelle Gebäude-Entwurf in einem CAD-System ermöglicht werden. Der Entwurf kann darüber hinaus gegen das definierte Fachwissen geprüft werden. Im Rahmen des graphbasierten Top-Down-Ansatzes wurde zunächst eine neue Spezifikationsmethode für die Sprache PROGRES entwickelt. Das PROGRES-System erlaubt die Spezifikation von Werkzeugen in deklarativer Form. Üblicherweise wird domänenspezifisches Fachwissen in der PROGRES-Spezifikation codiert, das daraus generierte visuelle Werkzeug stellt dann die entsprechende Funktionalität zur Verfügung. Mit dieser Methode sind am Lehrstuhl für Informatik III Werkzeuge für verschie-dene Anwendungsdomänen entstanden. In unserem Fall versetzen wir einen Domänen-Experten, z. B. einen erfahrenen Architekten, in die Lage, Fachwissen zur Laufzeit einzugeben, dieses zu evaluieren, abzuändern oder zu ergänzen. Im Rahmen der bisherigen Arbeit wurde dazu eine parametrisierte PROGRES-Spezifikation und zwei darauf aufbauende Werkzeuge entwickelt, welche die dynamische Eingabe von gebäude-technisch relevantem Fachwissen erlauben und einen graphbasierten, konzeptuellen Gebäude-Entwurf ermöglichen. In diesem konzeptuellen Gebäude-Entwurf wird von Raumgrößen und Positionen abstrahiert, um die funktionale Struktur eines Gebäudes zu beschreiben. Das Fachwissen kann von einem Architekten visuell definiert werden. Es können semantische Einheiten, im einfachsten Fall Räume, nach verschiedenen Kriterien kategorisiert und klassifiziert werden. Mit Hilfe von Attributen und Relationen können die semantischen Einheiten präziser beschrieben und in Beziehung zueinander gesetzt werden. Die in PROGRES spezifizierten Konsistenz-Analysen erlauben die Prüfung eines graphbasierten konzeptuellen Gebäude-Entwurfs gegen das dynamisch eingefügte Fachwissen. Im zweiten Teil des Forschungsprojekts, dem Bottom-Up-Ansatz, wird das CAD-System ArchiCAD erweitert, um den integrierten konzeptuellen Gebäude-Entwurf zu ermöglichen. Der Architekt erhält dazu neue Entwurfselemente, die Raumobjekte, welche die relevanten semantischen Einheiten während der frühen Entwurfsphase repräsentieren. Mit Hilfe der Raumobjekte kann der Architekt in ArchiCAD den Grundriss und das Raumprogramm eines Gebäudes entwerfen, ohne von konstruktiven Details in seiner Kreativität eingeschränkt zu werden. Die Arbeitsweise mit Raumobjekten entspricht dem informellen konzeptuellen Entwurf auf einer Papierskizze und ist daher für den Architekten intuitiv und einfach zu verwenden. Durch die Integration in ArchiCAD ergibt sich eine weitere Unterstützung: Das im Top-Down-Ansatz spezifizierte Fach-wissen wird verwendet, um den konzeptuellen Gebäude-Entwurf des Architekten auf Regelverletzungen zu überprüfen. Entwurfsfehler werden angezeigt. Zum Abschluss des konzeptuellen Gebäude-Entwurfs mit Raumobjekten wird durch ein weiteres neu entwickeltes Werkzeug eine initiale Wandstruktur automatisch erzeugt, die als Grundlage für die folgenden konstruktiven Entwurfsphasen dient. Alle beschriebenen Erwei-terungen sind in ArchiCAD integriert, sie sind für den Architekten daher leicht zu erlernen und einfach zu bedienen.
In: Computer Aided Architectural Design Futures 2005 2005, Part 4, 207-216, DOI: http://dx.doi.org/10.1007/1-4020-3698-1_19 The conceptual design at the beginning of the building construction process is essential for the success of a building project. Even if some CAD tools allow elaborating conceptual sketches, they rather focus on the shape of the building elements and not on their functionality. We introduce semantic roomobjects and roomlinks, by way of example to the CAD tool ArchiCAD. These extensions provide a basis for specifying the organisation and functionality of a building and free architects being forced to directly produce detailed constructive sketches. Furthermore, we introduce consistency analyses of the conceptual sketch, based on an ontology containing conceptual relevant knowledge, specific to one class of buildings.
Proc. of the 2005 ASCE Intl. Conf. on Computing in Civil Engineering (ICCC 2005) eds. L. Soibelman und F. Pena-Mora, Seite 1-14, ASCE (CD-ROM), Cancun, Mexico, 2005 Current CAD tools are not able to support the fundamental conceptual design phase, and none of them provides consistency analyses of sketches produced by architects. To give architects a greater support at the conceptual design phase, we develop a CAD tool for conceptual design and a knowledge specification tool allowing the definition of conceptually relevant knowledge. The knowledge is specific to one class of buildings and can be reused. Based on a dynamic knowledge model, different types of design rules formalize the knowledge in a graph-based realization. An expressive visual language provides a user-friendly, human readable representation. Finally, consistency analyses enable conceptual designs to be checked against this defined knowledge. In this paper we concentrate on the knowledge specification part of our project.
In: Forum Bauinformatik 2005 : junge Wissenschaftler forschen / [Lehrstuhl Bauinformatik, Brandenburgische Technische Universität Cottbus. Frank Schley ... (Hrsg.)]. - Cottbus : Techn. Universität 2005. S. 1-10 ISBN 3-934934-11-0
Mittels eines operationalen Ansatzes zur Semantikdefinition wird am Bei-spiel des konzeptuellen Gebäudeentwurfs ein Regelsystem formalisiert. Dazu werdenzwei Teile, zum einen das Regelwissen, zum anderen ein konzeptueller Entwurfsplan zunächst informell eingeführt und dann formal beschrieben. Darauf aufbauend wird die Grundlage für eine Konsistenzprüfung des konzeptuellen Entwurfs gegen das Regel-wissen formal angeben
In: Proceedings of the 39th Annual Hawaii International Conference on System Sciences, 2006. HICSS '06 http://dx.doi.org/10.1109/HICSS.2006.200 The conceptual design phase at the beginning of the building construction process is not adequately supported by any CAD-tool. Conceptual design support needs regarding two aspects: first, the architect must be able to develop conceptual sketches that provide abstraction from constructive details. Second, conceptually relevant knowledge should be available to check these conceptual sketches. The paper deals with knowledge to formalize for conceptual design. To enable domain experts formalizing knowledge, a graph-based specification is presented that allows the development of a domain ontology and design rules specific for one class of buildings at runtime. The provided tool support illustrates the introduced concepts and demonstrates the consistency analysis between knowledge and conceptual design.
In: Proc. of the 11th Intl. Conf. on Computing in Civil and Building Engineering (ICCCBE-XI) ed. Hugues Rivard, Montreal, Canada, Seite 1-12, ACSE (CD-ROM), 2006 Currently, the conceptual design phase is not adequately supported by any CAD tool. Neither the support while elaborating conceptual sketches, nor the automatic proof of correctness with respect to effective restrictions is currently provided by any commercial tool. To enable domain experts to store the common as well as their personal domain knowledge, we develop a visual language for knowledge formalization. In this paper, a major extension to the already existing concepts is introduced. The possibility to define rule dependencies extends the expressiveness of the knowledge definition language and contributes to the usability of our approach.
In: Advanced Engineering Informatics. Vol 21, Issue 1, 2007, Pages 67-83 http://dx.doi.org/10.1016/j.aei.2006.10.001 eds. J.C. Kunz, I.F.C. Smith and T. Tomiyama, Elsevier, Seite 1-22 Current CAD tools are not able to support the conceptual design phase, and none of them provides a consistency analysis for sketches produced by architects. This phase is fundamental and crucial for the whole design and construction process of a building. To give architects a better support, we developed a CAD tool for conceptual design and a knowledge specification tool. The knowledge is specific to one class of buildings and it can be reused. Based on a dynamic and domain-specific knowledge ontology, different types of design rules formalize this knowledge in a graph-based form. An expressive visual language provides a user-friendly, human readable representation. Finally, a consistency analysis tool enables conceptual designs to be checked against this formal conceptual knowledge. In this article, we concentrate on the knowledge specification part. For that, we introduce the concepts and usage of a novel visual language and describe its semantics. To demonstrate the usability of our approach, two graph-based visual tools for knowledge specification and conceptual design are explained.
Abstract of the authors: In many areas of computer science ontologies become more and more important. The use of ontologies for domain modeling often brings up the issue of ontology integration. The task of merging several ontologies, covering specific subdomains, into one united ontology has to be solved. Many approaches for ontology integration aim at automating the process of ontology alignment. However, a complete automation is not feasible, and user interaction is always required. Nevertheless, most ontology integration tools offer only very limited support for the interactive part of the integration process. In this paper, we present a novel approach for the interactive integration of ontologies. The result of the ontology integration is incrementally updated after each definition of a correspondence between ontology elements. The user is guided through the ontologies to be integrated. By restricting the possible user actions, the integrity of all defined correspondences is ensured by the tool we developed. We evaluated our tool by integrating different regulations concerning building design.
In: Advances in intelligent computing in engineering : proceedings of the 9.International EG-ICE Workshop ; Darmstadt, (01 - 03 August) 2002 / Martina Schnellenbach-Held ... (eds.) . - Düsseldorf: VDI-Verl., 2002 .- Fortschritt-Berichte VDI, Reihe 4, Bauingenieurwesen ; 180 ; S. 1-35 The paper describes a novel way to support conceptual design in civil engineering. The designer uses semantical tools guaranteeing certain internal structures of the design result but also the fulfillment of various constraints. Two different approaches and corresponding tools are discussed: (a) Visually specified tools with automatic code generation to determine a design structure as well as fixing various constraints a design has to obey. These tools are also valuable for design knowledge specialist. (b) Extensions of existing CAD tools to provide semantical knowledge to be used by an architect. It is sketched how these different tools can be combined in the future. The main part of the paper discusses the concepts and realization of two prototypes following the two above approaches. The paper especially discusses that specific graphs and the specification of their structure are useful for both tool realization projects.
WS GTaD-2003 - The 1st Workshop on Graph Transformations and Design ed Grabska, E., Seite 6-7, Jagiellonian University Krakow. 2 pages
Applications of Graph Transformations with Industrial Relevance Lecture Notes in Computer Science, 2004, Volume 3062/2004, 90-105, DOI: 10.1007/978-3-540-25959-6_7 In this paper we discuss how tools for conceptual design in civil engineering can be developed using graph transformation specifications. These tools consist of three parts: (a) for elaborating specific conceptual knowledge (knowledge engineer), (b) for working out conceptual design results (architect), and (c) automatic consistency analyses which guarantee that design results are consistent with the underlying specific conceptual knowledge. For the realization of such tools we use a machinery based on graph transformations. In a traditional PROGRES tool specification the conceptual knowledge for a class of buildings is hard-wired within the specification. This is not appropriate for the experimentation platform approach we present in this paper, as objects and relations for conceptual knowledge are due to many changes, implied by evaluation of their use and corresponding improvements. Therefore, we introduce a parametric specification method with the following characteristics: (1) The underlying specific knowledge for a class of buildings is not fixed. Instead, it is built up as a data base by using the knowledge tools. (2) The specification for the architect tools also does not incorporate specific conceptual knowledge. (3) An incremental checker guarantees whether a design result is consistent with the current state of the underlying conceptual knowledge (data base).
Applications of Graph Transformations with Industrial Relevance Lecture Notes in Computer Science, 2004, Volume 3062/2004, 434-439, DOI: http://dx.doi.org/10.1007/978-3-540-25959-6_33 This paper gives a brief overview of the tools we have developed to support conceptual design in civil engineering. Based on the UPGRADE framework, two applications, one for the knowledge engineer and another for architects allow to store domain specific knowledge and to use this knowledge during conceptual design. Consistency analyses check the design against the defined knowledge and inform the architect if rules are violated.
ITCE-2003 - 4th Joint Symposium on Information Technology in Civil Engineering ed Flood, I., Seite 1-12, ASCE (CD-ROM), Nashville, USA In this paper we discussed graph based tools to support architects during the conceptual design phase. Conceptual Design is defined before constructive design; the used concepts are more abstract. We develop two graph based approaches, a topdown using the graph rewriting system PROGRES and a more industrially oriented approach, where we extend the CAD system ArchiCAD. In both approaches, knowledge can be defined by a knowledge engineer, in the top-down approach in the domain model graph, in the bottom-up approach in the in an XML file. The defined knowledge is used to incrementally check the sketch and to inform the architect about violations of the defined knowledge. Our goal is to discover design error as soon as possible and to support the architect to design buildings with consideration of conceptual knowledge.
In: Net-distributed Co-operation : Xth International Conference on Computing in Civil and Building Engineering, Weimar, June 02 - 04, 2004 ; proceedings / [ed. by Karl Beuke ...] . - Weimar: Bauhaus-Univ. Weimar 2004. - 1. Aufl. . Seite 1-14 ISBN 3-86068-213-X International Conference on Computing in Civil and Building Engineering <10, 2004, Weimar> Summary In our project, we develop new tools for the conceptual design phase. During conceptual design, the coarse functionality and organization of a building is more important than a detailed worked out construction. We identify two roles, first the knowledge engineer who is responsible for knowledge definition and maintenance; second the architect who elaborates the conceptual de-sign. The tool for the knowledge engineer is based on graph technology, it is specified using PROGRES and the UPGRADE framework. The tools for the architect are integrated to the in-dustrial CAD tool ArchiCAD. Consistency between knowledge and conceptual design is en-sured by the constraint checker, another extension to ArchiCAD.
Agil ist im Trend und immer mehr Unternehmen, die ihre Projekte bisher nach klassischen Prinzipien durchführten, denken über den Einsatz agiler Methoden nach. Doch selbst wenn die Organisation bereits beide Philosophien unterstützt, gilt für ein Projekt meist die klare Vorgabe: agil oder klassisch. Es gibt aber noch einen anderen Ansatz, mit diesen "unterschiedlichen Welten" umzugehen: Und zwar die beiden Philosophien innerhalb eines Projekts zu kombinieren. Wie dies in der Praxis aussehen und gelingen kann, zeigen Dr. Michael Kirchhof und Prof. Dr. Bodo Kraft in diesem Beitrag.
This paper presents NLP Lean Programming
framework (NLPf), a new framework
for creating custom natural language processing
(NLP) models and pipelines by utilizing
common software development build systems.
This approach allows developers to train and
integrate domain-specific NLP pipelines into
their applications seamlessly. Additionally,
NLPf provides an annotation tool which improves
the annotation process significantly by
providing a well-designed GUI and sophisticated
way of using input devices. Due to
NLPf’s properties developers and domain experts
are able to build domain-specific NLP
applications more efficiently. NLPf is Opensource
software and available at https://
gitlab.com/schrieveslaach/NLPf.
Research collaborations provide opportunities for both practitioners and researchers: practitioners need solutions for difficult business challenges and researchers are looking for hard problems to solve and publish. Nevertheless, research collaborations carry the risk that practitioners focus on quick solutions too much and that researchers tackle theoretical problems, resulting in products which do not fulfill the project requirements.
In this paper we introduce an approach extending the ideas of agile and lean software development. It helps practitioners and researchers keep track of their common research collaboration goal: a scientifically enriched software product which fulfills the needs of the practitioner’s business model.
This approach gives first-class status to application-oriented metrics that measure progress and success of a research collaboration continuously. Those metrics are derived from the collaboration requirements and help to focus on a commonly defined goal.
An appropriate tool set evaluates and visualizes those metrics with minimal effort, and all participants will be pushed to focus on their tasks with appropriate effort. Thus project status, challenges and progress are transparent to all research collaboration members at any time.
Software Stories Guide
(2017)
Messenger apps like WhatsApp or Telegram are an integral part of daily communication. Besides the various positive effects, those services extend the operating range of criminals. Open trading groups with many thousand participants emerged on Telegram. Law enforcement agencies monitor suspicious users in such chat rooms. This research shows that text analysis, based on natural language processing, facilitates this through a meaningful domain overview and detailed investigations. We crawled a corpus from such self-proclaimed black markets and annotated five attribute types products, money, payment methods, user names, and locations. Based on each message a user sends, we extract and group these attributes to build profiles. Then, we build features to cluster the profiles. Pretrained word vectors yield better unsupervised clustering results than current
state-of-the-art transformer models. The result is a semantically meaningful high-level overview of the user landscape of black market chatrooms. Additionally, the extracted structured information serves as a foundation for further data exploration, for example, the most active users or preferred payment methods.
The integration of product data from heterogeneous sources and manufacturers into a single catalog is often still a laborious, manual task. Especially small- and medium-sized enterprises face the challenge of timely integrating the data their business relies on to have an up-to-date product catalog, due to format specifications, low quality of data and the requirement of expert knowledge. Additionally, modern approaches to simplify catalog integration demand experience in machine learning, word vectorization, or semantic similarity that such enterprises do not have. Furthermore, most approaches struggle with low-quality data. We propose Attribute Label Ranking (ALR), an easy to understand and simple to adapt learning approach. ALR leverages a model trained on real-world integration data to identify the best possible schema mapping of previously unknown, proprietary, tabular format into a standardized catalog schema. Our approach predicts multiple labels for every attribute of an inpu t column. The whole column is taken into consideration to rank among these labels. We evaluate ALR regarding the correctness of predictions and compare the results on real-world data to state-of-the-art approaches. Additionally, we report findings during experiments and limitations of our approach.
Multi-attribute relation extraction (MARE): simplifying the application of relation extraction
(2021)
Natural language understanding’s relation extraction makes innovative and encouraging novel business concepts possible and facilitates new digitilized decision-making processes. Current approaches allow the extraction of relations with a fixed number of entities as attributes. Extracting relations with an arbitrary amount of attributes requires complex systems and costly relation-trigger annotations to assist these systems. We introduce multi-attribute relation extraction (MARE) as an assumption-less problem formulation with two approaches, facilitating an explicit mapping from business use cases to the data annotations. Avoiding elaborated annotation constraints simplifies the application of relation extraction approaches. The evaluation compares our models to current state-of-the-art event extraction and binary relation extraction methods. Our approaches show improvement compared to these on the extraction of general multi-attribute relations.
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.