Generative AI and large language models (LLMs) offer significant potential for automating the extraction of structured information from web pages. In this work, we focus on food product pages from online retailers and explore schema-constrained extraction approaches to retrieve key product attributes, such as ingredient lists and nutrition tables. We compare two LLM-based approaches, direct extraction and indirect extraction via generated functions, evaluating them in terms of accuracy, efficiency, and cost on a curated dataset of 3,000 food product pages from three different online shops. Our results show that although the indirect approach achieves slightly lower accuracy (96.48\%, $-1.61\%$ compared to direct extraction), it reduces the number of required LLM calls by 95.82\%, leading to substantial efficiency gains and lower operational costs. These findings suggest that indirect extraction approaches can provide scalable and cost-effective solutions for large-scale information extraction tasks from template-based web pages using LLMs.

The enormous progress in the field of artificial intelligence (AI) enables retail companies to automate their processes and thus to save costs. Thereby, many AI-based automation approaches are based on machine learning and computer vision. The realization of such approaches requires high-quality training data. In this paper, we describe the creation process of an annotated dataset that contains 1,034 images of single food products, taken under studio conditions, annotated with 5 class labels and 30 object detection labels, which can be used for product recognition and classification tasks. We based all images and labels on standards presented by GS1, a global non-profit organisation. The objective of our work is to support the development of machine learning models in the retail domain and to provide a reference process for creating the necessary training data.

Due to the high volume of data and the increasing automation in retail, more and more companies are dealing with procedures to improve the quality of product data. A promising approach is the use of machine learning methods that support the user in master data management. The development of such procedures demands error-free training data. This means that product data must be cleaned and labelled which requires extensive data profiling. For typical retail company data bases with usually complex and convoluted structures this exploration step can take a huge and expensive amount of time. In order to speed up this process we present a reference model and best practices for the systematic and efficient profiling and exploration of product data.

We introduce a two-staged Genetic Algorithm for optimizing weighted random pattern testing in a Built-in-Self-Test (BIST) environment. The first stage includes the OBDD-based optimization of input probabilities with regard to the expected test length. The optimization itself is constrained to discrete weight values which can directly be integrated in a BIST environment. During the second stage, the hardware-design of the actual BIST-structure is optimized. Experimental results are given to demonstrate the quality of our approach.

Abstract| F ault simulation for synchronous sequential circuits is a very time-consuming task.The complexity of the task increases if there is no information about the initial state of the circuit.In this case an unknown initial state is assumed which is usually handled by i n troducing a three-valued logic.As it is well-known fault simulation based on this logic only determines a lower bound of the fault coverage.Recently it has been shown that fault simulation based on the multiple observation time test strategy can improve the accuracy of the fault coverage.In this paper we describe how this strategy can be successfully implemented based on Ordered Binary Decision Diagrams.Our experiments demonstrate the eciency of the fault simulation procedure developed.

Fault simulation for synchronous sequential circuits is a very time-consuming task. The complexity of the task increases if there is no information available about the initial state of the circuit. In this case, an unknown initial state is assumed which is usually handled by introducing a three-valued logic. It is known that fault simulation based upon this logic only determines a lower bound for the fault coverage achieved by a test sequence. Therefore, we developed a hybrid fault simulator H-FS combining the advantages of a fault simulator using the three-valued logic and of an exact symbolic fault simulator based upon binary decision diagrams. H-FS is able to handle even the largest benchmark circuits and thereby determines fault coverages much more accurately than previous algorithms using the three-valued logic.