52442 research outputs found
Sort by
Gendered beauty inequalities? A multiverse analysis of physical attractiveness, occupational gender-typicality and earnings in the German labour market
Get PDFA regularized variance-reduced modified extragradient method for stochastic hierarchical games
Get PDFEntity matching using deep neural networks: From discriminative pre-trained language models to generative large language models
Get PDFEntity matching is the task of identifying records that refer to the same entity across different datasets. It is a critical step in the data integration process. Supervised entity matching methods typically frame the problem as a binary classification task between record pairs. These methods require labeled record pairs, consisting of matches and non-matches, for training. Key challenges in entity matching include high heterogeneity among records referring to the same entity, scarcity of training data, and the continuous emergence of unseen entities in real-world applications.
This thesis introduces two novel benchmarks for product matching, created using semantically annotated product identifiers on the Web as distant supervision. These benchmarks, sourced from thousands of e-shops, are among the largest and most diverse publicly available product matching datasets. They enable a fine-grained evaluation of entity matching methods across different entity matching challenges.
The thesis presents two new neural approaches for entity matching based on pre-trained language models, which achieve state-of-the-art results on multiple benchmarks. Unlike existing methods, both approaches exploit entity group information alongside binary matching labels during training. The first method, JointBERT, employs a dual-objective fine-tuning strategy. The second method, R-SupCon, uses supervised contrastive learning and establishes new state-of-the-art results on multiple benchmarks, proving particularly effective on smaller training sets. In addition, the thesis explores the usefulness of multilingual Transformers for improving product matching performance in low-resource languages.
This work further investigates generative large language models for entity matching, comparing them with pre-trained language models. The investigations include an analysis of prompting techniques, such as zero-shot inference, in-context learning, and rule-based prompting, as well as fine-tuning for entity matching. The results highlight the potential of large language models to match or exceed the performance of fine-tuned pre-trained language models, while requiring no or minimal amounts of training data. Additionally, the experiments demonstrate better generalization to unseen entities compared to pre-trained language models.
The thesis also examines the explainability of matching decisions, introducing two methods for aggregating local explanations into global insights. The first method, based on LIME explanations, is broadly applicable to matching classifiers. The second method uses large language models to produce structured explanations that can be automatically parsed and aggregated. Finally, the thesis introduces a method for automating error analysis using large language models. This approach allows for the automatic generation of error classes, which can help data engineers in the process of improving entity matching pipelines
Perplexity-inspired metasearch-based alternatives to FAIR GPT: Open-source AI consultants for RDM
Get PDFFAIR GPT was recently proposed as a virtual consultant for research data management (RDM) designed to help researchers and institutions in making their data FAIR (Findable, Accessible, Interoperable, Reusable). To reduce hallucinations and improve accuracy for certain tasks, FAIR GPT uses external APIs (FAIR-Checker, FAIR Enough, TIB Terminology, and re3data) and uploaded RDM resources (Horizon 2020 guidelines and the awesome-RDM GitHub repository). Its functionalities include metadata enhancement, dataset organization, repository selection, FAIRness assessment, license recommendations, and generating documentation such as data management plans, README files, and codebooks.
However, FAIR GPT has limitations. It does not provide sources for its answers, which reduces transparency and trust in its outputs. As part of OpenAI's "Custom GPTs", FAIR GPT is not open source, which limits customization, and it lacks an API for integration into existing RDM workflows. Reliance on external cloud-based services leads to privacy concerns when dealing with sensitive (meta)data. These issues led us to explore alternative open-source solutions.
We specifically searched for open-source alternatives to Perplexity AI, a system known for its ability to provide citations for the information it retrieves. We identified three candidates available on GitHub: Perplexica, sensei, and farfalle. These tools use local instances of SearXNG to perform internet search, using the results as contextual input for large language models (LLMs). We modified each of these tools to focus specifically on RDM tasks, releasing the new versions on GitHub openly under the names FAIR-Perplexica, FAIR-sensei and FAIR-farfalle.
We conducted a comparative analysis of these open-source candidates against each other and FAIR GPT, including (but not limited to) the following criteria:
1. Provenance. Unlike FAIR GPT, all three tools provide clear links to the sources of their search results, which improves transparency and trust in their outputs.
2. Privacy. While these tools are designed to run locally, they also send requests to the internet for information retrieval, which leads to privacy concerns.
3. Up-to-dateness. FAIR GPT partly relies on a static knowledge base, which may become outdated. The other tools use internet searches that contain more up-to-date, RDM-specific information.
4. Customizability. The open-source nature of the new tools allows users to customize them according to their specific RDM needs, which contrasts with FAIR GPT.
5. Ease of installation and use. All tools are straightforward to install using Docker Compose, and they offer intuitive, user-friendly graphical interfaces.
6. Community support. Open-source tools benefit from upstream development and a community of contributors.
7. Accuracy and Completeness. Each tool's responses were evaluated for missing information and potential errors.
8. Performance. Due to the varying pre- and post-processing steps involved in each tool, their overall performance differs.
In this work, we introduce and compare the open-source solutions FAIR-Perplexica, FAIR-sensei, and FAIR-farfalle as alternatives to FAIR GPT. These tools are designed for users who prioritize transparency, customization, and control over their (meta)data workflows. However, these tools involve sending requests to the internet via metasearch engine SearXNG, which may lead to privacy challenges
Emotions and Keywords on 'The Long 2020': A view of pandemic reckonings from cultural studies
No full text
