• Publications
  • Influence
Interpretable & Time-Budget-Constrained Contextualization for Re-Ranking
TLDR
TK (Transformer-Kernel): a neural re-ranking model for ad-hoc search using an efficient contextualization mechanism that achieves the highest effectiveness in comparison to BERT and other re- ranking models is proposed. Expand
Improving Efficient Neural Ranking Models with Cross-Architecture Knowledge Distillation
TLDR
This work proposes a cross-architecture training procedure with a margin focused loss (Margin-MSE), that adapts knowledge distillation to the varying score output distributions of different BERT and non-BERT ranking architectures, and shows that across evaluated architectures it significantly improves their effectiveness without compromising their efficiency. Expand
Local Self-Attention over Long Text for Efficient Document Retrieval
TLDR
A local self-attention which considers a moving window over the document terms and for each term attends only to other terms in the same window resulting in increased retrieval of longer documents at moderate increase in compute and memory costs is proposed. Expand
Efficiently Teaching an Effective Dense Retriever with Balanced Topic Aware Sampling
TLDR
This work introduces an efficient topic-aware query and balanced margin sampling technique, called TAS-Balanced, and produces the first dense retriever that outperforms every other method on recall at any cutoff on TREC-DL and allows more resource intensive re-ranking models to operate on fewer passages to improve results further. Expand
Let's measure run time! Extending the IR replicability infrastructure to include performance aspects
TLDR
It is proposed to extend the OSIRRC docker-based replicability infrastructure with two performance focused benchmark scenarios and supply the argument with a case study exploring the performance of different neural re-ranking models. Expand
On the Effect of Low-Frequency Terms on Neural-IR Models
TLDR
This paper evaluates the neural IR models with various vocabulary sizes for their respective word embeddings, considering different levels of constraints on the available GPU memory, and investigates the use of subword-token embedding models, and in particular FastText, for Neural IR models. Expand
Conformer-Kernel with Query Term Independence for Document Retrieval
TLDR
It is demonstrated that incorporating explicit term matching signal into the model can be particularly useful in the full retrieval setting, and to reduce the memory complexity of the Transformer layers with respect to the input sequence length, a new Conformer layer is proposed. Expand
TU Wien @ TREC Deep Learning '19 - Simple Contextualization for Re-ranking
TLDR
The TK (Transformer-Kernel) model is submitted: a neural re-ranking model for ad-hoc search using an efficient contextualization mechanism and a document-length enhanced kernel-pooling, which enables users to gain insight into the model. Expand
Enriching Word Embeddings for Patent Retrieval with Global Context
TLDR
This study explores the use of word embeddings for patent retrieval, a challenging domain, especially for methods based on distributional semantics, and proposes a strategy to address this limitation by adapting the Skip-gram model’s vectors using global retrofitting and filtering word similarities using global context. Expand
Fine-Grained Relevance Annotations for Multi-Task Document Ranking and Question Answering
TLDR
This work extends the ranked retrieval annotations of the Deep Learning track of TREC 2019 with passage and word level graded relevance annotations for all relevant documents, and presents FiRA: a novel dataset of Fine-Grained Relevance Annotations. Expand
...
1
2
3
...