BERT Rankers are Brittle: a Study using Adversarial Document Perturbations

• URL: http://arxiv.org/abs/2206.11724v1
• Date: Thu, 23 Jun 2022 14:16:48 GMT
• Title: BERT Rankers are Brittle: a Study using Adversarial Document Perturbations
• Authors: Yumeng Wang, Lijun Lyu, Avishek Anand
• Abstract summary: Contextual ranking models based on BERT are well established for a wide range of passage and document ranking tasks. We argue that BERT-rankers are not immune to adversarial attacks targeting retrieved documents given a query.
• Score: 3.6704226968275258
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Contextual ranking models based on BERT are now well established for a wide range of passage and document ranking tasks. However, the robustness of BERT-based ranking models under adversarial inputs is under-explored. In this paper, we argue that BERT-rankers are not immune to adversarial attacks targeting retrieved documents given a query. Firstly, we propose algorithms for adversarial perturbation of both highly relevant and non-relevant documents using gradient-based optimization methods. The aim of our algorithms is to add/replace a small number of tokens to a highly relevant or non-relevant document to cause a large rank demotion or promotion. Our experiments show that a small number of tokens can already result in a large change in the rank of a document. Moreover, we find that BERT-rankers heavily rely on the document start/head for relevance prediction, making the initial part of the document more susceptible to adversarial attacks. More interestingly, we find a small set of recurring adversarial words that when added to documents result in successful rank demotion/promotion of any relevant/non-relevant document respectively. Finally, our adversarial tokens also show particular topic preferences within and across datasets, exposing potential biases from BERT pre-training or downstream datasets.

Related papers

• PRISM: Fine-Grained Paper-to-Paper Retrieval with Multi-Aspect-Aware Query Optimization [61.783280234747394]
  PRISM is a document-to-document retrieval method that introduces multiple, fine-grained representations for both the query and candidate papers.<n>We present SciFullBench, a novel benchmark in which the complete and segmented context of full papers for both queries and candidates is available.<n>Experiments show that PRISM improves performance by an average of 4.3% over existing retrieval baselines.
  arXiv  Detail & Related papers  (2025-07-14T08:41:53Z)
• Multi-Facet Blending for Faceted Query-by-Example Retrieval [5.156059061769101]
  We propose a multi-facet blending (FaBle) augmentation method, which exploits modularity by decomposing and recomposing to explicitly synthesize facet-specific training sets. Our modularization eliminates the need for pre-defined facet knowledge or labels. FaBle augmentation on 1K documents remarkably assists training in obtaining facet conditional embeddings.
  arXiv  Detail & Related papers  (2024-12-02T12:32:19Z)
• Contextual Document Embeddings [77.22328616983417]
  We propose two complementary methods for contextualized document embeddings. First, an alternative contrastive learning objective that explicitly incorporates the document neighbors into the intra-batch contextual loss. Second, a new contextual architecture that explicitly encodes neighbor document information into the encoded representation.
  arXiv  Detail & Related papers  (2024-10-03T14:33:34Z)
• Top-Down Partitioning for Efficient List-Wise Ranking [24.600506147325717]
  We propose a novel algorithm that partitions a ranking to depth k and processes documents top-down. Our algorithm is inherently parallelizable due to the use of a pivot element, which can be compared to documents down to an arbitrary depth concurrently.
  arXiv  Detail & Related papers  (2024-05-23T14:00:26Z)
• Generating Natural Language Queries for More Effective Systematic Review Screening Prioritisation [53.77226503675752]
  The current state of the art uses the final title of the review as a query to rank the documents using BERT-based neural rankers. In this paper, we explore alternative sources of queries for prioritising screening, such as the Boolean query used to retrieve the documents to be screened and queries generated by instruction-based large-scale language models such as ChatGPT and Alpaca. Our best approach is not only viable based on the information available at the time of screening, but also has similar effectiveness to the final title.
  arXiv  Detail & Related papers  (2023-09-11T05:12:14Z)
• DAPR: A Benchmark on Document-Aware Passage Retrieval [57.45793782107218]
  We propose and name this task emphDocument-Aware Passage Retrieval (DAPR) While analyzing the errors of the State-of-The-Art (SoTA) passage retrievers, we find the major errors (53.5%) are due to missing document context. Our created benchmark enables future research on developing and comparing retrieval systems for the new task.
  arXiv  Detail & Related papers  (2023-05-23T10:39:57Z)
• Shuffle & Divide: Contrastive Learning for Long Text [6.187839874846451]
  We propose a self-supervised learning method for long text documents based on contrastive learning. A key to our method is Shuffle and Divide (SaD), a simple text augmentation algorithm. We have empirically evaluated our method by performing unsupervised text classification on the 20 Newsgroups, Reuters-21578, BBC, and BBCSport datasets.
  arXiv  Detail & Related papers  (2023-04-19T02:02:29Z)
• GERE: Generative Evidence Retrieval for Fact Verification [57.78768817972026]
  We propose GERE, the first system that retrieves evidences in a generative fashion. The experimental results on the FEVER dataset show that GERE achieves significant improvements over the state-of-the-art baselines.
  arXiv  Detail & Related papers  (2022-04-12T03:49:35Z)
• Augmenting Document Representations for Dense Retrieval with Interpolation and Perturbation [49.940525611640346]
  Document Augmentation for dense Retrieval (DAR) framework augments the representations of documents with their Dense Augmentation and perturbations. We validate the performance of DAR on retrieval tasks with two benchmark datasets, showing that the proposed DAR significantly outperforms relevant baselines on the dense retrieval of both the labeled and unlabeled documents.
  arXiv  Detail & Related papers  (2022-03-15T09:07:38Z)
• Multilevel Text Alignment with Cross-Document Attention [59.76351805607481]
  Existing alignment methods operate at a single, predefined level. We propose a new learning approach that equips previously established hierarchical attention encoders for representing documents with a cross-document attention component.
  arXiv  Detail & Related papers  (2020-10-03T02:52:28Z)
• Fine-Grained Relevance Annotations for Multi-Task Document Ranking and Question Answering [9.480648914353035]
  We present FiRA: a novel dataset of Fine-Grained Relevances. We extend the ranked retrieval annotations of the Deep Learning track of TREC 2019 with passage and word level graded relevance annotations for all relevant documents. As an example, we evaluate the recently introduced TKL document ranking model. We find that although TKL exhibits state-of-the-art retrieval results for long documents, it misses many relevant passages.
  arXiv  Detail & Related papers  (2020-08-12T14:59:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.