Systematic Evaluation of Neural Retrieval Models on the Touché 2020 Argument Retrieval Subset of BEIR

The zero-shot effectiveness of neural retrieval models is often evaluated on the BEIR benchmark -- a combination of different IR evaluation datasets. Interestingly, previous studies found that particularly on the BEIR subset Touch\'e 2020, an argument retrieval task, neural retrieval models are considerably less effective than BM25. Still, so far, no further investigation has been conducted on what makes argument retrieval so "special". To more deeply analyze the respective potential limits of neural retrieval models, we run a reproducibility study on the Touch\'e 2020 data. In our study, we focus on two experiments: (i) a black-box evaluation (i.e., no model retraining), incorporating a theoretical exploration using retrieval axioms, and (ii) a data denoising evaluation involving post-hoc relevance judgments. Our black-box evaluation reveals an inherent bias of neural models towards retrieving short passages from the Touch\'e 2020 data, and we also find that quite a few of the neural models' results are unjudged in the Touch\'e 2020 data. As many of the short Touch\'e passages are not argumentative and thus non-relevant per se, and as the missing judgments complicate fair comparison, we denoise the Touch\'e 2020 data by excluding very short passages (less than 20 words) and by augmenting the unjudged data with post-hoc judgments following the Touch\'e guidelines. On the denoised data, the effectiveness of the neural models improves by up to 0.52 in nDCG@10, but BM25 is still more effective. Our code and the augmented Touch\'e 2020 dataset are available at \url{https://github.com/castorini/touche-error-analysis}.

A Systematic Investigation of Distilling Large Language Models into Cross-Encoders for Passage Re-ranking

Cross-encoders distilled from large language models are more effective re-rankers than cross-encoders fine-tuned using manually labeled data. However, the distilled models do not reach the language model's effectiveness. We construct and release a new distillation dataset, named Rank-DistiLLM, to investigate whether insights from fine-tuning cross-encoders on manually labeled data -- hard-negative sampling, deep sampling, and listwise loss functions -- are transferable to large language model ranker distillation. Our dataset can be used to train cross-encoders that reach the effectiveness of large language models while being orders of magnitude more efficient. Code and data is available at: https://github.com/webis-de/msmarco-llm-distillation

Set-Encoder: Permutation-Invariant Inter-Passage Attention for Listwise Passage Re-Ranking with Cross-Encoders

Cross-encoders are effective passage re-rankers. But when re-ranking multiple passages at once, existing cross-encoders inefficiently optimize the output ranking over several input permutations, as their passage interactions are not permutation-invariant. Moreover, their high memory footprint constrains the number of passages during listwise training. To tackle these issues, we propose the Set-Encoder, a new cross-encoder architecture that (1) introduces inter-passage attention with parallel passage processing to ensure permutation invariance between input passages, and that (2) uses fused-attention kernels to enable training with more passages at a time. In experiments on TREC Deep Learning and TIREx, the Set-Encoder is more effective than previous cross-encoders with a similar number of parameters. Compared to larger models, the Set-Encoder is more efficient and either on par or even more effective.

Analyzing Adversarial Attacks on Sequence-to-Sequence Relevance Models

Modern sequence-to-sequence relevance models like monoT5 can effectively capture complex textual interactions between queries and documents through cross-encoding. However, the use of natural language tokens in prompts, such as Query, Document, and Relevant for monoT5, opens an attack vector for malicious documents to manipulate their relevance score through prompt injection, e.g., by adding target words such as true. Since such possibilities have not yet been considered in retrieval evaluation, we analyze the impact of query-independent prompt injection via manually constructed templates and LLM-based rewriting of documents on several existing relevance models. Our experiments on the TREC Deep Learning track show that adversarial documents can easily manipulate different sequence-to-sequence relevance models, while BM25 (as a typical lexical model) is not affected. Remarkably, the attacks also affect encoder-only relevance models (which do not rely on natural language prompt tokens), albeit to a lesser extent.

Investigating the Effects of Sparse Attention on Cross-Encoders

Cross-encoders are effective passage and document re-rankers but less efficient than other neural or classic retrieval models. A few previous studies have applied windowed self-attention to make cross-encoders more efficient. However, these studies did not investigate the potential and limits of different attention patterns or window sizes. We close this gap and systematically analyze how token interactions can be reduced without harming the re-ranking effectiveness. Experimenting with asymmetric attention and different window sizes, we find that the query tokens do not need to attend to the passage or document tokens for effective re-ranking and that very small window sizes suffice. In our experiments, even windows of 4 tokens still yield effectiveness on par with previous cross-encoders while reducing the memory requirements to at most 78% / 41% and being 1% / 43% faster at inference time for passages / documents.

Evaluating Generative Ad Hoc Information Retrieval

Recent advances in large language models have enabled the development of viable generative information retrieval systems. A generative retrieval system returns a grounded generated text in response to an information need instead of the traditional document ranking. Quantifying the utility of these types of responses is essential for evaluating generative retrieval systems. As the established evaluation methodology for ranking-based ad hoc retrieval may seem unsuitable for generative retrieval, new approaches for reliable, repeatable, and reproducible experimentation are required. In this paper, we survey the relevant information retrieval and natural language processing literature, identify search tasks and system architectures in generative retrieval, develop a corresponding user model, and study its operationalization. This theoretical analysis provides a foundation and new insights for the evaluation of generative ad hoc retrieval systems.

The Information Retrieval Experiment Platform

We integrate ir_datasets, ir_measures, and PyTerrier with TIRA in the Information Retrieval Experiment Platform (TIREx) to promote more standardized, reproducible, scalable, and even blinded retrieval experiments. Standardization is achieved when a retrieval approach implements PyTerrier's interfaces and the input and output of an experiment are compatible with ir_datasets and ir_measures. However, none of this is a must for reproducibility and scalability, as TIRA can run any dockerized software locally or remotely in a cloud-native execution environment. Version control and caching ensure efficient (re)execution. TIRA allows for blind evaluation when an experiment runs on a remote server or cloud not under the control of the experimenter. The test data and ground truth are then hidden from public access, and the retrieval software has to process them in a sandbox that prevents data leaks. We currently host an instance of TIREx with 15 corpora (1.9 billion documents) on which 32 shared retrieval tasks are based. Using Docker images of 50 standard retrieval approaches, we automatically evaluated all approaches on all tasks (50 $\cdot$ 32 = 1,600~runs) in less than a week on a midsize cluster (1,620 CPU cores and 24 GPUs). This instance of TIREx is open for submissions and will be integrated with the IR Anthology, as well as released open source.

The Archive Query Log: Mining Millions of Search Result Pages of Hundreds of Search Engines from 25 Years of Web Archives

The Archive Query Log (AQL) is a previously unused, comprehensive query log collected at the Internet Archive over the last 25 years. Its first version includes 356 million queries, 166 million search result pages, and 1.7 billion search results across 550 search providers. Although many query logs have been studied in the literature, the search providers that own them generally do not publish their logs to protect user privacy and vital business data. Of the few query logs publicly available, none combines size, scope, and diversity. The AQL is the first to do so, enabling research on new retrieval models and (diachronic) search engine analyses. Provided in a privacy-preserving manner, it promotes open research as well as more transparency and accountability in the search industry.

The Infinite Index: Information Retrieval on Generative Text-To-Image Models

The text-to-image model Stable Diffusion has recently become very popular. Only weeks after its open source release, millions are experimenting with image generation. This is due to its ease of use, since all it takes is a brief description of the desired image to "prompt" the generative model. Rarely do the images generated for a new prompt immediately meet the user's expectations. Usually, an iterative refinement of the prompt ("prompt engineering") is necessary for satisfying images. As a new perspective, we recast image prompt engineering as interactive image retrieval - on an "infinite index". Thereby, a prompt corresponds to a query and prompt engineering to query refinement. Selected image-prompt pairs allow direct relevance feedback, as the model can modify an image for the refined prompt. This is a form of one-sided interactive retrieval, where the initiative is on the user side, whereas the server side remains stateless. In light of an extensive literature review, we develop these parallels in detail and apply the findings to a case study of a creative search task on such a model. We note that the uncertainty in searching an infinite index is virtually never-ending. We also discuss future research opportunities related to retrieval models specialized for generative models and interactive generative image retrieval. The application of IR technology, such as query reformulation and relevance feedback, will contribute to improved workflows when using generative models, while the notion of an infinite index raises new challenges in IR research.

Sparse Pairwise Re-ranking with Pre-trained Transformers

Pairwise re-ranking models predict which of two documents is more relevant to a query and then aggregate a final ranking from such preferences. This is often more effective than pointwise re-ranking models that directly predict a relevance value for each document. However, the high inference overhead of pairwise models limits their practical application: usually, for a set of $k$ documents to be re-ranked, preferences for all $k^2-k$ comparison pairs excluding self-comparisons are aggregated. We investigate whether the efficiency of pairwise re-ranking can be improved by sampling from all pairs. In an exploratory study, we evaluate three sampling methods and five preference aggregation methods. The best combination allows for an order of magnitude fewer comparisons at an acceptable loss of retrieval effectiveness, while competitive effectiveness is already achieved with about one third of the comparisons.