Exploring the Effectiveness of Multi-stage Fine-tuning for Cross-encoder Re-rankers

State-of-the-art cross-encoders can be fine-tuned to be highly effective in passage re-ranking. The typical fine-tuning process of cross-encoders as re-rankers requires large amounts of manually labelled data, a contrastive learning objective, and a set of heuristically sampled negatives. An alternative recent approach for fine-tuning instead involves teaching the model to mimic the rankings of a highly effective large language model using a distillation objective. These fine-tuning strategies can be applied either individually, or in sequence. In this work, we systematically investigate the effectiveness of point-wise cross-encoders when fine-tuned independently in a single stage, or sequentially in two stages. Our experiments show that the effectiveness of point-wise cross-encoders fine-tuned using contrastive learning is indeed on par with that of models fine-tuned with multi-stage approaches. Code is available for reproduction at https://github.com/fpezzuti/multistage-finetuning.

ECLIPSE: Contrastive Dimension Importance Estimation with Pseudo-Irrelevance Feedback for Dense Retrieval

Recent advances in Information Retrieval have leveraged high-dimensional embedding spaces to improve the retrieval of relevant documents. Moreover, the Manifold Clustering Hypothesis suggests that despite these high-dimensional representations, documents relevant to a query reside on a lower-dimensional, query-dependent manifold. While this hypothesis has inspired new retrieval methods, existing approaches still face challenges in effectively separating non-relevant information from relevant signals. We propose a novel methodology that addresses these limitations by leveraging information from both relevant and non-relevant documents. Our method, ECLIPSE, computes a centroid based on irrelevant documents as a reference to estimate noisy dimensions present in relevant ones, enhancing retrieval performance. Extensive experiments on three in-domain and one out-of-domain benchmarks demonstrate an average improvement of up to 19.50% (resp. 22.35%) in mAP(AP) and 11.42% (resp. 13.10%) in nDCG@10 w.r.t. the DIME-based baseline (resp. the baseline using all dimensions). Our results pave the way for more robust, pseudo-irrelevance-based retrieval systems in future IR research.

Maybe you are looking for CroQS: Cross-modal Query Suggestion for Text-to-Image Retrieval

Query suggestion, a technique widely adopted in information retrieval, enhances system interactivity and the browsing experience of document collections. In cross-modal retrieval, many works have focused on retrieving relevant items from natural language queries, while few have explored query suggestion solutions. In this work, we address query suggestion in cross-modal retrieval, introducing a novel task that focuses on suggesting minimal textual modifications needed to explore visually consistent subsets of the collection, following the premise of ''Maybe you are looking for''. To facilitate the evaluation and development of methods, we present a tailored benchmark named CroQS. This dataset comprises initial queries, grouped result sets, and human-defined suggested queries for each group. We establish dedicated metrics to rigorously evaluate the performance of various methods on this task, measuring representativeness, cluster specificity, and similarity of the suggested queries to the original ones. Baseline methods from related fields, such as image captioning and content summarization, are adapted for this task to provide reference performance scores. Although relatively far from human performance, our experiments reveal that both LLM-based and captioning-based methods achieve competitive results on CroQS, improving the recall on cluster specificity by more than 115% and representativeness mAP by more than 52% with respect to the initial query. The dataset, the implementation of the baseline methods and the notebooks containing our experiments are available here: https://paciosoft.com/CroQS-benchmark/

Static Pruning in Dense Retrieval using Matrix Decomposition

In the era of dense retrieval, document indexing and retrieval is largely based on encoding models that transform text documents into embeddings. The efficiency of retrieval is directly proportional to the number of documents and the size of the embeddings. Recent studies have shown that it is possible to reduce embedding size without sacrificing - and in some cases improving - the retrieval effectiveness. However, the methods introduced by these studies are query-dependent, so they can't be applied offline and require additional computations during query processing, thus negatively impacting the retrieval efficiency. In this paper, we present a novel static pruning method for reducing the dimensionality of embeddings using Principal Components Analysis. This approach is query-independent and can be executed offline, leading to a significant boost in dense retrieval efficiency with a negligible impact on the system effectiveness. Our experiments show that our proposed method reduces the dimensionality of document representations by over 50% with up to a 5% reduction in NDCG@10, for different dense retrieval models.

A Theoretical Analysis of Recommendation Loss Functions under Negative Sampling

Recommender Systems (RSs) are pivotal in diverse domains such as e-commerce, music streaming, and social media. This paper conducts a comparative analysis of prevalent loss functions in RSs: Binary Cross-Entropy (BCE), Categorical Cross-Entropy (CCE), and Bayesian Personalized Ranking (BPR). Exploring the behaviour of these loss functions across varying negative sampling settings, we reveal that BPR and CCE are equivalent when one negative sample is used. Additionally, we demonstrate that all losses share a common global minimum. Evaluation of RSs mainly relies on ranking metrics known as Normalized Discounted Cumulative Gain (NDCG) and Mean Reciprocal Rank (MRR). We produce bounds of the different losses for negative sampling settings to establish a probabilistic lower bound for NDCG. We show that the BPR bound on NDCG is weaker than that of BCE, contradicting the common assumption that BPR is superior to BCE in RSs training. Experiments on five datasets and four models empirically support these theoretical findings. Our code is available at \url{https://anonymous.4open.science/r/recsys_losses} .

Efficient Inference of Sub-Item Id-based Sequential Recommendation Models with Millions of Items

Transformer-based recommender systems, such as BERT4Rec or SASRec, achieve state-of-the-art results in sequential recommendation. However, it is challenging to use these models in production environments with catalogues of millions of items: scaling Transformers beyond a few thousand items is problematic for several reasons, including high model memory consumption and slow inference. In this respect, RecJPQ is a state-of-the-art method of reducing the models' memory consumption; RecJPQ compresses item catalogues by decomposing item IDs into a small number of shared sub-item IDs. Despite reporting the reduction of memory consumption by a factor of up to 50x, the original RecJPQ paper did not report inference efficiency improvements over the baseline Transformer-based models. Upon analysing RecJPQ's scoring algorithm, we find that its efficiency is limited by its use of score accumulators for each item, which prevents parallelisation. In contrast, LightRec (a non-sequential method that uses a similar idea of sub-ids) reported large inference efficiency improvements using an algorithm we call PQTopK. We show that it is also possible to improve RecJPQ-based models' inference efficiency using the PQTopK algorithm. In particular, we speed up RecJPQ-enhanced SASRec by a factor of 4.5 x compared to the original SASRec's inference method and by a factor of 1.56 x compared to the method implemented in RecJPQ code on a large-scale Gowalla dataset with more than a million items. Further, using simulated data, we show that PQTopK remains efficient with catalogues of up to tens of millions of items, removing one of the last obstacles to using Transformer-based models in production environments with large catalogues.

A Reproducible Analysis of Sequential Recommender Systems

Sequential Recommender Systems (SRSs) have emerged as a highly efficient approach to recommendation systems. By leveraging sequential data, SRSs can identify temporal patterns in user behaviour, significantly improving recommendation accuracy and relevance.Ensuring the reproducibility of these models is paramount for advancing research and facilitating comparisons between them. Existing works exhibit shortcomings in reproducibility and replicability of results, leading to inconsistent statements across papers. Our work fills these gaps by standardising data pre-processing and model implementations, providing a comprehensive code resource, including a framework for developing SRSs and establishing a foundation for consistent and reproducible experimentation. We conduct extensive experiments on several benchmark datasets, comparing various SRSs implemented in our resource. We challenge prevailing performance benchmarks, offering new insights into the SR domain. For instance, SASRec does not consistently outperform GRU4Rec. On the contrary, when the number of model parameters becomes substantial, SASRec starts to clearly dominate all the other SRSs. This discrepancy underscores the significant impact that experimental configuration has on the outcomes and the importance of setting it up to ensure precise and comprehensive results. Failure to do so can lead to significantly flawed conclusions, highlighting the need for rigorous experimental design and analysis in SRS research. Our code is available at https://github.com/antoniopurificato/recsys_repro_conf.

A Tale of Trust and Accuracy: Base vs. Instruct LLMs in RAG Systems

Retrieval Augmented Generation (RAG) represents a significant advancement in artificial intelligence combining a retrieval phase with a generative phase, with the latter typically being powered by large language models (LLMs). The current common practices in RAG involve using "instructed" LLMs, which are fine-tuned with supervised training to enhance their ability to follow instructions and are aligned with human preferences using state-of-the-art techniques. Contrary to popular belief, our study demonstrates that base models outperform their instructed counterparts in RAG tasks by 20% on average under our experimental settings. This finding challenges the prevailing assumptions about the superiority of instructed LLMs in RAG applications. Further investigations reveal a more nuanced situation, questioning fundamental aspects of RAG and suggesting the need for broader discussions on the topic; or, as Fromm would have it, "Seldom is a glance at the statistics enough to understand the meaning of the figures".

Graph Neural Re-Ranking via Corpus Graph

Re-ranking systems aim to reorder an initial list of documents to satisfy better the information needs associated with a user-provided query. Modern re-rankers predominantly rely on neural network models, which have proven highly effective in representing samples from various modalities. However, these models typically evaluate query-document pairs in isolation, neglecting the underlying document distribution that could enhance the quality of the re-ranked list. To address this limitation, we propose Graph Neural Re-Ranking (GNRR), a pipeline based on Graph Neural Networks (GNNs), that enables each query to consider documents distribution during inference. Our approach models document relationships through corpus subgraphs and encodes their representations using GNNs. Through extensive experiments, we demonstrate that GNNs effectively capture cross-document interactions, improving performance on popular ranking metrics. In TREC-DL19, we observe a relative improvement of 5.8% in Average Precision compared to our baseline. These findings suggest that integrating the GNN segment offers significant advantages, especially in scenarios where understanding the broader context of documents is crucial.

Generating Query Recommendations via LLMs

Query recommendation systems are ubiquitous in modern search engines, assisting users in producing effective queries to meet their information needs. However, these systems require a large amount of data to produce good recommendations, such as a large collection of documents to index and query logs. In particular, query logs and user data are not available in cold start scenarios. Query logs are expensive to collect and maintain and require complex and time-consuming cascading pipelines for creating, combining, and ranking recommendations. To address these issues, we frame the query recommendation problem as a generative task, proposing a novel approach called Generative Query Recommendation (GQR). GQR uses an LLM as its foundation and does not require to be trained or fine-tuned to tackle the query recommendation problem. We design a prompt that enables the LLM to understand the specific recommendation task, even using a single example. We then improved our system by proposing a version that exploits query logs called Retriever-Augmented GQR (RA-GQR). RA-GQr dynamically composes its prompt by retrieving similar queries from query logs. GQR approaches reuses a pre-existing neural architecture resulting in a simpler and more ready-to-market approach, even in a cold start scenario. Our proposed GQR obtains state-of-the-art performance in terms of NDCG@10 and clarity score against two commercial search engines and the previous state-of-the-art approach on the Robust04 and ClueWeb09B collections, improving on average the NDCG@10 performance up to ~4% on Robust04 and ClueWeb09B w.r.t the previous best competitor. RA-GQR further improve the NDCG@10 obtaining an increase of ~11%, ~6\% on Robust04 and ClueWeb09B w.r.t the best competitor. Furthermore, our system obtained ~59% of user preferences in a blind user study, proving that our method produces the most engaging queries.