Predictive models in natural language processing (NLP) have evolved from training models from scratch to fine-tuning pre-trained models with labelled data. An extreme form of this fine-tuning involves in-context learning (ICL), where the output of a pre-trained generative model (frozen decoder parameters) is controlled only with variations in the input strings (called instructions or prompts). An important component of ICL is the use of a small number of labelled data instances as examples in the prompt. While existing work uses a static number of examples during inference for each data instance, in this paper we propose a novel methodology of dynamically adapting the number of examples as per the data. This is analogous to the use of a variable-sized neighborhood in k-nearest neighbors (k-NN) classifier. In our proposed workflow of adaptive ICL (AICL), the number of demonstrations to employ during the inference on a particular data instance is predicted by the Softmax posteriors of a classifier. The parameters of this classifier are fitted on the optimal number of examples in ICL required to correctly infer the label of each instance in the training set with the hypothesis that a test instance that is similar to a training instance should use the same (or a closely matching) number of few-shot examples. Our experiments show that our AICL method results in improvement in text classification task on several standard datasets.
Information retrieval models have witnessed a paradigm shift from unsupervised statistical approaches to feature-based supervised approaches to completely data-driven ones that make use of the pre-training of large language models. While the increasing complexity of the search models have been able to demonstrate improvements in effectiveness (measured in terms of relevance of top-retrieved results), a question worthy of a thorough inspection is - "how explainable are these models?", which is what this paper aims to evaluate. In particular, we propose a common evaluation platform to systematically evaluate the explainability of any ranking model (the explanation algorithm being identical for all the models that are to be evaluated). In our proposed framework, each model, in addition to returning a ranked list of documents, also requires to return a list of explanation units or rationales for each document. This meta-information from each document is then used to measure how locally consistent these rationales are as an intrinsic measure of interpretability - one that does not require manual relevance assessments. Additionally, as an extrinsic measure, we compute how relevant these rationales are by leveraging sub-document level relevance assessments. Our findings show a number of interesting observations, such as sentence-level rationales are more consistent, an increase in complexity mostly leads to less consistent explanations, and that interpretability measures offer a complementary dimension of evaluation of IR systems because consistency is not well-correlated with nDCG at top ranks.
Pseudo-relevance feedback (PRF) can enhance average retrieval effectiveness over a sufficiently large number of queries. However, PRF often introduces a drift into the original information need, thus hurting the retrieval effectiveness of several queries. While a selective application of PRF can potentially alleviate this issue, previous approaches have largely relied on unsupervised or feature-based learning to determine whether a query should be expanded. In contrast, we revisit the problem of selective PRF from a deep learning perspective, presenting a model that is entirely data-driven and trained in an end-to-end manner. The proposed model leverages a transformer-based bi-encoder architecture. Additionally, to further improve retrieval effectiveness with this selective PRF approach, we make use of the model's confidence estimates to combine the information from the original and expanded queries. In our experiments, we apply this selective feedback on a number of different combinations of ranking and feedback models, and show that our proposed approach consistently improves retrieval effectiveness for both sparse and dense ranking models, with the feedback models being either sparse, dense or generative.
The recent success in language generation capabilities of large language models (LLMs), such as GPT, Bard, Llama etc., can potentially lead to concerns about their possible misuse in inducing mass agitation and communal hatred via generating fake news and spreading misinformation. Traditional means of developing a misinformation ground-truth dataset does not scale well because of the extensive manual effort required to annotate the data. In this paper, we propose an LLM-based approach of creating silver-standard ground-truth datasets for identifying misinformation. Specifically speaking, given a trusted news article, our proposed approach involves prompting LLMs to automatically generate a summarised version of the original article. The prompts in our proposed approach act as a controlling mechanism to generate specific types of factual incorrectness in the generated summaries, e.g., incorrect quantities, false attributions etc. To investigate the usefulness of this dataset, we conduct a set of experiments where we train a range of supervised models for the task of misinformation detection.
Research in scientific disciplines evolves, often rapidly, over time with the emergence of novel methodologies and their associated terminologies. While methodologies themselves being conceptual in nature and rather difficult to automatically extract and characterise, in this paper, we seek to develop supervised models for automatic extraction of the names of the various constituents of a methodology, e.g., `R-CNN', `ELMo' etc. The main research challenge for this task is effectively modeling the contexts around these methodology component names in a few-shot or even a zero-shot setting. The main contributions of this paper towards effectively identifying new evolving scientific methodology names are as follows: i) we propose a factored approach to sequence modeling, which leverages a broad-level category information of methodology domains, e.g., `NLP', `RL' etc.; ii) to demonstrate the feasibility of our proposed approach of identifying methodology component names under a practical setting of fast evolving AI literature, we conduct experiments following a simulated chronological setup (newer methodologies not seen during the training process); iii) our experiments demonstrate that the factored approach outperforms state-of-the-art baselines by margins of up to 9.257\% for the methodology extraction task with the few-shot setup.
Neural retrieval models (NRMs) have been shown to outperform their statistical counterparts owing to their ability to capture semantic meaning via dense document representations. These models, however, suffer from poor interpretability as they do not rely on explicit term matching. As a form of local per-query explanations, we introduce the notion of equivalent queries that are generated by maximizing the similarity between the NRM's results and the result set of a sparse retrieval system with the equivalent query. We then compare this approach with existing methods such as RM3-based query expansion and contrast differences in retrieval effectiveness and in the terms generated by each approach.
Due to the massive size of test collections, a standard practice in IR evaluation is to construct a 'pool' of candidate relevant documents comprised of the top-k documents retrieved by a wide range of different retrieval systems - a process called depth-k pooling. A standard practice is to set the depth (k) to a constant value for each query constituting the benchmark set. However, in this paper we argue that the annotation effort can be substantially reduced if the depth of the pool is made a variable quantity for each query, the rationale being that the number of documents relevant to the information need can widely vary across queries. Our hypothesis is that a lower depth for the former class of queries and a higher depth for the latter can potentially reduce the annotation effort without a significant change in retrieval effectiveness evaluation. We make use of standard query performance prediction (QPP) techniques to estimate the number of potentially relevant documents for each query, which is then used to determine the depth of the pool. Our experiments conducted on standard test collections demonstrate that this proposed method of employing query-specific variable depths is able to adequately reflect the relative effectiveness of IR systems with a substantially smaller annotation effort.
Despite the retrieval effectiveness of queries being mutually independent of one another, the evaluation of query performance prediction (QPP) systems has been carried out by measuring rank correlation over an entire set of queries. Such a listwise approach has a number of disadvantages, notably that it does not support the common requirement of assessing QPP for individual queries. In this paper, we propose a pointwise QPP framework that allows us to evaluate the quality of a QPP system for individual queries by measuring the deviations between each prediction versus the corresponding true value, and then aggregating the results over a set of queries. Our experiments demonstrate that this new approach leads to smaller variances in QPP evaluations across a range of different target metrics and retrieval models.
Motivated by the recent success of end-to-end deep neural models for ranking tasks, we present here a supervised end-to-end neural approach for query performance prediction (QPP). In contrast to unsupervised approaches that rely on various statistics of document score distributions, our approach is entirely data-driven. Further, in contrast to weakly supervised approaches, our method also does not rely on the outputs from different QPP estimators. In particular, our model leverages information from the semantic interactions between the terms of a query and those in the top-documents retrieved with it. The architecture of the model comprises multiple layers of 2D convolution filters followed by a feed-forward layer of parameters. Experiments on standard test collections demonstrate that our proposed supervised approach outperforms other state-of-the-art supervised and unsupervised approaches.
A query performance predictor estimates the retrieval effectiveness of an IR system for a given query. An important characteristic of QPP evaluation is that, since the ground truth retrieval effectiveness for QPP evaluation can be measured with different metrics, the ground truth itself is not absolute, which is in contrast to other retrieval tasks, such as that of ad-hoc retrieval. Motivated by this argument, the objective of this paper is to investigate how such variances in the ground truth for QPP evaluation can affect the outcomes of QPP experiments. We consider this not only in terms of the absolute values of the evaluation metrics being reported (e.g. Pearson's $r$, Kendall's $\tau$), but also with respect to the changes in the ranks of different QPP systems when ordered by the QPP metric scores. Our experiments reveal that the observed QPP outcomes can vary considerably, both in terms of the absolute evaluation metric values and also in terms of the relative system ranks. Through our analysis, we report the optimal combinations of QPP evaluation metric and experimental settings that are likely to lead to smaller variations in the observed results.