Review score prediction requires review text understanding, a critical real-world application of natural language processing. Due to dissimilar text domains in product reviews, a common practice is fine-tuning BERT models upon reviews of differing domains. However, there has not yet been an empirical study of cross-domain behaviors of BERT models in the various tasks of product review understanding. In this project, we investigate text classification BERT models fine-tuned on single-domain and multi-domain Amazon review data. In our findings, though single-domain models achieved marginally improved performance on their corresponding domain compared to multi-domain models, multi-domain models outperformed single-domain models when evaluated on multi-domain data, single-domain data the single-domain model was not fine-tuned on, and on average when considering all tests. Though slight increases in accuracy can be achieved through single-domain model fine-tuning, computational resources and costs can be reduced by utilizing multi-domain models that perform well across domains.
Language models (LMs) are being scaled and becoming powerful. Improving their efficiency is one of the core research topics in neural information processing systems. Tay et al. (2022) provided a comprehensive overview of efficient Transformers that have become an indispensable staple in the field of NLP. However, in the section of "On Evaluation", they left an open question "which fundamental efficient Transformer one should consider," answered by "still a mystery" because "many research papers select their own benchmarks." Unfortunately, there was not quantitative analysis about the performances of Transformers on any benchmarks. Moreover, state space models (SSMs) have demonstrated their abilities of modeling long-range sequences with non-attention mechanisms, which were not discussed in the prior review. This article makes a meta analysis on the results from a set of papers on efficient Transformers as well as those on SSMs. It provides a quantitative review on LM efficiency research and gives suggestions for future research.
In this paper, we propose a novel framework to pre-train language models for enhancing their abilities of comparative reasoning over texts. While recent research has developed models for NLP tasks that require comparative reasoning, they suffer from costly manual data labeling and limited generalizability to different tasks. Our approach involves a scalable method for collecting data for text-based entity comparison, which leverages both structured and unstructured data, and the design of three novel pre-training tasks. Evaluation on a range of downstream tasks including comparative question answering, question generation, and summarization shows that our pre-training framework significantly improves the comparative reasoning abilities of language models, especially under low-resource conditions. This work also releases the first integrated benchmark for comparative reasoning over texts.
Contrast consistency, the ability of a model to make consistently correct predictions in the presence of perturbations, is an essential aspect in NLP. While studied in tasks such as sentiment analysis and reading comprehension, it remains unexplored in open-domain question answering (OpenQA) due to the difficulty of collecting perturbed questions that satisfy factuality requirements. In this work, we collect minimally edited questions as challenging contrast sets to evaluate OpenQA models. Our collection approach combines both human annotation and large language model generation. We find that the widely used dense passage retriever (DPR) performs poorly on our contrast sets, despite fitting the training set well and performing competitively on standard test sets. To address this issue, we introduce a simple and effective query-side contrastive loss with the aid of data augmentation to improve DPR training. Our experiments on the contrast sets demonstrate that DPR's contrast consistency is improved without sacrificing its accuracy on the standard test sets.
Although counterfactual reasoning is a fundamental aspect of intelligence, the lack of large-scale counterfactual open-domain question-answering (QA) benchmarks makes it difficult to evaluate and improve models on this ability. To address this void, we introduce the first such dataset, named IfQA, where each question is based on a counterfactual presupposition via an "if" clause. For example, if Los Angeles was on the east coast of the U.S., what would be the time difference between Los Angeles and Paris? Such questions require models to go beyond retrieving direct factual knowledge from the Web: they must identify the right information to retrieve and reason about an imagined situation that may even go against the facts built into their parameters. The IfQA dataset contains over 3,800 questions that were annotated annotated by crowdworkers on relevant Wikipedia passages. Empirical analysis reveals that the IfQA dataset is highly challenging for existing open-domain QA methods, including supervised retrieve-then-read pipeline methods (EM score 36.2), as well as recent few-shot approaches such as chain-of-thought prompting with GPT-3 (EM score 27.4). The unique challenges posed by the IfQA benchmark will push open-domain QA research on both retrieval and counterfactual reasoning fronts.
Large language models (LLMs) exhibit remarkable performance across various NLP tasks. However, they often generate incorrect or hallucinated information, which hinders their practical applicability in real-world scenarios. Human feedback has been shown to effectively enhance the factuality and quality of generated content, addressing some of these limitations. However, this approach is resource-intensive, involving manual input and supervision, which can be time-consuming and expensive. Moreover, it cannot be provided during inference, further limiting its practical utility in dynamic and interactive applications. In this paper, we introduce ReFeed, a novel pipeline designed to enhance LLMs by providing automatic retrieval feedback in a plug-and-play framework without the need for expensive fine-tuning. ReFeed first generates initial outputs, then utilizes a retrieval model to acquire relevant information from large document collections, and finally incorporates the retrieved information into the in-context demonstration for output refinement, thereby addressing the limitations of LLMs in a more efficient and cost-effective manner. Experiments on four knowledge-intensive benchmark datasets demonstrate our proposed ReFeed could improve over +6.0% under zero-shot setting and +2.5% under few-shot setting, compared to baselines without using retrieval feedback.
Data imbalance is easily found in annotated data when the observations of certain continuous label values are difficult to collect for regression tasks. When they come to molecule and polymer property predictions, the annotated graph datasets are often small because labeling them requires expensive equipment and effort. To address the lack of examples of rare label values in graph regression tasks, we propose a semi-supervised framework to progressively balance training data and reduce model bias via self-training. The training data balance is achieved by (1) pseudo-labeling more graphs for under-represented labels with a novel regression confidence measurement and (2) augmenting graph examples in latent space for remaining rare labels after data balancing with pseudo-labels. The former is to identify quality examples from unlabeled data whose labels are confidently predicted and sample a subset of them with a reverse distribution from the imbalanced annotated data. The latter collaborates with the former to target a perfect balance using a novel label-anchored mixup algorithm. We perform experiments in seven regression tasks on graph datasets. Results demonstrate that the proposed framework significantly reduces the error of predicted graph properties, especially in under-represented label areas.
Document retrieval is a key stage of standard Web search engines. Existing dual-encoder dense retrievers obtain representations for questions and documents independently, allowing for only shallow interactions between them. To overcome this limitation, recent autoregressive search engines replace the dual-encoder architecture by directly generating identifiers for relevant documents in the candidate pool. However, the training cost of such autoregressive search engines rises sharply as the number of candidate documents increases. In this paper, we find that large language models (LLMs) can follow human instructions to directly generate URLs for document retrieval. Surprisingly, when providing a few {Query-URL} pairs as in-context demonstrations, LLMs can generate Web URLs where nearly 90\% of the corresponding documents contain correct answers to open-domain questions. In this way, LLMs can be thought of as built-in search engines, since they have not been explicitly trained to map questions to document identifiers. Experiments demonstrate that our method can consistently achieve better retrieval performance than existing retrieval approaches by a significant margin on three open-domain question answering benchmarks, under both zero and few-shot settings. The code for this work can be found at \url{https://github.com/Ziems/llm-url}.
Graph property prediction tasks are important and numerous. While each task offers a small size of labeled examples, unlabeled graphs have been collected from various sources and at a large scale. A conventional approach is training a model with the unlabeled graphs on self-supervised tasks and then fine-tuning the model on the prediction tasks. However, the self-supervised task knowledge could not be aligned or sometimes conflicted with what the predictions needed. In this paper, we propose to extract the knowledge underlying the large set of unlabeled graphs as a specific set of useful data points to augment each property prediction model. We use a diffusion model to fully utilize the unlabeled graphs and design two new objectives to guide the model's denoising process with each task's labeled data to generate task-specific graph examples and their labels. Experiments demonstrate that our data-centric approach performs significantly better than fourteen existing various methods on fifteen tasks. The performance improvement brought by unlabeled data is visible as the generated labeled examples unlike self-supervised learning.
Text data mining is the process of deriving essential information from language text. Typical text mining tasks include text categorization, text clustering, topic modeling, information extraction, and text summarization. Various data sets are collected and various algorithms are designed for the different types of tasks. In this paper, I present a blue sky idea that very large language model (VLLM) will become an effective unified methodology of text mining. I discuss at least three advantages of this new methodology against conventional methods. Finally I discuss the challenges in the design and development of VLLM techniques for text mining.