Search

Large language models (LLMs) represented by GPT family have achieved remarkable success. The characteristics of LLMs lie in their ability to accommodate a wide range of tasks through a generative approach. However, the flexibility of their output format poses challenges in controlling and harnessing the model's outputs, thereby constraining the application of LLMs in various domains. In this work, we present Sketch, an innovative toolkit designed to streamline LLM operations across diverse fields. Sketch comprises the following components: (1) a suite of task description schemas and prompt templates encompassing various NLP tasks; (2) a user-friendly, interactive process for building structured output LLM services tailored to various NLP tasks; (3) an open-source dataset for output format control, along with tools for dataset construction; and (4) an open-source model based on LLaMA3-8B-Instruct that adeptly comprehends and adheres to output formatting instructions. We anticipate this initiative to bring considerable convenience to LLM users, achieving the goal of ''plug-and-play'' for various applications. The components of Sketch will be progressively open-sourced at https://github.com/cofe-ai/Sketch.

Short-video recommender systems often exhibit a biased preference to recently released videos. However, not all videos become outdated; certain classic videos can still attract user's attention. Such bias along temporal dimension can be further aggravated by the matching model between users and videos, because the model learns from preexisting interactions. From real data, we observe that different videos have varying sensitivities to recency in attracting users' attention. Our analysis, based on a causal graph modeling short-video recommendation, suggests that the release interval serves as a confounder, establishing a backdoor path between users and videos. To address this confounding effect, we propose a model-agnostic causal architecture called Learning to Deconfound the Release Interval Bias (LDRI). LDRI enables jointly learning of the matching model and the video recency sensitivity perceptron. In the inference stage, we apply a backdoor adjustment, effectively blocking the backdoor path by intervening on each video. Extensive experiments on two benchmarks demonstrate that LDRI consistently outperforms backbone models and exhibits superior performance against state-of-the-art models. Additional comprehensive analyses confirm the deconfounding capability of LDRI.

Controlling the format of outputs generated by large language models (LLMs) is a critical functionality in various applications. Current methods typically employ constrained decoding with rule-based automata or fine-tuning with manually crafted format instructions, both of which struggle with open-domain format requirements. To address this limitation, we introduce a novel framework for controlled generation in LLMs, leveraging user-provided, one-shot QA pairs. This study investigates LLMs' capabilities to follow open-domain, one-shot constraints and replicate the format of the example answers. We observe that this is a non-trivial problem for current LLMs. We also develop a dataset collection methodology for supervised fine-tuning that enhances the open-domain format control of LLMs without degrading output quality, as well as a benchmark on which we evaluate both the helpfulness and format correctness of LLM outputs. The resulting datasets, named OIFC-SFT, along with the related code, will be made publicly available at https://github.com/cofe-ai/OIFC., Comment: 6 pages

The increasing availability of learning activity data in Massive Open Online Courses (MOOCs) enables us to conduct a large-scale analysis of learners' learning behavior. In this paper, we analyze a dataset of 351 million learning activities from 0.8 million unique learners enrolled in over 1.6 thousand courses within two years. Specifically, we mine and identify the learning patterns of the crowd from both temporal and course enrollment perspectives leveraging mutual information theory and sequential pattern mining methods. From the temporal perspective, we find that the time intervals between consecutive learning activities of learners exhibit a mix of power-law and periodic cosine function distribution. By qualifying the relationship between course pairs, we observe that the most frequently co-enrolled courses usually fall in the same category or the same university. We demonstrate these findings can facilitate manifold applications including recommendation tasks on courses. A simple recommendation model utilizing the course enrollment patterns is competitive to the baselines with 200$\times$ faster training time., Comment: 16 pages

In this paper, we propose the task of \textit{Ranked Video Moment Retrieval} (RVMR) to locate a ranked list of matching moments from a collection of videos, through queries in natural language. Although a few related tasks have been proposed and studied by CV, NLP, and IR communities, RVMR is the task that best reflects the practical setting of moment search. To facilitate research in RVMR, we develop the TVR-Ranking dataset, based on the raw videos and existing moment annotations provided in the TVR dataset. Our key contribution is the manual annotation of relevance levels for 94,442 query-moment pairs. We then develop the $NDCG@K, IoU\geq \mu$ evaluation metric for this new task and conduct experiments to evaluate three baseline models. Our experiments show that the new RVMR task brings new challenges to existing models and we believe this new dataset contributes to the research on multi-modality search. The dataset is available at \url{https://github.com/Ranking-VMR/TVR-Ranking}

Large Language Models (LLMs) represent a significant stride toward Artificial General Intelligence. As scaling laws underscore the potential of increasing model sizes, the academic community has intensified its investigations into LLMs with capacities exceeding 50 billion parameters. This technical report builds on our prior work with Tele-FLM (also known as FLM-2), a publicly available 52-billion-parameter model. We delve into two primary areas: we first discuss our observation of Supervised Fine-tuning (SFT) on Tele-FLM-52B, which supports the "less is more" approach for SFT data construction; second, we demonstrate our experiments and analyses on the best practices for progressively growing a model from 52 billion to 102 billion, and subsequently to 1 trillion parameters. We will open-source a 1T model checkpoint, namely Tele-FLM-1T, to advance further training and research., Comment: For the Tele-FLM-52B tech report, see also 2404.16645

Understanding documents with rich layouts and multi-modal components is a long-standing and practical task. Recent Large Vision-Language Models (LVLMs) have made remarkable strides in various tasks, particularly in single-page document understanding (DU). However, their abilities on long-context DU remain an open problem. This work presents MMLongBench-Doc, a long-context, multi-modal benchmark comprising 1,062 expert-annotated questions. Distinct from previous datasets, it is constructed upon 130 lengthy PDF-formatted documents with an average of 49.4 pages and 20,971 textual tokens. Towards comprehensive evaluation, answers to these questions rely on pieces of evidence from (1) different sources (text, image, chart, table, and layout structure) and (2) various locations (i.e. page number). Moreover, 33.2% of the questions are cross-page questions requiring evidence across multiple pages. 22.8% of the questions are designed to be unanswerable for detecting potential hallucinations. Experiments on 14 LVLMs demonstrate that long-context DU greatly challenges current models. Notably, the best-performing model, GPT-4o, achieves an F1 score of only 42.7%, while the second-best, GPT-4V, scores 31.4%. Furthermore, 12 LVLMs (all except GPT-4o and GPT-4V) even present worse performance than their LLM counterparts which are fed with lossy-parsed OCR documents. These results validate the necessity of future research toward more capable long-context LVLMs. Project Page: https://mayubo2333.github.io/MMLongBench-Doc

Our study aims to identify behavior patterns in cultural values exhibited by large language models (LLMs). The studied variants include question ordering, prompting language, and model size. Our experiments reveal that each tested LLM can efficiently behave with different cultural values. More interestingly: (i) LLMs exhibit relatively consistent cultural values when presented with prompts in a single language. (ii) The prompting language e.g., Chinese or English, can influence the expression of cultural values. The same question can elicit divergent cultural values when the same LLM is queried in a different language. (iii) Differences in sizes of the same model (e.g., Llama2-7B vs 13B vs 70B) have a more significant impact on their demonstrated cultural values than model differences (e.g., Llama2 vs Mixtral). Our experiments reveal that query language and model size of LLM are the main factors resulting in cultural value differences., Comment: 20 pages

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

Recommender Systems (RecSys) have become indispensable in numerous applications, profoundly influencing our everyday experiences. Despite their practical significance, academic research in RecSys often abstracts the formulation of research tasks from real-world contexts, aiming for a clean problem formulation and more generalizable findings. However, it is observed that there is a lack of collective understanding in RecSys academic research. The root of this issue may lie in the simplification of research task definitions, and an overemphasis on modeling the decision outcomes rather than the decision-making process. That is, we often conceptualize RecSys as the task of predicting missing values in a static user-item interaction matrix, rather than predicting a user's decision on the next interaction within a dynamic, changing, and application-specific context. There exists a mismatch between the inputs accessible to a model and the information available to users during their decision-making process, yet the model is tasked to predict users' decisions. While collaborative filtering is effective in learning general preferences from historical records, it is crucial to also consider the dynamic contextual factors in practical settings. Defining research tasks based on application scenarios using domain-specific datasets may lead to more insightful findings. Accordingly, viable solutions and effective evaluations can emerge for different application scenarios., Comment: Published in ACM SIGWEB Newsletter, Spring 2024: https://dl.acm.org/doi/10.1145/3663752.3663756

Due to the large number of parameters, the inference phase of Large Language Models (LLMs) is resource-intensive. However, not all requests posed to LLMs are equally difficult to handle. Through analysis, we show that for some tasks, LLMs can achieve results comparable to the final output at some intermediate layers. That is, not all layers of LLMs are necessary during inference. If we can predict at which layer the inferred results match the final results (produced by evaluating all layers), we could significantly reduce the inference cost. To this end, we propose a simple yet effective algorithm named AdaInfer to adaptively terminate the inference process for an input instance. AdaInfer relies on easily obtainable statistical features and classic classifiers like SVM. Experiments on well-known LLMs like the Llama2 series and OPT, show that AdaInfer can achieve an average of 17.8% pruning ratio, and up to 43% on sentiment tasks, with nearly no performance drop (<1%). Because AdaInfer does not alter LLM parameters, the LLMs incorporated with AdaInfer maintain generalizability across tasks.

Recommender systems have been widely used for various scenarios, such as e-commerce, news, and music, providing online contents to help and enrich users' daily life. Different scenarios hold distinct and unique characteristics, calling for domain-specific investigations and corresponding designed recommender systems. Therefore, in this paper, we focus on food delivery recommendations to unveil unique features in this domain, where users order food online and enjoy their meals shortly after delivery. We first conduct an in-depth analysis on food delivery datasets. The analysis shows that repeat orders are prevalent for both users and stores, and situations' differently influence repeat and exploration consumption in the food delivery recommender systems. Moreover, we revisit the ability of existing situation-aware methods for repeat and exploration recommendations respectively, and find them unable to effectively solve both tasks simultaneously. Based on the analysis and experiments, we have designed two separate recommendation models -- ReRec for repeat orders and ExpRec for exploration orders; both are simple in their design and computation. We conduct experiments on three real-world food delivery datasets, and our proposed models outperform various types of baselines on repeat, exploration, and combined recommendation tasks. This paper emphasizes the importance of dedicated analyses and methods for domain-specific characteristics for the recommender system studies., Comment: 11 pages, 5 figures

Scientific reasoning poses an excessive challenge for even the most advanced Large Language Models (LLMs). To make this task more practical and solvable for LLMs, we introduce a new task setting named tool-augmented scientific reasoning. This setting supplements LLMs with scalable toolsets, and shifts the focus from pursuing an omniscient problem solver to a proficient tool-user. To facilitate the research of such setting, we construct a tool-augmented training corpus named MathFunc which encompasses over 30,000 samples and roughly 6,000 tools. Building on MathFunc, we develop SciAgent to retrieve, understand and, if necessary, use tools for scientific problem solving. Additionally, we craft a benchmark, SciToolBench, spanning five scientific domains to evaluate LLMs' abilities with tool assistance. Extensive experiments on SciToolBench confirm the effectiveness of SciAgent. Notably, SciAgent-Mistral-7B surpasses other LLMs with the same size by more than 13% in absolute accuracy. Furthermore, SciAgent-DeepMath-7B shows much superior performance than ChatGPT.

Conversational recommender systems (CRS) aim to recommend relevant items to users by eliciting user preference through natural language conversation. Prior work often utilizes external knowledge graphs for items' semantic information, a language model for dialogue generation, and a recommendation module for ranking relevant items. This combination of multiple components suffers from a cumbersome training process, and leads to semantic misalignment issues between dialogue generation and item recommendation. In this paper, we represent items in natural language and formulate CRS as a natural language processing task. Accordingly, we leverage the power of pre-trained language models to encode items, understand user intent via conversation, perform item recommendation through semantic matching, and generate dialogues. As a unified model, our PECRS (Parameter-Efficient CRS), can be optimized in a single stage, without relying on non-textual metadata such as a knowledge graph. Experiments on two benchmark CRS datasets, ReDial and INSPIRED, demonstrate the effectiveness of PECRS on recommendation and conversation. Our code is available at: https://github.com/Ravoxsg/efficient_unified_crs., Comment: 9 pages, 4 figures, 8 tables, EACL 2024 conference, fixed typo

Large language models (LLMs) excel in abstractive summarization tasks, delivering fluent and pertinent summaries. Recent advancements have extended their capabilities to handle long-input contexts, exceeding 100k tokens. However, in question answering, language models exhibit uneven utilization of their input context. They tend to favor the initial and final segments, resulting in a U-shaped performance pattern concerning where the answer is located within the input. This bias raises concerns, particularly in summarization where crucial content may be dispersed throughout the source document(s). Besides, in summarization, mapping facts from the source to the summary is not trivial as salient content is usually re-phrased. In this paper, we conduct the first comprehensive study on context utilization and position bias in summarization. Our analysis encompasses 6 LLMs, 10 datasets, and 5 evaluation metrics. We introduce a new evaluation benchmark called MiddleSum on the which we benchmark two alternative inference methods to alleviate position bias: hierarchical summarization and incremental summarization. Our code and data can be found here: https://github.com/ntunlp/MiddleSum., Comment: ACL 2024. 9 pages, 7 figures, 3 tables

Although achieving great success, Large Language Models (LLMs) usually suffer from unreliable hallucinations. Although language attribution can be a potential solution, there are no suitable benchmarks and evaluation metrics to attribute LLMs to structured knowledge. In this paper, we define a new task of Knowledge-aware Language Model Attribution (KaLMA) that improves upon three core concerns with conventional attributed LMs. First, we extend attribution source from unstructured texts to Knowledge Graph (KG), whose rich structures benefit both the attribution performance and working scenarios. Second, we propose a new ``Conscious Incompetence" setting considering the incomplete knowledge repository, where the model identifies the need for supporting knowledge beyond the provided KG. Third, we propose a comprehensive automatic evaluation metric encompassing text quality, citation quality, and text citation alignment. To implement the above innovations, we build a dataset in biography domain BioKaLMA via evolutionary question generation strategy, to control the question complexity and necessary knowledge to the answer. For evaluation, we develop a baseline solution and demonstrate the room for improvement in LLMs' citation generation, emphasizing the importance of incorporating the "Conscious Incompetence" setting, and the critical role of retrieval accuracy., Comment: acl findings 2024

Recommender systems (RS) have achieved significant success by leveraging explicit identification (ID) features. However, the full potential of content features, especially the pure image pixel features, remains relatively unexplored. The limited availability of large, diverse, and content-driven image recommendation datasets has hindered the use of raw images as item representations. In this regard, we present PixelRec, a massive image-centric recommendation dataset that includes approximately 200 million user-image interactions, 30 million users, and 400,000 high-quality cover images. By providing direct access to raw image pixels, PixelRec enables recommendation models to learn item representation directly from them. To demonstrate its utility, we begin by presenting the results of several classical pure ID-based baseline models, termed IDNet, trained on PixelRec. Then, to show the effectiveness of the dataset's image features, we substitute the itemID embeddings (from IDNet) with a powerful vision encoder that represents items using their raw image pixels. This new model is dubbed PixelNet.Our findings indicate that even in standard, non-cold start recommendation settings where IDNet is recognized as highly effective, PixelNet can already perform equally well or even better than IDNet. Moreover, PixelNet has several other notable advantages over IDNet, such as being more effective in cold-start and cross-domain recommendation scenarios. These results underscore the importance of visual features in PixelRec. We believe that PixelRec can serve as a critical resource and testing ground for research on recommendation models that emphasize image pixel content. The dataset, code, and leaderboard will be available at https://github.com/westlake-repl/PixelRec.

Large language models (LLMs) have achieved remarkable success in NLP and multimodal tasks, among others. Despite these successes, two main challenges remain in developing LLMs: (i) high computational cost, and (ii) fair and objective evaluations. In this paper, we report a solution to significantly reduce LLM training cost through a growth strategy. We demonstrate that a 101B-parameter LLM with 0.31T tokens can be trained with a budget of 100K US dollars. Inspired by IQ tests, we also consolidate an additional range of evaluations on top of existing evaluations that focus on knowledge-oriented abilities. These IQ evaluations include symbolic mapping, rule understanding, pattern mining, and anti-interference. Such evaluations minimize the potential impact of memorization. Experimental results show that our model, named FLM-101B, trained with a budget of 100K US dollars, achieves performance comparable to powerful and well-known models, e.g., GPT-3 and GLM-130B, especially on the additional range of IQ evaluations. The checkpoint of FLM-101B is released at https://huggingface.co/CofeAI/FLM-101B.

A typical benchmark dataset for recommender system (RecSys) evaluation consists of user-item interactions generated on a platform within a time period. The interaction generation mechanism partially explains why a user interacts with (e.g., like, purchase, rate) an item, and the context of when a particular interaction happened. In this study, we conduct a meticulous analysis of the MovieLens dataset and explain the potential impact of using the dataset for evaluating recommendation algorithms. We make a few main findings from our analysis. First, there are significant differences in user interactions at the different stages when a user interacts with the MovieLens platform. The early interactions largely define the user portrait which affects the subsequent interactions. Second, user interactions are highly affected by the candidate movies that are recommended by the platform's internal recommendation algorithm(s). Third, changing the order of user interactions makes it more difficult for sequential algorithms to capture the progressive interaction process. We further discuss the discrepancy between the interaction generation mechanism that is employed by the MovieLens system and that of typical real-world recommendation scenarios. In summary, the MovieLens platform demonstrates an efficient and effective way of collecting user preferences to address cold-starts. However, models that achieve excellent recommendation accuracy on the MovieLens dataset may not demonstrate superior performance in practice, for at least two kinds of differences: (i) the differences in the contexts of user-item interaction generation, and (ii) the differences in user knowledge about the item collections. While results on MovieLens can be useful as a reference, they should not be solely relied upon as the primary justification for the effectiveness of a recommendation system model.

Fast-growing scientific publications present challenges to the scientific community. In this paper, we describe their implications to researchers. As references form explicit foundations for researchers to conduct a study, we investigate the evolution in reference patterns based on 60.8 million papers published from 1960 to 2015. The results demonstrate that recent papers contain more references than older ones, especially the well-cited papers compared with other papers. Well-cited papers receive 10 or more citations within 5 years of publication. Their references cover a longer period from classic research to very recent studies. Authors of well-cited papers are also farsighted to discover the reference papers with good potential to receive high citation numbers in near future. We also discover that the number of accumulative publications has a negative impact on next-5-year citation count for most fields except Chemistry, Materials science, Environmental science, Biology, and Engineering. Our findings suggest that researchers are expected to devote more effort to producing impactful research. Based on all these findings, we strongly advise against judging researchers simply based on the number of their publications. On the other hand, authors and reviewers should ensure that published papers contain adequate contributions. The code for our analysis is on GitHub: https://github.com/ECNU-Text-Computing/Research-Explosion., Comment: 21 pages, 24 figures

Knowledge Graph Completion (KGC) often requires both KG structural and textual information to be effective. Pre-trained Language Models (PLMs) have been used to learn the textual information, usually under the fine-tune paradigm for the KGC task. However, the fine-tuned PLMs often overwhelmingly focus on the textual information and overlook structural knowledge. To tackle this issue, this paper proposes CSProm-KG (Conditional Soft Prompts for KGC) which maintains a balance between structural information and textual knowledge. CSProm-KG only tunes the parameters of Conditional Soft Prompts that are generated by the entities and relations representations. We verify the effectiveness of CSProm-KG on three popular static KGC benchmarks WN18RR, FB15K-237 and Wikidata5M, and two temporal KGC benchmarks ICEWS14 and ICEWS05-15. CSProm-KG outperforms competitive baseline models and sets new state-of-the-art on these benchmarks. We conduct further analysis to show (i) the effectiveness of our proposed components, (ii) the efficiency of CSProm-KG, and (iii) the flexibility of CSProm-KG., Comment: Accepted by ACL 2023 Findings, Long Paper

Given a query, the task of Natural Language Video Localization (NLVL) is to localize a temporal moment in an untrimmed video that semantically matches the query. In this paper, we adopt a proposal-based solution that generates proposals (i.e., candidate moments) and then select the best matching proposal. On top of modeling the cross-modal interaction between candidate moments and the query, our proposed Moment Sampling DETR (MS-DETR) enables efficient moment-moment relation modeling. The core idea is to sample a subset of moments guided by the learnable templates with an adopted DETR (DEtection TRansformer) framework. To achieve this, we design a multi-scale visual-linguistic encoder, and an anchor-guided moment decoder paired with a set of learnable templates. Experimental results on three public datasets demonstrate the superior performance of MS-DETR., Comment: Accepted by ACL 2023

Goal-oriented Script Generation is a new task of generating a list of steps that can fulfill the given goal. In this paper, we propose to extend the task from the perspective of cognitive theory. Instead of a simple flat structure, the steps are typically organized hierarchically - Human often decompose a complex task into subgoals, where each subgoal can be further decomposed into steps. To establish the benchmark, we contribute a new dataset, propose several baseline methods, and set up evaluation metrics. Both automatic and human evaluation verify the high-quality of dataset, as well as the effectiveness of incorporating subgoals into hierarchical script generation. Furthermore, We also design and evaluate the model to discover subgoal, and find that it is a bit more difficult to decompose the goals than summarizing from segmented steps., Comment: Accepted by ACL 2023 Findings

Open Information Extraction (OIE) aims to extract factual relational tuples from open-domain sentences. Downstream tasks use the extracted OIE tuples as facts, without examining the certainty of these facts. However, uncertainty/speculation is a common linguistic phenomenon. Existing studies on speculation detection are defined at sentence level, but even if a sentence is determined to be speculative, not all tuples extracted from it may be speculative. In this paper, we propose to study speculations in OIE and aim to determine whether an extracted tuple is speculative. We formally define the research problem of tuple-level speculation detection and conduct a detailed data analysis on the LSOIE dataset which contains labels for speculative tuples. Lastly, we propose a baseline model OIE-Spec for this new research task.

In a practical recommender system, new interactions are continuously observed. Some interactions are expected, because they largely follow users' long-term preferences. Some other interactions are indications of recent trends in user preference changes or marketing positions of new items. Accordingly, the recommender needs to be periodically retrained or updated to capture the new trends, and yet not to forget the long-term preferences. In this paper, we propose a novel and generic retraining framework called Disentangled Incremental Learning (DIL) for graph-based recommenders. We assume that long-term preferences are well captured in the existing model, in the form of model parameters learned from past interactions. New preferences can be learned from the user-item bipartite graph constructed using the newly observed interactions. In DIL, we design an Information Extraction Module to extract historical preferences from the existing model. Then we blend the historical and new preferences in the form of node embeddings in the new graph, through a Disentanglement Module. The essence of the disentanglement module is to decorrelate the historical and new preferences so that both can be well captured, via carefully designed losses. Through experiments on three benchmark datasets, we show the effectiveness of DIL in capturing dynamics of useritem interactions. We also demonstrate the robustness of DIL by attaching it to two base models - LightGCN and NGCF.

Open Information Extraction (OIE) aims to extract relational tuples from open-domain sentences. Existing OIE systems split a sentence into tokens and recognize token spans as tuple relations and arguments. We instead propose Sentence as Chunk sequence (SaC) and recognize chunk spans as tuple relations and arguments. We argue that SaC has better quantitative and qualitative properties for OIE than sentence as token sequence, and evaluate four choices of chunks (i.e., CoNLL chunks, simple phrases, NP chunks, and spans from SpanOIE) against gold OIE tuples. Accordingly, we propose a simple BERT-based model for sentence chunking, and propose Chunk-OIE for tuple extraction on top of SaC. Chunk-OIE achieves state-of-the-art results on multiple OIE datasets, showing that SaC benefits OIE task.

Few-shot event detection (ED) has been widely studied, while this brings noticeable discrepancies, e.g., various motivations, tasks, and experimental settings, that hinder the understanding of models for future progress.This paper presents a thorough empirical study, a unified view of ED models, and a better unified baseline. For fair evaluation, we compare 12 representative methods on three datasets, which are roughly grouped into prompt-based and prototype-based models for detailed analysis. Experiments consistently demonstrate that prompt-based methods, including ChatGPT, still significantly trail prototype-based methods in terms of overall performance. To investigate their superior performance, we break down their design elements along several dimensions and build a unified framework on prototype-based methods. Under such unified view, each prototype-method can be viewed a combination of different modules from these design elements. We further combine all advantageous modules and propose a simple yet effective baseline, which outperforms existing methods by a large margin (e.g., 2.7% F1 gains under low-resource setting)., Comment: Accepted by ACL 2023 main conference

Pre-trained language models (PLMs) have achieved remarkable success in NLP tasks. Despite the great success, mainstream solutions largely follow the pre-training then finetuning paradigm, which brings in both high deployment costs and low training efficiency. Nevertheless, fine-tuning on a specific task is essential because PLMs are only pre-trained with language signal from large raw data. In this paper, we propose a novel fine-tuning-free strategy for language models, to consider both language signal and teacher signal. Teacher signal is an abstraction of a battery of downstream tasks, provided in a unified proposition format. Trained with both language and strong task-aware teacher signals in an interactive manner, our FreeLM model demonstrates strong generalization and robustness. FreeLM outperforms large models e.g., GPT-3 and InstructGPT, on a range of language understanding tasks in experiments. FreeLM is much smaller with 0.3B parameters, compared to 175B in these models.

As language models scale up, it becomes increasingly expensive to verify research ideas because conclusions on small models do not trivially transfer to large ones. A possible solution is to establish a generic system that accurately predicts certain metrics for large models without training them. Existing scaling laws require hyperparameter search on the largest models, limiting their predicative capability. In this paper, we present an approach (namely {\mu}Scaling) to predict the pre-training loss, based on our observations that Maximal Update Parametrization ({\mu}P) enables accurate fitting of scaling laws close to common loss basins in hyperparameter space. With {\mu}Scaling, different model designs can be compared on large scales by training only their smaller counterparts. Further, we introduce nanoLM: an affordable LLM pre-training benchmark that facilitates this new research paradigm. With around 14% of the one-time pre-training cost, we can accurately forecast the loss for models up to 52B. Our goal with nanoLM is to empower researchers with limited resources to reach meaningful conclusions on large models. We also aspire for our benchmark to serve as a bridge between the academic community and the industry. Code for {\mu}Scaling is available at https://github.com/cofe-ai/Mu-scaling. Code for nanoLLM will be available later., Comment: This is a modified and extended version of our previous Mu-scaling work released in April 2023 (see v1)

Mainstream solutions to Sequential Recommendation (SR) represent items with fixed vectors. These vectors have limited capability in capturing items' latent aspects and users' diverse preferences. As a new generative paradigm, Diffusion models have achieved excellent performance in areas like computer vision and natural language processing. To our understanding, its unique merit in representation generation well fits the problem setting of sequential recommendation. In this paper, we make the very first attempt to adapt Diffusion model to SR and propose DiffuRec, for item representation construction and uncertainty injection. Rather than modeling item representations as fixed vectors, we represent them as distributions in DiffuRec, which reflect user's multiple interests and item's various aspects adaptively. In diffusion phase, DiffuRec corrupts the target item embedding into a Gaussian distribution via noise adding, which is further applied for sequential item distribution representation generation and uncertainty injection. Afterward, the item representation is fed into an Approximator for target item representation reconstruction. In reverse phase, based on user's historical interaction behaviors, we reverse a Gaussian noise into the target item representation, then apply a rounding operation for target item prediction. Experiments over four datasets show that DiffuRec outperforms strong baselines by a large margin.

Given comparative text, comparative relation extraction aims to extract two targets (\eg two cameras) in comparison and the aspect they are compared for (\eg image quality). The extracted comparative relations form the basis of further opinion analysis.Existing solutions formulate this task as a sequence labeling task, to extract targets and aspects. However, they cannot directly extract comparative relation(s) from text. In this paper, we show that comparative relations can be directly extracted with high accuracy, by generative model. Based on GPT-2, we propose a Generation-based Comparative Relation Extractor (GCRE-GPT). Experiment results show that \modelname achieves state-of-the-art accuracy on two datasets., Comment: 6 pages, 6 tables, 1 figure

Large Language Models (LLMs) have made remarkable strides in various tasks. Whether LLMs are competitive few-shot solvers for information extraction (IE) tasks, however, remains an open problem. In this work, we aim to provide a thorough answer to this question. Through extensive experiments on nine datasets across four IE tasks, we demonstrate that current advanced LLMs consistently exhibit inferior performance, higher latency, and increased budget requirements compared to fine-tuned SLMs under most settings. Therefore, we conclude that LLMs are not effective few-shot information extractors in general. Nonetheless, we illustrate that with appropriate prompting strategies, LLMs can effectively complement SLMs and tackle challenging samples that SLMs struggle with. And moreover, we propose an adaptive filter-then-rerank paradigm to combine the strengths of LLMs and SLMs. In this paradigm, SLMs serve as filters and LLMs serve as rerankers. By prompting LLMs to rerank a small portion of difficult samples identified by SLMs, our preliminary system consistently achieves promising improvements (2.4% F1-gain on average) on various IE tasks, with an acceptable time and cost investment., Comment: Accepted by EMNLP 2023 Findings

Deep learning technologies have brought us many models that outperform human beings on a few benchmarks. An interesting question is: can these models well solve real-world problems with similar settings (e.g., identical input/output) to the benchmark datasets? We argue that a model is trained to answer the same information need for which the training dataset is created. Although some datasets may share high structural similarities, e.g., question-answer pairs for the question answering (QA) task and image-caption pairs for the image captioning (IC) task, they may represent different research tasks aiming for answering different information needs. To support our argument, we use the QA task and IC task as two case studies and compare their widely used benchmark datasets. From the perspective of information need in the context of information retrieval, we show the differences in the dataset creation processes, and the differences in morphosyntactic properties between datasets. The differences in these datasets can be attributed to the different information needs of the specific research tasks. We encourage all researchers to consider the information need the perspective of a research task before utilizing a dataset to train a model. Likewise, while creating a dataset, researchers may also incorporate the information need perspective as a factor to determine the degree to which the dataset accurately reflects the research task they intend to tackle., Comment: 19 pages, 5 figures

Open Information Extraction (OpenIE) aims to extract relational tuples from open-domain sentences. Traditional rule-based or statistical models have been developed based on syntactic structures of sentences, identified by syntactic parsers. However, previous neural OpenIE models under-explore the useful syntactic information. In this paper, we model both constituency and dependency trees into word-level graphs, and enable neural OpenIE to learn from the syntactic structures. To better fuse heterogeneous information from both graphs, we adopt multi-view learning to capture multiple relationships from them. Finally, the finetuned constituency and dependency representations are aggregated with sentential semantic representations for tuple generation. Experiments show that both constituency and dependency information, and the multi-view learning are effective., Comment: To appear in EMNLP 2022

Recently, amounts of works utilize perplexity~(PPL) to evaluate the quality of the generated text. They suppose that if the value of PPL is smaller, the quality(i.e. fluency) of the text to be evaluated is better. However, we find that the PPL referee is unqualified and it cannot evaluate the generated text fairly for the following reasons: (i) The PPL of short text is larger than long text, which goes against common sense, (ii) The repeated text span could damage the performance of PPL, and (iii) The punctuation marks could affect the performance of PPL heavily. Experiments show that the PPL is unreliable for evaluating the quality of given text. Last, we discuss the key problems with evaluating text quality using language models.

Recommendation has become a prominent area of research in the field of Information Retrieval (IR). Evaluation is also a traditional research topic in this community. Motivated by a few counter-intuitive observations reported in recent studies, this perspectives paper takes a fresh look at recommender systems from an evaluation standpoint. Rather than examining metrics like recall, hit rate, or NDCG, or perspectives like novelty and diversity, the key focus here is on how these metrics are calculated when evaluating a recommender algorithm. Specifically, the commonly used train/test data splits and their consequences are re-examined. We begin by examining common data splitting methods, such as random split or leave-one-out, and discuss why the popularity baseline is poorly defined under such splits. We then move on to explore the two implications of neglecting a global timeline during evaluation: data leakage and oversimplification of user preference modeling. Afterwards, we present new perspectives on recommender systems, including techniques for evaluating algorithm performance that more accurately reflect real-world scenarios, and possible approaches to consider decision contexts in user preference modeling., Comment: Accepted for SIGIR 2023 (Perspectives Paper Track)

Quotation extraction aims to extract quotations from written text. There are three components in a quotation: source refers to the holder of the quotation, cue is the trigger word(s), and content is the main body. Existing solutions for quotation extraction mainly utilize rule-based approaches and sequence labeling models. While rule-based approaches often lead to low recalls, sequence labeling models cannot well handle quotations with complicated structures. In this paper, we propose the Context and Former-Label Enhanced Net (CofeNet) for quotation extraction. CofeNet is able to extract complicated quotations with components of variable lengths and complicated structures. On two public datasets (i.e., PolNeAR and Riqua) and one proprietary dataset (i.e., PoliticsZH), we show that our CofeNet achieves state-of-the-art performance on complicated quotation extraction., Comment: Accepted by COLING 2022

Multi-scenario learning (MSL) enables a service provider to cater for users' fine-grained demands by separating services for different user sectors, e.g., by user's geographical region. Under each scenario there is a need to optimize multiple task-specific targets e.g., click through rate and conversion rate, known as multi-task learning (MTL). Recent solutions for MSL and MTL are mostly based on the multi-gate mixture-of-experts (MMoE) architecture. MMoE structure is typically static and its design requires domain-specific knowledge, making it less effective in handling both MSL and MTL. In this paper, we propose a novel Automatic Expert Selection framework for Multi-scenario and Multi-task search, named AESM^{2}. AESM^{2} integrates both MSL and MTL into a unified framework with an automatic structure learning. Specifically, AESM^{2} stacks multi-task layers over multi-scenario layers. This hierarchical design enables us to flexibly establish intrinsic connections between different scenarios, and at the same time also supports high-level feature extraction for different tasks. At each multi-scenario/multi-task layer, a novel expert selection algorithm is proposed to automatically identify scenario-/task-specific and shared experts for each input. Experiments over two real-world large-scale datasets demonstrate the effectiveness of AESM^{2} over a battery of strong baselines. Online A/B test also shows substantial performance gain on multiple metrics. Currently, AESM^{2} has been deployed online for serving major traffic., Comment: Accepted by SIGIR 2022; 10 pages, 8 figures

Open Information Extraction (OpenIE) facilitates domain-independent discovery of relational facts from large corpora. The technique well suits many open-world natural language understanding scenarios, such as automatic knowledge base construction, open-domain question answering, and explicit reasoning. Thanks to the rapid development in deep learning technologies, numerous neural OpenIE architectures have been proposed and achieve considerable performance improvement. In this survey, we provide an extensive overview of the-state-of-the-art neural OpenIE models, their key design decisions, strengths and weakness. Then, we discuss limitations of current solutions and the open issues in OpenIE problem itself. Finally we list recent trends that could help expand its scope and applicability, setting up promising directions for future research in OpenIE. To our best knowledge, this paper is the first review on this specific topic., Comment: Accepted by IJCAI22 survey track

Transformer-based models have delivered impressive results on many tasks, particularly vision and language tasks. In many model training situations, conventional configurations are typically adopted. For example, we often set the base model with hidden dimensions (i.e. model width) to be 768 and the number of transformer layers (i.e. model depth) to be 12. In this paper, we revisit these conventional configurations. Through theoretical analysis and experimental evaluation, we show that the masked autoencoder is effective in alleviating the over-smoothing issue in deep transformer training. Based on this finding, we propose Bamboo, an idea of using deeper and narrower transformer configurations, for masked autoencoder training. On ImageNet, with such a simple change in configuration, re-designed model achieves 87.1% top-1 accuracy and outperforms SoTA models like MAE and BEiT. On language tasks, re-designed model outperforms BERT with default setting by 1.1 points on average, on GLUE datasets., Comment: Accepted at ICML 2023

In academic research, recommender systems are often evaluated on benchmark datasets, without much consideration about the global timeline. Hence, we are unable to answer questions like: Do loyal users enjoy better recommendations than non-loyal users? Loyalty can be defined by the time period a user has been active in a recommender system, or by the number of historical interactions a user has. In this paper, we offer a comprehensive analysis of recommendation results along global timeline. We conduct experiments with five widely used models, i.e., BPR, NeuMF, LightGCN, SASRec and TiSASRec, on four benchmark datasets, i.e., MovieLens-25M, Yelp, Amazon-music, and Amazon-electronic. Our experiment results give an answer "No" to the above question. Users with many historical interactions suffer from relatively poorer recommendations. Users who stay with the system for a shorter time period enjoy better recommendations. Both findings are counter-intuitive. Interestingly, users who have recently interacted with the system, with respect to the time point of the test instance, enjoy better recommendations. The finding on recency applies to all users, regardless of users' loyalty. Our study offers a different perspective to understand recommender accuracy, and our findings could trigger a revisit of recommender model design. The code is available in \url{https://github.com/putatu/recommenderLoyalty., Comment: Accepted to ICTIR2022

Many studies on dialog emotion analysis focus on utterance-level emotion only. These models hence are not optimized for dialog-level emotion detection, i.e. to predict the emotion category of a dialog as a whole. More importantly, these models cannot benefit from the context provided by the whole dialog. In real-world applications, annotations to dialog could fine-grained, including both utterance-level tags (e.g. speaker type, intent category, and emotion category), and dialog-level tags (e.g. user satisfaction, and emotion curve category). In this paper, we propose a Context-based Hierarchical Attention Capsule~(Chat-Capsule) model, which models both utterance-level and dialog-level emotions and their interrelations. On a dialog dataset collected from customer support of an e-commerce platform, our model is also able to predict user satisfaction and emotion curve category. Emotion curve refers to the change of emotions along the development of a conversation. Experiments show that the proposed Chat-Capsule outperform state-of-the-art baselines on both benchmark dataset and proprietary dataset. Source code will be released upon acceptance.

A sentence may express sentiments on multiple aspects. When these aspects are associated with different sentiment polarities, a model's accuracy is often adversely affected. We observe that multiple aspects in such hard sentences are mostly expressed through multiple clauses, or formally known as elementary discourse units (EDUs), and one EDU tends to express a single aspect with unitary sentiment towards that aspect. In this paper, we propose to consider EDU boundaries in sentence modeling, with attentions at both word and EDU levels. Specifically, we highlight sentiment-bearing words in EDU through word-level sparse attention. Then at EDU level, we force the model to attend to the right EDU for the right aspect, by using EDU-level sparse attention and orthogonal regularization. Experiments on three benchmark datasets show that our simple EDU-Attention model outperforms state-of-the-art baselines. Because EDU can be automatically segmented with high accuracy, our model can be applied to sentences directly without the need of manual EDU boundary annotation., Comment: Accepted in PAKDD2022

Temporal sentence grounding in videos (TSGV), \aka natural language video localization (NLVL) or video moment retrieval (VMR), aims to retrieve a temporal moment that semantically corresponds to a language query from an untrimmed video. Connecting computer vision and natural language, TSGV has drawn significant attention from researchers in both communities. This survey attempts to provide a summary of fundamental concepts in TSGV and current research status, as well as future research directions. As the background, we present a common structure of functional components in TSGV, in a tutorial style: from feature extraction from raw video and language query, to answer prediction of the target moment. Then we review the techniques for multimodal understanding and interaction, which is the key focus of TSGV for effective alignment between the two modalities. We construct a taxonomy of TSGV techniques and elaborate the methods in different categories with their strengths and weaknesses. Lastly, we discuss issues with the current TSGV research and share our insights about promising research directions., Comment: Accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)

Conducting a systematic review (SR) is comprised of multiple tasks: (i) collect documents (studies) that are likely to be relevant from digital libraries (eg., PubMed), (ii) manually read and label the documents as relevant or irrelevant, (iii) extract information from the relevant studies, and (iv) analyze and synthesize the information and derive a conclusion of SR. When researchers label studies, they can screen ranked documents where relevant documents are higher than irrelevant ones. This practice, known as screening prioritization (ie., document ranking approach), speeds up the process of conducting a SR as the documents labelled as relevant can move to the next tasks earlier. However, the approach is limited in reducing the manual workload because the total number of documents to screen remains the same. Towards reducing the manual workload in the screening process, we investigate the quality of document ranking of SR. This can signal researchers whereabouts in the ranking relevant studies are located and let them decide where to stop the screening. After extensive analysis on SR document rankings from different ranking models, we hypothesize 'topic broadness' as a factor that affects the ranking quality of SR. Finally, we propose a measure that estimates the topic broadness and demonstrate that the proposed measure is a simple yet effective method to predict the qualities of document rankings for SRs., Comment: 9 pages, work done in 2019, revised in 2021

When medical researchers conduct a systematic review (SR), screening studies is the most time-consuming process: researchers read several thousands of medical literature and manually label them relevant or irrelevant. Screening prioritization (ie., document ranking) is an approach for assisting researchers by providing document rankings where relevant documents are ranked higher than irrelevant ones. Seed-driven document ranking (SDR) uses a known relevant document (ie., seed) as a query and generates such rankings. Previous work on SDR seeks ways to identify different term weights in a query document and utilizes them in a retrieval model to compute ranking scores. Alternatively, we formulate the SDR task as finding similar documents to a query document and produce rankings based on similarity scores. We propose a document matching measure named Mirror Matching, which calculates matching scores between medical abstract texts by incorporating common writing patterns, such as background, method, result, and conclusion in order. We conduct experiments on CLEF 2019 eHealth Task 2 TAR dataset, and the empirical results show this simple approach achieves the higher performance than traditional and neural retrieval models on Average Precision and Precision-focused metrics., Comment: 10 pages, work done in 2019

A payment network contains transactions between sellers and buyers. Detecting risky (or bad) sellers on such a payment network is crucial to payment service providers for risk management and legal compliance. In this research, we formulate this task as a node classification task. Specifically, we aim to predict a label for each seller in a payment network, by analysing its properties and/or interactions. Nodes residing in different parts of a payment network can have similar local topology structures. Such local topology structures reveal sellers' business roles, eg., supplier, drop-shipper, or retailer. We note that many existing solutions for graph-based node classification only consider node connectivity but not the similarity between node's local topology structure. Motivated by business need, we present a dual-path graph convolution network, named DP-GCN, for node classification. DP-GCN considers both node connectivity and topology structure similarity. The proposed model consists of three main modules: (i) a C-GCN module to capture connectivity relationships between nodes, (ii) a T-GCN module to capture topology structure similarities between nodes, and (iii) a multi-head self-attention module to align both properties. We evaluate DP-GCN on seven benchmark datasets against diverse baselines. We also provide a case study of running DP-GCN on three large-scale payment networks from PayPal, one of the leading payment service providers. Experimental results demonstrate DP-GCN's effectiveness and practicability.

The temporal sentence grounding in video (TSGV) task is to locate a temporal moment from an untrimmed video, to match a language query, i.e., a sentence. Without considering bias in moment annotations (e.g., start and end positions in a video), many models tend to capture statistical regularities of the moment annotations, and do not well learn cross-modal reasoning between video and language query. In this paper, we propose two debiasing strategies, data debiasing and model debiasing, to "force" a TSGV model to capture cross-modal interactions. Data debiasing performs data oversampling through video truncation to balance moment temporal distribution in train set. Model debiasing leverages video-only and query-only models to capture the distribution bias, and forces the model to learn cross-modal interactions. Using VSLNet as the base model, we evaluate impact of the two strategies on two datasets that contain out-of-distribution test instances. Results show that both strategies are effective in improving model generalization capability. Equipped with both debiasing strategies, VSLNet achieves best results on both datasets., Comment: 13 pages, 6 figures, 11 tables

Sarcasm employs ambivalence, where one says something positive but actually means negative, and vice versa. The essence of sarcasm, which is also a sufficient and necessary condition, is the conflict between literal and implied sentiments expressed in one sentence. However, it is difficult to recognize such sentiment conflict because the sentiments are mixed or even implicit. As a result, the recognition of sophisticated and obscure sentiment brings in a great challenge to sarcasm detection. In this paper, we propose a Dual-Channel Framework by modeling both literal and implied sentiments separately. Based on this dual-channel framework, we design the Dual-Channel Network~(DC-Net) to recognize sentiment conflict. Experiments on political debates (i.e. IAC-V1 and IAC-V2) and Twitter datasets show that our proposed DC-Net achieves state-of-the-art performance on sarcasm recognition. Our code is released to support research., Comment: Accepted by Findings of NAACL 2022