Ghadir Alselwi, Hao Xue, Shoaib Jameel, Basem Suleiman, Flora D. Salim, Imran Razzak. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Abstract Language models often underperform when fine tuned with limited labelled data. To reduce dependence on costly annotation, recent studies have explored clustering and pseudo labelling to leverage unlabeled text with only a few labelled examples for fine tuning. However, these methods face persistent challenges: clustering errors and pseudo-label mismatches can degrade fine-tuning performance. We propose Cluster-SCP , a novel framework that improves intra-cluster coherence and reduces cluster quantity to generate more reliable pseudo labels. The framework begins with K-Means initialization and applies an iterative refinement process, Intermediate Pseudo Labels , comprising two stages: embedding reassignment to minimize clustering errors, and cluster merging via contrastive learning with graph readouts. The refined pseudo labels are first used to fine tune language model, which is subsequently refined using a small set of labels. Experiments on three benchmark datasets show that Cluster-SCP consistently outperforms baseline and state-of-the-art methods, achieving at least a 5.3% accuracy improvement while reducing the number of clusters by 22%.

Personality detection from text is commonly performed by analysing users' social media posts. However, existing methods heavily rely on large-scale annotated datasets, making it challenging to obtain high-quality personality labels. Moreover, most studies treat emotion and personality as independent variables, overlooking their interactions. In this paper, we propose a novel self-supervised framework, EmoPerso, which improves personality detection through emotion-aware modelling. EmoPerso first leverages generative mechanisms for synthetic data augmentation and rich representation learning. It then extracts pseudo-labeled emotion features and jointly optimizes them with personality prediction via multi-task learning. A cross-attention module is employed to capture fine-grained interactions between personality traits and the inferred emotional representations. To further refine relational reasoning, EmoPerso adopts a self-taught strategy to enhance the model's reasoning capabilities iteratively. Extensive experiments on two benchmark datasets demonstrate that EmoPerso surpasses state-of-the-art models. The source code is available at https://github.com/slz0925/EmoPerso.

NLP models often face challenges with under-represented languages due to a lack of sufficient training data and language complexities. This can result in inaccurate predictions and a failure to capture the inherent uncertainties within these languages. This paper introduces a new method for modelling uncertainty in under-represented languages by employing deep Bayesian Gaussian Processes. We develop a novel framework that integrates prior knowledge and leverages kernel functions. This helps enable the quantification of uncertainty in predictions to overcome the data limitations in under-represented languages. The efficacy of our approach is validated through various experiments, and the results are benchmarked against existing methods to highlight the enhancements in prediction accuracy and measurement of uncertainty.

Multi-level Hierarchical Classification (MLHC) tackles the challenge of categorizing items within a complex, multi-layered class structure. However, traditional MLHC classifiers often rely on a backbone model with n independent output layers, which tend to ignore the hierarchical relationships between classes. This oversight can lead to inconsistent predictions that violate the underlying taxonomy. Leveraging Large Language Models (LLMs), we propose novel taxonomy-embedded transitional LLM-agnostic framework for multimodality classification. The cornerstone of this advancement is the ability of models to enforce consistency across hierarchical levels. Our evaluations on the MEP-3M dataset - a Multi-modal E-commerce Product dataset with various hierarchical levels- demonstrated a significant performance improvement compared to conventional LLMs structure.

Graph-based personality detection constructs graph structures from textual data, particularly social media posts. Current methods often struggle with sparse or noisy data and rely on static graphs, limiting their ability to capture dynamic changes between nodes and relationships. This paper introduces LL4G, a self-supervised framework leveraging large language models (LLMs) to optimize graph neural networks (GNNs). LLMs extract rich semantic features to generate node representations and to infer explicit and implicit relationships. The graph structure adaptively adds nodes and edges based on input data, continuously optimizing itself. The GNN then uses these optimized representations for joint training on node reconstruction, edge prediction, and contrastive learning tasks. This integration of semantic and structural information generates robust personality profiles. Experimental results on Kaggle and Pandora datasets show LL4G outperforms state-of-the-art models.

Personality detection automatically identifies an individual’s personality from various data sources, such as social media texts. However, as the parameter scale of language models continues to grow, the computational cost becomes increasingly difficult to manage. Fine-tuning also grows more complex, making it harder to justify the effort and reliably predict outcomes. We introduce a novel parameter-efficient fine-tuning framework, PersLLM, to address these challenges. In PersLLM, a large language model (LLM) extracts high-dimensional representations from raw data and stores them in a dynamic memory layer. PersLLM then updates the downstream layers with a replaceable output network, enabling flexible adaptation to various personality detection scenarios. By storing the features in the memory layer, we eliminate the need for repeated complex computations by the LLM. Meanwhile, the lightweight output network serves as a proxy for evaluating the overall effectiveness of the framework, improving the predictability of results. Experimental results on key benchmark datasets like Kaggle and Pandora show that PersLLM significantly reduces computational cost while maintaining competitive performance and strong adaptability.

Cultural awareness in language models is the capacity to understand and adapt to diverse cultural contexts. However, most existing approaches treat culture as static background knowledge, overlooking its dynamic and evolving nature. This limitation reduces their reliability in downstream tasks that demand genuine cultural sensitivity. In this work, we introduce CALM, a novel framework designed to endow language models with cultural self-awareness. CALM disentangles task semantics from explicit cultural concepts and latent cultural signals, shaping them into structured cultural clusters through contrastive learning. These clusters are then aligned via cross-attention to establish fine-grained interactions among related cultural features and are adaptively integrated through a Mixture-of-Experts mechanism along culture-specific dimensions. The resulting unified representation is fused with the model's original knowledge to construct a culturally grounded internal identity state, which is further enhanced through self-prompted reflective learning, enabling continual adaptation and self-correction. Extensive experiments conducted on multiple cross-cultural benchmark datasets demonstrate that CALM consistently outperforms state-of-the-art methods.

Pre-trained language models (PLMs), despite showing strong performance, can carry and increase biases, which can limit the development of fair NLP and IR systems. This research investigates the foundational origins of bias within PLMs, moving beyond detection to a detailed analysis of its formation and propagation across diverse architectures. Through a novel attention weight analysis, we reveal distinct attention patterns for biased versus neutral content, offering insights into the internal representations learned by PLMs. Our findings demonstrate a complex interplay between training data and model architecture, revealing that while the transformer's self-attention mechanism amplifies existing biases, the training data plays a crucial role in the initial encoding of bias within the model's representations.

In critical domains such as healthcare and law, accurately modelling the uncertainty of automatic computational models is essential. For instance, healthcare models must produce reliable estimates to guide human decision-making. However, modelling uncertainty remains challenging, particularly for models handling low-resource datasets and complex, domain-specific vocabulary. Most existing predictive models model point estimates rather than probability distributions, limiting our ability to quantify model uncertainty. This paper introduces a novel model, BAKER, designed to address these limitations. BAKER combines the strengths of Bayesian inference, known for its effectiveness in modelling uncertainty, and kernel methods, which excel at capturing complex data relationships. Incorporating kernel functions enhances model performance, particularly by reducing overfitting in data-limited scenarios. Our experimental analysis shows that BAKER significantly improves uncertainty reasoning compared to existing models.

Multimodal fake news detection often involves modelling heterogeneous data sources, such as vision and language. Existing detection methods typically rely on fusion effectiveness and cross-modal consistency to model the content, complicating understanding how each modality affects prediction accuracy. Additionally, these methods are primarily based on static feature modelling, making it difficult to adapt to the dynamic changes and relationships between different data modalities. This paper develops a significantly novel approach, GAMED, for multimodal modelling, which focuses on generating distinctive and discriminative features through modal decoupling to enhance cross-modal synergies, thereby optimizing overall performance in the detection process. GAMED leverages multiple parallel expert networks to refine features and pre-embed semantic knowledge to improve the experts' ability in information selection and viewpoint sharing. Subsequently, the feature distribution of each modality is adaptively adjusted based on the respective experts' opinions. GAMED also introduces a novel classification technique to dynamically manage contributions from different modalities, while improving the explainability of decisions. Experimental results on the Fakeddit and Yang datasets demonstrate that GAMED performs better than recently developed state-of-the-art models. The source code can be accessed at https://github.com/slz0925/GAMED.

User-generated content on social media platforms provides a valuable resource for developing automated computational methods to detect mental health issues online leading to suicidal thoughts automatically. Although current fully automated methods show promise, they may produce uncertain predictions, leading to flawed conclusions. To address this, we propose a novel model called DiGrI, or Distorted Greedy Approach for Human-Assisted Online Suicide Ideation Detection, which reformulates suicide ideation assessment as a selective, prioritized prediction problem. The model incorporates a novel multi-classifier distorted greedy model that is optimized to operate under various levels of automation and abstains from making uncertain predictions with theoretical guarantees. Our results show that DiGrI outperforms strong comparative models including large language models in detecting mental health issues on a publicly available Reddit dataset. We discuss the empirical and practical implications, including the ethical considerations of using DiGrI for online automatic suicide ideation detection involving humans, if it were to be translated for use in clinical and public health practice.

The aim of multi-label few-shot image classification (ML-FSIC) is to assign semantic labels to images, in settings where only a small number of training examples are available for each label. A key feature of the multi-label setting is that an image often has several labels, which typically refer to objects appearing in different regions of the image. When estimating label prototypes, in a metric-based setting, it is thus important to determine which regions are relevant for which labels, but the limited amount of training data and the noisy nature of local features make this highly challenging. As a solution, we propose a strategy in which label prototypes are gradually refined. First, we initialize the prototypes using word embeddings, which allows us to leverage prior knowledge about the meaning of the labels. Second, taking advantage of these initial prototypes, we then use a Loss Change Measurement (LCM) strategy to select the local features from the training images (i.e., the support set) that are most likely to be representative of a given label. Third, we construct the final prototype of the label by aggregating these representative local features using a multi-modal cross-interaction mechanism, which again relies on the initial word embedding-based prototypes. Experiments on COCO, PASCAL VOC, NUS-WIDE, and iMaterialist show that our model substantially improves the current state-of-the-art.

Language models such as Bidirectional Encoder Representations from Transformers (BERT) have been very effective in various Natural Language Processing (NLP) and text mining tasks including text classification. However, some tasks still pose challenges for these models, including text classification with limited labels. This can result in a cold-start problem. Although some approaches have attempted to address this problem through single-stage clustering as an intermediate training step coupled with a pre-trained language model, which generates pseudo-labels to improve classification, these methods are often error-prone due to the limitations of the clustering algorithms. To overcome this, we have developed a novel two-stage intermediate clustering with subsequent fine-tuning that models the pseudo-labels reliably, resulting in reduced prediction errors. The key novelty in our model, IDoFew, is that the two-stage clustering coupled with two different clustering algorithms helps exploit the advantages of the complementary algorithms that reduce the errors in generating reliable pseudo-labels for fine-tuning. Our approach has shown significant improvements compared to strong comparative models.

Abstract Question‐driven automatic text summarization is a popular technique to produce concise and informative answers to specific questions using a document collection. Both query‐based and question‐driven summarization may not produce reliable summaries nor contain relevant information if they do not take advantage of extractive and abstractive summarization mechanisms to improve performance. In this article, we propose a novel extractive and abstractive hybrid framework designed for question‐driven automatic text summarization. The framework consists of complimentary modules that work together to generate an effective summary: (1) discovering appropriate non‐redundant sentences as plausible answers using an open‐domain multi‐hop question answering system based on a convolutional neural network, multi‐head attention mechanism and reasoning process; and (2) a novel paraphrasing generative adversarial network model based on transformers rewrites the extracted sentences in an abstractive setup. Experiments show this framework results in more reliable abstractive summary than competing methods. We have performed extensive experiments on public datasets, and the results show our model can outperform many question‐driven and query‐based baseline methods (an R1, R2, RL increase of 6%–7% for over the next highest baseline).

People across the globe have felt and are still going through the impact of COVID-19. Some of them share their feelings and suffering online via different online social media networks such as Twitter. Due to strict restrictions to reduce the spread of the novel virus, many people are forced to stay at home, which significantly impacts people's mental health. It is mainly because the pandemic has directly affected the lives of the people who were not allowed to leave home due to strict government restrictions. Researchers must mine the related human-generated data and get insights from it to influence government policies and address people's needs. In this paper, we study social media data to understand how COVID-19 has impacted people's depression. We share a large-scale COVID-19 dataset that can be used to analyze depression. We also have modeled the tweets of depressed and non-depressed users before and after the start of the COVID-19 pandemic. To this end, we developed a new approach based on Hierarchical Convolutional Neural Network (HCN) that extracts fine-grained and relevant content on user historical posts. HCN considers the hierarchical structure of user tweets and contains an attention mechanism that can locate the crucial words and tweets in a user document while also considering the context. Our new approach is capable of detecting depressed users occurring within the COVID-19 time frame. Our results on benchmark datasets show that many non-depressed people became depressed during the COVID-19 pandemic.

Product reviews on e-commerce platforms play a critical role in shaping users’ purchasing decisions. Unfortunately, online reviews sometimes can be intentionally misleading to manipulate the ecosystem. To date, existing methods to automatically detect “spam reviews” either focus on sophisticated feature engineering with traditional classification models or rely on tuning neural networks with aggregated features. In this article, we develop a novel graph-based model, namely, graph-aware deep fusion networks (GDFNs) that use information from relevant metadata (review text, features of users, and items) and relational data (network) to capture the semantic information from their complex heterogeneous interactions via graph convolutional networks (GCNs). Besides, GDFN also uses a novel fusion technique to synthesize low- and high-order interactions with propagated information across multiple review-related subgraphs. Extensive experiments on publicly available datasets show that our proposed model is effective and outperforms several strong state-of-the-art baselines.

Visualizing and examining the intellectual landscape and evolution of scientific communities to support collaboration is crucial for multiple research purposes. In some cases, measuring similarities and matching patterns between research publication document sets can help to identify people with similar interests for building research collaboration networks and university–industry linkages. The premise of this work is assessing feasibility for resolving ambiguous cases in similarity detection to determine authorship with natural language processing (NLP) techniques so that crowdsourcing is applied only in instances that require human judgment. Using an NLP-crowdsourcing convergence strategy, we can reduce the costs of microtask crowdsourcing while saving time and maintaining disambiguation accuracy over large datasets. This article contributes a next-gen crowd-artificial intelligence framework that used an ensemble of term frequency-inverse document frequency and bidirectional encoder representation from transformers to obtain similarity rankings for pairs of scientific documents. A sequence of content-based similarity tasks was created using a crowd-powered interface for solving disambiguation problems. Our experimental results suggest that an adaptive NLP-crowdsourcing hybrid framework has advantages for inter-researcher similarity detection tasks where fully automatic algorithms provide unsatisfactory results, with the goal of helping researchers discover potential collaborators using data-driven approaches.

Multi-label few-shot image classification (ML-FSIC) is the task of assigning descriptive labels to previously unseen images, based on a small number of training examples. A key feature of the multi-label setting is that images often have multiple labels, which typically refer to different regions of the image. When estimating prototypes, in a metric-based setting, it is thus important to determine which regions are relevant for which labels, but the limited amount of training data makes this highly challenging. As a solution, in this paper we propose to use word embeddings as a form of prior knowledge about the meaning of the labels. In particular, visual prototypes are obtained by aggregating the local feature maps of the support images, using an attention mechanism that relies on the label embeddings. As an important advantage, our model can infer prototypes for unseen labels without the need for fine-tuning any model parameters, which demonstrates its strong generalization abilities. Experiments on COCO and PASCAL VOC furthermore show that our model substantially improves the current state-of-the-art.

Abstract The ability to explain why the model produced results in such a way is an important problem, especially in the medical domain. Model explainability is important for building trust by providing insight into the model prediction. However, most existing machine learning methods provide no explainability, which is worrying. For instance, in the task of automatic depression prediction, most machine learning models lead to predictions that are obscure to humans. In this work, we propose explainable Multi-Aspect Depression Detection with Hierarchical Attention Network MDHAN, for automatic detection of depressed users on social media and explain the model prediction. We have considered user posts augmented with additional features from Twitter. Specifically, we encode user posts using two levels of attention mechanisms applied at the tweet-level and word-level, calculate each tweet and words’ importance, and capture semantic sequence features from the user timelines (posts). Our hierarchical attention model is developed in such a way that it can capture patterns that leads to explainable results. Our experiments show that MDHAN outperforms several popular and robust baseline methods, demonstrating the effectiveness of combining deep learning with multi-aspect features. We also show that our model helps improve predictive performance when detecting depression in users who are posting messages publicly on social media. MDHAN achieves excellent performance and ensures adequate evidence to explain the prediction.

With the increasing development of human-AI teaming structures within and across geographies, the time is ripe for a continuous and objective look at the predictors, barriers, and facilitators of human-AI scientific collaboration from a multidisciplinary point of view. This paper aims at contributing to this end by exploiting a set of factors affecting attitudes towards the adoption of human-AI interaction into scientific work settings. In particular, we are interested in identifying the determinants of trust and acceptability when considering the combination of hybrid human-AI approaches for improving research practices. This includes the way as researchers assume human-centered artificial intelligence (AI) and crowdsourcing as valid mechanisms for aiding their tasks. Through the lens of a unified theory of acceptance and use of technology (UTAUT) combined with an extended technology acceptance model (TAM), we pursue insights on the perceived usefulness, potential blockers, and adoption drivers that may be representative of the intention to use hybrid intelligence systems as a way of unveiling unknown patterns from large amounts of data and thus enabling novel scientific discoveries.

Advertising is critical to many online e-commerce platforms such as e-Bay and Amazon. One of the important signals that these platforms rely upon is the click-through rate (CTR) prediction. The recent popularity of multi-modal sharing platforms such as TikTok has led to an increased interest in online micro-videos. It is, therefore, useful to consider micro-videos to help a merchant target micro-video advertising better and find users' favourites to enhance user experience. Existing works on CTR prediction largely exploit unimodal content to learn item representations. A relatively minimal effort has been made to leverage multi-modal information exchange among users and items. We propose a model to exploit the temporal user-item interactions to guide the representation learning with multi-modal features, and further predict the user click rate of the micro-video item. We design a Hypergraph Click-Through Rate prediction framework (HyperCTR) built upon the hyperedge notion of hypergraph neural networks, which can yield modal-specific representations of users and micro-videos to better capture user preferences. We construct a time-aware user-item bipartite network with multi-modal information and enrich the representation of each user and item with the generated interests-based user hypergraph and item hypergraph. Through extensive experiments on three public datasets, we demonstrate that our proposed model significantly outperforms various state-of-the-art methods.

Despite decades of research and development in named entity resolution, dealing with name ambiguity is still a challenging issue for many bibliometric-enhanced information retrieval (IR) tasks. As new bibliographic datasets are created as a result of the upward growth of publication records worldwide, more problems arise when considering the effects of errors resulting from missing data fields, duplicate entities, misspellings, extra characters, etc. As these concerns tend to be of large-scale, both the general consistency and the quality of electronic data are largely affected. This paper presents an approach to handle these name ambiguity problems through the use of crowdsourcing as a complementary means to traditional unsupervised approaches. To this end, we present “AuthCrowd”, a crowdsourcing system with the ability to decompose named entity disambiguation and entity matching tasks. Experimental results on a real-world dataset of publicly available papers published in peer-reviewed venues demonstrate the potential of our proposed approach for improving author name disambiguation. The findings further highlight the importance of adopting hybrid crowd-algorithm collaboration strategies, especially for handling complexity and quantifying bias when working with large amounts of data.

It has been a quarter of a century since the publication of the first edition of the IEEE International Conference on Computer Supported Cooperative Work in Design (CSCWD) held in 1996 in Beijing, China. Despite some attempts to empirically examine the evolution and identity of the field of CSCW and its related communities and disciplines, the scarcity of scientometric studies on the IEEE CSCWD research productivity is noteworthy. To fill this gap, this study reports on an exploratory quantitative analysis of the literature published in the IEEE CSCWD conference proceedings with the purpose of visualizing and understanding its structure and evolution for the 2001-2019 period. The findings offer valuable insights into the paper and author distribution, country and citation-level productivity indicators, degree of collaboration, and collaboration index. Through this analysis we also expect to get an initial overview of the IEEE CSCWD conference concerning the main topics being presented, most cited papers, and variances in the number of keywords, full-text views, and references.

The aim of few-shot learning (FSL) is to learn how to recognize image categories from a small number of training examples. A central challenge is that the available training examples are normally insufficient to determine which visual features are most characteristic of the considered categories. To address this challenge, we organise these visual features into facets, which intuitively group features of the same kind (e.g. features that are relevant to shape, color, or texture). This is motivated from the assumption that (i) the importance of each facet differs from category to category and (ii) it is possible to predict facet importance from a pre-trained embedding of the category names. In particular, we propose an adaptive similarity measure, relying on predicted facet importance weights for a given set of categories. This measure can be used in combination with a wide array of existing metric-based methods. Experiments on miniImageNet and CUB show that our approach improves the state-of-the-art in metric-based FSL.

Few-shot learning (FSL) is the task of learning to recognize previously unseen categories of images from a small number of training examples. This is a challenging task, as the available examples may not be enough to unambiguously determine which visual features are most characteristic of the considered categories. To alleviate this issue, we propose a method that additionally takes into account the names of the image classes. While the use of class names has already been explored in previous work, our approach differs in two key aspects. First, while previous work has aimed to directly predict visual prototypes from word embeddings, we found that better results can be obtained by treating visual and text-based prototypes separately. Second, we propose a simple strategy for learning class name embeddings using the BERT language model, which we found to substantially outperform the GloVe vectors that were used in previous work. We furthermore propose a strategy for dealing with the high dimensionality of these vectors, inspired by models for aligning cross-lingual word embeddings. We provide experiments on miniImageNet, CUB and tieredImageNet, showing that our approach consistently improves the state-of-the-art in metric-based FSL.

Twitter is currently a popular online social media platform which allows users to share their user-generated content. This publicly-generated user data is also crucial to healthcare technologies because the discovered patterns would hugely benefit them in several ways. One of the applications is in automatically discovering mental health problems, e.g., depression. Previous studies to automatically detect a depressed user on online social media have largely relied upon the user behaviour and their linguistic patterns including user's social interactions. The downside is that these models are trained on several irrelevant content which might not be crucial towards detecting a depressed user. Besides, these content have a negative impact on the overall efficiency and effectiveness of the model. To overcome the shortcomings in the existing automatic depression detection methods, we propose a novel computational framework for automatic depression detection that initially selects relevant content through a hybrid extractive and abstractive summarization strategy on the sequence of all user tweets leading to a more fine-grained and relevant content. The content then goes to our novel deep learning framework comprising of a unified learning machinery comprising of Convolutional Neural Network (CNN) coupled with attention-enhanced Gated Recurrent Units (GRU) models leading to better empirical performance than existing strong baselines.

Since early 2020, the COVID-19 global pandemic has significantly impacted almost every aspect of the human society throughout the world. Until now, middle of 2021, although with all the efforts on pandemic intervention and vaccination, COVID-19 is still hovering around the world, resulting in more than 177 million confirmed cases and 3.8 million deaths.

The recent Coronavirus Infectious Disease 2019 (COVID-19) pandemic has caused an unprecedented impact across the globe. We have also witnessed millions of people with increased mental health issues, such as depression, stress, worry, fear, disgust, sadness, and anxiety, which have become one of the major public health concerns during this severe health crisis. Depression can cause serious emotional, behavioral, and physical health problems with significant consequences, both personal and social costs included. This article studies community depression dynamics due to the COVID-19 pandemic through user-generated content on Twitter. A new approach based on multimodal features from tweets and term frequency-inverse document frequency (TF-IDF) is proposed to build depression classification models. Multimodal features capture depression cues from emotion, topic, and domain-specific perspectives. We study the problem using recently scraped tweets from Twitter users emanating from the state of New South Wales in Australia. Our novel classification model is capable of extracting depression polarities that may be affected by COVID-19 and related events during the COVID-19 period. The results found that people became more depressed after the outbreak of COVID-19. The measures implemented by the government, such as the state lockdown, also increased depression levels.

With cutting edge scientific breakthroughs, human-centred algorithmic approaches have proliferated in recent years and information technology (IT) has begun to redesign socio-technical systems in the context of human-AI collaboration. As a result, distinct forms of interaction have emerged in tandem with the proliferation of infrastructures aiding interdisciplinary work practices and research teams. Concomitantly, large volumes of heterogeneous datasets are produced and consumed at a rapid pace across many scientific domains. This results in difficulties in the reliable analysis of scientific production since current tools and algorithms are not necessarily able to provide acceptable levels of accuracy when analyzing the content and impact of publication records from large continuous scientific data streams. On the other hand, humans cannot consider all the information available and may be adversely influenced by extraneous factors. Using this rationale, we propose an initial design of a human-AI enabled pipeline for performing scientometric analyses that exploits the intersection between human behavior and machine intelligence. The contribution is a model for incorporating central principles of human-machine symbiosis (HMS) into scientometric workflows, demonstrating how hybrid intelligence systems can drive and encapsulate the future of research evaluation.

Recently, many KB-to-text generation tasks have been proposed to bridge the gap between knowledge bases and natural language by directly converting a group of knowledge base triples into human-readable sentences.However, most of the existing models suffer from the off-topic problem, namely, the models are prone to generate some unrelated clauses that are somehow involved with certain input terms regardless of the given input data.This problem seriously degrades the quality of the generation results.In this paper, we propose a novel dynamic topic tracker for solving this problem.Different from existing models, our proposed model learns a global hidden representation for topics and recognizes the corresponding topic during each generation step.The recognized topic is used as additional information to guide the generation process and thus alleviates the off-topic problem.The experimental results show that our proposed model can enhance the performance of sentence generation and the off-topic problem is significantly mitigated.

Word embedding models typically learn two types of vectors: target word vectors and context word vectors.These vectors are normally learned such that they are predictive of some word co-occurrence statistic, but they are otherwise unconstrained.However, the words from a given language can be organized in various natural groupings, such as syntactic word classes (e.g.nouns, adjectives, verbs) and semantic themes (e.g.sports, politics, sentiment).Our hypothesis in this paper is that embedding models can be improved by explicitly imposing a cluster structure on the set of context word vectors.To this end, our model relies on the assumption that context word vectors are drawn from a mixture of von Mises-Fisher (vMF) distributions, where the parameters of this mixture distribution are jointly optimized with the word vectors.We show that this results in word vectors which are qualitatively different from those obtained with existing word embedding models.We furthermore show that our embedding model can also be used to learn high-quality document representations.

The structure and evolution of a scientific research community can be quantitatively assessed taking into account the interactions between scientific agents dispersed geographically. In the recent years, CSCW has stabilized as a cross-disciplinary field suffering significant changes in its core structure, and there is limited understanding about the factors influencing the nature and progress of collaborative computing research. In this paper, we measure the correlation between a set of features related to the influence of collaboration types on the number of citations as well as the geographical distribution of the accumulated contribution to the CSCW literature. Overall, our work can represent a starting point to demonstrate how the study of scientific collaboration can partly explain the variations in the number of citations, frequency of papers, and topics addressed.

Recently a number of unsupervised approaches have been proposed for learning vectors that capture the relationship between two words. Inspired by word embedding models, these approaches rely on co-occurrence statistics that are obtained from sentences in which the two target words appear. However, the number of such sentences is often quite small, and most of the words that occur in them are not relevant for characterizing the considered relationship. As a result, standard co-occurrence statistics typically lead to noisy relation vectors. To address this issue, we propose a latent variable model that aims to explicitly determine what words from the given sentences best characterize the relationship between the two target words. Relation vectors then correspond to the parameters of a simple unigram language model which is estimated from these words.

A substantial amount of work is often overlooked due to the exponential rate of growth in global scientific output across all disciplines. Current approaches for addressing this issue are usually limited in scope and often restrict the possibility of obtaining multidisciplinary views in practice. To tackle this problem, researchers can now leverage an ecosystem of citizens, volunteers and crowd workers to perform complex tasks that are either difficult for humans and machines to solve alone. Motivated by the idea that human crowds and computer algorithms have complementary strengths, we present an approach where the machine will learn from crowd behavior in an iterative way. This approach is embodied in the architecture of SciCrowd, a crowd-powered human-machine hybrid system designed to improve the analysis and processing of large amounts of publication records. To validate the proposal's feasibility, a prototype was developed and an initial evaluation was conducted to measure its robustness and reliability. We conclude this paper with a set of implications for design.

Crowdsourcing has shown to be a valuable problem-solving approach to handle the increasing complexity and scale of tasks for which the current AI algorithms are still struggling. Crowd intelligence can be particularly useful to train and supervise AI systems in a symbiotic, co-evolutionary relationship that raises long-term research challenges to the hybrid, crowd-computing design space. With the increase in the scale of mixed-initiative approaches, we need to gain a better understanding of the implications of crowd-powered systems as a scaffold for AI through the study of massive crowd-machine interactions. In this paper, we identify some open challenges and design implications for future crowd-AI hybrid systems. A framework is also proposed based on the practical challenges of addressing human-centered AI methods and processes.

Word embedding models such as GloVe rely on co-occurrence statistics to learn vector representations of word meaning.While we may similarly expect that cooccurrence statistics can be used to capture rich information about the relationships between different words, existing approaches for modeling such relationships are based on manipulating pre-trained word vectors.In this paper, we introduce a novel method which directly learns relation vectors from co-occurrence statistics.To this end, we first introduce a variant of GloVe, in which there is an explicit connection between word vectors and PMI weighted co-occurrence vectors.We then show how relation vectors can be naturally embedded into the resulting vector space.

Given a set of instances of some relation, the relation induction task is to predict which other word pairs are likely to be related in the same way. While it is natural to use word embeddings for this task, standard approaches based on vector translations turn out to perform poorly. To address this issue, we propose two probabilistic relation induction models. The first model is based on translations, but uses Gaussians to explicitly model the variability of these translations and to encode soft constraints on the source and target words that may be chosen. In the second model, we use Bayesian linear regression to encode the assumption that there is a linear relationship between the vector representations of related words, which is considerably weaker than the assumption underlying translation based models.

Structured knowledge about concepts plays an increasingly important role in areas such as information retrieval. The available ontologies and knowledge graphs that encode such conceptual knowledge, however, are inevitably incomplete. This observation has led to a number of methods that aim to automatically complete existing knowledge bases. Unfortunately, most existing approaches rely on black box models, e.g. formulated as global optimization problems, which makes it difficult to support the underlying reasoning process with intuitive explanations. In this paper, we propose a new method for knowledge base completion, which uses interpretable conceptual space representations and an explicit model for inductive inference that is closer to human forms of commonsense reasoning. Moreover, by separating the task of representation learning from inductive reasoning, our method is easier to apply in a wider variety of contexts. Finally, unlike optimization based approaches, our method can naturally be applied in settings where various logical constraints between the extensions of concepts need to be taken into account.

Vector representations of word meaning have found many applications in the field of natural language processing.Word vectors intuitively represent the average context in which a given word tends to occur, but they cannot explicitly model the diversity of these contexts.Although region representations of word meaning offer a natural alternative to word vectors, only few methods have been proposed that can effectively learn word regions.In this paper, we propose a new word embedding model which is based on SVM regression.We show that the underlying ranking interpretation of word contexts is sufficient to match, and sometimes outperform, the performance of popular methods such as Skip-gram.Furthermore, we show that by using a quadratic kernel, we can effectively learn word regions, which outperform existing unsupervised models for the task of hypernym detection.

We propose a new class of methods for learning vector space embeddings of entities. While most existing methods focus on modelling similarity, our primary aim is to learn embeddings that are interpretable, in the sense that query terms have a direct geometric representation in the vector space. Intuitively, we want all entities that have some property (i.e. for which a given term is relevant) to be located in some well-defined region of the space. This is achieved by imposing max-margin constraints that are derived from a bag-of-words representation of the entities. The resulting vector spaces provide us with a natural vehicle for identifying entities that have a given property (or ranking them according to how much they have the property), and conversely, to describe what a given set of entities have in common. As we show in our experiments, our models lead to a substantially better performance in a range of entity-oriented search tasks, such as list completion and entity ranking.

Word embedding models such as Skip-gram learn a vector-space representation for each word, based on the local word collocation patterns that are observed in a text corpus. Latent topic models, on the other hand, take a more global view, looking at the word distributions across the corpus to assign a topic to each word occurrence. These two paradigms are complementary in how they represent the meaning of word occurrences. While some previous works have already looked at using word embeddings for improving the quality of latent topics, and conversely, at using latent topics for improving word embeddings, such "two-step'' methods cannot capture the mutual interaction between the two paradigms. In this paper, we propose STE, a framework which can learn word embeddings and latent topics in a unified manner. STE naturally obtains topic-specific word embeddings, and thus addresses the issue of polysemy. At the same time, it also learns the term distributions of the topics, and the topic distributions of the documents. Our experimental results demonstrate that the STE model can indeed generate useful topic-specific word embeddings and coherent latent topics in an effective and efficient way.

We propose a new word embedding model, inspired by GloVe, which is formulated as a feasible least squares optimization problem. In contrast to existing models, we explicitly represent the uncertainty about the exact definition of each word vector. To this end, we estimate the error that results from using noisy co-occurrence counts in the formulation of the model, and we model the imprecision that results from including uninformative context words. Our experimental results demonstrate that this model compares favourably with existing word embedding models.

Conceptual spaces are geometric representations of conceptual knowledge in which entities correspond to points, natural properties correspond to convex regions, and the dimensions of the space correspond to salient features. While conceptual spaces enable elegant models of various cognitive phenomena, the lack of automated methods for constructing such representations have so far limited their application in artificial intelligence. To address this issue, we propose a method which learns a vector-space embedding of entities from Wikipedia and constrains this embedding such that entities of the same semantic type are located in some lower-dimensional subspace. We experimentally demonstrate the usefulness of these subspaces as approximate conceptual space representations by showing, among others, that important features can be modelled as directions and that natural properties tend to correspond to convex regions.

While most methods for learning-to-rank documents only consider relevance scores as features, better results can often be obtained by taking into account the latent topic structure of the document collection. Existing approaches that consider latent topics follow a two-stage approach, in which topics are discovered in an unsupervised way, as usual, and then used as features for the learning-to-rank task. In contrast, we propose a learning-to-rank framework which integrates the supervised learning of a maximum margin classifier with the discovery of a suitable probabilistic topic model. In this way, the labelled data that is available for the learning-to-rank task can be exploited to identify the most appropriate topics. To this end, we use a unified constrained optimization framework, which can dynamically compute the latent topic similarity score between the query and the document. Our experimental results show a consistent improvement over the state-of-the-art learning-to-rank models.

This work presents our attempt to understand the research topics that characterize the papers submitted to a conference, by using topic modeling and data visualization techniques. We infer the latent topics from the abstracts of all the papers submitted to Interspeech2014 by means of Latent Dirichlet Allocation. Pertopic word distributions thus obtained are visualized through word clouds. We also compare the automatically inferred topics against the expert-defined topics (also known as tracks for Interspeech2014). The comparison is based on an information retrieval framework, where we use each latent topic as a query and each track as a document. For each latent topic, we retrieve a ranked list of tracks scored by the degree of word overlap. Each latent topic is associated with the top-scoring track. This analytic procedure was applied to all submissions to Interspeech2014 and sheds some interesting light in terms of providing an overview of topic categorization in the conference, popular versus unpopular topics, emerging topics and topic compositions. Such insights are potentially valuable for understanding the technical content of a field and planning the future development of its conference(s).

An important way to improve users’ satisfaction in Web search is to assist them by issuing more effective queries. One such approach is query reformulation, which generates new queries according to the current query issued by users. A common procedure for conducting reformulation is to generate some candidate queries first, then a scoring method is employed to assess these candidates. Currently, most of the existing methods are context based. They rely heavily on the context relation of terms in the history queries and cannot detect and maintain the semantic consistency of queries. In this article, we propose a graphical model to score queries. The proposed model exploits a latent topic space, which is automatically derived from the query log, to detect semantic dependency of terms in a query and dependency among topics. Meanwhile, the graphical model also captures the term context in the history query by skip-bigram and n-gram language models. In addition, our model can be easily extended to consider users’ history search interests when we conduct query reformulation for different users. In the task of candidate query generation, we investigate a social tagging data resource—Delicious bookmark—to generate addition and substitution patterns that are employed as supplements to the patterns generated from query log data.

We present a new unsupervised topic discovery model for a collection of text documents. In contrast to the majority of the state-of-the-art topic models, our model does not break the document's structure such as paragraphs and sentences. In addition, it preserves word order in the document. As a result, it can generate two levels of topics of different granularity, namely, segment-topics and word-topics. In addition, it can generate n-gram words in each topic. We also develop an approximate inference scheme using Gibbs sampling method. We conduct extensive experiments using publicly available data from different collections and show that our model improves the quality of several text mining tasks such as the ability to support fine grained topics with n-gram words in the correlation graph, the ability to segment a document into topically coherent sections, document classification, and document likelihood estimation.

We propose a novel framework for determining the conceptual difficulty of a domain-specific text document without using any external lexicon. Conceptual difficulty relates to finding the reading difficulty of domain-specific documents. Previous approaches to tackling domain-specific readability problem have heavily relied upon an external lexicon, which limits the scalability to other domains. Our model can be readily applied in domain-specific vertical search engines to re-rank documents according to their conceptual difficulty. We develop an unsupervised and principled approach for computing a term's conceptual difficulty in the latent space. Our approach also considers transitions between the segments generated in sequence. It performs better than the current state-of-the-art comparative methods.

Users look for information that can suit their level of expertise, but it often takes a mammoth effort to trace such information. One has to sift through multiple pages to look for one that fits the appropriate technical background. In this paper, a query-independent ranking system is proposed for technical web pages. The pages returned by the system are sorted by their relative technical difficulty in either ascending or descending order specified by the user. The technical difficulty of a document i.e. terms in sequence, is first computed by the combination of each individual term's geometry in the low-dimensional latent semantic indexing (LSI) space, which can be visualized as a conceptual terrain. Then the pages are ranked based on the expected cost to get over the terrain. Results indicate that our terrain based method outperforms traditional readability measures.