Fine-grained Analysis Research Articles

How Do First-Year Engineering Students’ Emotions Change while Working on Programming Problems?

Emotions are a complex multi-faceted phenomenon. To assess the complexity of emotions from different facets, multi-modal approaches are necessary. However, multi-modal approaches are rarely used for assessing emotions, especially in the context of computer programming. This study adopts a multi-modal approach to understand the changes in students’ perception of emotions before and after working on programming problems. Understanding these changes in students’ perceptions may enable educators to devise interventions that help students adjust their perceptions and regulate their emotions as per their skills. We conducted a one-on-one programming session and retrospective think-aloud interview with 17 students from an introductory programming course. During the programming session, students filled surveys and performed four programming tasks. While working on these tasks, students’ eye gaze, video of face and screen, and electrodermal activity data were also collected using a non-invasive device. The data collection for this study was multi-modal, with a mix of both qualitative and quantitative data collection methods. Data analysis was primarily qualitative, with additional triangulation of qualitative and biometric data for select exemplars. The findings of this study suggest that students experience changes in emotions because of many reasons, for instance, they encountered repeated errors, they set high standards for their performance, or they could not manage time. For some students, negative emotions changed to positive emotions when they solved errors without any external help or achieved more than what they expected going into the task. Moreover, the triangulation of qualitative and biometric data of two participants provides a fine-grained analysis of their emotions and behaviors and confirmed the change in the perception of their emotions while performing the programming tasks.

Cross-Domain Aspect-Based Sentiment Classification with a Pre-Training and Fine-Tuning Strategy for Low-Resource Domains

Aspect-based sentiment classification (ABSC) is a crucial sub-task of fine-grained sentiment analysis, which aims to predict the sentiment polarity of the given aspects in a sentence as positive, negative, or neutral. Most existing ABSC methods are based on supervised learning. However, these methods rely heavily on fine-grained labeled training data, which can be scarce in low-resource domains, limiting their effectiveness. To overcome this challenge, we propose a low-resource cross-domain aspect-based sentiment classification (CDABSC) approach based on a pre-training and fine-tuning strategy. This approach applies the pre-training and fine-tuning strategy to an advanced deep learning method designed for ABSC, namely the attention-based encoding graph convolutional network (AEGCN) model. Specifically, a high-resource domain is selected as the source domain, and the AEGCN model is pre-trained using a large amount of fine-grained annotated data from the source domain. The optimal parameters of the model are preserved. Subsequently, a low-resource domain is used as the target domain, and the pre-trained model parameters are used as the initial parameters of the target domain model. The target domain is fine-tuned using a small amount of annotated data to adapt the parameters to the target domain model, improving the accuracy of sentiment classification in the low-resource domain. Finally, experimental validation on two domain benchmark datasets, restaurant and laptop, demonstrates significant outperformance of our approach over the baselines in CDABSC Micro-F1.

Room-scale Location Trace Tracking via Continuous Acoustic Waves

The increasing prevalence of smart devices spurs the development of emerging indoor localization technologies for supporting diverse personalized applications at home. Given marked drawbacks of popular chirp signal-based approaches, we aim at developing a novel device-free localization system via the continuous wave of the inaudible frequency. To achieve this goal, solutions are developed for fine-grained analyses, able to precisely locate moving human traces in the room-scale environment. In particular, a smart speaker is controlled to emit continuous waves at inaudible 20kHz , with a co-located microphone array to record their Doppler reflections for localization. We first develop solutions to remove potential noises and then propose a novel idea by slicing signals into a set of narrowband signals, each of which is likely to include at most one body segment’s reflection. Different from previous studies, which take original signals themselves as the baseband, our solutions employ the Doppler frequency of a narrowband signal to estimate the velocity first and apply it to get the accurate baseband frequency, which permits a precise phase measurement after I-Q (i.e., in-phase and quadrature) decomposition. A signal model is then developed, able to formulate the phase with body segment’s velocity, range, and angle. We next develop novel solutions to estimate the motion state in each narrowband signal, cluster the motion states for different body segments corresponding to the same person, and locate the moving traces while mitigating multi-path effects. Our system is implemented with commodity devices in room environments for performance evaluation. The experimental results exhibit that our system can conduct effective localization for up to three persons in a room, with the average errors of 7.49 cm for a single person, with 24.06 cm for two persons, with 51.15 cm for three persons.

To Diverge or Not to Diverge: A Morphosyntactic Perspective on Machine Translation vs Human Translation

Abstract We conduct a large-scale fine-grained comparative analysis of machine translations (MTs) against human translations (HTs) through the lens of morphosyntactic divergence. Across three language pairs and two types of divergence defined as the structural difference between the source and the target, MT is consistently more conservative than HT, with less morphosyntactic diversity, more convergent patterns, and more one-to-one alignments. Through analysis on different decoding algorithms, we attribute this discrepancy to the use of beam search that biases MT towards more convergent patterns. This bias is most amplified when the convergent pattern appears around 50% of the time in training data. Lastly, we show that for a majority of morphosyntactic divergences, their presence in HT is correlated with decreased MT performance, presenting a greater challenge for MT systems.

Spatial roles in hockey special teams

Abstract Special teams (i.e. power play and penalty kill) situations play an outsized role in determining the outcome of ice hockey games. Yet, quantitative methods for characterizing special teams tactics are limited. This work focuses on team structure and player deployment during in-zone special teams possessions. Leveraging player and puck tracking data from the National Hockey League (NHL), a framework is developed for describing player positioning during 5-on-4 power play and 4-on-5 penalty kill possessions. More specifically, player roles are defined directly from the player tracking data using non-negative matrix factorization, and every player is allocated a unique role at every frame of tracking data by solving a linear assignment problem. Team formations naturally arise through the combination of roles occupied in a frame. Roles that vary on a per-frame basis allow for a fine-grained analysis of team structure. This property of the roles-based representation is used to group together similar power play possessions using latent Dirichlet allocation, a topic modelling technique. The concept of assignments, which remain constant over an entire possession, is also introduced. Assignments provide a more stable measure of player positioning, which may be preferable when assessing deployment over longer periods of time.

How Positive Sentiment of Telemedicine Applications using Naïve Bayes and Support Vector Machine?

The telemedicine software lets users schedule appointments with physicians or hospitals, purchase medications online, read health-related news and publications, and have online consultations with doctors. Public opinion is necessary for this application to meet societal needs. The public's perception of contemporary health applications can be ascertained by looking to public opinion. Fine-grained sentiment analysis is used in public opinion taking to extract information from evaluations posted by the public and categorize it into positive and negative groups. Using RapidMiner as a supporting tool, sentiment analysis uses the Naïve Bayes Classification algorithm and Support Vector Machine to determine an accuracy value. After analyzing data from two different sources—Twitter and Playstore—it can be said that the Support Vector Machine algorithm performs 90.99% more accurately in predicting public opinion than the Naïve Bayes algorithm, which achieves 88.55% accuracy.

EPri-MDAS: An efficient privacy-preserving multiple data aggregation scheme without trusted authority for fog-based smart grid

With the increasingly pervasive deployment of fog servers, fog computing extends data processing and analysis to network edges. At the same time, as the next-generation power grid, the smart grid should meet the requirements of security, efficiency, and real-time monitoring of user energy consumption. By utilizing the low-latency and distributed properties of fog computing, it can improve communication efficiency and user service satisfaction in smart grids. For the sake of providing adequate functionality for the power grid, various schemes have been proposed. Whereas, many methods are vulnerable to privacy leakage since the existence of trusted authority may increase the exposure to threats. In this paper, we propose the EPri-MDAS: an Efficient Privacy-preserving Multiple Data Aggregation Scheme without trusted authority based on the ElGamal homomorphic cryptosystem, which achieves both data integrity verification and data source authentication with the most efficient block cipher-based authenticated encryption algorithm. It performs well in energy efficiency with strong security. Especially, the proposed multidimensional aggregation statistics scheme can perform the fine-grained data analyses; it also allows for fault tolerance while protecting personal privacy. The security analysis and simulation experiments show that EPri-MDAS can satisfy the security requirements and work efficiently in the smart grid.

Processing of visual social-communication cues during a social-perception of action task in autistic and non-autistic observers

Social perception and communication differ between those with and without autism, even when verbal fluency and intellectual ability are equated. Previous work found that observers responded more quickly to an actor's points if the actor had chosen by themselves where to point instead of being directed where to point. Notably, this ‘choice-advantage’ effect decreased across non-autistic participants as the number of autistic-like traits and tendencies increased (Pesquita et al., 2016). Here, we build on that work using the same task to study individuals over a broader range of the spectrum, from autistic to non-autistic, measuring both response initiation and mouse movement times, and considering the response to each actor separately. Autistic and non-autistic observers viewed videos of three different actors pointing to one of two locations, without knowing that the actors were sometimes freely choosing to point to one target and other times being directed where to point. All observers exhibited a choice-advantage overall, meaning they responded more rapidly when actors were freely choosing versus when they were directed, indicating a sensitivity to the actors' postural cues and movements. Our fine-grained analyses found a more robust choice-advantage to some actors than others, with autistic observers showing a choice-advantage only in response to one of the actors, suggesting that both actor and observer characteristics influence the overall effect. We briefly explore existing actor characteristics that may have contributed to this effect, finding that both duration of exposure to pre-movement cues and kinematic cues of the actors likely influence the choice advantage to different degrees across the groups. Altogether, the evidence suggested that both autistic and non-autistic individuals could detect the choice-advantage signal, but that for autistic observers the choice-advantage was actor specific. Notably, we found that the influence of the signal, when present, was detected early for all actors by the non-autistic observers, but detected later and only for one actor by the autistic observers. Altogether, we have more accurately characterized the ability of social-perception in autistic individuals as intact, but highlighted that detection of signal is likely delayed/distributed compared to non-autistic observers and that it is important to investigate actor characteristics that may influence detection and use of their social-perception signals.

Accurate and fast congestion feedback in MEC-enabled RDMA datacenters

Mobile edge computing (MEC) is a novel computing paradigm that pushes computation and storage resources to the edge of the network. The interconnection of edge servers forms small-scale data centers, enabling MEC to provide low-latency network services for mobile users. Nowadays, Remote Direct Memory Access (RDMA) has been widely deployed in such data centers to reduce CPU overhead and network latency. Plenty of congestion control mechanisms have been proposed for RDMA data centers, aiming to provide low-latency data delivery and high throughput network services. However, our fine-grained experimental analysis reveals that existing congestion control mechanisms still have performance limitations due to inappropriate congestion notifications and the long congestion feedback cycle. In this paper, we propose Mercury, which is an accurate and fast congestion feedback mechanism. Mercury comprises two key components: (1) the state-driven congestion detection and (2) the window-based congestion notification. Specifically, the state-driven congestion detection monitors the queue length and the number of packets received at the switch egress port when the PFC is triggered. It determines the states of egress ports and identifies flows that really contribute to congestion. Then, window-based congestion notification calculates the window sizes for congested flows and rapidly returns Congestion Notification Packets (CNPs) with the window information to the sender. It facilitates the rate adjustment of congested flows. Mercury is compatible with existing RDMA CC mechanisms and can be easily implemented in switches. We employ real-world data sets and conduct both micro-benchmark and large-scale simulations to evaluate the performance of Mercury. The results indicate that, thanks to the accurate and fast congestion feedback, Mercury achieves a reduction in the 99th tail flow completion time by up to 45.1%, 41.8%, 38.7%, 30.9%, and 37.9% compared with Timely, DCQCN, DCQCN+TCD, PACC, and HPCC, respectively.

ABSA for Chinese review with fine-tuned BERT and zero-shot classification

In recent years, there has been a trend towards utilizing Large Language Models (LLMs) to address diverse issues in computational linguistics. More parameters and layers make them better suited to approximating the objective function in multi-domain tasks. In the twenty-first century, since e-commerce has grown quickly, people's interest in gleaning insightful information from other people's evaluations has grown. Conducting fine-grained sentiment analysis is becoming more beneficial. Aspect-Based Sentiment Analysis (ABSA) is a sub-problem within the field of Sentiment Analysis. The ABSA problem encompasses two sub-problems: Identification of targets and Opinion Mining. Existing methods address ABSA well with a relationship between the aspect and the labels. It is the most common situation in application scenarios. This paper innovatively proposes an ABSA method based on zero-shot classification combined and fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model. The method is used experimentally in five tags (1-5) at the "service" and "environment" levels. In the experiment, the predictions of the method-derived model are contrasted with those of the original model. The results show a substantial improvement in the model's prediction accuracy in different aspects.

Twin Towers End to End model for aspect-based sentiment analysis

Aspect-based sentiment analysis (ABSA) aims to conduct fine-grained sentiment analysis, necessitating the extraction of three key components: target entity, aspect category and sentiment polarity. These three components collectively form an integrated ABSA task known as TASD (Target-Aspect-Sentiment jointly Detection). Most of existing approaches on ABSA usually employ Recurrent neural networks(RNNs), Convolutional neural networks(CNNs) or pre-training models such as Bidirectional Encoder Representations from Transformers(BERT). However, they have some common weaknesses. First, most of the existing methods focus on one or two sub-tasks instead of triplet detection, thus they do not establish an end-to-end (training a complex learning system represented by a single model) ABSA model and cannot utilize the relevance of multiple ABSA sub-tasks during training. Second, they cannot achieve accuracy and efficiency simultaneously due to the coupling of context and given aspects. Third, they are poor in recognizing implicit targets. To tackle these limitations, this paper proposes a novel method named the Twin Towers End to End model (TTEE) to solve TASD task. It transforms complex TASD task into a simple end-to-end multi-task framework, simultaneously conducting target and aspect-sentiment detection. It builds twin towers system based on BERT or its updated versions to decouple context and given aspects, which can reduce redundant calculation to improve computational efficiency significantly. It offers great advantage to identify implicit target entity and its associated aspect-sentiment in the context without introducing extra model architecture. Experiments on three public datasets in different domains demonstrate that our approach not only achieves better performance on various evaluation metrics, but also has high efficiency in both training and inference phases, over a wide range of sample size and number of aspect categories.

A network analysis of mechanisms of change during exposures over the course of intensive OCD treatment

Exposure and response prevention (ERP) is an evidence-based treatment for obsessive-compulsive disorder (OCD). Theories for how it works vary in their emphasis on active mechanisms of change. The current study aimed to clarify mechanisms of change in ERP for OCD using network analysis, comparing ERP networks at the start and end of intensive treatment (partial hospital and residential). In our sample of 182 patients, the most central node in both networks was engagement with exposure, which was consistently related to greater understanding of ERP rationale, higher willingness, and less ritualization, accounting for all other variables in the network. There were no significant differences in networks between the start and end of treatment. These results suggest that nonspecific parameters like facilitating engagement in exposures without ritualizing and providing a clear rationale to clients may be key to effective treatment. As such, it may be useful for clinicians to spend adequate time underscoring the need to eliminate rituals to fully engage in exposure tasks and explaining the rationale for ERP prior to doing exposures, regardless of theoretical orientation. Nonetheless, findings represent group-level statistics and more fine-grained idiographic analyses may reveal individual-level differences with respect to central mechanisms of change. Other limitations include demographic homogeneity of our sample.

Genomics Research of Lifetime Depression in the Netherlands: The BIObanks Netherlands Internet Collaboration (BIONIC) Project.

In this cohort profile article we describe the lifetime major depressive disorder (MDD) database that has been established as part of the BIObanks Netherlands Internet Collaboration (BIONIC). Across the Netherlands we collected data on Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) lifetime MDD diagnosis in 132,850 Dutch individuals. Currently, N = 66,684 of these also have genomewide single nucleotide polymorphism (SNP) data. We initiated this project because the complex genetic basis of MDD requires large population-wide studies with uniform in-depth phenotyping. For standardized phenotyping we developed the LIDAS (LIfetime Depression Assessment Survey), which then was used to measure MDD in 11 Dutch cohorts. Data from these cohorts were combined with diagnostic interview depression data from 5 clinical cohorts to create a dataset of N = 29,650 lifetime MDD cases (22%) meeting DSM-5 criteria and 94,300 screened controls. In addition, genomewide genotype data from the cohorts were assembled into a genomewide association study (GWAS) dataset of N = 66,684 Dutch individuals (25.3% cases). Phenotype data include DSM-5-based MDD diagnoses, sociodemographic variables, information on lifestyle and BMI, characteristics of depressive symptoms and episodes, and psychiatric diagnosis and treatment history. We describe the establishment and harmonization of the BIONIC phenotype and GWAS datasets and provide an overview of the available information and sample characteristics. Our next step is the GWAS of lifetime MDD in the Netherlands, with future plans including fine-grained genetic analyses of depression characteristics, international collaborations and multi-omics studies.

Understanding Creative Entrepreneurs’ Work Practices: The Varying Conversation between Artistic and Economic Rationales

Research on creative organizations often questions how artistic practices can be squared within the rational decision-making of economic thinking. This paper examines how the relational language, or conversation, between artistic and economic rationales unfolds for creative entrepreneurs. Through ethnographic work with a designer-entrepreneur, this paper presents a fine-grained analysis of the conversation the designer cultivates between artistic and economic rationales through work practices. We contribute to the literature about artistic and economic rationales at work, and more specifically to the concept of conversation. First, we show that high levels of conversing make way for low levels of conversing and vice versa. In the studio, the designer’s engagement with either rationale varies as the creative process progresses. Second, on a more global dynamic, we demonstrate the conversation is continuous. It relies on its variations, which ensure the balance between rationales in the long run. We also contribute to the field of creative entrepreneurship research. We identify here one type of creative entrepreneur, with what we call a ‘small is beautiful’ attitude. Far from the mythical figure of the entrepreneur, this unconventional entrepreneur aims for sustainable use of creative resources rather than growth at all costs.

Passive and Causative in Sanskrit

To what extent can fine-grained statistical analysis provide evidence regarding syntactic patterns in corpus languages like Sanskrit, particularly in cases where the interaction of multiple syntactic phenomena obscures the evidence? We investigate the value of a correlation matrix for bivariate data analysis in relation to varying syntactic patterns in a relatively poorly attested yet productive construction in Classical Sanskrit: the passive of the causative (‘passive causative’). The interaction of causative and passive is complex in Sanskrit, but we show that even in the case of low frequency data, syntactic conclusions can be drawn from such interactions when detailed statistical analysis is employed. In particular, our analysis speaks to the status of the ergative in Sanskrit.

Diabetic Retinopathy Detection Using InceptionResnet-V2 and Densenet121

This project addresses the global health challenge posed by the prevalence of diabetic retinopathy (DR) by developing an efficient automated diagnostic system. The dataset, consisting of diverse high-resolution retinal images, underwent preprocessing to categorize images into No DR (0) and DR (1-4) classes. The First initial binary classification model using a Convolutional Neural Network (CNN) discriminated between healthy and diseased retinas. Subsequently, The second multi-class CNN model was designed to predict the severity of diabetic retinopathy (DR) across a spectrum from mild (1) to proliferative DR (4), enabling a fine-grained analysis for early identification of cases requiring urgent intervention. To address real-world complexities, potential noise in the dataset, including artifacts and exposure variations, was acknowledged. The CNN models were designed to exhibit resilience to these challenges, ensuring robust performance in clinical settings. Preprocessing is considered the common occurrence of image inversion in retinal imaging by incorporating anatomical features, such as macula position and notches, to correctly identify image orientation and enhance result interpretability. The proposed automated analysis system demonstrated promising results in accurately categorizing retinal images into No DR and DR, as well as assigning severity scores for diabetic retinopathy. This project contributes significantly to computer-aided diagnostics, Supplying a dependable instrument for promptly identifying and addressing cases of diabetic retinopathy.

Neighbourhood structure and environmental quality: A fine-grained analysis of spatial inequalities in urban Germany

Urban environments are characterised by sparsity of space, elevated levels of air pollution and limited exposure to natural environments. Yet, residential environmental quality varies substantially both between and within cities. This study combines information on the socio-economic and demographic composition of 243,607 urban neighbourhoods with administrative and remote sensing data on the spatial distribution of industrial plants and urban green space to investigate patterns of environmental inequality in urban Germany at unprecedented levels of spatial granularity. It disentangles neighbourhood disadvantages experienced by foreign minorities (non-nationals) from those experienced by low-income households in order to assess the plausibility of economic explanations of residential sorting. The high level of spatial granularity makes it possible to examine patterns of environmental inequality not only between the relatively large areas that have been used as units of analysis in previous work but also within them, while reducing the threat of ecological bias. Results indicate that non-nationals are more likely to be exposed to industrial air pollution and less likely to live close to green spaces. This association holds even after adjusting for neighbourhood income composition and in fixed-effects specifications that restrict the analysis to within-city variation. I find no evidence for environmental inequality by socio-economic status. Exploratory sub-sample analyses show that neighbourhood disadvantages for non-nationals are higher in cities characterised by high levels of anti-foreigner sentiment, pointing towards housing market discrimination as a potentially important driver of foreign residents’ neighbourhood disadvantage.

Remote Sensing Micro-Object Detection under Global and Local Attention Mechanism

With the rapid advancement of technology, satellite and drone technologies have had significant impacts on various fields, creating both opportunities and challenges. In areas like the military, urban planning, and environmental monitoring, the application of remote sensing technology is paramount. However, due to the unique characteristics of remote sensing images, such as high resolution, large-scale scenes, and small, densely packed targets, remote sensing object detection faces numerous technical challenges. Traditional detection methods are inadequate for effectively detecting small targets, rendering the accurate and efficient detection of objects in complex remote sensing images a pressing issue. Current detection techniques fall short in accurately detecting small targets compared to medium and large ones, primarily due to limited feature information, insufficient contextual data, and poor localization capabilities for small targets. In response, we propose an innovative detection method. Unlike previous approaches that often focused solely on either local or contextual information, we introduce a novel Global and Local Attention Mechanism (GAL), providing an in-depth modeling method for input images. Our method integrates fine-grained local feature analysis with global contextual information processing. The local attention concentrates on details and spatial relationships within local windows, enabling the model to recognize intricate details in complex images. Meanwhile, the global attention addresses the entire image’s global information, capturing overarching patterns and structures, thus enhancing the model’s high-level semantic understanding. Ultimately, a specific mechanism fuses local details with global context, allowing the model to consider both aspects for a more precise and comprehensive interpretation of images. Furthermore, we have developed a multi-head prediction module that leverages semantic information at various scales to capture the multi-scale characteristics of remote sensing targets. Adding decoupled prediction heads aims to improve the accuracy and robustness of target detection. Additionally, we have innovatively designed the Ziou loss function, an advanced loss calculation, to enhance the model’s precision in small target localization, thereby boosting its overall performance in small target detection. Experimental results on the Visdrone2019 and DOTA datasets demonstrate that our method significantly surpasses traditional methods in detecting small targets in remote sensing imagery.

Fine-grained Sentiment-enhanced Collaborative Filtering-based Hybrid Recommender System

Developing online educational platforms necessitates the incorporation of new intelligent procedures in order to improve long-term student experience. Presently, e-learning recommender systems rely on deep learning methods to recommend appropriate e-learning materials to the students based on their learner profiles. Fine-grained sentiment analysis (FSA) can be leveraged to enrich the recommender system. User-posted reviews and rating data are vital in accurately directing the student to the appropriate e-learning resources based on posted comments by comparable learners. In this work, a new e-learning recommendation system is proposed based on individualization and FSA. A hybrid framework is provided by integrating alternating least square (ALS) based collaborative filtering (CF) with FSA to generate an effective e-content recommendation named HCFSAR. ALS attempts to capture the learner’s latent factors based on their selections of interest to build the learner profile. Three FSA models based on attention mechanisms and bidirectional long short-term memory (bi-LSTM) are suggested and used to train twelve models in order to predict new ratings from learner-posted book reviews based on the extracted learner profile. HCFSAR used multiplication word embeddings for stronger corpus representation that were trained on a dataset generated for an educational context and showed a better accuracy of 93.39% for the best model entitled MHAM based ABHR-2 with multiplication (MHAAM), which performed better than other models. A tailored dataset that has been created by scraping reviews of different e-learning resources is leveraged to train different proposed models and validate against public datasets.

Energy-Motion Features Aggregation Network for Players’ Fine-Grained Action Analysis in Soccer Videos

Rich and complex events in sports have led to the development of a wide-variety of techniques for interpreting content of sports videos in terms of players’ actions, poses, gait, performance, etc. This is due to the requirements from coaches, trainers and players who expect to analyze actions in top sports events, as well as sports fans who practice to imitate professional playing skills, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., dribbling, shooting, etc. However, this poses two key challenges for automated sports analysis community. Firstly, there are extremely limited public sports datasets. Secondly, recent advances in interpretations of sports activities, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., soccer, are predominantly made through analyzing coarse-grained contents. Players’ fine-grained skills analysis still remains under-explored. To alleviate these problems, this paper (a) collects the dataset of highlight videos of soccer players, including two coarse-grained action types of soccer players and six fine-grained actions of players. Detailed annotations are provided for the collected dataset, in terms of action classes, bounding boxes, segmentation maps, and body keypoints of soccer players, and positions of a soccer ball in a game. (b) leverages the understanding of complex highlight videos by proposing an energy-motion features aggregation network- <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">EMA-Net</i> to fully exploit energy-based representation of soccer players movements in video sequences and explicit motion dynamics of soccer players in videos for soccer players’ fine-grained action analysis. Experimental results and ablation studies validate the proposed approach in recognizing soccer players actions using the collected soccer highlight video datasets.