Journal articles on the topic 'Auto-picking'

Seedling auto-picking device is one of the most important working parts of a vegetable automatic transplanter. Its performance determines the transplanting quality, reliability and operation efficiency. The current developments of the seedling auto-picking device were discussed, and several typical devices were introduced in this paper. The existing problems of seedling auto-picking devices of different picking methods were pointed out. With the advantages of accurate location, high efficiency and easy picking, the mandril-type seedling picking was thought as the better transplanting mode of vegetable plug seedling. At the same time, the development direction of auto-picking devices of vegetable automatic transplanter was provided.

Abstract In this paper, we propose a novel approach of deep-learning-based seismic horizon auto-picking that introduces a modified vector quantized variational autoencoder (VQ-VAE) framework to improve the accuracy of seismic horizon interpretation and, for the first time, quantitatively evaluate the uncertainty of the auto-picked horizon by exploiting the concept of entropy. Compared with the conventional VQ-VAE approach, the proposed method not only modifies the VQ-VAE model with more deep-learning channels at each layer of the network to enhance the performance of horizon auto-picking within the VQ-VAE framework, but also extends the 1D seismic labels with more continuous samplings within a single trace to boost the stability of auto-picked horizon in geologically complex settings and also significantly suppress the resulting uncertainty. To further improve the resulting accuracy in geologically complex settings, we introduce the directional structure tensor to extract a more reliable initial horizon and, moreover, a dilated horizon searching strategy to extend the capacity of the proposed method in dealing with the large fault displacement and reducing the computational cost simultaneously. Additionally, the resulting uncertainty quantitatively measured by entropy can also serve as an effective indicator to enable a further refinement of the auto-picked result accordingly. Both 2D example and 3D field applications are carried out to validate the effectiveness of the proposed method.

The sandglass neural network is often used for nonlinear auto-association, where the principal information can be extracted by picking up the values of the middle layer. However, the boundary of the classes on this 2-1) surface tends to be complicated because no class information is used. In this paper, the hybridization of auto-associative network and the multi-layer perceptron for classification is proposed. The usefulness of this method is demonstrated by using clinical data.

This Paper designs a mechanical device for the manipulator of the watermelon picking machine against the low mechanical degree of watermelon picking machine. The mechanical device utilizes a mechanical arm to drive the end effector to run and the end effector is responsible for clamping and shearing watermelon vines, which avoids vine disturbance and sorts out vines to be easily cut down through the process design of clamping, promoting, and re-shearing. In addition, this Paper applies Pro/E modeling, finite element analysis, and simulation analysis to complete the 3D model design of the mechanical device and transforms the 3D model into 2D drawings in Auto CAD to complete the manufacturing and assembly of the manipulator, and the test result verifies the mechanical device may realize the reliable picking of watermelons.

Stacking velocity is a very important parameter in seismic data processing. Until now the determination of stacking velocity has been done manually. This article proposes an automatic algorithm for picking stacking velocity. The algorithm uses artificial intelligence and pattern recognition techniques.

Grounded on Labview platform, the grid of technology of solution pH value contributes to a new device----the auto-control technology of solution pH value. This device is easier to operate just by the parameter input in the interface. And it breaks the limit of single admeasuring apparatus, meanwhile adopts the method of picking the average value by multi-metering, which avoids the delay effect caused by solution mixing link. The Acid precipitation-Flocculation process in paper making sewage sets an example for the pH value auto-control and supports PAC control of pH value with software concept..

Grounded on Labview platform, the grid of technology of solution pH value contributes to a new device----the auto-control technology of solution pH value. This device is easier to operate just by the parameter input in the interface. And it breaks the limit of single admeasuring apparatus, meanwhile adopts the method of picking the average value by multi-metering, which avoids the delay effect caused by solution mixing link. The Acid precipitation-Flocculation process in paper making sewage sets an example for the pH value auto-control and supports PAC control of pH value with software concept.

Abstract An energy- and predominant-period-dependent (EDP) P-wave onset automatic picking (EDP-picker) algorithm is proposed to deal with the problem of inaccurate P-wave onset picking in cases in which the P-wave onset is hidden in high-amplitude ambient noise or the energy difference between the seismic P-wave and ambient noise is indistinguishable. The algorithm evaluates the energy change using a characteristic variable ΔE, which describes the energy increment of the P wave above ambient noise. The period change is evaluated using two variables with respect to the predominant period, namely Tpd as proposed by Hildyard et al. (2008) and ΔTpd as the gradient of Tpd. The EDP-picker algorithm has two steps: (1) threshold-based cursory P-wave onset picking and (2) precise P-wave onset picking using an Akaike information criterion function, in which both energy information and period information are considered. All three parameters are determined in a 1 s sliding window. The proposed algorithm is verified on a large dataset comprising 13,481 vertical strong ground motion records for 570 events selected from K-NET (Japan) and China Strong Motion Networks Center data. For all records with an epicentral distance of less than 150 km, 93.5% of residuals of manual picks and auto picks are within ±0.5 s. The results demonstrate that EDP-picker is robust and suitable for real-time systems.

This research investigates adverse events and patient safety in healthcare due to poor supply chain management practices, and inadequate and disorganized product validation procedures. Focusing on commodity medical and surgical products, this research investigates correct product validation points for maximum patient safety. This study also explores benefits of standard product identifying technologies such as HIBC or GS1 data standards as well as automated validation systems such as barcode or Auto ID to minimize workflow interruptions. Site visits and phone interviews are conducted with six healthcare providers to document common product validation practices and procedures. Based on observations and collected data, a simulation model is developed. Different scenarios are compared for patient safety, care delay, and system efficiency. The results show that validation points during PAR picking or bedside product administration, and warehouse picking operations provide optimal overall system performance. The results also indicate that standard product identifying technologies and automated validation systems significantly impact the efficiency of supply chain.

This is a writing story about becoming. It is therefore about change and about identifying myself—deconstructing myself—as learner always: “Getting smart” “getting lost” and “getting real” eventually as doing what we consider to be the ideal; moral perfectibility and learning as both function and fiction. It is a Deleuzian stumbling nomadic and rhizomatic inquiry into creating community through not and supplements and the displacement of terms: Subject/subjectivity/reconstruction/deconstruction/intersubjectivity/ co-construction/co-deconstruction…—being under erasure. Sentence (de) construction might therefore be sometimes a bit stumbling too. Thinking Deleuze and Derrida and a little bit of Dewey together: DDD + assemblage. A deconstructive auto ethnography, autobiography, youto(o)biography: Writing community, school and ultimately research together hopefully picking up speed in the middle.

Order picking, which is collecting a set of products from different locations in a warehouse, has repeatedly been described as one of the most laborious and time-consuming internal logistic processes. Each order is issued to pick some products located at given locations in the warehouse. In this paper, we consider an order picking problem, in which a number of orders with different delivery due dates are going to be retrieved by a limited number of order pickers in multiperiods such that the total tardiness is minimized. The aim is to determine a retrieval plan in terms of order batching and order picker multitrip routing as decision variables. Besides, products are arrived and replenished at the predetermined locations at different periods. Therefore, products sitting in those locations should be delivered soon to provide empty rooms for replenishment. A mixed integer linear programming formulation is proposed for this new problem. The model is optimally solved for small-size problems. For larger instances, grouping metaheuristic algorithms are proposed based on particle swarm optimization and the league championship algorithm that use group-based operators to generate reasonable batches of orders. Improvement heuristics are designed as well. The performance of the MILP formulation and metaheuristic algorithms is analyzed for different problem instances whose designs are based on real data gathered from an auto parts warehouse. Results indicate that our algorithms can stably solve large instances of the problem in a reasonable time.

La inteligencia de enjambres biológicos ha tenido un alto desarrollo en diferentes campos de la ingeniería aplicada, como es el caso de la robótica colaborativa en donde, a través de la bioinspiración, se desarrollan algoritmos que buscan imitar los comportamientos emergentes que suceden en la naturaleza cuando interactúan los integrantes de un enjambre de manera local, generando una inteligencia para resolver una problemática de manera auto organizada. Dentro de las aplicaciones en campos de la ingeniera está el desarrollo de procesos automatizados en ambientes de toda la cadena productiva de las empresas. Es por ello que este articulo busca dar un marco de referencia en la automatización de almacenes de picking, aplicando algoritmos de inteligencia artificial, específicamente la técnica de inteligencia de enjambres, que posibilite el uso de agentes que interactúen de forma colaborativa (cero colisiones, manipulación de objetos, entre otras) y competitiva (menor gasto de energía para el desarrollo de actividades), bajo un esquema de comunicación ambiente – agente, agente.

In order to improve the productivity of the filing form materials continuously and automatically packaged line, the control methods for the manipulator and the mechanism which has the capacity of weight self-measuring and auto–filling to the packaged material were studied. The manipulator is a pneumatic control system, it is consists of three pneumatic cylinder circuits to implement the horizontal movements, vertical movements and the activities of picking up or putting down the materials on line. The material to be picked up by manipulator was weighed and displayed by a single chip microcomputer. The general control function of the system was carried out by a PLC unit. The structure changeable type of simulating human being intelligence control methods resulting in the real observing at the objective changes were studied and implemented to achieve the plant high in positioning accuracy and fast in response speed. The experiments were realized and the results showed that the working period time is less than 5.8s,the weight variation measured by single chip microcomputer is not more than 2 grams in 300 grams.

The large structural diversity of specialized metabolites represents a substantial challenge in untargeted metabolomics. Modern LC–QTOF instruments can provide three- to four-digit numbers of auto-MS/MS spectra from sample sets. This case study utilizes twelve structurally closely related flavonol glycosides, characteristic specialized metabolites of plant tissues, some of them isomeric and isobaric, to illustrate the possibilities and limitations of their identification. This process requires specific software tools that perform peak picking and feature alignment after spectral deconvolution and facilitate molecular structure base searching with subsequent in silico fragmentation to obtain initial ideas about possible structures. The final assignment of a putative identification, so long as spectral databases are not complete enough, requires structure searches in a chemical reference database, such as SciFindern, in attempts to obtain additional information about specific product ions of a metabolite candidate or check its feasibility. The highlighted problems in this process not only apply to specialized metabolites in plants but to those occurring in other organisms as well. This case study is aimed at providing guidelines for all researchers who obtain data from such analyses but are interested in deeper information than just Venn diagrams of the feature distribution in their sample groups.

AbstractAssembly optimization of printed circuit boards (PCBs) has received considerable research attention because of efforts to improve productivity. Researchers have simplified complexities associated with PCB assembly; however, they have overlooked hardware constraints, such as pick-and-place restrictions and simultaneous pickup restrictions. In this study, a hybrid group search optimizer (HGSO) was proposed. Assembly optimization of PCBs for a multihead placement machine is segmented into three problems: the (1) auto nozzle changer (ANC) assembly problem, (2) nozzle setup problem, and (3) component pick-and-place sequence problem. The proposed HGSO proportionally applies a modified group search optimizer (MGSO), random-key integer programming, and assigned number of nozzles to an ANC to solve the component picking problem and minimize the number of nozzle changes, and the place order is treated as a traveling salesman problem. Nearest neighbor search is used to generate an initial place order, which is then improved using a 2-opt method, where chaos local search and a population manager improve efficiency and population diversity to minimize total assembly time. To evaluate the performance of the proposed HGSO, real-time PCB data from a plant were examined and compared with data obtained by an onsite engineer and from other related studies. The results revealed that the proposed HGSO has the lowest total assembly time, and it can be widely employed in general multihead placement machines.

In order to prevent illegal intrusion, theft, and destruction, important places require stable and reliable human intrusion detection technology to maintain security. In this paper, a combined sensing system using anti-jamming random code signals is proposed and demonstrated experimentally to detect the human intruder in the protected area. This sensing system combines the leaky coaxial cable (LCX) sensor and the single-transmitter-double-receivers (STDR) radar sensor. They transmit the orthogonal physical random code signals generated by Boolean chaos as the detection signals. The LCX sensor realizes the early intrusion alarm at the protected area boundary by comparing the correlation traces before and after intrusion. Meanwhile, the STDR radar sensor is used to track the intruder’s moving path inside the protected area by correlation ranging and ellipse positioning, as well as recognizing intruder’s activities by time-frequency analysis, feature extraction, and support vector machine. The experimental results demonstrate that this combined sensing system not only realizes the early alarm and path tracking for the intruder with the 13 cm positioning accuracy, but also recognizes the intruder’s eight activities including squatting, picking up, jumping, waving, walking forward, running forward, walking backward, and running backward with 98.75% average accuracy. Benefiting from the natural randomness and auto-correlation of random code signal, the proposed sensing system is also proved to have a large anti-jamming tolerance of 27.6 dB, which can be used in the complex electromagnetic environment.

Efficient transplanting has been identified as one of the essential steps towards achieving an increased yield in the farm. However, many factors are affecting these processes such as soil moisture content and the speed of pickup. This study was carried out to investigate the effect of different soil moisture content and pickup speeds on pickup force, balance, resistance, and lump damage during transplanting of seedlings. The results showed that penetration resistance was inversely proportional to the speed and soil moisture content. The highest penetration resistance (38 N) values were recorded under the lowest speed (0.5 mm/s) at the low moisture content; whereas, the lowest penetration resistance was obtained at highest speed (10 mm/s) under high moisture content. The highest pick-up force resistance values were recorded under the lowest speed (0.5 mm/s) at low moisture content than the lowest pick-up force resistance of 1.4 N at (10 mm/s) under the high moisture content. On the other hand, an increase of pick-up force led to a decrease in the pick-up force resistance. The pick-up damage and the pick-up speed are directly proportional—nevertheless, the former increased with a decreasing soil moisture content. The highest pick-up damage values (82%) were observed under the top-most speed (10 mm/s) at high moisture content. It can be concluded that for successful auto-transplanting of seedling the soil condition, the force applied and speed should be taken into consideration. This work will implement an effective seedling-picking performance and basis for the optimal design of end-effectors.

Onset detection of P-wave in seismic signals is of vital importance to seismologists because it is not only crucial to the development of early warning systems but it also aids in estimating the seismic source parameters. All the existing P-wave onset detection methods are based on a combination of statistical signal processing and time-series modeling ideas. However, these methods do not adequately accommodate some advanced ideas that exist in fault detection literature, especially those based on predictive analytics. When combined with a time-frequency (t-f) / temporal-spectral localization method, the effectiveness of such methods is enhanced significantly. This work proposes a novel real-time automatic P-wave detector and picker in the prediction framework with a time-frequency localization feature. The proposed approach brings a diverse set of capabilities in accurately detecting the P-wave onset, especially in low signal-to-noise ratio (SNR) conditions that all the existing methods fail to attain. The core idea is to monitor the difference in squared magnitudes of one-step-ahead predictions and measurements in the time-frequency bands with a statistically determined threshold. The proposed framework essentially accommodates any suitable prediction methodology and time-frequency transformation. We demonstrate the proposed framework by deploying auto-regressive integrated moving average (ARIMA) models for predictions and the well-known maximal overlap discrete wavelet packet transform (MODWPT) for the t-f projection of measurements. The ability and efficacy of the proposed method, especially in detecting P-waves embedded in low SNR measurements, is illustrated on a synthetic data set and 200 real-time data sets spanning four different geographical regions. A comparison with three prominently used detectors, namely, STA/LTA, AIC, and DWT-AIC, shows improved detection rate for low SNR events, better accuracy of detection and picking, decreased false alarm rate, and robustness to outliers in data. Specifically, the proposed method yields a detection rate of 89% and a false alarm rate of 11.11%, which are significantly better than those of existing methods.

Ongoing studies have anticipated that in 2030, car crashes will be the fifth driving reason for death around the world. The main cause of car crashes is difficult to decide these days because of a complex mix of qualities like the mental condition of the driver, road conditions, climate conditions, traffic, and infringement of traffic rules to give some examples. The expenses of fatalities and driver wounds because of car crashes incredibly influence the general public. The use of machine learning methods in the field of road accidents is picking up speed nowadays. The organization of machine learning classifiers has swapped conventional data mining methods for creating higher outcomes and exactness. This work presents a review of different existing businesses related to accident prediction utilizing the machine learning area. Wounds because of road accidents are one of the most pervasive reasons for death separated from health-related issues. The investigation of road accident seriousness was finished by running an accident dataset through a few machine learning arrangement calculations to see which model played out the best in characterizing the accidents into severity classes, for example, slight, extreme, and fatal. It was seen that calculated relapse to perform multilevel order gave the most noteworthy exactness score. It was additionally seen that variables, for example, the number of vehicles, lighting conditions, and road highlights assumed a part in deciding the seriousness of the accident. Engineers and analysts in the car business have attempted to plan and manufacture more secure vehicles, yet auto collisions are unavoidable. Examples associated with hazardous accidents could be identified by building up a prediction model that naturally orders the sort of injury severity of different traffic accidents. These social and roadway designs are valuable in the improvement of traffic security control strategies. Significantly, estimates be founded on logical and target reviews of the reasons for accidents and the seriousness of injuries. This paper presents a few models to predict the seriousness of the injury that happened during traffic accidents utilizing machine-learning paradigms. We considered networks prepared to utilize machine learning methods. Analysis results uncover that among the machine learning ideal models considered different standards paradigm approaches.

Picking an appropriate parameter setting (meta-parameters) for visualization and embedding techniques is a tedious task. However, especially when studying the latent representation generated by an autoencoder for unsupervised data analysis, it is also an indispensable one. Here we present a procedure using a cross-correlative take on the meta-parameters. This ansatz allows us to deduce meaningful meta-parameter limits using OPTICS, DBSCAN, UMAP, t-SNE, and k-MEANS. We can perform first steps of a meaningful visual analysis in the unsupervised case using a vanilla autoencoder on the MNIST and DeepVALVE data sets.

AbstractIn this paper, we describe the hardware and software components of a fully autonomous prototype delivery vehicle. Equipped with a robotic arm, the demonstrator is capable of delivering packages and picking up new ones by interacting with custom-made delivery boxes. As highly accurate positioning w. r. t. a box is required for successful handover of packages, we track the pose (position and orientation) of the box using a high-resolution on-board camera. The resulting estimate is relayed to our planning and control modules, which ensure that the vehicle reaches its required pose with centimeter-level accuracy.In order to deliver packages, the car needs to autonomously navigate our test facility, avoiding static and dynamic obstacles while obeying simple traffic rules. As one focus is on the practical challenges encountered when building a prototype, we cover issues ranging from sensor calibration and system identification to perception, planning, control, and the implementation of high-level behaviors. While some of the proposed solutions to these problems are not necessarily novel, they allowed us to demonstrate the vehicle’s capabilities after a development phase of less than 12 months.

AbstractSensor-based ore sorting is not a new technology. It has been around since more than 70 years, mainly for diamond concentration, where it was applied to eliminate the security risk of diamonds being stolen from the previously applied grease-tables [13]. Despite a few installations in uranium ore processing, it had no further widespread acceptance in the minerals industry, mainly due to low design capacity. Besides that, sensor-based colour sorters were used in the food industry for small particle sizes (e. g., rice cleaning). It is fact that the first machine designs appropriate for coarse bulk materials were not developed for the minerals industry, but for the upcoming recycling industry for plastics, glass, paper, metals in the late 1980s. In this sector, besides some magnetic separators, all the work was done by manual hand-picking, and it needed automation. After some years of optimization, these machines showed reliable performance under harsh conditions in scrap yards and recycling plants. Then, finally, the minerals industry, which at first was not convinced that this rather complicated machines were suited to be used with minerals, began with the first applications. These first installations of sensor-based ore sorters around the late 1990, all of them equipped with line-scan optical cameras, were mainly in industrial minerals, such as calcite, magnesite, quartz or rock salt. Since then, the technology has seen an enormous development in terms of available sensors, design capacity and availability, and the number of installations for minerals is growing – steadily but slower than expected, considering the many advantages it brings.

AbstractBecause there is no complete three-dimensional (3D) hybrid detector integrated PET+MRI internationally, this study aims to investigate a registration approach for a two-dimensional (2D) hybrid based on characteristic localization to achieve a 3D fusion from the images of PET and MRI as a whole.A cubic-oriented scheme of “9-point and 3-plane” for a coregistration design was verified to be geometrically practical. Through 3D reconstruction and virtual dissection, human internal feature points were sorted to combine with preselected external feature points for matching process. By following the procedure of feature extraction and image mapping, the processes of “picking points to form planes” and “picking planes for segmentation” were executed. Eventually, the image fusion was implemented at the real-time workstation Mimics based on auto-fuse techniques called “information exchange” and “signal overlay”.A complementary 3D image across PET+MRI modalities, which simultaneously present metabolic activities and anatomic structures, was created with a detectable rate of 56%. This is equivalent to the detectable rate of PET+CT or MRI+CT with no statistically significant difference, and it facilitates a 3D vision that is not yet functional for 2D hybrid imaging.This cross-modality fusion is doubtless an essential complement for the existing toolkit of a 2D hybrid device. Thus, it would potentially improve the efficiency of diagnosis and therapy for oncology.

Deep learning has been widely used for inferring robust grasps. Although human-labeled RGB-D datasets were initially used to learn grasp configurations, preparation of this kind of large dataset is expensive. To address this problem, images were generated by a physical simulator, and a physically inspired model (e.g., a contact model between a suction vacuum cup and object) was used as a grasp quality evaluation metric to annotate the synthesized images. However, this kind of contact model is complicated and requires parameter identification by experiments to ensure real world performance. In addition, previous studies have not considered manipulator reachability such as when a grasp configuration with high grasp quality is unable to reach the target due to collisions or the physical limitations of the robot. In this study, we propose an intuitive geometric analytic-based grasp quality evaluation metric. We further incorporate a reachability evaluation metric. We annotate the pixel-wise grasp quality and reachability by the proposed evaluation metric on synthesized images in a simulator to train an auto-encoder–decoder called suction graspability U-Net++ (SG-U-Net++). Experiment results show that our intuitive grasp quality evaluation metric is competitive with a physically-inspired metric. Learning the reachability helps to reduce motion planning computation time by removing obviously unreachable candidates. The system achieves an overall picking speed of 560 PPH (pieces per hour).

AbstractToday, with the development of 3-D studies and the increase in seismic data volume, there is a growing need to expand interpretation techniques for achieving higher speed and accuracy of interpretation tasks. Determining seismic faults and horizons is vital to accomplish the process as one of the essential stages of data interpretation. With the recent development of computational methods in seismic interpretation and their benefits, different approaches have been promoted. The specialist can make the understanding much faster with higher accuracy. In this research, a fully automated dual horizon and fault selection approach in the presence of semi-vertical faults is presented using a structural smoothing condition. Geological faults make it challenging to map sedimentary layers appropriately which is targeted for review in this work. Unlike Image processing techniques that determine the location of faults only, the proposed approach gives the benefit of the estimated fault displacement. In this method, faults are modeled as a displacement vector field. Despite traditional methods (such as similarity and coherence), in this method, the vector field of the estimated fault displacement determines the displacement and its location. This vector field can be used for auto-determination of fault-related layers displacement. As a result, automatic horizon picking in the presence of such faults is possible, thereby simplifying the mapping of sedimentary layers.

Abstract Background Computerised Physician Order Entry (CPOE) is considered to enhance the safety of prescribing. However, it can have unintended consequences and new forms of prescribing error have been reported. Objective The aim of this study was to explore the causes and contributing factors associated with prescribing errors reported by multidisciplinary prescribers working within a CPOE system. Main Outcome Measure Multidisciplinary prescribers experience of prescribing errors in an CPOE system. Method This qualitative study was conducted in a hospital with a well-established CPOE system. Semi-structured qualitative interviews were conducted with prescribers from the professions of pharmacy, nursing, and medicine. Interviews analysed using a mixed inductive and deductive approach to develop a framework for the causes of error. Results Twenty-three prescribers were interviewed. Six main themes influencing prescribing were found: the system, the prescriber, the patient, the team, the task of prescribing and the work environment. Prominent issues related to CPOE included, incorrect drug name picking, default auto-population of dosages, alert fatigue and remote prescribing. These interacted within a complex prescribing environment. No substantial differences in the experience of CPOE were found between the professions. Conclusion Medical and non-medical prescribers have similar experiences of prescribing errors when using CPOE, aligned with existing published literature about medical prescribing. Causes of electronic prescribing errors are multifactorial in nature and prescribers describe how factors interact to create the conditions errors. While interventions should focus on direct CPOE issues, such as training and design, socio-technical, and environmental aspects of practice remain important.

Background: Metabolomics is one of the main areas to understand cellular process at molecular level by analyzing metabolites. In recent years metabolomics has been emerged as key tool to understand molecular basis of disease, find diagnostic and prognostic biomarkers, and develop new treatment opportunities and drug molecules. Objective: In this study, an untargeted metabolite and lipid analysis were performed to identify potential biomarkers on premature ovarian insufficiency plasma samples. 43 POI subject plasma samples were compared with 32 healthy subject plasma samples. Methods: Plasma samples were pooled and extracted using chloroform:methanol:water (3:3:1 v/v/v) mixture. Agilent 6530 LC/MS Q-TOF instrument equipped with ESI source was used for analysis. A C18 column (Agilent Zorbax 1.8 μM, 50 x 2.1 mm) was used for separation of metabolites and lipids. XCMS, an “R software” based freeware program, was used for peak picking, grouping and comparing the findings. Isotopologue Parameter Optimization (IPO) software was used in order to optimize XCMS parameters. The analytical methodology and data mining process were validated according to the literature. Results: 83 metabolite peaks and 213 lipid peaks were found to be in semi-quantitatively and statistically different (fold change >1.5, p <0.05) between the POI plasma samples and control subjects. Conclusion: According to the results, two groups were successfully separated through principal component analysis. Among the peaks, phenyl alanine, decanoyl-L-carnitine, 1-palmitoyllysophosphatidylcholine and PC(O-16:0/2:0) were identified through auto MS/MS and matched with human metabolome database and proposed as plasma biomarker for POI and monitoring the patients in treatment period.

Summary Earthquake monitoring workflows are designed to detect earthquake signals and to determine source characteristics from continuous waveform data. Recent developments in deep learning seismology have been used to improve tasks within earthquake monitoring workflows that allow the fast and accurate detection of up to orders of magnitude more small events than are present in conventional catalogs. To facilitate the application of machine-learning algorithms to large-volume seismic records at scale, we developed a cloud-based earthquake monitoring workflow, QuakeFlow, that applies multiple processing steps to generate earthquake catalogs from raw seismic data. QuakeFlow uses a deep learning model, PhaseNet, for picking P/S phases and a machine learning model, GaMMA, for phase association with approximate earthquake location and magnitude. Each component in QuakeFlow is containerized, allowing straightforward updates to the pipeline with new deep learning/machine learning models, as well as the ability to add new components, such as earthquake relocation algorithms. We built QuakeFlow in Kubernetes to make it auto-scale for large datasets and to make it easy to deploy on cloud platforms, which enables large-scale parallel processing. We used QuakeFlow to process three years of continuous archived data from Puerto Rico within a few hours, and found more than a factor of ten more events that occurred on much the same structures as previously known seismicity. We applied Quakeflow to monitoring earthquakes in Hawaii and found over an order of magnitude more events than are in the standard catalog, including many events that illuminate the deep structure of the magmatic system. We also added Kafka and Spark streaming to deliver real-time earthquake monitoring results. QuakeFlow is an effective and efficient approach both for improving realtime earthquake monitoring and for mining archived seismic data sets.

Content warning: although it was kept to a minimum, this text displays instances of (anti-Semitic) hate speech. During the 2016 U.S. election and its aftermath, multiple journalistic accounts reported on “alt-right trolls” emanating from anonymous online spaces like the imageboard 4chan (e.g. Abramson; Ellis). Having gained infamy for its nihilist trolling subcultures (Phillips, This Is Why) and the loose hacktivist movement Anonymous (Coleman), 4chan now drew headlines because of the alt-right’s “genuinely new” concoction of white supremacy, ironic Internet humour, and a lack of clear leadership (Hawley 50). The alt-right “anons”, as imageboard users call themselves, were said to primarily manifest on the “Politically Incorrect” subforum of 4chan: /pol/. Gradually, a sentiment arose in the titles of several news articles that the pro-Trump “alt-right trolls” had successfully won the metapolitical battle intertwined with the elections (Phillips, Oxygen 5). For instance, articles titled that “trolls” were “The Only True Winners of this Election” (Dewey) or even “Plotting a GOP Takeover” (Stuart).The headlines were as enticing as questionable. As trolling-expert Whitney Phillips headlined herself, the alt-right did not attain political gravity solely through its own efforts but rather was “Conjured Out of Pearl Clutching and Media Attention” (“The Alt-Right”), with news outlets being provoked to criticise, debunk, or sensationalise its trolling activities (Faris et al. 131; Phillips, “Oxygen” 5-6). Even with the right intentions, attempts at denouncement through using vague, structuralist notions–from “alt-right” and “trolls” to “the basket of deplorables” (Robertson) – arguably only strengthened the coherence of those it was meant to disavow (Phillips, Oxygen; Phillips et al.; Marantz). Phillips et al. therefore lamented such generalisations, arguing attributing Trump’s win to vague notions of “4chan”, “alt-right”, or “trolls” actually bestowed an “atemporal, almost godlike power” to what was actually an “ever-reactive anonymous online collective”. Therefore, they called to refrain from making claims about opaque spaces like 4chan without first “plotting the landscape” and “safeguarding the actual record”. Indeed, “when it comes to 4chan and Anonymous”, Phillips et al. warned, “nobody steps in the same river twice”.This text answers the call to map anonymous online groups by engaging with the complexity of testing the muddy waters of the ever-changing and dissimulative 4chan-current. It first argues how anti-structuralist research outlooks can answer to many of the pitfalls arising from this complex task. Afterwards, it traces the word trump as it was used on 4chan/pol/ to problematise some of the above-mentioned media narratives. How did anons consider Trump, and how did the /pol/-current change during the build-up of the 2016 U.S. elections and afterwards?On Researching Masked and Dissimulative ExtremistsWhile potentially playing into the self-imagination of malicious actors (Phillips et al.), the frequent appearance of overblown narratives on 4chan is unsurprising considering the peculiar affordances of imageboards. Imageboards are anonymous – no user account is required to post – and ephemeral – posts are deleted after a certain amount of activity, sometimes after days, sometimes after minutes (Bernstein et al.; Hagen). These affordances complicate studying collectives on imageboards, with the primary reasons being that 1) they prevent insights into user demographics, 2) they afford particularly dissimulative, playful discourse that can rarely be taken at face value (Auerbach; de Zeeuw and Tuters), and 3) the sheer volume of auto-deleted activity means one has to stay up-to-date with a rapid waterfall of subcultural ephemera. Additionally, the person stepping into the muddy waters of the chan-river also changes their gaze over time. For instance, Phillips bravely narrates how she once saw parts of the 4chan-stream as “fun” to only later realise the blatantly racist elements present from the start (“It Wasn’t Just”).To help render legible the changing currents of imageboard activity without relying on vague understandings of the “alt-right”, “trolls”, or “Anonymous”, anti-structuralist research outlooks form a possible answer. Around 1900, sociologists like Gabriel Tarde already argued to refrain from departing from structuralist notions of society and instead let social compositions arise through iterative tracing of minute imitations (11). As described in Bruno Latour’s Reassembling the Social, actor-network theory (ANT) revitalises the Tardean outlook by similarly criticising the notion of the “social” and “society” as distinct, sui-generis entities. Instead, ANT advocates tracing “flat” networks of agency made up of both human and non-human actors (165-72). By tracing actors and describing the emerging network of heterogeneous mediators and intermediaries (105), one can slowly but surely get a sense of collective life. ANT thus takes a page from ethnomethodology, which advocates a similar mapping of how participants of a group produce themselves as such (Garfinkel).For multiple reasons, anti-structuralist approaches like ANT can be useful in tracing elusive anonymous online groups and their changing compositions. First, instead of grasping collectives on imageboards from the outset through structuralist notions, as networked individuals, or as “amorphous and formless entities” (see e.g. Coleman 113-5), it only derives its composition after following where its actors lead. This can result in an empirical and literally objective mapping of their collectivity while refraining from mystifications and non-existent connections–so often present in popular narratives about “trolls” and the “alt-right”. At the same time, it allows prominent self-imaginations and mythologizations – or, in ANT-parlance, “localisations of the global” (Latour 173-190) – rise to the surface whenever they form important actors, which, as we will see, tends to happen on 4chan.Second, ANT offers a useful lens with which to consider how non-human actors can uphold a sense of collectivity within anonymous imageboards. This can include digital objects as part of the infrastructure–e.g. the automatically assigned post numbers having mythical value on 4chan (Beran, It Came From 69)–but also cultural objects like words or memes. Considering 4chan’s anonymity, this focus on objects instead of individuals is partly a necessity: one cannot know the exact amount and flow of users. Still, as this text seeks to show, non-human actors like words or memes can form suitable actors to map the changing collectivity of anonymous imageboard users in the absence of demographic insights.There are a few pitfalls worth noting when conducting ANT-informed research into extremist spaces like 4chan/pol/. The aforementioned ironic and dissimulative rhetoric of anonymous forum culture (de Zeeuw and Tuters) means tracing is complicated by implicit (yet omnipresent) intertextual references undecipherable to the untrained eye. Even worse, when misread or exaggerated, such tracing efforts can play into trolling tactics. This can in turn risk what Phillips calls “giving oxygen” to bigoted narratives by amplifying their presence (“Oxygen”). Since ANT does not prescribe what sort of description is needed (Latour 149), this exposure can be limited and/or critically engaged with by the researcher. Still, it is inevitable that research on extremist collectives adds at least some garbage to already polluted information ecologies (Phillips and Milner 2020), even when “just” letting the actors speak (Venturini). Indeed, this text will unfortunately also show hate speech terms below.These complications of irony and amplification can be somewhat mitigated by mixing ethnographic involvement with computational methods. Together, they can render implicit references explicit while also mapping broad patterns in imitation and preventing singular (misleading) actors from over-dominating the description. When done well, such descriptions do not only have to amplify but can also marginalise and trivialise. An accurate mapping can thereby counter sensationalist media narratives, as long as that is where the actors lead. It because of this potentiality that anti-structuralist tracing of extremist, dissimulative online groups should not be discarded outright.Stopping Momentarily to Test the WatersTo put the above into practice, what follows is a brief case study on the term trump on 4chan/pol/. Instead of following users, here the actor trump is taken an entry point for tracing various assemblages: not only referring to Donald J. Trump as an individual and his actions, but also to how /pol/-anons imagine themselves in relation to Trump. In this way, the actor trump is a fluid one: each of its iterations contains different boundaries and variants of its environment (de Laet and Mol 252). By following these environments, can we make sense of how the delirious 2016 U.S. election cycle played out on /pol/, a space described as the “skeleton key to the rise of Trump” (Beran, 4chan)?To trace trump, I use the 4plebs.com archive, containing almost all posts made on /pol/ between late-2013 and early 2018 (the time of research). I subsequently use two text mining methods to trace various connections between trump and other actors and use this to highlight specific posts. As Latour et al. note, computational methods allow “navigations” (593) of different data points to ensure diverse empirical perspectives, preventing both structuralist “zoomed-out” views and local contexts from over-dominating. Instead of moving between micro and macro views, such a navigation should therefore be understood as a “circulation” around the data, deploying various perspectives that each assemble the actors in a different way. In following this, the case study aims to demonstrate how, instead of a lengthy ethnographic account, a brief navigation using both quali- and quantitative perspectives can quickly demystify some aspects of seemingly nebulous online groups.Tracing trump: From Meme-Wizard to Anti-Semitic TargetTo get a sense of the centrality of Trump on /pol/, I start with post frequencies of trump assembled in two ways. The first (Figure 1) shows how, soon after the announcement of Trump’s presidential bid on 16 June 2015, around 100,000 comments mention the word (2% of the total amount of posts). The frequencies spike to a staggering 8% of all comments during the build-up to Trump’s win of the Republican nomination in early 2016 and presidential election in November 2016. Figure 1: The absolute and relative amount of posts on 4chan/pol/ containing the word trump (prefixes and suffixes allowed).To follow the traces between trump and the more general discourse surrounding it, I compiled a more general “trump-dense threads” dataset. These are threads containing thirty or more posts, with at least 15% of posts mentioning trump. As Figure 2 shows, at the two peaks, 8% of any thread on /pol/ was trump-dense, accounting for approximately 15,000 monthly threads. While Trump’s presence is unsurprising, these two views show just how incredibly central the former businessman was to /pol/ at the time of the 2016 U.S. election. Figure 2: The absolute and relative amount of threads on 4chan/pol/ that are “trump-dense”, meaning they have thirty comments or more, out of which at least 15% contain the word trump (prefixes and suffixes allowed).Instead of picking a certain moment from these aggregate overviews and moving to the “micro” (Latour et al.), I “circulate” further with Figure 3, showing another perspective on the trump­-dense thread dataset. It shows a scatter plot of trump-dense threads grouped per week and plotted according to how similar their vocabulary is. First, all the words per week are weighted with tf-idf, a common information retrieval algorithm that scores units on the basis if they appear a lot in one of the datasets but not in others (Spärck-Jones). The document sets are then plotted according to the similarity of their weighted vocabulary (cosine similarity). The five highest-scoring terms for the five clusters (identified with K-means) are listed in the bottom-right corner. For legibility, the scatterplot is compressed by the MDS algorithm. To get a better sense of specific vocabulary per week, terms that appeared in all weeks are filtered out (like trump or hillary). Read counterclockwise, the nodes roughly increase in time, thus showing a clear temporal change of discourse, with the first clusters being more similar in vocabulary than the last, and the weeks before and after the primary election (orange cluster) showing a clear gap. Figure 3: A scatterplot showing cosine distances between tf-idf weighted vocabularies of trump-dense threads per week. Compressed with MDS and coloured by five K-means clusters on the underlying tf-idf matrix (excluding terms that appeared in all weeks). Legend shows the top five tf-idf terms within these clusters. ★ denotes the median week in the cluster.With this map, we can trace other words appearing around trump as significant actors in the weekly documents. For instance, Trump-supportive words like stump (referring to “Can’t Stump the Trump”) and maga (“Make America Great Again”) are highly ranked in the first two clusters. In later weeks, less clearly pro-Trump terms appear: drumpf reminds of the unattractive root of the Trump family name, while impeached and mueller show the Russia probe in 2017 and 2018 were significant in the trump-dense threads of that time. This change might thus hint at growing scepticism towards Trump after his win, but it is not shown how these terms are used. Fortunately, the scatterplot offers a rudder with which to navigate to further perspectives.In keeping with Latour’s advice to keep “aggregate structures” and “local contexts” flat (165-72), I contrast the above scatterplot with a perspective on the data that keeps sentence structures intact instead of showing abstracted keyword sets. Figure 4 uses all posts mentioning trump in the median weeks of the first and last clusters in the scatterplot (indicated with ★) and visualises word trees (Wattenberg and Viégas) of most frequent words following “trump is a”. As such, they render explicit ontological associations about Trump; what is Trump, according to /pol/-anons? The first word tree shows posts from 2-8 November 2015, when fifteen Republican competitors were still in the race. As we have seen in Figure 1, Trump was in this month still “only” mentioned in around 50,000 posts (2% of the total). This word tree suggests his eventual nomination was at this point seen as an unlikely and even undesirable scenario, showing derogatory associations like retard and failure, as well as more conspiratorial words like shill, fraud, hillary plant, and hillary clinton puppet. Notably, the most prominent association, meme, and others like joke and fucking comic relief, imply Trump was not taken too seriously (see also Figure 5). Figure 4: Word trees of words following “trump is a” in the median weeks of the first and last clusters of the scatterplot. Made with Jason Davies’s Word Tree application. Figure 5: Anons who did not take Trump seriously. Screencapture taken from archive.4plebs.org (see post 1 and post 2 in context).The first word tree contrast dramatically with the one from the last median week from 18 to 24 December 2017. Here, most associations are anti-Semitic or otherwise related to Judaism, with trump most prominently related to the hate speech term kike. This prompts several questions: did /pol/ become increasingly anti-Semitic? Did already active users radicalise, or were more anti-Semites drawn to /pol/? Or was this nefarious current always there, with Trump merely drawing anti-Semitic attention after he won the election? Although the navigation did not depart from a particular critical framework, by “just following the actors” (Venturini), it already stumbled upon important questions related to popular narratives on 4chan and the alt-right. While it is tempting to stop here and explain the change as “radicalisation”, the navigation should continue to add more empirical perspectives. When doing so, the more plausible explanation is that the unlikely success of Trump briefly attracted (relatively) more diverse and playful visitors to /pol/, obscuring the presence and steady growth of overt extremists in the process.To unpack this, I first focus on the claim that a (relatively) diverse set of users flocked to /pol/ because of the Trump campaign. /pol/’s overall posting activity rose sharply during the 2016 election, which can point to already active users becoming more active, but is likely mostly caused by new users flocking to /pol/. Indeed, this can be traced in actor language. For instance, many anons professed to be “reporting in” from other 4chan boards during crucial moments in the campaing. One of the longest threads in the trump-dense threads dataset (4,504 posts) simply announces “Cruz drops out”. In the comments below, multiple anons state they arrived from other boards to join the Trump-infused activity. For instance, Figure 6 shows an anon replying “/v/ REPORTING IN”, to which sixty other users reacted by similarly affirming themselves as representatives from other boards (e.g. “/mu/ here. Ready to MAGA”). While but another particular view, this implies Trump’s surprising nomination stimulated a crowd-like gathering of different anons jumping into the vortex of trump-related activity on /pol/. Figure 6: Replies by outside-anons “reporting in” the sticky thread announcing Ted Cruz's drop out, 4 May 2016. Screenshots taken from 4plebs.org (see post 1 and post 2 in context).Other actor-language further expresses Trump’s campaign “drew in” new and unadjusted (or: less extreme) users. Notably, many anons claimed the 2016 election led to an “invasion of Reddit users”. Figure 7 shows one such expression: an annotated timeline of /pol/’s posting activity graph (made by 4plebs), posted to /pol/ on 26 February 2016 and subsequently reposted 34 times. It interprets 2016 as a period where “Trump shit goes into overdrive, meme shit floods /pol/, /pol/ is now reddit”. Whether these claims hold any truth is difficult to establish, but the image forms an interesting case of how the entirety “/pol/” is imagined and locally articulated. Such simplistic narratives relate to what Latour calls “panoramas”: totalising notions of some imagined “whole” (188-90) that, while not to be “confused with the collective”, form crucial data since they express how actors understand their own composition (190). Especially in the volatile conditions of anonymous and ephemeral imageboards, repeated panoramic narratives can help in constructing a sense of cohesion–and thereby also form interesting actors to trace. Indeed, following the panoramic statement “/pol/ is now reddit”, other gatekeeping-efforts are not hard to find. For instance, phrases urging other anons to go “back to reddit” (occurring in 19,069 posts in the total dataset) or “back to The_Donald” (a popular pro-Trump subreddit, 1,940 posts) are also particularly popular in the dataset. Figure 7: An image circulated on /pol/ lamenting that "/pol/ is now reddit" by annotating 4plebs’s posting metrics. Screenshot taken from archive.4plebs.org (see posts).Did trump-related activity on /pol/ indeed become more “meme-y” or “Reddit-like” during the election cycle, as the above panorama articulates? The activity in the trump-dense threads seems to suggest so. Figure 8 again uses the tf-idf terms from these threads, but here with the columns denoting the weeks and the rows the top scoring tf-idf terms of their respective week. To highlight relevant actors, all terms are greyed out (see the unedited sheet here), except for several keywords that indicate particularly playful or memetic vernacular: the aforementioned stump, emperor, referring to Trump’s nickname as “God Emperor”; energy, referring to “high energy”, a common catchphrase amongst Trump supporters; magic, referring to “meme magic”, the faux-ironic belief that posting memes affects real-life events; and pepe, the infamous cartoon frog. In both the tf-idf ranking and the absolute frequencies, these keywords flourish in 2016, but disappear soon after the presidential election passes. The later weeks in 2017 and 2018 rarely contain similarly playful and memetic terms, and if they do, suggest mocking discourse regarding Trump (e.g. drumpf). This perspective thus pictures the environment around trump in the run-up to the election as a particularly memetic yet short-lived carnival. At least from this perspective, “meme shit” thus indeed seemed to have “flooded /pol/”, but only for a short while. Figure 8: tf-idf matrix of trump-dense threads, columns denoting weeks and rows denoting the top hundred most relevant terms per week. Download the full tf-idf matrix with all terms here.Despite this carnivalesque activity, further perspectives suggest it did not go at the expense of extremist activity on /pol/. Figure 9 shows the absolute and relative counts of the word "jew" and its derogatory synonym "kike". Each of these increases from 2015 onwards. As such, it seems to align with claims that Trump’s success and /pol/ becoming increasingly extremist were causally related (Thompson). However, apart from possibly confusing correlation with causation, the relative presence remains fairly stable, even slightly decreasing during the frenzy of the Trump campaign. Since we also saw Trump himself become a target for anti-Semitic activity, these trendlines rather imply /pol/’s extremist current grew proportionally to the overall increase in activity, and increased alongside but not but necessarily as a partisan contingent as a result of Trump’s campaign. Figure 9: The absolute and relative frequency of the terms "jew" and "kike" on 4chan/pol/.ConclusionCombined, the above navigation implies two main changes in 4chan/pol/’s trump-related current. First, the climaxes of the 2016 Republican primaries and presidential elections seem to have invoked crowd-like influxes of (relatively) heterogeneous users joining the Trump-delirium, marked by particularly memetic activity. Second, /pol/ additionally seemed to have formed a welcoming hotbed for anti-Semites and other extremists, as the absolute amount of (anti-Semitic) hate speech increased. However, while already-present and new users might have been energised by Trump, they were not necessarily loyal to him, as professed by the fact that Trump himself eventually became a target. Together with the fact that anti-Semitic hate speech stayed relatively consistent, instead of being “countercultural” (Nagle) or exclusively pro-Trump, /pol/ thus seems to have been composed of quite a stable anti-Semitic and Trump-critical contingent, increasing proportionally to /pol/’s general growth.Methodologically, this text sought to demonstrate how a brief navigation of trump on 4chan/pol/ can provide provisional yet valuable insights regarding continuously changing current of online anonymous collectives. As the cliché goes, however, this brief exploration has left more many questions, or rather, it did not “deploy the content with all its connections” (Latour 147). For instance, I have not touched on how many of the trump-dense threads are distinctly separated and pro-Trump “general threads” (Jokubauskaitė and Peeters). Considering the vastness of such tasks, the necessity remains to find appropriate ways to “accurately map” the wild currents of the dissimulative Web–despite how muddy they might get.NoteThis text is a compressed and edited version of a longer MA thesis available here.ReferencesAbramson, Seth. “Listen Up, Progressives: Here’s How to Deal with a 4Chan (“Alt-Right”) Troll.” Medium, 2 May 2017. <https://medium.com/@Seth_Abramson/listen-up-progressives-heres-how-to-deal-with-a-4chan-alt-right-troll-48594f59a303>.Auerbach, David. “Anonymity as Culture: Treatise.” Triple Canopy, n.d. 22 June 2020 <https://www.canopycanopycanopy.com/contents/anonymity_as_culture__treatise>.Beran, Dale. “4chan: The Skeleton Key to the Rise of Trump”. Medium, 14 Feb. 2017. <https://medium.com/@DaleBeran/4chan-the-skeleton-key-to-the-rise-of-trump-624e7cb798cb>.Beran, Dale. It Came from Something Awful: How a Toxic Troll Army Accidentally Memed Donald Trump into Office. New York: All Points Books, 2019.Bernstein, Michael S, Andrés Monroy-Hernández, Drew Harry, Paul André, Katrina Panovich, and Greg Vargas. “4chan and /b/: An Analysis of Anonymity and Ephemerality in a Large Online Community.” Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media, 2011.Coleman, Gabriella. Hacker, Hoaxer, Whistleblower, Spy: The Many Faces of Anonymous. London: Verso Books, 2014.De Laet, Marianne, and Annemarie Mol. “The Zimbabwe Bush Pump: Mechanics of a Fluid Technology.” Social Studies of Science 30.2 (2000): 225–263. 1 May 2020 <https://journals.sagepub.com/doi/10.1177/030631200030002002>. De Zeeuw, Daniel, and Marc Tuters. “Teh Internet Is Serious Business: On the Deep Vernacular Web Imaginary.” Cultural Politics 16.2 (2020).Dewey, Caitlin. “The Only True Winners of this Election are Trolls.” The Washington Post, 3 Nov. 2016. <https://www.washingtonpost.com/news/the-intersect/wp/2016/11/03/the-only-true-winners-of-this-election-are-trolls/>.Faris, Robert, Hal Roberts, Bruce Etling, Nikki Bourassa, Ethan Zuckerman, and Yochai Benkler. “Partisanship, Propaganda, and Disinformation: Online Media and the 2016 U.S. Presidential Election.” Berkman Klein Center Research Publication, 2017. <http://nrs.harvard.edu/urn-3:HUL.InstRepos:33759251>.Garfinkel, Harold. Studies in Ethnomethodology. New Jersey: Prentice-Hall, 1967.Hagen, Sal. “Rendering Legible the Ephemerality of 4chan/pol/.” OILab.eu, 12 Apr. 2020. <https://oilab.eu/rendering-legible-the-ephemerality-of-4chanpol/>.Hawley, George. Making Sense of the Alt-Right. New York: Columbia UP, 2017.Jokubauskaitė, Emilija, and Stijn Peeters. “Generally Curious: Thematically Distinct Datasets of General Threads on 4chan/Pol/”. Proceedings of the International AAAI Conference on Web and Social Media 14.1 (2020): 863-7. <https://www.aaai.org/ojs/index.php/ICWSM/article/view/7351>.Latour, Bruno. Reassembling the Social: An Introduction to Actor-Network Theory. New York: Oxford UP, 2005.Latour, Bruno, Pablo Jensen, Tommaso Venturini, Sébastian Grauwin, and Dominique Boullier. “‘The Whole Is Always Smaller than Its Parts’. A Digital Test of Gabriel Tarde’s Monads.” British Journal of Sociology 63.4 (2012): 590-615.Marantz, Andrew. Antisocial: Online Extremists, Techno-Utopians, and the Hijacking of the American Conversation. New York: Penguin Random House, 2019.Nagle, Angela. Kill All Normies: Online Culture Wars from 4chan and Tumblr to Trump and the White House. Winchester: Zero Books, 2017.Phillips, Whitney. This Is Why We Can’t Have Nice Things: Mapping the Relationship between Online Trolling and Mainstream Culture. Cambridge: MIT Press, 2015.———. “The Alt-Right Was Conjured Out of Pearl Clutching and Media Attention.” Motherboard, 12 Oct. 2016 <https://www.vice.com/en_us/article/jpgaeb/conjuring-the-alt-right>.———. “The Oxygen of Amplification: Better Practices for Reporting on Extremists, Antagonists, and Manipulators Online.” Data & Society, 2018. <https://datasociety.net/wp-content/uploads/2018/05/1_PART_1_Oxygen_of_Amplification_DS.pdf>.———. “It Wasn’t Just the Trolls: Early Internet Culture, ‘Fun,’ and the Fires of Exclusionary Laughter.” Social Media + Society (2019). <https://journals.sagepub.com/doi/10.1177/2056305119849493>.Phillips, Whitney, Gabriella Coleman, and Jessica Beyer. “Trolling Scholars Debunk the Idea That the Alt-Right’s Shitposters Have Magic Powers.” Motherboard, 22 Mar. 2017. <https://motherboard.vice.com/en_us/article/z4k549/trolling-scholars-debunk-the-idea-that-the-alt-rights-trolls-have-magic-powers>.Robertson, Adi. “Hillary Clinton Exposing Pepe the Frog Is the Death of Explainers.” The Verge, 15 Sep. 2016. <https://www.theverge.com/2016/9/15/12926976/hillary-clinton-trump-pepe-the-frog-alt-right-explainer>.Spärck Jones, Karen. “A Statistical Interpretation of Term Specificity and its Application in Retrieval.” Journal of Documentation 28.1 (1972): 11-21.Stuart, Tessa. “Inside the DeploraBall: The Trump-Loving Trolls Plotting a GOP Takeover.” Rolling Stone, 20 Jan. 2017. <https://www.rollingstone.com/politics/politics-features/inside-the-deploraball-the-trump-loving-trolls-plotting-a-gop-takeover-128128/>.Tarde, Gabriel. The Laws of Imitation. Ed. and trans. Elsie Clews Parsons. New York: Henry Holt and Company, 1903.Thompson, Andrew. “The Measure of Hate on 4chan.” Rolling Stone, 10 May 2018. <https://www.rollingstone.com/politics/politics-news/the-measure-of-hate-on-4chan-627922/>.Venturini, Tommaso. “Diving in Magma: How to Explore Controversies with Actor-Network Theory.” Public Understanding of Science 19.3 (2010): 258-273.Wattenberg, Martin, and Fernanda Viégas. “The Word Tree, an Interactive Visual Concordance.” IEEE Transactions on Visualization and Computer Graphics 14.6 (2008): 1221-1228.