Abstract: Methods and systems for collecting and analyzing an audience response are provided. An example method commences with determining that a media playing device has played a question within a media stream. The method further includes recording, via an acoustic sensor, an ambient acoustic signal for a pre-determined time interval. The method further includes detecting a presence of a voice of a user in the ambient acoustic signal, providing the ambient acoustic signal to a remote computing system, and performing, by the remote computing system, speech recognition of the ambient acoustic signal to obtain a text response. The method further includes adding, by the remote computing system, the text response to a set of text responses and analyzing the set of text responses to obtain a statistics concerning text results. The method then continues with providing, by the remote computing system, the statistics to a provider of the media stream.

Abstract: An apparatus (5) manages resources of a resource-controlled system (1), e.g. data processing resources in a data processing system. The resource-controlled system (1) has resource-elasticity, namely it may operate in several states involving different amounts of system resources. The resource management apparatus (5) may select a modification action having at least one resource variation parameter. The resource variation parameter represents a resource quantity to be added or removed in the resource-controlled system (1). The apparatus (5) may update a value function of an adaptive learning agent (7) as a function of performance metrics (4) measured after the resources of the resource-controlled system have been modified by effecting the modification action. Responsive to a safety criterion not being fulfilled, a finer resource variation parameter may be substituted.

Abstract: Sequential hypothesis testing in a digital medium environment is described using continuous data. To begin, a model is received that defines at least one data distribution. Testing data is also received that describes an effect of user interactions with the plurality of options of digital content on achieving an action using continuous non-binary data. Values of parameters of the model are then estimated for each option of the plurality of options based on the testing data. In one example. A variance estimate is then generated based on the estimated values of the parameters of the model for each option of the plurality of options. From this, a determination is made as to a decision boundary based on the variance estimate and an estimate for a mean value of each option of the plurality of options based on the testing data.

Abstract: A learning device (10) includes an acquirer (11), a learner (12), and a generator (14). The acquirer (11) acquires a learning signal. The learner (12) performs, in accordance with similarities indicating degrees of similarity between waveforms, clustering of partial signals cut out from the learning signal acquired by the acquirer (11), and learns reference waveforms that each indicate a waveform of a corresponding partial signal of the clustered partial signals. The generator (14) generates, based on at least one of a distribution of the similarities or characteristics of clusters that each include a corresponding partial signal of the clustered partial signals, progress information indicating a progress status of the learning by the learner (12), and outputs the progress information.

Abstract: At a first computing device, a target image is identified. The target image is provided in a display on the first computing device in a communication area displaying a conversation between a first user using the first computing device and a second user using a second computing device. An indication is received that the first user has initiated editing of the target image. An edit control listing editing options is provided for display in the communication area. At the first computing device, a user selection is received of a particular editing option selected from the editing options. An edited target image is created by editing the target image using an edit operation associated with the particular editing option. The edited target image is provided to the second computing device and the communication area is updated to display the edited target image.

Abstract: Predicting probable activity consequences is provided. Information is collected from data sources to identify various activities. Patterns of how any identified activity is linked with a corresponding event are detected based on analyzing the information. The patterns are indexed with data having a relationship to a particular event. Activity context information associated with a set of identified activities corresponding to the particular event is extracted from the information. A cognitive model of how the set of identified activities corresponding to the particular event are related to a set of activity consequences is generated. Probable activity consequences with degree of severity corresponding to the activity context information is predicted based on the cognitive model. A recommendation to perform a set of action steps to reduce impact of the probable activity consequences on different aspects of the activity context information associated with the set of identified activities is generated.

Abstract: According to one or more embodiments of the present invention, a computer-implemented method includes generating, by a cognitive system, an answer for a user-provided query using an analytics algorithm. The answer is based on a set of data sources. The method further includes determining an influence weightage of each data source from the set of data sources. The method further includes generating and presenting a rationale for the answer based on the influence weightage.

Abstract: The disclosure discloses a method and apparatus for identifying a traffic light. An embodiment of the method comprises: zooming a to-be-processed image acquired by an image acquisition device by at least one preset ratio to obtain at least one zoomed image; inputting the at least one zoomed image into a pre-trained convolutional neural network to obtain location information and category information of a traffic light corresponding to each zoomed image of the at least one zoomed image, wherein the convolutional neural network is used for retrieving location information and category information of a traffic light displayed in an image; and analyzing the obtained location information and category information to generate at least one candidate traffic light identification result, and fusing the generated candidate traffic light identification result to generate a traffic light identification result corresponding to the to-be-processed image. The embodiment improves the accuracy in identifying a traffic light.

Abstract: Certain embodiments include an electronic device comprising a housing, a touchscreen display viewable through a portion of the housing, at least one processor disposed inside the housing and operatively connected with the touchscreen display; and a memory disposed within the housing and operatively connected with the processor, wherein the memory stores instructions that, when executed, cause the at least one processor to perform a plurality of operations, the plurality of operations comprising displaying a first augmented reality (AR) emoji on a window capable of receiving at least one first stroke input, through the touchscreen display, receiving the at least one first stroke input while displaying the first AR emoji on the window, and displaying a second AR emoji, based at least partially on a position and a shape of the at least one first stroke input.

Abstract: In an embodiment, a computer-implemented method comprises, in response to receiving lead data identifying an entity associated with a health care claim relating to suspected fraud, determining one or more data sources that were used to identify the entity or the suspected fraud; determining a subset of a plurality of data display elements, based on the determined one or more data sources, wherein each of the plurality of data display elements is configured to cause displaying health care claims data associated with the entity in a designated format; automatically obtaining, from a data repository, specific health care claims data associated with the entity for each of the plurality of data display elements in the subset; generating a lead summary report associated with the entity using a report template, the subset, and the obtained specific health care claims data.

Abstract: Improved systems and methods for automatically discovering and filtering electronic messages. These systems and methods improve the operation of computer apparatus to achieve dramatic reductions in processing resources, data storage resources, network resources, and filter production times compared to conventional approaches. In some examples, improvements result from configuring computer apparatus to perform a unique sequence of specific electronic message processing rules in a network communications environment. In this regard, these examples are able to automatically learn the structures and semantics of machine generated electronic message headers, accelerating the ability to support new message sources and new markets. These examples provide a purchase related electronic message discovery and filtering service that is able to identify and filter purchase related electronic messages with high accuracy across a wide variety of electronic message formats.

Abstract: A people-flow analysis system includes an image source, a computing device, and a host. The image source captures a first image and a second image. The computing device is connected to the image source. The computing device identifies the first image according to a data set to generate a first detecting image. The first detecting image has a position box corresponding to a pedestrian in the first image. The computing device generates a tracking image according to the data set and a difference between the first detecting image and the second image. The tracking image has another position box corresponding to a pedestrian in the second image. The host is connected to the computing device and generates a people-flow list according to the first detecting image and the tracking image.

Abstract: The disclosed embodiments include a method performed by server computer(s). The method includes obtaining private healthcare insurance claims data and public healthcare procedure code data from one or more data source(s), generating a model based the training data, and determining a multiplier for each of the many facilities based on the model. The model provides predictive results for variability between healthcare facilities thereby enabling consumer research in light of considerations in facility variability. The multiplier is indicative of a value used to scale the public healthcare procedure code data for a facility such that a healthcare service associated with the facility can be estimated based on the multiplier.

Abstract: Methods, apparatus, systems and articles of manufacture to classify a first file are disclosed herein. Example apparatus include a feature hash generator to generate respective sets of one or more feature hashes for respective features of the first file. The number of the one or more feature hashes to be generated is based on an ability of the feature to distinguish the first file from a second file. The apparatus also includes a bit setter to set respective bits of a first fuzzy hash value based on respective ones of the one or more feature hashes, a classifier to assign the first file to a class associated with a second file based on a similarity between the first fuzzy hash value and a second fuzzy hash value for a second file.

Abstract: Aspects of the subject disclosure may include, for example, a process that performs a statistical, natural-language processing analysis on a group of text documents to determine a group of topics. The topics are determined according to parameters obtained by training on a sample of documents. One or more topics in a subset of topics are associated to each document, resulting in topic-document pairs. A bias is identified for each topic-document pair, and clusters of topics are created from the subset of topics. Each cluster of topics is determined from a value for each bias of each topic-document pair and from a frequency of occurrence of each topic. Each cluster is presentable according to a corresponding image configuration based on all or a subset of the bias dimensions and the frequency of occurrence of topics in a cluster that distinguishes the cluster from other clusters. Other embodiments are disclosed.

Abstract: Some embodiments of the methods provided herein relate to sample analysis and particle characterization methods. Some such embodiments include receiving, from a particle analyzer, measurements for a first portion of particles associated with an experiment. Some embodiments also include generating a tree representing groups of related particles based at least in part on the measurements, wherein the tree includes at least three groups. Some embodiments also include generating a measure of relatedness between a first group and a second group of the tree based at least in part on the measurements. Some embodiments also include and configuring the particle analyzer to classify a subsequent particle associated with the experiment with the first group real-time, wherein the subsequent particle is not included in the first portion of particles. Some embodiments also include sorting the subsequent particle.

Abstract: A traveling assistance method detects another vehicle and objects around a host vehicle, sets a blind spot area from the other vehicle, specifies an object present in the blind spot area among the detected objects, and predicts an action that the other vehicle takes in accordance with the specified object.

Abstract: A method, apparatus and computer program product for using machine learning to evaluate medical risks are described. A graphical user interface is provided for entering a set of patient symptoms for a first patient. A query is sent to a medical system containing the set of patient symptoms. A set of causal factors related to the patient symptoms and a set of association link strengths are received in response. Each association link strength is representative of a strength of a relationship between a respective patient symptom and a respective causal factor. The graphical user interface displays the set of patient symptoms and the set of causal factors according to the respective association link strength to show which causal factors have likely caused the patient symptoms for the first patient and are relevant for selecting among different treatment regimens.

Abstract: An example method is provided for a computing system in a private cloud computing environment to deploy an application in a hybrid cloud computing environment. The method may comprise generating a request to deploy an application in a public cloud environment, the request including a deployment plan that specifies one or more tasks to be executed by a virtual computing resource from a cloud provider to deploy the application. The method may comprise sending, to an orchestration node in the public cloud computing environment, the request to instruct the orchestration node to provision the virtual computing resource from the cloud provider and to cause the virtual computing resource to execute the one or more tasks. The method may further comprise receiving, from the orchestration node in the public cloud environment, status data and determining whether the application is successfully deployed in the public cloud computing environment.

Abstract: A machine-learning mechanism of a disaster-avoidance system trains a knowledgebase to associate characteristics of a data-center component with corresponding degrees of vulnerability to failure and with remedial steps that may be undertaken to avoid failure or to reduce adverse effects of a failure. This training is performed as a function of inferences derived from historical records and from extrinsic information sources. The historical records identify past failures of similar components, component characteristics associated with past failures, and results of remedial procedures undertaken in response to past failures or to previous occurrences of the characteristics. The extrinsic sources identify the current existence of external conditions known to be associated with past failures.

Abstract: An inter-variable dependency analyzing unit analyzes a relation between an argument and a return value between instructions of a plurality of instructions included in a sequence control program, which is a control program written in a language for sequence control. A PLC instruction sorting/division unit at least either changes an instruction execution order of the plurality of instructions or divides the sequence control program in units of instructions, based on the relation between the argument and the return value between the instructions analyzed by the inter-variable dependency analyzing unit.

Abstract: In some embodiments, a system comprises a game controller module, a trigger satisfaction module, a metadata module, and a marker module. The game controller module may be configured to receive user control signals and execute gameplay based on the user control signals. The trigger satisfaction module may be configured to determine whether one or more events that occur during the gameplay satisfy one or more trigger conditions of a trigger. The metadata module may be configured to generate one or more trigger satisfaction indicators based on the satisfaction of the one or more trigger conditions of the trigger. The marker module may be configured to associate the one or more trigger satisfaction indicators with video of the gameplay to generate one or more game event markers to be used to enable a viewer to jump to one or more game segments associated with the trigger.

Abstract: Methods and systems for estimating latent service and latent wait times include extracting transition times between activities from a partial event log. Parameters for respective gamma distributions are estimated that characterize latent waiting time and latent service time for each activity. A latent waiting time and latent service time for each activity is estimated based on the estimated parameters using a processor.

Abstract: Method and system is provided for detecting faults and/or providing optimization measures to enhance the performance of one or more buildings, especially building automation equipment of the one or more buildings. The system may include a database for storing a data model comprising asset and performance data of the one or more buildings. The data model may be represented in one or more hierarchical tree-structures, in which the nodes of the tree-structures represent asset data and the leaves of the tree-structures represent performance data of the one or more buildings. The performance data may be represented in each case by data points. The data points may include time series of measured or derived field values and meta information.

Abstract: An inter-variable dependency analyzing unit analyzes a relation between an argument and a return value between instructions of a plurality of instructions included in a sequence control program, which is a control program written in a language for sequence control. A PLC instruction sorting/division unit at least either changes an instruction execution order of the plurality of instructions or divides the sequence control program in units of instructions, based on the relation between the argument and the return value between the instructions analyzed by the inter-variable dependency analyzing unit.

Abstract: A device may receive a first sample. The device may identify a plurality of second samples that are similar to the first sample. The plurality of second samples may correspond to service tickets that have previously been processed. The device may obtain or determine log information identifying a plurality of events associated with the first sample. The device may determine respective probabilities of occurrence corresponding to the plurality of second samples. A probability of occurrence, for a particular second sample, may identify a likelihood of receiving a corresponding service ticket within a particular time period of occurrence of the plurality of events. The device may assign ranks to the plurality of second samples based on the respective probabilities of occurrence. The device may provide information identifying the ranks and the plurality of second samples.

Abstract: Method and system are provided for generating a chatbot interface for an application programming interface (API) that interacts with networked applications. The method may include: receiving as an input a definition document for an API that interacts with networked applications and parsing the definition document to identify intents and entities and obtain examples of the identified intents and entities. The method may convert the definition document to a chatbot data structure including: extracting the intents and entities and their relationship to objects and fields in the API from the definition document; and training the chatbot data structure with the example intents and entities to generate a conversation specification in the chatbot data structure. The method may then generate a chatbot interface for the API.

Abstract: An image is accepted by one or more processing circuits from a user depicting the user's facial skin. Machine learning models stored in one or more memory circuits are applied to the image to classify facial skin characteristics. A regimen recommendation is provided to the user based on the classified facial skin characteristics.

Abstract: A method and system for storage exhaustion estimation. Specifically, the method and system disclosed herein entail deriving a timeline for the depletion of available storage capacity on a backup storage array based on the data backup dynamics of various user clients. The timeline may deduce storage capacity availability in terms of future successful backup cycles, which may serve to address critical issues involving the administration of the backup storage array.

Abstract: A control network, system and method is disclosed for control of a plurality of entities. The control network comprises a control hub, a data repository and a plurality of control agents. The data repository models each entity as a plurality of granular components, each granular component having a type selected from a set of granular component types. The control hub is configured to receive performance data on each entity, translate the performance data into data on the granular components and store the transformed data in the data repository in association with its respective granular component. The control hub is further configured to determine an optimal control action in respect of a granular component of one of the entities in dependence on the granular component's transformed data and on transformed data in the data repository for other granular components having the same type.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for asynchronous deep reinforcement learning. One of the systems includes a plurality of workers, wherein each worker is configured to operate independently of each other worker, and wherein each worker is associated with a respective actor that interacts with a respective replica of the environment during the training of the deep neural network.

Abstract: A system is provided for analyzing item description attributes and user profile attributes, and dynamically generating item description content based on the analysis. The system can obtain item descriptions for a particular item from any number of different sources and/or in any number of different formats. The system can process the item descriptions to determine item attributes. The system can also obtain data regarding user profile attributes. The system may dynamically generate item description content for the particular item and user based on analyzing the item attributes and user profile attributes. The item description content may be further tailored to a particular channel of delivery.

Abstract: A method for computing a dominant class of a scene includes: receiving an input image of a scene; generating a segmentation map of the input image, the segmentation map including a plurality of pixels, each of the pixels being labeled with a corresponding class of a plurality of classes; computing a plurality of area ratios based on the segmentation map, each of the area ratios corresponding to a different class of the plurality of classes of the segmentation map; applying inference to generate a plurality of ranked labels based on the area ratios; and outputting a detected dominant class of the scene based on the plurality of ranked labels.

Abstract: A method for amending or adding functionality to an automation device in an automation systems domain is provided. A semantic representation is provided as a basis for representing the functionality to be amended or added. The semantic model for semantically representing the functionality is not only interpretable but also executable. A machine interpretation and execution is carried out by a semantic reasoner and an event processing engine, respectively, which are both part of a runtime environment implemented on the automation device. The need of implementing a code skeleton is eliminated by an engineer. Instead, the semantic model may be directly executed in a device that has an embedded runtime with reasoning capabilities.

Abstract: A heterogeneous log pattern editing recommendation system and computer-implemented method are provided. The system has a processor configured to identify, from heterogeneous logs, patterns including variable fields and constant fields. The processor is also configured to extract a category feature, a cardinality feature, and a before-after n-gram feature by tokenizing the variable fields in the identified patterns. The processor is additionally configured to generate target similarity scores between target fields to be potentially edited and other fields from among the variable fields in the heterogeneous logs using pattern editing operations based on the extracted category feature, the extracted cardinality feature, and the extracted before-after n-gram feature. The processor is further configured to recommend, to a user, log pattern edits for at least one of the target fields based on the target similarity scores between the target fields in the heterogeneous logs.

Abstract: A system and method for tracking a plurality of objects. Unlabeled measurement data identifying a plurality of targets corresponding to the plurality of objects is received. A multi-target likelihood function is generated using a persistent target density, a birth target density, and a clutter density. The multi-target likelihood function is used to associate persistent targets with the unlabeled measurement data to update persistent tracks and to initiate a new track for each target in the plurality of targets in the measurement data that is not associated with a persistent target.

Abstract: A “Data Transfer Tool” extracts, labels and stores user data or information that may be confined within application silos during user interaction with arbitrary apps. The Data Transfer Tool enables sharing of this otherwise siloed data shared across and between authorized apps. The Data Transfer Tool provides a task-centric approach to define and extract structured and semantically meaningful information from source applications by providing multiple semantic models that are individually tailored to particular source apps. The Data Transfer Tool applies an accessibility API or the like of the OS in combination with the semantic model for the source app to scrape user entered or selected data or information from the source app. The Data Transfer Tool enables and creates new user experiences and increases user efficiency when interacting with various apps by making the scraped data or information available to subscribing destination apps approved to receive that data or information.

Abstract: Technologies are described herein to compile a Turing-complete quantum programming language program into a quantum circuit. The techniques described and recited herein include compiling TCQPL source code to generate a quantum circuit by generating a function object ensemble, generating an abstract syntax tree from received source code, and annotating nodes corresponding to the abstract syntax tree with corresponding function objects.

Abstract: An alertness prediction bio-mathematical model for use in devices such as a wearable device that improves upon previous models of predicting fatigue and alertness by gathering data from the individual being monitored to create a more accurate estimation of alertness levels. The bio-mathematical model may be a two-process algorithm which incorporates a sleep-wake homeostasis aspect and a circadian rhythm aspect. The sleep-wake homeostasis aspect of the model is improved by using actigraphy measures in conjunction with distal skin, ambient light and heart rate measures to improve the accuracy of the sleep and wake estimations. The circadian rhythm model aspect improves fatigue prediction and estimation by using distal skin, heart rate and actigraphy data. The sleep-wake homeostasis and circadian rhythm aspects may also be combined with additional objective and subjective measures as well as information from a user to improve the accuracy of the alertness estimation even further.

Abstract: A method includes receiving an alert representing that application of at least one rule to a scenario occurring with a financial account indicates that the scenario is consistent with illegal activity. The method includes, in response to the alert, using a fuzzy inference derived from activity of the financial account to determine a likelihood that the scenario is associated with illegal activity.

Abstract: Technology is disclosed herein for handing approximate (or uncertain) values in spreadsheet applications. More specifically, the technology describes spreadsheet applications that support arrays or sets of approximate (or uncertain) values as native entities. An approximate (or uncertain) value may be the value of a cell of the spreadsheet that is resolvable by formula, charts and other functionalities. In some implementations, approximate values may include a range of data and a probability distribution that can be automatically created by the spreadsheet application, generated based on context, input by the user, etc. Because the approximate (or uncertain) value is natively available, it can be incorporated in a spreadsheet like other values. Additionally, the approximate (or uncertain) values can automatically propagate through a spreadsheet calculation to obtain a final approximate result.

Abstract: One embodiment provides a method comprising classifying one or more objects present in an input comprising visual data by executing a first set of models associated with a domain on the input. Each model corresponds to an object category. Each model is trained to generate a visual classifier result relating to a corresponding object category in the input with an associated confidence value indicative of accuracy of the visual classifier result. The method further comprises aggregating a first set of visual classifier results based on confidence value associated with each visual classifier result of each model of the first set of models. At least one other model is selectable for execution on the input based on the aggregated first set of visual classifier results for additional classification of the objects. One or more visual classifier results are returned to an application running on an electronic device for display.

Abstract: Methods and systems for detecting and correcting anomalous inputs include training a neural network to embed high-dimensional input data into a low-dimensional space with an embedding that preserves neighbor relationships. Input data items are embedded into the low-dimensional space to form respective low-dimensional codes. An anomaly is determined among the high-dimensional input data based on the low-dimensional codes. The anomaly is corrected.

Abstract: A first cognitive instance receives information about other cognitive instances and from this compiles a cognitive community map that associates individual ones of the other cognitive instances with specific capabilities of said respective other cognitive instances. The first cognitive instance stores that map in a local memory of the first cognitive instance; and when the first cognitive instance executes a cognitive computing program it checks the cognitive community map for at least one of the specific capabilities relevant for executing that program to address/satisfy a user request that caused the program to execute. In various embodiment these cognitive instances share their respective cognitive capabilities via cognitive capability maps, which may be refreshed in their local memories for example by sending a broadcast message. Thus any given cognitive instance can identify cognitive peers with the capabilities most relevant to assist itself in solving a given problem/request.

Abstract: Techniques are described for classifying seasonal patterns in a time series. In an embodiment, a set of time series data is decomposed to generate a noise signal and a dense signal, where the noise signal includes a plurality of sparse features from the set of time series data and the dense signal includes a plurality of dense features from the set of time series data. A set of one or more sparse features from the noise signal is selected for retention. After selecting the sparse features, a modified set of time series data is generated by combining the set of one or more sparse features with a set of one or more dense features from the plurality of dense features. At least one seasonal pattern is identified from the modified set of time series data. A summary for the seasonal pattern may then be generated and stored.

Abstract: The methods and apparatus disclosed herein can diagnose or identify a subject as at risk of having one or more developmental disorders with fewer questions, decreased amounts of time, and determine a plurality of developmental disorders, and provide clinically acceptable sensitivity and specificity in a clinical environment. The methods and apparatus disclosed herein can be configured to diagnose or determine the subject as at risk of a developmental disorder among a plurality of developmental disorders, and decreasing the number of questions presented can be particularly helpful where a subject presents with a plurality of possible developmental disorders. A processor can be configured with instructions to identify a most predictive next question, such that a person can be diagnosed or identified as at risk with fewer questions.

Abstract: Methods and systems are described for generating integrated intelligent content overlays for media content streams. A server computing device receives a video content stream from a video data source. The server extracts a corpus of machine-recognizable text from the video content stream, the corpus of machine-recognizable text corresponding to at least one of audio or closed captioning text associated with the video content stream. The server identifies one or more entity names contained in the corpus of machine-recognizable text. The server determines a set of content keywords associated with each of the identified entity names. The server generates a content overlay for the video content stream comprising one or more layers that include graphical content relating to at least one of the sets of content keywords. The server integrates the content overlay into the video content stream to generate a customized video content stream.

Abstract: Apparatus and methods are provided for graphically defining a real-world cybersecurity protocol of an entity. The graphical platform includes searchable, manipulatable, graphs mapping cybersecurity threats. Manipulating nodes and relationships within the graphs translates into real-time modification of a cybersecurity protocol in effect for the entity. An ability to map known cybersecurity threats and analyze them (even according to known frameworks) may streamline and integrate efforts of cybersecurity defense teams. Graphical representation of a security protocol facilitates proactive threat hunting as well as expediting incident response activities by providing evidence-based pathways to inform impact analysis and source event analysis.

Abstract: Systems and methods for analyzing an unknown sample are disclosed. The system includes at least one subsystem to obtain molecular information about the sample and at least one other subsystem to obtain elemental information about the sample. The system also includes a data collection component to collect and combine the information from the subsystems to create combined analytical information and a multivariate model that relates known attributes to information previously generated with at least two analytical systems that are the same types of systems as the at least two analytical subsystems. A prediction engine applies the multivariate model to the combined analytical information to produce predictions of attributes in the unknown sample.

Abstract: Disclosed are systems, methods, and non-transitory computer-readable media for an improved text classification model. A text classification system determines a first embedding for a first set of characters in a text, and a second embedding for a second set of characters in the text. The text classification system applies a first coefficient to the first embedding and a second coefficient to the second embedding, yielding a first weighted embedding and a second weighted embedding. The first coefficient is different than the second coefficient. The text classification system determines a weighted average embedding for the text based on the first weighted embedding, the second weighted embedding, the first coefficient and the second coefficient. The text classification system identifies at least a first classification label and a second classification label for the text based on the weighted average embedding for the text.