JP2022161033A - Method and system of generating chain of alerts based on a plurality of critical indicators - Google Patents

Abstract

To provide reliable data, and reliable methods for analyzing the data to determine present and future trends in this data.SOLUTION: A computer-implemented method comprising at least one hardware processor to: at each of one or more times, receive, from a user, a company identifier and one or more conditions, generate an alert for the company identifier and the one or more conditions, and associate the alert with the user; receive multiple pieces of data from a plurality of data sources in real time; update a plurality of signals based on the multiple pieces of data, each of the plurality of signals representing one of a plurality of critical indicators for one of a plurality of companies; and for each alert associated with the user, apply the one or more conditions of the alert to at least one of the plurality of signals to determine whether or not the alert is triggered.SELECTED DRAWING: Figure 3

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application is a continuation of U.S. patent application Ser. Claiming priority to U.S. Provisional Application No. 63/021,550, filed May 7, and U.S. Provisional Application No. 63/171,967, filed April 7, 2021, the contents of which are is incorporated herein by reference in its entirety.

TECHNICAL FIELD The present disclosure relates to systems and methods for providing assistance to users in making decisions based on single-user actions, and more particularly, to pushes provided in response to data conditions and user profiles. Providing systems, methods, and user experiences related to users making decisions based on notifications.

In related art systems, data analysts typically manually study and analyze trends in economic data. For example, those analysts study data to manually determine when to buy or sell stocks, or to determine what the unemployment rate looks like in any given period of time. can.

However, the data used to determine these indicators may be unreliable due to the unavailability of such unique information. For example, data that are not necessarily correlated with a particular economic indicator may not have been filtered out when analyzing the data, thereby creating inaccuracies in the analysis.

Furthermore, in related art, users may not be able to receive push notifications for making decisions through single user actions. Approaches in the related art may provide information related to surroundings, such as stock prices or overall market performance, by email or other means. However, the approaches in the related art do not consider multiple data streams, nor do they remove inaccuracies from the data as described above. Thus, any information received by the user is based on data that may contain inaccuracies. Moreover, when provided with such information, the user cannot make a decision simply by providing a single user action. For example, but not limited to, the shares you wish to purchase, including the name of the company, the number of shares, the timing of the purchase, whether you want to make a single order or multiple orders, etc. No detailed instructions shall be provided to the service provider. Such requirements are inconvenient for users, such as users with limited time not being able to execute a desired number of trades. may miss the opportunity to complete the transaction before

Accordingly, there is a need for reliable data and reliable methods for analyzing data to determine its current and future trends. Additionally, there is an unmet need to allow users to order or order purchases with a single action.

An aspect is a method using at least one hardware processor, the hardware processor receiving a company identifier and one or more conditions from a user at each of one or more times; generate alerts on conditions, associate alerts with users, receive multiple data in real time from multiple data sources, and update multiple signals based on multiple data, where each of the multiple signals is: for each alert associated with a user, representing one of a plurality of key indicators for one of a plurality of companies, applying one or more conditions of the alert to at least one of a plurality of signals; determining whether the alert has been triggered.

According to other aspects, the at least one hardware processor further comprises, for each alert associated with the user, pushing a notification of the alert to the user upon determining that the alert has been triggered, the notification comprising: Includes recommendations.

According to yet another aspect, the recommendation includes one of buying stock or selling stock.

According to yet another aspect, the notification has a single input that, when selected, performs the recommendation without further input from the user.

According to yet another aspect, executing the recommendation comprises submitting the order to the brokerage service according to the recommendation.

According to yet another aspect, notifications of alerts are pushed by email messages to an email address associated with the user.

According to yet another aspect, the alert notification is pushed to the user's mobile device by text message.

According to yet another aspect, the notification of the alert is pushed by a voice message through the speaker of the user's device.

According to yet another aspect, generating a recommendation includes: applying a predictive model to at least one signal to predict a value of the at least one signal; Determining the strength of the at least one signal compared to a benchmark value of the one signal and classifying the strength of the at least one signal into a recommendation, thereby generating the recommendation.

According to yet another aspect, receiving the company identifier and the one or more conditions at least one of the one or more times comprises receiving a selection of alerts used by a plurality of other users. and creating an alert comprises copying an alert that has been used by multiple other users.

According to yet another aspect, the one or more times is a plurality of times, and associating the alert with the user includes adding the alert to the user's chain of alerts.

According to yet another aspect, the at least one hardware processor further comprises pushing a notification of the triggered alert to the user for all triggered alerts in the chain of alerts, the notification being triggered by For each alert triggered, a single input that, when selected, executes the recommended stock order for the company identified by the company identifier in the triggered alert and, when selected, all and a single input that executes recommended stock orders for all companies identified by company identifiers in triggered alerts.

According to yet another aspect, further comprising, for each alert associated with the user, after determining that the alert has been triggered, initiating an order to buy or sell stocks based on the alert.

According to yet another aspect, an order is automatically initiated when an alert is triggered.

According to yet another aspect, initiating the order includes generating the order based on one or more predefined user settings.

According to yet another aspect, initiating the order includes applying a trading model to the plurality of signals and the current state of the user's account to generate the order.

According to yet another aspect, initiating the order includes distributing the quantity of the order over multiple trades over a period of time.

According to yet another aspect, applying one or more conditions of an alert to a plurality of signals includes applying a predictive model to the plurality of signals to predict values of the plurality of signals; to the predicted values of the plurality of signals.

Yet another aspect is at least one hardware processor and one or more software modules that, when executed by the at least one hardware processor, each of one or more times, from a user, a company identifier and Receive one or more conditions, generate alerts on company identifiers and one or more conditions, associate alerts with users, receive multiple data in real time from multiple data sources, multiple signals based on multiple data where each of the plurality of signals represents one of a plurality of key indicators for one of the plurality of companies, and for each alert associated with the user, one or more of the alerts one or more software modules containing computer-executable instructions configured to apply a condition to at least one of a plurality of signals to determine whether an alert has been triggered. It's okay.

Yet another aspect is a non-transitory computer-readable medium storing instructions that, when executed by a hardware processor, cause the hardware processor, at each of one or more times, to: receiving a company identifier and one or more conditions; generating an alert for the company identifier and one or more conditions; associating the alert with a user; receiving multiple data from multiple data sources in real time; Update a plurality of signals, where each of the plurality of signals represents one of a plurality of key indicators for one of the plurality of companies, and for each alert associated with the user: A non-transitory computer readable medium may be included that causes one or more conditions to be applied to at least one of the plurality of signals to determine whether an alert has been triggered.

One or more exemplary embodiments of the invention are described in detail below with reference to the drawings.

FIG. 1 illustrates architecture features according to an exemplary implementation of the present application. FIG. 2 shows the architecture structure according to an exemplary implementation of the present application. FIG. 3 illustrates processes associated with an architecture according to an exemplary implementation of this application. FIG. 4 illustrates exemplary alternate data types according to exemplary implementations of the present application. FIG. 5 illustrates exemplary alternative dataset processing according to exemplary implementations of the present application. FIG. 6 illustrates exemplary automatic tagging according to exemplary implementations of the present application. FIG. 7 illustrates the user experience with an exemplary implementation of the present application regarding watch lists. FIG. 8 illustrates the process associated with creating a user experience according to an exemplary implementation of the present application with respect to watchlists. FIG. 9 illustrates the user experience with an exemplary implementation of this application regarding predictive analytics. FIG. 10 illustrates the processes associated with generating a user experience according to exemplary implementations of this application for predictive analytics. FIG. 11 illustrates the user experience with an exemplary implementation of this application regarding predictive analytics. FIG. 12 illustrates user experience with an exemplary implementation of the present application regarding alert chains. FIG. 13 illustrates the processes associated with generating a user experience according to exemplary implementations of the present application with respect to alert chains. FIG. 14 illustrates processes associated with generating a user experience according to exemplary implementations of the present application with respect to automated trading. FIG. 15 illustrates processes associated with generating a mobile user experience according to exemplary implementations of the present application for automated trading. FIG. 16 illustrates processes associated with generating a mobile user experience according to exemplary implementations of the present application for automated trading. FIG. 17 shows an exemplary implementation. FIG. 18 shows an exemplary implementation. FIG. 19 shows an exemplary implementation. FIG. 20 shows an exemplary implementation. FIG. 21 shows an exemplary implementation. FIG. 22A shows an exemplary implementation. FIG. 22B shows an exemplary implementation. FIG. 23 shows an exemplary implementation. FIG. 24 shows an exemplary implementation. FIG. 25 shows an example environment according to an example implementation of the present application. FIG. 26 illustrates an exemplary computing environment with exemplary computing devices suitable for use in some exemplary implementations. FIG. 27 illustrates the processes associated with executing orders based on single-user action instructions and user settings in response to data notifications provided by an exemplary implementation of the present application. FIG. 28 illustrates an exemplary user experience associated with exemplary implementations disclosed herein. FIG. 29 illustrates an exemplary user experience associated with exemplary implementations disclosed herein.

The detailed description below provides further details regarding the drawings and exemplary implementations of the present application. Among the figures, reference numerals and descriptions of redundant parts have been omitted for clarity. The terminology used throughout the description is provided by way of illustration and is not intended to be limiting. For example, the use of the term "automatic" may refer to fully automatic or semi-automatic, including user or operator control over certain aspects of the implementation, depending on the desired implementation to be made by the person skilled in the art implementing the present application. May contain an implementation.

A more complex and more dynamic economy requires accurate data that accurately reflects what is happening in the economy at any given time. More specifically, to be able to accurately determine the current state of the economy and subsequently determine the current and future trends of the economy, accurate data should be used so that the recommendations themselves are accurate. is required.

For example, when considering the purchase of a particular stock, a purchaser, such as a trader, may consider company information regarding the company associated with the stock. This company information may include, at a very high level, how much money the company makes or how much it loses in a particular time period (e.g. quarter, month, year). good.

However, this general and pervasive data often doesn't accurately portray how companies behave in reality. For example, a clothing company's revenues may appear to be high overall, but breaking down the data may indicate that revenues are primarily from online sales. Typically, stores that are highly profitable online will close some brick-and-mortar stores because they are no longer profitable. Thus, traders considering buying shares in the clothing company may not want to buy shares if they know that those stores are likely to close at some point in the future. do not have.

Unfortunately, without accurate data, traders have little way of knowing what a particular trend will look like in the future. Therefore, by collecting data and cleaning and normalizing that data, we can determine what a particular trend might look like in the future by comparing this cleaned and normalized data to historical data. can do. Similarly, traders may not be able to evaluate current information to make decisions or recommendations. According to this example implementation, the cleaned and normalized data may be current data that is compared with historical data to provide an analysis of current trends.

Although the example implementation above described determining trends related to stock markets, regional, national, or global unemployment rates, public and private company earnings and market share, multiple companies consumer behavior across markets, baskets of stocks (e.g. two or more specific stocks, such as an overall stock portfolio), electronic indices, restaurant indices, how specific sectors in the workforce are performing; Other indicators may be determined (eg, obtained deterministically), including inflation and mutual fund performance. In particular, any data or information that may have an economic impact may be measured.

Aspects of example implementations are directed to apparatus, methods, and systems for receiving multiple data inputs from different data providers having different types of data. Different types of data are received within a big data system, which standardizes the data by performing normalization, deduplication, and classification, and optionally other processes.

The normalized data is provided to a panel building multi-panel generator to build predictions and generate predictions. The predictions are output to access points and users may access the predictions. The user experience may include receiving predictive results as recommendations on a watchlist showing multiple entities and/or receiving detailed performance and recommendation reports for one of multiple entities and/or receiving one or receiving an alert chain that provides the user with an alert that can be triggered by multiple circumstances.

In the exemplary implementations above, the data is processed continuously and in real-time such that users are automatically provided with real-time updates on recommendations, watchlists, and/or alert chains. It has become so. Users may have the opportunity to execute trades either manually or automatically, and may have the opportunity to provide real-time feedback to the system.

To implement the above exemplary implementations, an architecture is provided, as described in more detail below. Additionally, the user experience related to the architecture is also discussed in more detail.

architecture

As shown in FIGS. 1-3, for example, the exemplary implementations described herein relate to architecture, including system components and associated functionality and operations.

FIG. 1 shows a schematic description of the architecture according to an exemplary implementation. At 101, data is received from multiple alternative data providers. Details of multiple alternative data providers are described in more detail below. For example, some of the data providers include, but are not limited to, financial transaction information providers, location information such as GPS data, consumer behavior information such as social media or online publication information, or other data. It's okay.

At 103, an automatic tagging system is provided that receives data from multiple data providers and performs tagging or labeling of the data. More specifically, data may be standardized across different types of data received from different data providers. Standardization processing can include, but is not limited to, filtering, de-duplication, normalization, and classification or labeling. We provide further details on the automatic tagging system.

At 105 an update and feedback process is provided. More specifically, manual or automatic processes may be provided for updating the automatic tagging system and for checking and auditing the process. Optionally, updates and inputs to the oversight process may be provided to manual sources such as data miners or analysts. Alternatively, in some example implementations, artificial intelligence systems employing machine learning, such as neural networks, may use backpropagation techniques for periodic updates to the training phase.

At 107, financial data associated with the enterprise is provided. More specifically, annual reports, press releases, information from publicists or executive social media, published pricing information, and other information as would be understood by one of ordinary skill in the art; Published information is provided.

At 109 , the standardized data of the auto-tagging system 103 and the company's financial data 107 are provided to the modeling system 109 . Modeling system 109 includes one or more handle generators and prediction builders that are applied to the model. Models may include, but are not limited to, artificial intelligence-based systems that apply machine learning in the form of neural networks, such systems include standardized data for automated tagging system 103 and corporate financial data. 107 and apply them to a neural network to produce predictions as outputs.

Also, as noted at 111, inputs to the modeling system may include, but are not limited to, proprietary data related to a company's mergers and acquisitions information, and a company's fiscal quarterly calendar.

At 113, the forecast output is provided as one or more revenue forecasts. As noted above, company financial data based on standardized data and unique data are fed into a machine learning model to generate a revenue forecast.

At 115, the final output is provided in the form of recommendations. For example, without limitation, the final output may be a recommendation to buy, hold, or sell as a transaction. Alternatively, the final output may be a score. Still alternatively, the final output may be the execution of the transaction itself, without user involvement. Optionally, rule-based or other deterministic approaches may be employed in combination with machine learning probabilistic approaches such as those described above.

FIG. 2 shows architecture 200 according to an exemplary implementation. Input 201 includes data providers 201 - 207 provided to big data system 211 , which in turn is associated with access point 213 .

More specifically, as shown at 200, data is received as input 201 from a plurality of N data providers 203-207. When the term "data source" is used herein, data can come from one or more data providers, such as N data providers 203-209, as well as from any obtained from other data providers in

In the big data system 209, there are N corresponding adapters 215-221 for each of the data providers 203-209. Adapters 215-221 may normalize 223, deduplicate 225, and classify 227 the data as described herein.

As described above, data is provided by data providers 203-209 and processed by data adapters 215-221. The outputs of data adapters 215-221 are provided to an automatic tagging system so that the resulting output data can be normalized, de-duped, and classified as described above. Although these aspects of the exemplary implementation are automated, a data miner or analyst may optionally oversee and keep the process updated. The resulting output data may be combined against financial data of the enterprise, such as financial reports, or against other publicly available information relating to historical or current characteristics of the enterprise. good.

The output of the auto-tagging system and the company's financial data are provided to the modeling system. The modeling system may include a plurality of multi-panel generators 229, 231, 233, multi-prediction builders 235, 237, 239 and prediction modules 241, 243, 245. The modeling system generates revenue projections based on financial data, proprietary data, and machine learning as its output projections. More specifically, the financial and proprietary data includes the above inputs for the auto-tagging system and the financial data of the company. For example, without limitation, the data may include financial reports, mergers and acquisitions information, and quarterly financial results.

The output is a prediction and may be provided for output in the form of a file, API, console, or custom format. The output is thus provided to the user at the user access point. Exemplary user experiences associated with access points are described in greater detail herein.

FIG. 3 illustrates a process 300 associated with the above structures and functions according to an exemplary implementation. More specifically, the process consists of a first phase 301 related to data processing, a second phase 303 related to panel development and calibration, and a third It is divided into phases 305 and .

With respect to the first phase 301, at 307 input data is received. As noted above and described herein, the input data is from multiple sources of different types. For example, without limitation, input data may include time series data from multiple vendors such as credit card transactions, debit card transactions, or other electronic purchase transactions.

At 309, the data is normalized. More specifically, the input data of operation 307 is mapped from the organic vendor data format as received to an internal data format that provides consistency between vendors. For example, but not by way of limitation, normalization includes considering differences between different alternative data sets and normalizing the data to a common standard. Additional aspects may include replication or other processes for normalizing data.

At 311, the brand associated with the purchase is tagged. Tagging may include applying tagging rules to the normalized data. The rules may be rules specific to credit or debit cards, geofencing rules for GPS data, rules related to browser history or use of online applications, or other rules as would be understood by those skilled in the art. can include, but are not limited to, For example, without limitation, a series of financial transactions may be labeled or tagged for each financial transaction. One or more tagging rules 313 may be applied to perform tagging. Tagging rules may be thought of as inclusive or exclusive filters. Furthermore, in addition to rule-based approaches such as filters, natural language processing approaches incorporating artificial intelligence may be employed, such as machine learning models involving neural networks.

After the input data has been normalized and tagged, in the second phase 303 a panelize operation 315 is performed. According to an exemplary implementation, a sample of users is established as a panel. Users are selected to match criteria associated with the panel. For example, without limitation, a panel may be created in a manner that relates to the churn rate of the input data. Other examples of channelization rules are shown at 317 and are based on user trading patterns, census data balancing, or other rules that associate characteristics of a user's trading behavior to trading. can be anything.

After the panels are generated at 315 , grouping processing is performed at 319 . More specifically, panels are grouped by symbols. In the tagging process, data-related brands are assigned to company-related symbols. Please note that this data may change over time due to mergers, acquisitions, spin-offs, bankruptcies, rebrands, listings, delistings, or other financial events. Events are thus assigned in real-time, as included at the time of assigning brands to symbols at 319 . Symbols can be selected from a database such as that shown at 321 .

At 323, one or more corrections are applied to the grouped ones. For example, and not by way of limitation, patterns specifically related to financial institutions are applied to calibrate the grouping, as shown at 325 . Examples of such patterns may include weekend postings of information, delayed postings typically associated with financial institutions, and pending but not yet posted transactions. Additionally, as indicated at 327, anomalies are removed. Examples of anomalies that are removed may include, but are not limited to, anomalous transactions or anomalous users.

After the panel has been completely generated in the second phase 303, the process proceeds to the third phase 305 as follows. At 329, a predictive model is generated. For example, without limitation, during the training phase historical data related to the company, such as company fundamentals and stock prices, as well as historical measurements that may have been previously provided by an exemplary implementation, used as training data. A predictive model is therefore trained based on this training data, as shown at 331 .

At 333, predictions are generated. The forecast model generated at 329 is used to derive future stock price forecasts for future company fundamental indicators by applying feature points associated with open quotas 335 . Although the features are disclosed as directed to open quotas, example implementations are not so limited and other open periods may be substituted.

The above exemplary implementations of the architecture may be implemented in hardware, as described below with respect to FIGS. 26 and 27. FIG. In addition, the hardware provides continuous large-scale data input so that the model continuously receives data and is updated automatically and in real time. Therefore, hardware processing power must be sufficient to enable such processing. For example, without limitation, GPUs may be used for artificial intelligence processing. Alternatively, or in combination with a GPU (Graphics Processing Unit), an NPU (Neural Processing Unit) may be used. Either or both of these units may be used in remotely located processors, such as distributed server systems for cloud computing systems.

For example, without limitation, external data fetching according to exemplary implementations described herein copies data from an external third party (e.g., vendor) and stores the data in a cloud storage container. may be performed by doing The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources.

Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction.

Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

Fungible dataset

As noted above, the architecture is configured to receive multiple alternative datasets. As shown in FIG. 4, examples of alternative data inputs 400 include, but are not limited to, single credit and debit card transactions, mobile device usage information 403, geolocation data 405, social data and sentiment. Data 407 and web traffic and internet data 409 may be included.

The hardware implementation described above may be used to process the alternative datasets of FIG. 4, as will be appreciated by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

FIG. 5 illustrates processing of alternative datasets according to an exemplary implementation. More specifically, at 500, receiving and processing alternative datasets is disclosed. The fungible data set 501 may include, but is not limited to, data having types that are fungible. For example, without limitation, GPS data 503, financial transaction related data 505, vaccination data 507, satellite imagery data 509, app usage data 511, and browsing history information 513 may be included in alternative data sets 501. All types of data that can be part of a However, the alternative data set 501 is not limited to the above exemplary case, and may include other data sets, as will be appreciated by those skilled in the art.

At 515, multiple data sources perform ETL (extract, transform, introduce) operations to extract, transform, and introduce data. Thus, at 515 , the feature points associated with data types 1 through m are extracted into corresponding feature points 1 through m, as shown at 517 .

At 519, the process undergoes modeling operations. Modeling operations include model selection, feature point selection, performing a training phase on the model, and performing model testing, as shown at 521 . Modeling operations 519 may be performed on an artificial intelligence system using neural networks and machine learning.

At 523, a verification step, also called backtesting, is performed. For example, as disclosed at 525, historical data related to stock prices, company fundamentals, and past decisions or events are applied to the generated model. Based on the verification of 523, a determination is made whether the application of the past information was successful in verifying the model. If the model was not successfully validated, or in other words if the backtest results were found to be unacceptable, the process returns to 519 and modeling is performed again. be.

On the other hand, if the verification in operation 523 succeeds, then the operation proceeds. More specifically, operations proceed to modeling operations 527 and prediction operations 529, as described above.

The hardware implementations described above may be used to process the alternative datasets of FIG. 5, as will be appreciated by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

automatic tagging

When an alternative data set has been processed as described above, with respect to operation 101, auto-tagging 103 is performed on the data. FIG. 6 shows automatic tagging 600 according to an exemplary implementation.

At 601, a trade is received and the fungible data set is processed as described above. At 603, normalization is performed. More specifically, feature points are extracted from the transactions associated with the brand classifier.

At 605, brand classification is performed. More specifically, the extracted feature points are normalized 603 before being applied to the classification layered model. More specifically, rule-based approaches that perform extraction in a deterministic manner are applied, which are inherently probabilistic in artificial intelligence, such as machine learning based on neural networks. Mixed with approach. Thus, brand classification 605 is performed with a mixture of deterministic and probabilistic models.

At 607, a verification step is performed to verify that the classifier is correct. More specifically, a sample of classified data is validated to confirm that the labeling was applied correctly with respect to the brand. If necessary, perform retraining of the classifier. Optionally, this validation and retraining operations at 607 may be performed iteratively until the brand's classification is validated at a threshold confidence level.

At 609, the brand is classified with respect to the enterprise. For example, a company may be a private company, a public company, or another trading company organization with characteristics similar to a public company or a private company. In a manner similar to operations 605 and 607 above, rule-based deterministic operations and probabilistic artificial intelligence operations such as machine learning and neural networks to classify brands into companies. A mixture is employed. Also similar to operation 607, at 611, the firm classifier is verified to ensure correct labeling of the sample data with respect to firm classification. If necessary, retraining may be performed until the firm classifier is validated at a threshold level of confidence, and optionally repeated gratuitous training may be performed.

At 613, the tagged transaction is considered generated after the brand taxonomy 605 and company taxonomy 609 have been verified. For example, and not by way of limitation, labeling is performed in such a manner that the data is automatically labeled. The output of this process may be used in panel generation, prediction construction, and prediction.

Exemplary implementations may have various advantages and benefits compared to approaches in related art. For example, but not by way of limitation, approaches in the related art include generic names (e.g., DENNYS or SPRINT), short names (e.g., AMC or BOOT), specific names (e.g., TACO or TARGET), and Problems or shortcomings such as incorrect tagging with similar names (eg, FIVE BELOW or BIG FIVE) cannot be avoided. Additionally, entities may be omitted if they do not follow clear criteria. Examples of companies being omitted for not following clear standards are using abbreviations instead of full names, names that have changed over time, and slight variations in names due to different acquisition times of stores. , or typographical errors in the store name. Other examples of errors in the related art include assigning a deal to the wrong ticker (company indicator) even though the deal was clearly associated with a different company.

As will be appreciated by those skilled in the art, the hardware implementation described above may be used to handle the automatic tagging of FIG. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer rates. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

user experience

The output of the example implementations described herein may be provided in the user experience in ways that provide the user with visualization of the information generated by the architecture. Although the example implementations above can be run in the cloud, as described above, an online application that allows users to access a user interface associated with the output of the example implementations above is the user may be provided for Its user interfaces include, but are not limited to, online applications on mobile devices, applications on laptops, interactive applications on home devices such as Amazon Alexa, and automotive applications run by processors in automobiles. or any other aspect or structure that executes computer readable instructions to provide the user experience described herein.

For example, a user may be provided with watchlists generated to provide the user with decision support information related to architecture outputs such as, but not limited to, forecasts, predictions, recommendations, or the like. may be provided to In addition, a detailed analysis of the entries on the watchlist may be provided along with the specific recommendations and detailed indicators on the rationale for the recommendations, optionally provided by external benchmarks. May be compared with recommendations. Additionally, an alert chain may be provided by using the exemplary implementations described herein, or optionally including techniques of deterministic rule-based and artificial intelligence approaches. Decision support or automated decision-making tools may be provided in combination with, but not limited to, the above aspects.

Data processing related to user experience may be processed using the hardware disclosed herein. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer rates. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

watch list

FIG. 7 shows the user experience associated with an exemplary implementation. As shown at 700, the user experience provides a watch list of companies, which may be selected by the user or suggested to the user based on preferences associated with the user, for example. For example, but not by way of limitation, the watch list may include company names in first column 701, including company names and transaction symbols. In another column 703, the stock price is provided, along with information about performance such as stock price change as a total amount and as a percentage.

In yet another column 705, performance-related critical data types used as inputs in the exemplary implementation are identified. As shown in FIG. 7, the first row is associated with the first company, which is a large retail company with a physical and online presence for retail sales. and revenue is the key indicator of performance for this company. The second line is an online subscription company, where the number of other users associated with the product is a key indicator of performance. In the third row, which retail stores the user may visit, same-store visits are provided as a key indicator of performance. Line 4 is a major automaker with an international presence, whose sales in the US are provided as a key indicator of performance. For social media companies, line 5, social media company engagement of users (eg, time spent on the platform) is provided as a key indicator of performance.

Yet another column 707 shows performance associated with key indicators of performance. As shown in Figure 7, in the first row the key indicator of revenue shows a 0.4% performance increase, in the second row the number of users shows a 3% increase, and in the third row , showing a 10% decrease in same-store sales, the fourth row shows a 6.2% increase in US revenue, and the last row shows a 3% increase in user engagement. Accordingly, the exemplary user experience provides the user with information regarding key indicators of performance and actual performance based on input data as described above.

According to this exemplary implementation, the system generates scores in yet another column 709 based on key indicators and performance. A score is a presentation of a company's performance to the user, who can apply it as a form of recommendation. For example, and not by way of limitation, in the case of the first row in FIG. The increase is associated with a score of overperformance. In the third row, a 10% decline in same-store sales is associated with a score of underperformance. In row 4, a 6.2% increase in sales in the US is associated with a score of overperformance. In the last row, a 3% increase in engagement is associated with a score of overperformance.

Thus, the user experience according to the exemplary implementation is based on the type of company and based on available data among multiple data streams, information about key indicators of performance determined by the system, Provide to users. Additionally, actual performance on key indicators by company is provided along with the score determination. For example, without limitation, first line information such as earnings information may be generated based on input credit card information received as a data source, as described elsewhere in this disclosure. . That information may be used to determine revenue associated with the enterprise and may be used to calculate performance.

Although the scores illustrated in the figures are single words such as neutral, overperforming, or underperforming, other scores may be substituted without departing from the scope of the invention. For example, and without limitation, a numerical score such as a performance rating from 1 to 10 may be provided. This information and scores are generated based on the input data provided and based on observing what is characterized as an appropriate score for the company. Thus, each company may determine a different score based on company attributes such as industry or company size to determine the amount of performance variation needed to provide the score.

The example implementations described herein provide a method for determining key indicators such as those shown in watchlists. For example, without limitation, historical data may be applied against an artificial intelligence machine learning model to determine the criteria with the highest correlation with respect to stock price changes. For example, and not by way of limitation, approaches in some related art may attempt to directly measure the total number of users to determine revenue. However, doubling the number of users does not necessarily mean doubling the amount of revenue, so the approach in such related art may not be accurate. The reason this may be the case is that new users may not have the same expressions, the same usage patterns, or the same spending habits as compared to early or previous users.

Similarly, geographic location of use, user, or other information may be important in determining whether the data is representative of the user, and key indicators can be correctly determined. As described above with respect to exemplary implementations, GPS data is received as input and stabilized in the auto-tagging system, so the data can be representative. As another example, the data need to be stabilized not only to show the overall number of users, but also to show the time they spend on online apps, because new users are currently If users spend more or less time in the app than users, this may reflect different user actions with respect to revenue, purchases, advertising, or the like, making the data more accurate. Because it can be

Similarly, when a company acquires Target, the number of users may increase, but those new users may not be representative of previous users. Months or years may pass between the announcement of the merger and the completion of the merger process, during which time publicly available information is incorporated into these exemplary implementations, resulting in significant Indicators, performance presentations, and/or recommendations can be adjusted to calibrate such information. Legal or regulatory obstacles, such as antitrust laws or export controls, may stall or prevent a merger, in which case the presence of certain terms in publicly available information relating to the merger It can be used to tailor predictions and recommendations. In addition, the predictive aspect of the tool makes predictions about post-merger performance based on similar patterns used to train the models regarding how such models are characterized, how they are processed, and how they generate output expectations. may be provided.

As another example, for restaurant-related stocks, the shift from a more severe to a less severe coronavirus pandemic situation has driven consumer preferences to return to private dining. can have an impact. Posting such information may be more sensitive to certain industries or companies. In the case of coronavirus, the travel, leisure, dining, or other industries may be more sensitive than other industries, and this example implementation considers such changes to account for data Stabilization can be performed. Without the above approaches in exemplary implementations that perform normalization, de-duplication, classification, and data stabilization, the data lacks the necessary precision in determining key indicators with a sufficiently high degree of confidence. (eg, clustering to better represent the user population).

FIG. 8 illustrates a process for generating and updating information to watchlists, according to an example implementation. As related to the user experience described above, key indicator types, as well as scores, are continuously updated in real-time based on data input. For example, in the first row, revenue is listed as the type of indicator that matters. However, if there is a change in the input data associated with the company, the exemplary implementation may change the type of key indicator from revenue to same store sales, customer base, or other key indicator. may

More specifically, as shown at 800, inputs are provided to the architecture described herein. For example, without limitation, for each company there is a separate pipeline processing for each type of signal associated with the key indicators. Thus, in the example above, there may be multiple pipeline processes for each of the candidate key indicators for the first company. One pipeline process may relate to signals about revenue, another pipeline process may relate to consumer spending, or indicators of business performance.

More specifically, at 801 real-time data events from alternative data sources such as those described above are provided. Each pipeline process is triggered by receiving new data associated with the model. Input of new data from alternative data sources has been described above. These real-time data events are processed according to exemplary implementations of the architecture and signals are generated based on current data. Thus, at 803 an updated signal is provided. More specifically, the new data received at 801 is provided to the model of the exemplary implementation. Therefore, predictions are generated and provided as output signals as described above.

At 805, the strength of the updated signal is evaluated. For example, but not by way of limitation, output signals, which may be predictions of potentially important indicators, are compared to benchmark consensus signals. According to one example implementation, the updated signal may be the firm's projected earnings, which are forecasts generated by one or more analysts from available information, i.e. compared to consensus. Based on the comparison between the updated signal and the benchmark signal, a signal strength vector is generated, which indicates the degree of relative closeness between the updated signal and the benchmark signal.

At 807, the strength of the updated signal is classified. More specifically, the signal strength vectors generated at 805 are classified into recommendations. For example, without limitation, the recommendation may be a one-dimensional actionable recommendation such as hold, buy, or sell.

At 809, the signal strength related classifications are used to generate any updates, which are then provided and displayed to the user in the user experience. More specifically, user interfaces such as web interfaces, mobile interfaces, or pushed alerts may be updated with the results of signal strength classifications and with the results of watch list updates. Thus, according to an exemplary implementation, depending on which of the pipeline processes has the strongest vector, changes in important indicator types, and also recommended changes, are provided to the user. may be provided for Optionally, multiple pipeline processes may be blended or weighted based on signal strength, thereby providing a blended result, with key indicators listed as the most heavily weighted indicators. be done. According to another example implementation, all of the multiple indicators may be listed together with the resulting multiple outputs, followed by each of the signals for each company, for each pipeline process. Recommendations based on matrix operations or multiplexing of intensity vectors, weighted or unweighted, may be listed.

According to this exemplary implementation, real-time changes regarding key indicators, performance, and recommendations may be provided to the user in actionable form. As will be discussed below, recommendations may also be provided within the company's detailed analysis or within the alert chain, and may be used for automatically made recommendations or actual actions.

The hardware implementation described above may be used to process the watchlists of FIGS. 7-8, as will be appreciated by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of specific inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer rates. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

performance and recommendations

As shown in FIG. 9, according to another exemplary user experience 900, any of the companies illustrated in the above user experience of FIG. 7 may be selected to provide further summaries. Here, row 4 is selected to provide further information related to prediction. Recommendations 905 are provided, in this case "strong buys" based on decreasing revenue.

Further details 903 are provided to the user, such as the relative impact of earnings on stock price (e.g., correlation between earnings and stock price) and predictive accuracy for stocks in relation to exemplary implementations. be done. Information is provided regarding forecasts based on analyst estimates compared to forecasts based on actual data measurements, and scores related to performance based on actual data measurements are also provided. Chart 907 graphically provides a comparison of returns based on actual data measurements and expected returns based on public information provided by the company. In addition, stock price information is also provided.

FIG. 10 shows a process 1000 for generating output visualizing differences between expected and actual values, according to an example implementation.

At 1001, measured sales are provided. As initially applicable, but not limited to, as described above, measured sales are calculated such as in the second phase 303 as illustrated in FIG. 3 and described above. May include sales.

At 1003, projected sales are provided. Projected sales are defined as analyst estimates, analyst consensus, industry publications, or other available information related to providing projected sales information, as opposed to actual sales information. may be based on

At 1005 , an aggregation operation is performed on the measured sales figures obtained at 1001 . More specifically, the measured sales are aggregated based on a time series and at intervals equivalent to those associated with expected sales signals. For example, without limitation, if the expected sales signal has a known time interval such as daily, weekly, hourly, quarterly, etc., the measured sales are aggregated over the equivalent time period. be.

At 1007, scaling operations are performed. More specifically, since the measured sales figures provide information from a sample of the population, the measured sales figures need to be scaled. For example, but not by way of limitation, scaling involves extrapolating a representative population of related samples to a sample, and similarly, calculating the aggregate sales of that population of samples to represent the representative population. extrapolating to the population.

At 1009 , the period associated with the scaled and aggregated information on measured sales from 1001 , 1005 and 1007 is combined with the period of expected sales at 1003 . Therefore, the aggregated and scaled measured sales are matched against the expected sales by time interval.

At 1011, visualization operations are performed. Accordingly, output may be provided to the user in the form of bar graphs 1013, line graphs 1015, or any other visualization technique 1017 as will be understood by those skilled in the art. Visualization techniques are not limited to a single approach, and different visualization techniques may be combined, added, mixed or blended with each other. Further, as an alternative to the graphical visualization approach, the textual output may be provided in the form of written charts that simply display the results.

FIG. 11 shows an exemplary graphical visualization associated with the exemplary implementations described above. More specifically, at 1100, a chart is provided displaying sales revenue on a weekly basis over a period of twelve weeks. In 1101, tabulated information on sales measured on a weekly basis is displayed. In addition, at 1103, an expected sales amount provided on a weekly basis is displayed. Accordingly, a display is provided to the user graphically illustrating the difference between the measured sales associated with the exemplary implementations herein and the expected sales, such as provided by analysts, analyst consensus, etc. can provide.

Returning to FIG. 9, chart 907 includes a revenue comparison. Thus, the user can see from the visualization by looking at detailed information and 903 that the measured profit is about 6.2% higher than the forecast based on analyst forecasts. Thus, an exemplary implementation provides a determination that an entity is overperforming relative to analyst expectations. As a result, the exemplary implementation generates a "strong buy" recommendation at 905 in the manner described above.

The hardware implementation described above may be used to process the evaluation and recommendation operations associated with FIGS. 9-10, as will be appreciated by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of specific inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

alert chain

As shown in FIG. 12, according to yet another user experience 1200 associated with an exemplary implementation, alerts are provided to a user. Selected alerts are listed as multiple rows, divided into alerts specific to user 1201 and aggregated list 1213 of the most popular alerts across all users. As described above, columns are provided for the company name and stock symbol, the type of information the alert is based on 1203, the circumstances that trigger the alert 1205, and the frequency with which the alert should be provided 1207. The user is provided with tools 1209 for setting alerts to active or inactive, tools 1211 for editing alerts, and tools for deleting alerts.

For example, without limitation, the alert may be based on performance being greater than 5%, as shown in the first row. As shown in the second row, multiple alerts are provided in association with a single firm, in this case price movement of more than 5%, and a score of "overperforming" is to have Similarly, time-based alerts may be provided, for example, as shown in line 3, the user is provided with an alert 10 days prior to the earnings call. Thus, the user is provided with a series of alerts that can be triggered by one or more circumstances related to the enterprise.

Also, as shown in FIG. 12, users can see the most popular alerts 1213 being generated across their user base and how many users 1215 are generating and using those alerts. can be Detailed information is provided about each of those popular alerts, including type, status and frequency. The user is provided with the option of adding those alerts to their own user alert chain, as indicated by the '+' symbol 1217 .

FIG. 13 shows operations 1300 associated with processing an alert chain. For example, as described above, type, status, performance, price, and score can be processed as data to determine whether to provide an alert to the user. Additionally, users may view and review alerts that are being used throughout their community, thereby benefiting from alerts generated by others. According to an exemplary implementation, users may subscribe to such rules and receive alerts in response to changes in the above signals.

At 1301, a signal update is performed. More specifically, signal updating is performed as described above with respect to 803 of FIG. For example, output of a watch list including, but not limited to, signal strength vectors and signal strength classifications may be provided.

At 1303, the output of 1301, such as the absolute value of the signal, or the relative value of the signal compared to the previous signal update, is applied against the deterministic rule. For example, without limitation, provide an alert when revenue has increased by 2% compared to the last signal update and the user has increased revenue by more than 2% for a given company The rule may be triggered if you have set the rule to Similar or other deterministic rules as described above with respect to FIG. 12 are processed in real-time as updates at 1301 are provided. Thus, the probabilistic output of the watch list is applied against the user's deterministic rule base, and alerts are either triggered or not triggered by each real-time update.

At 1305, a notification generation operation is performed. More specifically, a custom notification payload is generated for the user based on the user's rule base. For example, without limitation, the user has determined that they should receive an alert that they should sell their stock if their earnings increased by 2% or more compared to the previous signal update. In some cases, such notification payloads are generated.

At 1307, a notification is delivered to the user. A delivery channel may be set to one or more modes including, but not limited to, mobile application notifications, email simplified messaging services, or other means of communication. Accordingly, notifications containing rule-related recommendations, triggered based on real-time signal updates provided to the rule base, are pushed to the user in the manner desired by the user.

Although some example implementations may provide real-time recommendations based on real-time measurements of data collection, price checks, and calculations, example implementations are not limited to recommendations. The above example implementations may be integrated with approaches that provide automated trading such that the user does not need to provide input and execution for decision making.

FIG. 14 illustrates a process 1400 associated with making decisions according to an example implementation. More specifically, instead of users receiving and viewing signal classifications and signal strength vectors to receive alerts, signal classification is achieved by applying artificial intelligence techniques such as machine learning within neural networks. and signal strength vectors into decision-making signals that can be submitted for decision-making execution. For example, without limitation, signals may be converted into trading orders for a brokerage service.

At 1401, measurement data is received from multiple alternative data sources and processed on an automated basis in real time, according to the exemplary implementations described above, thereby providing real-time updates for key indicators. .

At 1403, updates are input to an equivalent but not limited trading model, which employs artificial intelligence approaches such as machine learning within neural networks to provide trading decisions. may contain. According to an exemplary implementation, the trading model may be a neural network trained on past stock prices and a history of signal updates associated with an exemplary Tatian measured signal. Inputs to the trading model are signal data and the current state of the user's account. For example, if the user has open or pending orders related to purchases or sales, the model will take this information into account. As an output, trading model 1403 provides decisions or actions instead of alerts or recommendations as done in other exemplary implementations.

At 1405, an order is created. For example, an order is created by submitting an automated request to the merchant, including but not limited to brokerage services, and optionally confirming the order to the user. may User confirmation may optionally be provided by the notification delivery channel, as described above with respect to FIG.

At 1407, the system performs an operation to determine if user viewing is required. If no user viewing is required, the order is executed at 1409 . For example, the order created at 1405 may be executed by a merchant such as a brokerage service.

Alternatively, if at 1407 it is determined that the user is a purchaser requiring viewing, at 1411 a confirmation request is sent via the notification delivery channel as described above. . If the user confirms the order at 1411 , the order is fulfilled at 1409 . If the user does not confirm the order at 1411 , the order is considered rejected and canceled at 1413 .

More specifically, periodic notifications, such as push notifications, may be provided to users. Periodic notifications may occur on a regular basis, such as running daily or at the same time each morning. Push notifications may include, but are not limited to, information related to data conditions. For example, based on the exemplary implementation above, and based on the predictions associated with the signals, it is predicted that the expected circumstances will not be met, such as the company missing earnings reports. The user may be notified of the data condition when it is determined that Additionally, reliability may be presented to the user. In one exemplary implementation, the user may be provided with a confidence score in units of percentage.

In addition, according to an exemplary implementation, push notifications provide selectable choices (also known as "user prompts") that require only a single user action to make a decision based on the information provided in the push notification. ) may be included. For example, and not by way of limitation, the user may be asked the question "Would you like to invest?" followed by a selectable choice of "yes" or "no."

If the user selects the "yes" option, the purchase order is executed based on pre-defined user settings such as user preferences. According to one exemplary implementation, if the user has included a preference in the user settings associated with the online application to place the order immediately, the order will be placed as soon as the user makes a selection without delay. will be submitted. According to another example implementation, orders may be executed over a period of time, such as business days. According to one exemplary implementation, the user may include a preference in the user preferences to make purchases of a predetermined amount, such as $1 million, at even intervals throughout the day.

Optionally, the default settings are clustered at 20 minute intervals on certain parts of the day and spread at longer intervals of 1 or 2 hours on other parts of the day. The user may be allowed to adjust the percentage of orders for the day at uneven intervals. Also optionally, the defaults fit a user-defined profile curve based on the output of recommendations in the above exemplary implementation of providing predictions, or based on other user-determined defaults. and to match expectations of stock prices rising or falling in the next period; It may be possible to allow the user to adjust the quantity of each order, evenly over the course of the day, so that it gradually increases over the course of the day and aggregates toward the end of the day.

Similarly, user settings may have a limit order in the default settings such that if the price exceeds the limit price, the order will no longer be executed for the remainder of the day. Thus, according to one exemplary implementation, the user selects "yes" to buy $1000 of stock in equal installments over 5 hours and place an order of $200 every hour. In fact, if the stock price exceeds the price limit in hour 3, the order for $200 in hour 4 will not be executed if the price continues to exceed the price limit. Similarly, if the stock price exceeds the limit price in the 5th hour, the order in the 5th hour will not be fulfilled.

For order execution, a user may provide one or more default service providers, such as brokerage firms, by default. Additionally, the user may provide one or more stock exchanges for executing orders. The user may specify the percentage of total orders to place for each brokerage firm, split evenly or unevenly or automatically proportionally based on past performance against past recommendations. .

For order size, user preferences may provide a predetermined order size, such as 100 shares, or a predetermined amount, such as the equivalent of $1000. Therefore, the user does not need to decide the number of shares, the amount, which securities company or stock exchange to inquire about, the timing of placing and executing an order, the ratio or evenness of the timing or amount in terms of time. , because the parameters have already been determined based on preset user settings. Thus, the user can simply select "yes" or "no" to execute the order according to user preferences determined by the user.

Although the example implementations above refer to a single security and the single-user actions associated with purchasing that single security, example implementations are not limited to do not have. For example, without limitation, a user may receive multiple push notifications associated with each of different companies, either at the same time in time or distributed in time based on desired user preferences. good. The companies provided within the push notification to the user, such as those with a credibility score above a certain percentage, are also of interest to the user, as shown in the User Experience Implementation above. or a company on a list of companies the user is tracking, or a predetermined threshold between the predicted value determined by the exemplary implementation and the published projected earnings report. It may be determined based on the user's preference, such as companies with a difference of .

In addition to the one-to-one relationship between user actions and trade executions, selecting an option such as “yes to all” will allow multiple push notifications delivered together to be combined into a single intention. An option to act on the decision may be provided to the user. If such an option is selected, all orders will be fulfilled according to pre-defined user preferences, as described above.

The example implementations described above may provide push notifications through modes known to those skilled in the art. For example, without limitation, push notifications may be provided by email, text message or SMS (Short Message Service), chat via an online social networking service, or the like. Further, push notifications may not only be provided by visual presentation, but may alternatively or additionally be provided as voice messages, such as through speakers on home devices such as Amazon Alexa or the like. good.

Further, the above exemplary implementations provide options during transit, such as drivers receiving signals, such as via communication networks including but not limited to 5G networks. good too. Thus, the driver may issue commands to carry out orders as described above by voice communication with a speaker and microphone, or by other input/output devices as will be appreciated by those skilled in the art. may be executed by a voice command of Similarly, passengers may also use the system. Thus, example implementations may provide hands-free ordering.

Additionally, although example implementations may be implemented on mobile communication devices such as smart phones, example implementations are not so limited. For example, without limitation, user settings may require users to participate from an authenticated device, such as by requiring two-factor authentication, before accessing recommendations or push notifications. Additionally, the user may be provided with control over a single-action purchase mode, or require other forms of authentication, such as login, biometrics, or two-factor authentication, to receive push notifications. Privacy protection aspects may be provided as described. Similarly, instead of the user providing a voice response or selecting a particular option on the screen, the user may use gestures, signals, or other biometrics to make decisions. may be presented.

For example, without limitation, the user has determined that in the user settings the user has indicated that an order should be placed when the first finger is placed on the fingerprint detector. On the other hand, it may be determined that if a finger other than the first finger is placed on the fingerprint detector, the user indicates that the order should not be fulfilled. Such an exemplary implementation would allow the user to hear and see the user interface, but understand if the user's single action is a decision to execute or not to execute an order. It allows you to make your own decisions in the presence of others who cannot. Although the above examples refer to fingerprints, other gestures or user signals defined in user settings may be employed. Similarly, use of voice commands other than "yes" or "no" also prevents third parties from knowing what the user has decided just by hearing the user's voice response. be able to. Thus, the user's privacy is protected even in situations where the user interface is seen or heard by others.

In terms of hardware, the exemplary implementations described above may be implemented within client devices such as smartphones, laptops, or the like. Further, as described herein above, the above exemplary implementations may include, for example, devices, processors, memories, etc., relating to automobiles or other vehicles, or other devices as will be appreciated by those skilled in the art. It may be integrated with other devices, such as devices. The user experience described above can be applied to online applications as parameters entered by text input, radio buttons, checkboxes, sliders, or other visual aspects of user input and user output as will be understood by those skilled in the art. may be provided by the user in the "user-preferred" aspect of

The example implementations described herein may be implemented in the form of machine-readable executable instructions stored in memory, which instructions may be used to access predefined user settings. These instructions can be configured and configured such that one or more devices authorized by the user to execute the instructions can access user settings, for example locally or remotely in the cloud. It may be stored in a database or the like. Instructions may also be provided in the form of an online user application.

Before executing the above exemplary implementation to provide push notifications, the user must enter user preferences. When an online application is installed, a set of default settings may be provided based on a profile of information associated with the user. For example, without limitation, if a user self-identifies their risk level as high, medium, or low, a response may be made based on aggregating other user information related to their risk profile. Default settings may be selected from a set of pre-defined profiles that match the user's default settings to the average default settings for other users with similar risk profiles. good. Factors other than risk may be used to determine initial settings.

In addition, a user may have multiple preferences, and different preferences may be applied according to different user situations, which may be automatically accessed by the online application. For example, without limitation, if the online application recognizes that the user's available cash funds in a bank account exceed a predetermined level, the default is to set the user-defined amount, frequency , price limits, or even other aspects may be set to a default setting that is adjusted to account for increased availability of investment funds or to account for increased risk profiles. good.

Optionally, the user adjusts the settings according to time and date, such as before or after the earnings call, when the fiscal year starts/ends, or other significant times as will be appreciated by those skilled in the art. You may Similarly, timing may be based on the user's state, such as after an employer's payday, or after monthly debt payments such as mortgages, credit cards, or the like. Similarly, an online application may choose to change among default settings based on changes in the user's financial situation, such as a new job, layoff, major purchase such as a home or car, vacation, or other financial event. The user may be prompted to do so. Thus, user selection of default settings may be partially or fully assisted or automated. Alternatively, instead of entering one or more user settings, the user may generate a rule base that applies to the user settings, or for a "yes" decision on a push notification. may generate a rule base that is directly applied to

FIG. 27 shows an example process 2700 associated with the above example implementations. More specifically, exemplary process 2700 is directed to executing orders based on single-user actions, where orders are placed based on user context and to a single user. Executed in response to a push notification containing a data condition prompting an action.

At 2701, after the online application is installed, information associated with the user is received as user preference information. For example, and without limitation, information related to a user's profile, such as, but not limited to, risk information, financial information, demographic information, or other information related to the user, may be provided by the online application. received.

At 2703, user settings are determined. More specifically, the user preference information received from the user at 2701 is aggregated and stored as a set of preferences, thresholds, parameters, or the like, such as in a database. Additionally, as noted above, additional user settings may be used to determine user preferences such as the current point in time during a period, such as quarter end, year end, payday, before or after major purchase events, or the like. Support tools or automated processes may be considered. As a result of operation 2703, the user settings are saved in machine-readable form so that they can be accessed and applied to an order to provide the necessary instruction or order execution without requiring user input. be.

At 2705, the exemplary implementation performs data processing to determine data conditions. More specifically, as described above, multiple alternative data inputs of different types are processed, normalized, and applied to a series of operations to provide predictions or expectations or recommendations. to generate Additionally, recommendations, predictions, or expected values may also have an associated reliability. Thus, a data term is generated that includes a forecast or expected value or recommendation compared to forecast information, such as expected quarterly earnings, and a confidence associated with the recommended forecast or expected value.

At 2707, data conditions are provided to the user in the form of push notifications, such as SMS, email, or the like, as described above. Additionally, the user asks for a single user action such as "yes" or "no". The user only has to decide whether or not to execute the order based on the data conditions provided. Information at 2707 is provided to the user via a user interface and may include, but is not limited to, visual input and audio input as described above. .

At 2709, single-user instructions are received. Single-user instructions are based on single-user actions. For example, and not by way of limitation, a single user action is the user selecting "yes" or "no". This may be done by visual or audio input to the client device, as described above, which may include a mobile phone or laptop, also as described above. good.

At 2711, a determination is made about the instructions associated with the single-user action. For example, without limitation, it is determined whether the single user action is to place an order or not to place an order. If the operation at 2711 determines that the single user action is an instruction not to execute an order, the process ends.

On the other hand, if the single user action is determined to be an order to execute order, then at 2713 the order is executed based on the user settings determined at 2703 . More specifically, user settings determine whether orders are executed immediately or at a later time, whether orders are executed as a single trade or as multiple trades, and whether multiple orders are executed. if such a transaction is requested, the timing and amounts to be distributed evenly or unevenly over the day or other period, and whether to distribute them unevenly over the day or other period. If so, are there any price restrictions or other restrictions or patterns, and is there an identification of one or more brokerage firms or markets that constitute the service provider, or other Provide information including, but not limited to, whether user settings exist. Thus, orders are executed based on single user actions in conjunction with user settings.

At 1415, an operation is performed to update the user's portfolio to show that the order has been executed. Optionally, to confirm that orders have been executed and/or to provide portfolio updates and/or to remind the user of pending or unresolved orders. Reports may be provided to users through the communication channels or distribution channels described above.

The hardware implementation described above may be used to process the alert chains disclosed in FIGS. 12-14, as will be understood by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

The above exemplary implementations of automated trading aspects are performed continuously and in real time. Thus, multiple orders may be executed simultaneously or may be in different stages of execution. To the extent that confirmation by the user is required, the user may decide to execute some orders, reject other orders, and not yet place other orders on hold. You may To the extent no confirmation by the user is required, automatic execution may continue without providing any visibility to the user prior to execution, and notification of pending orders or current portfolio immediately after execution. may be presented. Execution without user viewing is another non-strategic subject. Changes are required in the availability that is the identity for justice or the availability on the entity side of the order, such as exceeding a predetermined amount for individual or subscription purchases.

The exemplary implementation above automatically determines the types of indicators of interest, calculates performance, and generates scores for each of the companies listed. In the above exemplary implementations of the user experience, leading indicator determination and performance comparisons are provided on a real-time basis. As noted above, a mixture of rule-based deterministic approaches are mixed with probabilistic approaches such as the artificial intelligence neural network approach described herein. The architecture may run on a big data server, such as in a server farm or in the cloud. Thus, signals are processed in real-time as they are received, unlike approaches in the related art that perform batch processing. Additionally, cloud computing may also be used. To the extent the necessary processing power is available, signal processing associated with exemplary implementations may be performed on a self-contained machine. Because this example implementation can provide real-time information, users can receive recommendations and make decisions in a timely manner that has a significant impact on their return on investment. can. In contrast, the approaches in the related art do not consider the processing of real-time measured data, instead only using predictive data information.

Although the above exemplary implementations are shown on the screen of a display such as a laptop computer or a desktop computer, exemplary implementations are not limited to those two in terms of user experience. For example, without limitation, user experiences associated with mobile devices may replace the exemplary user experiences described above, or may be used in combination with such user experiences. Figures 15-16 illustrate such exemplary implementations associated with mobile devices.

Additionally, Figures 28 and 29 illustrate additional user experiences associated with exemplary implementations. For example, FIG. 28 shows interface 2800 . More specifically, at 2801, a comparison of company performance to industry performance is shown. More specifically, a profit and loss comparison is generated based on the output provided by the exemplary implementation above and based on alternative data sources processed by the architecture and operations described herein. , which generates an expected profit/loss value based on the signal. As can be seen, companies are performing well above industry levels. At 2803, P&L comparisons for industry sections provide analyst comparisons (e.g., overperformance, underperformance, neutral) when compared to expected P&L values generated by signals according to exemplary implementations.

FIG. 29 provides a user interface associated with comparing companies on a watch list 2900 as disclosed herein. More specifically, stock-by-stock results are shown over a period of time, such as the most recent quarter. As can be seen, each company is presented with the metrics underlying the valuation, strategy choices, and profit and loss results for the period. Comparison results are shown graphically. Thus, although multiple companies can be compared using data that has been processed and normalized as described above, the data that is most important and/or most sensitive to the financial performance of the companies can also be compared, as described above. The ability to rate each company based on parameters is also retained.

The above example implementations may also be used to provide users with discounts on company-related goods and services. Discounts may be in the form of rebates, rewards, markdowns, or other forms of benefits or compensation to the user. For example, and not by way of limitation, discounts may be based on the user's trading volume on the company's account as illustrated in the exemplary user experience in which the exemplary implementation performs the above operations on the above structure. , may be provided to the user.

Optionally, according to example implementations, a feedback loop may be provided. More specifically, the user may provide input regarding model results and recommendations, which may be used to calibrate the model. For example, without limitation, based on key parameters or indicators in combination with performance, if a recommendation is set that the user does not agree with, the user may respond to feedback within the system with suggestions. may provide. The user may suggest that the key indicator is not appropriate, may instead suggest another key indicator, or may suggest that the recommendation is not the desired recommendation. According to one exemplary implementation, a user may indicate that a suitable key indicator of a social media or online service is not only user engagement or user numbers, but also online advertisements, clicks, or some other parameter. You can suggest things that may be related. This feedback may be used by the system to retrain or recalibrate the forecasting tools to adjust according to consumer preferences and demands.

Optionally, automated trading tools may be provided that implement specific recommendations under specific circumstances. For example, without limitation, for items on the watch list that are also on the alert chain, instructions from the user in the user preferences may indicate that when one or more conditions in the alert chain are satisfied, a particular stock should be purchased or sold. Optionally, the user may create an entirely different and separate alert chain for automatically executing trades and for providing alerts. Thus, trades may be executed automatically by the user, thereby avoiding any delays and providing such orders immediately. Such systems can be of value in certain use cases. For example, but not by way of limitation, automated trading tools may allow users to quickly or timely respond to alerts or recommendations during periods of peak activity or intense financial news, when the user is traveling or on vacation. It can be of value in cases where it cannot deal with the memory, or simply for the convenience of the user.

According to an exemplary implementation, the automated trading tool receives user preference information via an online application associated with the user device and sets one or more user settings based on the received user preference information. determining, wherein one or more user settings as a set of preferences or thresholds or parameters in machine-readable form that can be accessed and applied to automated user requests; Aggregate and store in a database so that at the time of the automated user request, the automated user request shall be executed without the user inputting user preferences and determine data conditions wherein in determining data conditions, a plurality of alternative data inputs of different types are processed and normalized and applied to a series of operations, whereby generating reliable predictions or expectations or recommendations and comparing data terms with third party entity information; and generating electronic notices indicative of data terms, where electronic notices are , a data condition compared with third-party entity information, a confidence indicator associated with the data condition, and a query to perform a single action; and an automated responsive to the determination, the user device generating an automated instruction to perform the automated user request; Execution of a user request, where the user settings include information as to whether the order will be executed immediately or at a later time, and whether the order will be executed as a single trade or multiple orders. and if multiple transactions are requested, whether the timing and amounts are spread evenly or unevenly over the period; if unevenly distributed, shall include information as to whether there are any price limits or patterns, and information as to the identification of one or more brokerage firms or one or more markets; It's okay.

Also, optionally, in addition to using the company's financial data as described above, other publicly available data related to the company may be used. For example, but not limited to, one or more social media accounts, industry events, news releases or publications, or other information related company leaders, public facing announcements; Social media posts, public presentations, or other information may be sensed or detected. This information may be combined within sentiment analysis.

The example implementations described above may have various advantages and/or benefits. For example, without limitation, an exemplary implementation may provide a way to determine performance automatically and in real time. Approaches in the related art may involve manually running surveys, manually generating performance presentations over many days and many hours, and manually determining scores by investment advisors. However, approaches in the related art involve obtaining heterogeneous data from different data sources, performing operations on the data such as normalization, de-duplication, classification and optionally other operations, and performing sophisticated operations on the data. applying the obtained data to predictive tools and generating predictions or recommendations, all automatically and all in real time, based on the determination that particular indicators are important; not

This difference is important because it allows users to effectively make decisions and save time before predictions change again with the passage of time, dispersion of information, new information, or other events. information and recommendations must be provided in real-time so that they can make decisions. Actual continuous decisions, as well as recalculations and re-predictions, provide information and recommendations that are accurate and precise and can be used to assist users in making decisions.

Data cleaning and pipelining

According to the example implementation above, the user is prompted to create a panel. As provided herein, panels may be selected according to the following selection process.

An optimal panel to be used for prediction may be generated. More specifically, the selection process includes performing a filtering operation on candidate users for the panel. For example, and without limitation, a filter operation may be performed by applying a filter to produce panels that meet one or more criteria. According to this exemplary implementation, the criteria are users who are representative of the population (e.g., the US population) for their geographic location and from the start date to the end date (e.g., from 2011 to the current date). It may include, but is not limited to, panels containing users with a substantially steady number of purchases and a certain number of transactions. Optionally, outliers and duplicates may be removed.

A selection process 1700 may be performed as shown in FIG. Initially, the pool of candidate users is a fragmented panel of 60 million users at 1701 . After the first filter process performs filtering operations so that the pool of candidate users is representative of the population for geographic location, the filtered pool is narrowed down to 15 million users at 1703 . Another filtering operation is then performed to see a stable number of purchases over the time interval, and at 1705 the pool of candidate users is further narrowed down to 4 million users. At 1707, outliers and duplicates may be removed and a filter operation is performed on a constant number of transactions to generate an optimal panel of 1.5 million users.

FIG. 18 illustrates data pipeline processing according to an exemplary implementation. At 1801, data normalization is performed. Data received from different institutions may have different structures. In order to properly use data from different institutions, the data must be normalized by a fixed format. More specifically, data is transformed from textual form into a distributed database designed to operate on transactions.

At 1803, data cleaning operations are performed. More specifically, data that is not needed or desired for data pipeline processing 1800 is removed. For example, without limitation, duplicate transactions, duplicate accounts, and duplicate users are identified and removed. Additionally, incomplete data, such as incomplete transactions with missing data, are also removed from the data. Therefore, also as described above with respect to panel selection, the base of users who satisfy the selection process and filter operations are maintained in the user database, and outliers and duplicates are eliminated.

After data cleaning 1803 is completed, a classification operation 1805 is performed. This operation classifies each transaction in the database. Classification is performed by analyzing the transaction descriptor and associating the correct merchant name with the descriptor. In this operation, the quality of association has a direct and decisive impact on the quality of correlation and rejection, which will be discussed further below. More specifically, a combination of automated machine learning algorithms are combined against human manual control to reach a desired accuracy, such as close to 100%.

An exemplary extraction is provided as an example of the classification operation 1805 above. According to one exemplary implementation, the average throughput per month may be 200 million or more transactions, each of which must be associated with the correct company. .

FIG. 19 shows an exemplary extraction. As seen in exemplary extract 1900, the name of the merchant is found in each transaction.

After the cleaned data has been classified as described above, modeling operations 1807 are performed. More specifically, modeling operations 1807 generate predictions. As inputs, data output from classification 1805 and third party inputs, such as Bloomberg data 1809, are applied to modeling operations 1807 . The modeling operation 1807 obtains an optimal result by combining many predictions. Modeling operations 1807 are specialized for categories related to businesses. In addition, modeling operations 1807 compensate for various bias factors such as seasonality and the like.

FIG. 20 shows an exemplary implementation 2000 associated with modeling operations 1807. FIG. More specifically, a set of panels 2001 subjected to the above operations in data pipeline processing 1800 is provided to a corresponding set of predictions 2003, and the output of the set of predictions 2003 is It is provided to a plurality of assemblers 2005, which produce a final prediction 2007.

In modeling operations 1807, different categories of companies may require different algorithms to generate forecasts. For example, without limitation, algorithms should incorporate different consumer actions for different revenue structures within a company. Some of the categories may include, but are not limited to, wholly-owned restaurants, franchised restaurants, supermarket chains, insurance companies, and other businesses.

Modeling operations 1807 also provide bias correction. For example, without limitation, panel data may include years of historical data from thousands of users who have shared data in exchange for tokens or rewards. Such an approach to obtaining data allows us to more fully understand the biases that non-randomized panels impose and allows algorithms to correct for biases.

FIG. 21 shows a user experience 2100 associated with obtaining information. At 2101, the user is provided with input screens for entering data, entering a referral code, and the number of points that may be associated with completing the survey. At 2103, upon completion of the survey by the user, the number of points increases, and an option for selecting a bank where the points can be deposited and a privacy policy are displayed.

To perform bias correction, analyze historical data points to determine how the bias behaves over time. According to one exemplary implementation, nine algorithms were created to adjust data bias. We then combined those 9 algorithms with 3 other algorithms that are used for tickers and are less subject to bias.

At 1811, the output of the modeling is provided to perform prediction. More specifically, data is aggregated and inserted into the database. The data in the database can be accessed and used by other rules, such as traders, and further captured for validation, backtesting, and the like.

Additionally, data pipeline processing 1800 includes anomaly detection 1813-1819. More specifically, the output of each component of the data pipeline processing is subjected to anomaly detection to identify anomalies that could compromise the final prediction. As an example of anomaly detection 1813-1819, data flow anomalies within a pipeline processing chain may be checked for technical anomalies and data anomalies.

For example, but not by way of limitation, anomalies may be analyzed with respect to the behavior of the company against which the forecast is run, providing the following, but not limited to:

1. Acquisitions or sales informed by third party news sources.

2. For example, but not limited to, requirements changes that would change corporate behavior, such as accounting standards, as informed by SEC (Securities and Exchange Commission) files.

3. Changes in the ratio of franchised to privately owned stores.

4. Different periods of the quarter, such as changing from 90 days to 97 days.

5. Special promotions or other promotional activities over several weeks, which may vary from quarter to quarter.

6. A new product release by a company.

According to an exemplary implementation, representativeness is considered. More specifically, according to a particular example, the population of the United States in 2019 is approximately 330 million and the average family size is 3.14, thus approximately 100 million. There are 5 million households, and some studies put the number as high as 128 million. In an exemplary implementation, the ratio of US households to best panel is approximately 70/85. Also, the ratio between the company's reported income and the total amount of panel purchases is about 70-90. This ratio is in line with expectations and consistent across different companies, so consumer ratios are well-maintained panels with exemplary implementations.

Figures 22A and 22B show various examples of comparisons with data. Figure 22A includes an example of a franchise and Figure 22B includes an example of a company with a history of acquisitions or an anomalous history.

FIG. 23 shows a comparison between results from an exemplary implementation and approaches in related art. As can be seen from the column labeled "DF" for the exemplary implementation and the column labeled "1010 data" for the approach in the related art, the substantial difference in prediction results is It shows a significant performance improvement in the implementation.

FIG. 24 shows the technical basis following the statistical method for determining the panel size and measuring the actual statistical error.

From the central limit theorem, the theoretical % error on prediction is given by:

The % error in the prediction is proportional to the sample standard deviation of purchases and inversely proportional to the number of purchases and the square root of the mean amount.

This means that the panel size is sufficient to produce accurate predictions.

Correcting and adjusting the data by using previously published earnings further reduces the actual statistical error.

The hardware implementations described above may be used to process the operations of FIGS. 18-24, as will be appreciated by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of specific inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

Exemplary Environment FIG. 25 illustrates an exemplary environment suitable for some exemplary implementations. Environment 2500 includes devices 2510-2555, each of which is communicatively connected to at least one other device, eg, via network 2560 (eg, by wired and/or wireless connections). there is A number of devices may be communicatively connected to one or more storage devices 2540 and 2545 .

An example of one or more of devices 2510-2555 may each be computing device 2600 described in FIG. Devices 2505-2555 include a computer 2510 with a monitor (eg, a laptop computing device), a mobile device 2515 (eg, a smart phone or tablet), a television 2520, a device associated with a vehicle 2525, a server computer 2530 (simply “server ), computing devices 2535 and 2550, storage devices 2540 and 2545, and smart watches or other smart devices 2555.

In some implementations, devices 2510-2525 and 2555 may be considered user devices associated with enterprise users. Devices 2530-2550 may be devices associated with service providers (eg, used by external hosts to provide services such as those described above and used for data collection and storage).

The hardware implementation described above may be used in the environment of FIG. 25, as will be appreciated by those skilled in the art. More specifically, for example, but not by way of limitation, external data fetching according to example implementations described herein copies data from an external third party (e.g., vendor) and stores that data in It may also be implemented by storing it in a cloud storage container. The data fetching process may be managed by a scheduling server and/or by a serverless computing service that performs operations to manage external data storage and associated computing resources. Further, the data extraction, transformation, and introduction described herein may be performed by a batch management processor or batch management service. Batch computing is the execution of a series of executable instructions (a "job") on one or more processors, eg automatically, without requiring manual intervention by a user. Input parameters may be predefined by scripts, command line arguments, control files, or job control language. Batch jobs may be related to the completion of preceding jobs or to the availability of particular inputs. Therefore, sequencing and scheduling of multiple jobs becomes important. Optionally, batch processing may not be performed together with interactive processing. For example, a batch management processor or batch management service may allow users to create job queues and job definitions, as well as run job definitions and review results. According to an exemplary implementation, the batch cluster includes 256 CPUs and an ETL dedicated server with 64 cores and 312 GB of RAM. The number of running instances may be one. The ETL infrastructure described above may also be applied to the process of perspective extraction. Additionally, an API is provided for data access. For example, without limitation, a REST API that conforms to a REST-style architecture and allows interaction with RESTful resources may run on the service. A service may include hardware such as, but not limited to, 1 vCPU, 2 GB RAM, 10 GB SSD disk, and a minimum of 2 running instances. APIs may be exposed to the Internet via an online application load balancer that is resilient and configures incoming end-users to online applications based in the cloud. It can route and optionally push traffic across multiple targets in multiple Availability Zones. A caching layer may be provided by a high-speed content delivery network (CDN) service, which may securely deliver the data described herein with low latency and high transfer speeds. According to an exemplary implementation, a cluster of servers or instances need not be set up, configured, or scaled on virtual machines to perform container-related operations. You can run containers without having to manage a cluster.

Exemplary Computing Environment FIG. 26 illustrates an exemplary computing environment with exemplary computing devices suitable for implementing at least one exemplary embodiment. A computing device 2605 in computing environment 2600 includes one or more processing units, cores, or processors 2610, memory 2615 (eg, RAM, ROM, and/or the like), internal storage 2620 (eg, magnetic storage, optical storage, solid state storage, and/or organic storage), and an I/O interface 2625, all of which are coupled over a communication mechanism or bus 2630 for communicating information. be able to. Processor 2610 can be general-purpose processors (CPUs) and/or special-purpose processors (eg, digital signal processors (DSPs), graphics processing units (GPUs), and others).

In some exemplary embodiments, computing environment 2600 may include one or more devices used as analog-to-digital converters, digital-to-analog converters, and/or radio frequency handlers. .

Computing device 2605 can be communicatively coupled to input/user interface 2635 and output device/interface 2640 . One or both of input/user interface 2635 and output device/interface 2640 may be wired or wireless interfaces and may be removable. Input/user interface 2635 is any device, component, sensor, or interface, physical or virtual, that can be used to provide input (e.g., keyboard, pointing/cursor control, microphone, camera, Braille, motion sensors, optical readers, and/or the like). Output devices/interfaces 2640 may include displays, monitors, printers, speakers, or the like. In some exemplary embodiments, input/user interface 2635 and output device/interface 2640 include computing device 2605 (eg, buttons or touch screen input/user interface and output or print display). 2605 or can be physically coupled to the computing device 2605 .

Computing device 2605 may use external storage 2645 and network 2650 to communicate with any number of networked components, devices, and systems, including one or more computing devices of the same or different configuration. can be communicatively coupled to the Computing device 2605 or any connected computing device may function as a server, client, thin server, general machine, special purpose machine, or another label, or such or may be referred to as such.

I/O interface 2625 may be any communication or I/O protocol or standard for communicating information across at least all connected components, devices, and networks within computing environment 2600 (e.g., Ethernet 2600). (registered trademark), 802.1 1x, Universal System Bus, WiMax, modems, cellular network protocols, and the like) and/or wireless interfaces. . Network 2650 may be any network or combination of networks (eg, the Internet, local area networks, wide area networks, telephone networks, cellular networks, satellite networks, and the like).

Computing device 2605 may use and/or communicate using computer-usable or computer-readable media, including transitory and non-transitory media. can. Transient media include transmission media (eg, metallic cables, optical fibers), signals, carrier waves, and the like. Non-transitory media include magnetic media (eg, discs, tapes), optical media (eg, CD ROM, digital video discs, Blu-ray discs), solid state media (eg, RAM, ROM, flash memory, solid state storage). , and other non-volatile storage or memory.

Computing device 2605 can be used to implement techniques, methods, applications, processes, or computer-executable instructions for implementing at least one embodiment (e.g., the described embodiments). Computer-executable instructions can be obtained from transitory media and can be stored on and retrieved from non-transitory media. Executable instructions can be in any programming, scripting, and machine language (e.g., C, C++, C#, Java, Visual Basic, Python, Perl, JavaScript, and others) can be derived from one or more of

One or more processors 2610 can run under any operating system (OS) (not shown), either in a native environment or in a virtual environment. To implement the described embodiment, a logic unit 2655, an application programming interface (API) unit 2660, an input unit 2665, an output unit 2670, a service processing unit 2690, and different units communicate with each other and the OS One or more online applications can be deployed, including an inter-unit communication mechanism 2695, for communicating with and to communicate with other applications (not shown). For example, alternative data processing unit 2675, tagging unit 2680, and modeling/prediction unit 2685 may implement one or more of the processes described above. The units and components described may vary in design, function, configuration, or implementation and are not limited to the descriptions provided.

In some exemplary embodiments, when information or executable instructions are received by API unit 2660, the information or executable instructions are passed to one or more other units (eg, logic unit 2655, input unit 2665). , output unit 2670, service processing unit 2690). For example, the input unit 2665 may connect with other data sources by using the API unit 2660 so that the service processing unit 2690 can process the information. Service processing unit 2690 performs panelist filtering, data filtering and cleaning/normalization, and results generation, as described above.

In some examples, logic unit 2660 may be configured to control information flow between units, and to implement the embodiments described above, API unit 2660, input unit 2665, output It may be configured to guide services provided by unit 2670, alternative data processing unit 2675, tagging unit 2680 and modeling/prediction unit 2685. For example, one or more processes or implementation flows may be controlled by logic unit 2655 alone or in conjunction with API unit 2665 .

Although several example implementations have been illustrated and described, these example implementations are provided to convey the subject matter described herein to those skilled in the art. there is It will be appreciated that the subject matter described herein may be implemented in various ways without being limited to the example implementations described. The subject matter described herein may be practiced without any matter specifically defined or described, or may be practiced with other different components or matter not described. . It will be appreciated by those skilled in the art that changes may be made with respect to these exemplary implementations without departing from the subject matter of this specification as defined by the appended claims and equivalents thereof. be.

Claims (20)

A method using at least one hardware processor, the hardware processor comprising:
In each of one or more times,
receiving a company identifier and one or more conditions from a user;
generate an alert for the company identifier and the one or more conditions;
associate the alert with the user;
Receive multiple data in real time from multiple data sources,
updating a plurality of signals based on the plurality of data, wherein each of the plurality of signals represents one of a plurality of key indicators for one of the plurality of companies;
for each alert associated with the user, applying the one or more conditions of the alert to at least one of the plurality of signals to determine whether the alert has been triggered;
How to have that. The at least one hardware processor further comprises, for each alert associated with the user, pushing a notification of the alert to the user upon determining that the alert has been triggered, wherein the notification is a recommended 2. The method of claim 1, comprising: 3. The method of claim 2, wherein the recommendation includes one of buying stock or selling stock. 4. The method of claim 3, wherein said notification has a single input, said single input, when selected, executes said recommendation without further input from said user. 5. The method of claim 4, wherein executing the recommendation comprises submitting an order to a brokerage service according to the recommendation. 3. The method of claim 2, wherein said notification of said alert is pushed by an email message to an email address associated with said user. 3. The method of claim 2, wherein the notification of the alert is pushed by text message to the user's mobile device. 3. The method of claim 2, wherein said notification of said alert is pushed by a voice message through a speaker of said user's device. generating the recommendation,
applying a predictive model to the at least one signal to predict the value of the at least one signal;
comparing the predicted value of the at least one signal to a benchmark value of the at least one signal to determine the strength of the at least one signal;
classifying the intensity of the at least one signal into the recommendation;
3. The method of claim 2, further comprising generating the recommendation by: Receiving the company identifier and one or more conditions at least one of the one or more times comprises receiving a selection of alerts used by a plurality of other users to generate the alert. 2. The method of claim 1, wherein doing comprises copying the alert used by the plurality of other users. 2. The method of claim 1, wherein the one or more times is a plurality of times, and wherein associating the alert with the user comprises adding the alert to the user's chain of alerts. The at least one hardware processor further comprising pushing notifications of the triggered alerts to the user for all triggered alerts in the chain of alerts, the notifications comprising:
for each triggered alert, a single input that, when selected, executes a recommended stock order for a company identified by said company identifier in said triggered alert;
a single input that, when selected, executes recommended stock orders for all companies identified by said company identifiers in all said triggered alerts in said notifications;
12. The method of claim 11, comprising: 2. The method of claim 1, further comprising, for each alert associated with the user, initiating an order to buy or sell stocks based on the alert after determining that the alert has been triggered. 14. The method of claim 13, wherein the order is automatically initiated when the alert is triggered. 15. The method of claim 14, wherein initiating the order includes generating the order based on one or more predefined user settings. 15. The method of claim 14, wherein initiating the order comprises applying a trading model to the plurality of signals and the current state of the user's account to generate the order. 14. The method of claim 13, wherein initiating the order comprises distributing the quantity of the order over multiple trades over a period of time. applying said one or more conditions of said alert to said plurality of signals;
Applying a prediction model to the plurality of signals to predict values of the plurality of signals;
applying the one or more conditions to the predicted values of the plurality of signals;
2. The method of claim 1, comprising: at least one hardware processor;
one or more software modules, which when executed by the at least one hardware processor,
In each of one or more times,
receiving a company identifier and one or more conditions from a user;
generate an alert for the company identifier and the one or more conditions;
associate the alert with the user;
Receive multiple data in real time from multiple data sources,
updating a plurality of signals based on the plurality of data, wherein each of the plurality of signals represents one of a plurality of key indicators for one of the plurality of companies;
for each alert associated with the user, applying the one or more conditions of the alert to at least one of the plurality of signals to determine whether the alert has been triggered;
one or more software modules comprising computer-executable instructions configured to
A system with A non-transitory computer-readable medium storing instructions that, when executed by a hardware processor, cause the hardware processor to:
In each of one or more times,
receiving a company identifier and one or more conditions from a user;
generate an alert for the company identifier and the one or more conditions;
associate the alert with the user;
Receive multiple data in real time from multiple data sources,
updating a plurality of signals based on the plurality of data, wherein each of the plurality of signals represents one of a plurality of key indicators for one of the plurality of companies;
for each alert associated with the user, applying the one or more conditions of the alert to at least one of the plurality of signals to determine whether the alert has been triggered;
A non-transitory computer-readable medium that allows

Images