Modeling user trends based on location data

Abstract

A method and system for modeling user tendencies based on location data are provided. The method involves training at least one model to determine user tendencies toward specific behaviors using location data from multiple users, inferring activity data from location data, and transforming the activity and location data into time-varying and static behavioral attributes for each user. The trained model is then used to predict individual users' tendencies toward specific behaviors by acquiring their location data, inputting it into the trained model, and receiving assigned quantified tendencies toward those behaviors. This method can be applied to various behaviors such as purchase intent, hospitalization risk, employment, job change, travel intent, relocation intent, and healthcare risk.

Description

[Technical Field]

[0001] [Cross-reference of related applications] This application claims priority and interest with respect to all subject matter common to both applications with respect to U.S. Provisional Application No. 63 / 523,874, which was filed on 28 June 2023 and is concurrently pending. The disclosure of the aforementioned Provisional Patent Application is incorporated herein by reference in its entirety.

[0002] This invention relates to the fields of data analysis and machine learning, which are suitable for behavioral modeling. Specifically, this invention relates to modeling user trends based on location data. [Background technology]

[0003] In recent years, the proliferation of mobile devices and the rapid development of location-based services have led to an exponential increase in the availability of location data. This data, which includes information about users' geographical location and movement, has opened up new possibilities for understanding and predicting human behavior. One area of ​​interest is modeling user trends based on location data, which can provide valuable insights into users' preferences, habits, and potential future actions or experiences.

[0004] Traditional methods for modeling user trends often relied on self-reported data such as surveys and questionnaires. However, these methods can be subject to various biases and inaccuracies because users may not accurately remember or report their past behaviors and experiences. Furthermore, self-reported data may not capture the full range of factors influencing user trends, such as environmental and contextual factors that can be derived from location data.

[0005] Machine learning techniques are increasingly being adopted to model user trends based on location data. These techniques can automatically learn patterns and relationships in the data, enabling the development of more accurate and robust models. However, the process of training machine learning models is complex and can be computationally intensive, requiring careful selection and tuning of model parameters to achieve optimal performance. Furthermore, current methods can result in models that lack flexibility, making it difficult to repurpose or adapt the models for other uses.

[0006] Furthermore, integrating time-varying and static behavioral attributes in the modeling process can be challenging. Time-varying attributes, such as the frequency and duration of visits to specific locations, can change over time and be influenced by various factors, including user schedules and preferences. Static attributes, on the other hand, represent more stable characteristics, such as user demographic information or home location. Combining these different types of attributes in a meaningful and effective way is crucial for developing accurate models of user trends. [Overview of the Initiative]

[0007] In light of these challenges, there is a need for a system that determines user tendencies based on location data, which can effectively integrate time-varying and static behavioral attributes, and can automatically learn patterns and relationships in the data by leveraging the power of machine learning techniques. The present invention enables the development of a more accurate and robust system that predicts user tendencies toward specific behaviors or experiences / states with greater precision and reliability, while maintaining the ability to adapt or repurpose models to predict different behaviors, experiences, or states in a much more efficient manner as needed.

[0008] The present invention addresses this need by providing a method for determining user tendencies based on location data, which includes the steps of: training at least one model to determine a user's tendencies toward a particular behavior or experience / state; and predicting the tendencies toward a particular behavior or experience / state for individual users. The method involves determining the tendencies toward a particular behavior or experience / state by: acquiring location data for multiple users; formatting the location data into trajectories; inferring activity data from the location data; converting the activity data and location data into time-varying and static behavioral attributes; modeling the time-varying attributes; combining the modeled time-varying attributes with static attributes; and assigning quantifiable tendencies. The method also includes adjusting model parameters based on the results, obtaining a trained model as a result, and using the trained model to predict those tendencies based on the location data of individual users.

[0009] Location data can be collected from various sources, including GPS-enabled devices, Wi-Fi® access points, and cell towers. This data can be used to generate trajectories, which are sequences of geographical locations and timestamps representing the user's movement over time. By analyzing these trajectories, it is possible to infer various types of activity data, such as the types of places visited, the duration of visits, and the frequency of visits to specific locations.

[0010] Analysis of location and activity data can reveal patterns and trends in user behavior, which can be used to model those tendencies toward specific behaviors or experiences / states. These trends can be quantified and used to predict the likelihood that users will engage in certain activities or experience specific states in the future. Such predictions can be valuable for a variety of applications, including personalized recommendations, targeted advertising, and public health interventions.

[0011] In certain embodiments, the techniques described herein relate to a method for modeling user tendencies based on location data, the method comprising: I. Training at least one model to determine a user's tendencies toward a particular behavior or experience / state, the method comprising: A) acquiring location data of multiple users; B) formatting the location data into one or more trajectories for each of the multiple users; C) inferring activity data from the location data; D) converting the activity data and location data into time-varying and static behavioral attributes for each of the multiple users; and E) determining the tendencies toward a particular behavior or experience / state, the method comprising modeling the time-varying attributes for each user; I. A training step comprising: a determination step, which includes: a step of combining a digitized time-varying attribute with a static attribute; a step of assigning a quantitative trend to a specific behavior or experience / state; and a step of adjusting parameters based on the results to produce a trained model; and II. A prediction step for an individual user, comprising: A) a step of acquiring location data for the individual user; B) a step of inputting the location data into at least one trained model generated by the training step; and C) a step of receiving from the trained model an assigned quantitative trend for the individual user.

[0012] In certain aspects, a tendency towards a particular behavior or experience / state is purchase intent; a tendency towards a particular behavior or experience / state is hospitalization risk; a tendency towards a particular behavior or experience / state is employment; a tendency towards a particular behavior or experience / state is job change; a tendency towards a particular behavior or experience / state is travel intent; a tendency towards a particular behavior or experience / state is relocation intent; a tendency towards a particular behavior or experience / state is healthcare risk; and / or a tendency towards a particular behavior or experience / state is associated with an opportunity window.

[0013] In a certain aspect, the location data is geospatial data of a user's hardware device over a period of time. In a certain aspect, the location data of the user's device is provided by a location provider / vendor. In a certain aspect, the user's location data is identified without using one or more of the following: Personally Identifying Information (PII), demographic information, and socioeconomic information about the user.

[0014] In certain embodiments, the step of inferring activity data from location data further includes: i) mapping the location data to location types; ii) grouping the location types into activity groups based on the function of the location type; and iii) using the activity groups to transform the user's one or more trajectories into the user's one or more activity trajectories. In some such embodiments, the activity groups include one or more selected from a group including hospitals, health, essential shopping, fitness, public transport, private transport, religion, entertainment, travel, personal care, leisure shopping, unhealthy activities, restaurants, home, and workplace.

[0015] In a certain embodiment, for each of the multiple users, the step of converting activity data and location data into behavioral attributes includes a step of defining time-varying attributes and a step of defining static attributes.

[0016] In some such embodiments, the step of defining time-varying attributes includes the steps of determining lifestyle attributes; activity attributes; and mobility attributes. In further embodiments, the step of determining lifestyle attributes includes the step of using an unsupervised learning model that can identify similarities in activity patterns among users. In further embodiments, the unsupervised learning model includes clustering and dimensionality reduction models, hidden Markov models, or LDA and topic models. The unsupervised learning model includes LDA and topic models, and the model may further include an Author Topic Model (ATM) where the user is the author, the activity is a word, the duration of the activity is a document, and the lifestyle is the topic.

[0017] In other such embodiments, the step of defining static attributes includes the step of determining accessibility attributes; and the step of determining socio-demographic attributes.

[0018] In other such embodiments, the step of defining time-varying attributes includes: embedding the lifestyle attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters; concatenating the embedded lifestyle attributes with the attributes of the activity and mobility; and transforming the concatenated embedded lifestyle attributes and time-varying numerical attributes into a hidden representation having shared learnable weight and bias parameters and tunable model hyperparameters. In yet another embodiment, the step of transforming the concatenated embedded lifestyle attributes and time-varying numerical attributes into a hidden representation includes using a supervised machine learning model that models both spatial and temporal information from the user trajectory. In yet another embodiment, the supervised machine learning model includes a non-deep learning regression model or a classification model. The non-deep learning regression or classification model may include a decision tree-based model, a random forest-based model, or a gradient boosting model. Alternatively, the supervised machine learning model includes a deep learning model or a deep neural network-based model. The deep neural network-based model may include at least one of the following: a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory (LSTM) model, a radial basis function network (RBFN), or a transformer-based attention model. The deep neural network-based model may also include a convoluted long short-term memory (CLSTM) model that takes the time-varying attribute as input.

[0019] In a further embodiment of this, the step of combining the modeled time-varying attribute with the static attribute includes: embedding the categorical static attribute into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters; and concatenating the embedded categorical static attribute, the numerical static attribute, and the hidden representation of the modeled time-varying attribute.

[0020] In a certain manner, the model is tuned through cross-validation.

[0021] In certain embodiments, the method further comprises a step of outputting the assigned quantifiable trend for a specific behavior or experience / state for the specific user. In some such embodiments, the outputting step includes a step of providing a graphical representation of the assigned quantifiable trend for the specific behavior or experience / state for the specific user. In other such embodiments, the outputting step includes a step of providing a web service API that provides access to the assigned quantifiable trend for the specific behavior or experience / state for the specific user.

[0022] According to certain embodiments, a system is provided for modeling user trends based on location data. The system comprises a data acquisition module configured to acquire location data of multiple users; a model training module configured to train a model for determining user trends to specific behaviors using the location data; and a prediction module configured to predict trends to specific behaviors for individual users using the trained model and the individual user's location data.

[0023] In a certain embodiment, the model training module is further configured to infer activity data from location data, convert activity data and location data into behavioral attributes, and determine tendencies for specific behaviors for each user.

[0024] In a certain embodiment, the prediction module is further configured to input the location data into the trained model and receive from the trained model an assigned quantifiable trend for a particular user's specific behavior.

[0025] The present invention offers several advantages over conventional methods. Instead of predicting user types or user groups based on group demographics, the present invention provides accurate predictions for specific users. The present invention provides a flexible model, in a far more efficient manner as needed, that can be adapted or repurposed to predict different behaviors, different experiences, or different states. That is, iterative training of the model allows the system to be tuned to provide accurate predictions for specific behaviors, experiences, or states, and then quickly and easily repurposed and retrained to predict different behaviors, experiences, or / states, while using the same location data source (in a broad sense) as input. Furthermore, the present invention can also use publicly available location data to infer user activity and therefore does not require personally identifiable information (PII) about users for predictions. [Brief explanation of the drawing]

[0026] These and other features of the present invention will be better understood by referring to the following detailed description in conjunction with the accompanying drawings.

[0027] [Figure 1] This is a diagram illustrating an exemplary environment in which the present invention is used, according to embodiments of the present invention.

[0028] [Figure 2] This is an exemplary flowchart illustrating the process for modeling user trends based on location data, according to an embodiment of the present invention.

[0029] [Figure 3]This is an exemplary flowchart illustrating the steps involved in the inference stage of activity data according to an embodiment of the present invention.

[0030] [Figure 4] This is an exemplary flowchart illustrating the steps involved in converting activity data and location data into behavioral attributes, according to an embodiment of the present invention.

[0031] [Figure 5] This is an exemplary flowchart illustrating the steps involved in determining the trend and tuning the trend according to an embodiment of the present invention.

[0032] [Figure 6] This is an exemplary dashboard that graphically displays user trends in a domain, according to an embodiment of the present invention.

[0033] [Figure 7] This is an exemplary dashboard according to an embodiment of the present invention, which graphically displays user trends in a specific domain.

[0034] [Figure 8] This is an exemplary dashboard that graphically displays a real-time trend model for a user according to an embodiment of the present invention.

[0035] [Figure 9] This is an exemplary dashboard graphically illustrating such real-time potential use according to an embodiment of the present invention.

[0036] [Figure 10] This is a schematic example of a high-level architecture for implementing a process according to an embodiment of the present invention.

[0037] [Figure 11]This is an exemplary probabilistic graphical model of an author topic model using plate notation according to an embodiment of the present invention.

[0038] [Figure 12] This is an exemplary architectural diagram of the proposed sequential deep-learning model according to an embodiment of the present invention.

[0039] [Figure 13] This is an exemplary graphical model of a CLSTM and a concatenate layer according to an embodiment of the present invention.

[0040] [Figure 14] This is an exemplary heatmap of user activity according to an embodiment of the present invention. [Modes for carrying out the invention]

[0041] An exemplary embodiment of the present invention relates to a technology that leverages large amounts of crowdsourced location data to identify lifestyles and quantify their relevance to future behaviors or experiences. Analyzing large amounts of location data from hundreds of thousands to millions of users reveals several key insights. An individual's lifestyle choices are a more significant predictor of future behaviors and experiences than other factors generally considered to have a greater influence (such as healthcare accessibility, socioeconomic factors, age, and hobbies). For example, individuals with busy and variable work routines and limited gym visits are 2.01 times more likely to be hospitalized within a year compared to the population average. This technology enables the prediction of a person's tendencies toward certain behaviors or experiences. It leverages crowdsourced location data and information representing the location history of numerous users, and then correlates the data with the various behaviors and experiences these users have. The correlations are then relied upon to predict a particular person's future behaviors or experiences based on their real-time and / or recent location history. The applications of this technology are vast. It can be applied to a variety of actions or experiences, such as purchasing a product, requesting hospitalization, experiencing a job change, participating in certain travel adventures, moving their homes, receiving certain healthcare, and other actions or experiences.

[0042] Figures 1 to 14, in which similar parts are given the same reference numerals throughout, illustrate exemplary embodiments or a number of embodiments of a method and system for modeling user trends based on location data according to the present invention. While the present invention is described with reference to exemplary embodiments or a number of embodiments illustrated in the drawings, it should be understood that many alternative forms can embody the present invention. Those skilled in the art will additionally recognize different ways of modifying parameters of one or more of the disclosed embodiments, such as the size, shape, or type of elements or materials, in a manner that still maintains the spirit and scope of the present invention.

[0043] Figure 1 depicts an exemplary environment 100 in which the present invention is utilized. Environment 100 includes a plurality of users 102, a server or cloud service 104, and a company 106. Each user 102 has hardware or a mobile device 108, such as a mobile phone, which is connected to a network or the internet 110. The server or cloud service 104 is a system having a data acquisition module 112 configured to acquire location data of a plurality of users, a model training module 114 configured to train a model to determine the user's tendencies toward specific behaviors using the location data, and a prediction module 116 configured to predict the tendencies toward specific behaviors for individual users using the trained model and the location data of individual users, and receives location data of a plurality of users 102 based on their respective hardware or mobile devices 108, which can be located using GPS or network connectivity. In some embodiments, the location data is collected by a vendor 118 that collects and sells location data and provided to the server or cloud service 104. Using the location data received by user 102, the server or cloud service 104 models the user's behavior and provides the company 106 with predictions about user 102's behavior. The company can be any type of business entity, such as a retail company or a hospital, and its business model depends on its customers. The company can then use these predictions to tailor its advertising, marketing, or the experience it delivers.

[0044] Figure 2 illustrates a flowchart 200 illustrating the various steps involved in a method for modeling user trends based on location data, as described in the claims of the patent. The flowchart consists of the following elements and their respective functions.

[0045] The model training stage (stage 202) involves training at least one model to determine user tendencies toward specific behaviors. The training process includes stages of acquiring location data from multiple users (stage 206), formatting the data into trajectories (stage 208), inferring activity data (stage 210), converting activity and location data into behavioral attributes (stage 212), and determining tendencies toward specific behaviors or experiences / states (stage 214).

[0046] In certain embodiments, user location data is raw geospatial data collected over a period of time from a user hardware device such as a mobile device 108. In some such embodiments, location data is provided by a location provider / vendor 118. In some embodiments, location data is identified without using personally identifiable information (PII), demographic information, or socioeconomic information about the user 102.

[0047] The step of formatting the location data (step 208) involves processing the collected location data into one or more trajectories for each user. In certain embodiments, the trajectories may be mathematical representations. For example, the trajectory T of individual i. i This is a set of tuples ordered in time.

number

number

[0048] The stage of inferring activity data (stage 210) involves the stages of mapping geospatial location data to location types, grouping location types into activity groups, and transforming the trajectories for each user into activity trajectories. Figure 3 depicts the stages associated with the stage of inferring activity data (stage 210). It includes the stage of mapping geospatial location data to location types (stage 300), the stage of grouping location types into activity groups based on the functions of the location types (stage 302), and the stage of using the activity groups to transform one or more trajectories of a user into one or more activity trajectories of the user (stage 30).

[0049] The stage of mapping location data to location types (stage 300) may involve mapping the collected geospatial location data to specific location types such as hospitals, restaurants, or shopping centers.

[0050] The stage of grouping location types into activity groups (stage 302) may involve grouping location types into activity groups based on the functions of the location types.Examples of activity groups include hospitals, health, essential shopping, fitness, public transportation, private transportation, religion, entertainment, travel, personal care, leisure shopping, unhealthy activities, restaurants, home, and workplace.

[0051] The stage of using activity groups to transform one or more trajectories of a user into one or more activity trajectories of the user (stage 304) may involve mathematical expressions. For example, the activity trajectory D i of individual i can be defined as a mapping T i to activities exhibiting patterns of behavior, consumption, and leisure.D i is a set of temporally ordered tuples

Number

[0052] Figure 4 illustrates the steps (stage 212) involved in converting activity data and location data into behavioral attributes for each of multiple users. This involves defining time-varying attributes (stage 400) and static attributes (stage 402).

[0053] The step of defining time-varying attributes (step 400) may further include the steps of determining lifestyle attributes (step 404), determining activity attributes (step 406), and determining mobility attributes (step 408).

[0054] Lifestyle attributes can be determined using unsupervised learning models. The models identify similarities in activity patterns among users. In certain embodiments, this may be a clustering and dimensionality reduction model, a hidden Markov model, or an LDA and topic model. In some certain embodiments, the LDA and topic model includes an author-topic model (ATM) where the user is the author, the activity is a word, the duration of the activity is a document, and the lifestyle is the topic.

[0055] Mobility attributes represent the user's mobility patterns based on their location data.

[0056] The step of defining static attributes (step 402) may further include the step of determining accessibility attributes (step 410) and the step of determining socio-demographic attributes (step 412).

[0057] Figure 5 illustrates the stages that accompany the stage (stage 214) of determining and tuning tendencies for specific behaviors or experiences / states. For each user 102, this involves the stages of modeling time-varying attributes (stage 500), combining the modeled time-varying attributes with static attributes (stage 502), assigning quantitative tendencies for specific behaviors or experiences / states (stage 504), and adjusting parameters based on the results to produce a trained model.

[0058] In certain embodiments, the step of modeling time-varying attributes (step 500) includes the steps of embedding lifestyle attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters (step 508), concatenating the embedded lifestyle attributes with activity and mobility attributes (step 510), and transforming the concatenated embedded lifestyle attributes and time-varying numerical attributes into a hidden representation having shared learnable weight and bias parameters and tunable model hyperparameters (step 512).

[0059] In some such embodiments, the step of transforming concatenated embedded time-varying categorical and time-varying numerical attributes into hidden representations (step 512) includes the step of using a supervised machine learning model that models both spatial and temporal information from the user trajectory. In certain embodiments, the supervised machine learning model includes non-deep learning regression or classification models such as decision tree-based models, random forest-based models, or gradient boosting models. In other embodiments, the supervised machine learning model includes deep learning models. In other embodiments, the supervised machine learning model includes deep neural network-based models such as one or more of convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM) models, radial basis function networks (RBFNs), and transformer-based attention models. In one particular embodiment, the deep neural network-based model includes a convolutional long short-term memory (CLSTM) model that takes time-varying attributes as input.

[0060] In certain embodiments, the step of combining modeled time-varying attributes with static attributes (step 502) includes the step of embedding the categorical static attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters (step 514), and the step of concatenating the embedded categorical static attributes, numerical static attributes, and hidden representations of the modeled time-varying attributes (step 516).

[0061] The stage of assigning quantifiable tendencies (stage 504) involves a stage of making probabilistic decisions about specific behaviors or experiences / states. These tendencies can be any number of behaviors, experiences, or states. For example, tendencies might include purchase intent, hospitalization risk, employment, job change, travel intent, relocation intent, or healthcare risk.

[0062] In certain embodiments, tendencies toward specific behaviors or experiences / states are associated with opportunity windows such as months, weeks, days, or hours.

[0063] The parameter tuning step (step 506) allows the model to be tuned so that the assigned quantification trend (step 504) reflects the desired behavior or experience / state that should be predicted. In certain embodiments, the model is tuned through cross-validation.

[0064] Returning to Figure 2, after the model has been trained (step 202), the trained model can then be used to predict trends for specific behaviors or experiences / states for individual users (step 204). This step includes obtaining location data for individual users (step 216), inputting the location data into the trained model (step 218), and receiving the quantified trends assigned from the trained model for specific behaviors or experiences / states for the users (step 220).

[0065] In certain embodiments, trends are then output (step 222). In some such embodiments, the output step involves providing a graphical representation of the assigned quantitative trends for specific behaviors or experiences / states for a particular user. In other embodiments, the output involves providing a web service API that provides access to the assigned quantitative trends for specific behaviors or experiences / states for a particular user.

[0066] Figures 6–9 illustrate various examples of the graphical outputs that may be provided. Figure 6 is an exemplary dashboard graphically showing user trends in a region. Figure 7 is an exemplary dashboard, where it is zoomed in to graphically show user trends in a more specific region. Figure 8 is an exemplary dashboard graphically showing real-time trend modeling about users. Figure 9 is an exemplary dashboard graphically showing the potential real-time use of such an embodiment of the present invention.

[0067] Any suitable and specially configured electronic or computing device may be used to implement the functions of the present invention, including the hardware or mobile device 108, server or cloud service 104 (including modules 112, 114, and 116), enterprise 106, and vendor 118 described herein. One exemplary example of such an electronic or computing device 1000 is depicted in Figure 10. The computing device 1000 is merely an example for the purpose of illustrating a suitable computing environment and does not limit the scope of the present invention in any way. The “computing device” as represented in Figure 10 may include, as understood by those skilled in the art, a “workstation,” “server,” “laptop,” “desktop,” “device,” “smart device,” “tablet,” “smartphone,” “ECR,” or other specially configured computing device. Given that the computing device 1000 is depicted for illustrative purposes, embodiments of the present invention may utilize any number of computing devices 1000 in any number of different ways to implement a single embodiment of the present invention. Accordingly, embodiments of the present invention are not limited to a single computing device 1000, nor are they limited to a single type of implementation or configuration of an exemplary computing device 1000, as would be recognized by those skilled in the art.

[0068] The computing device 1000 may include a bus or network 1010, which may be directly or indirectly connected to one or more of the following exemplary components: memory 1012, one or more processors 1014, one or more presentation components 1016, input / output ports 1018, input / output components 1020, and power supplies 1024.

[0069] Those skilled in the art will recognize that bus 1010 may include one or more buses, such as an address bus, a data bus, a network, or any combination thereof. Furthermore, those skilled in the art will recognize that, depending on the intended use and application of a particular embodiment, several of these components may be implemented by a single device. Similarly, in some cases, a single component may be implemented by multiple devices. Therefore, Figure 10 is merely an example of exemplary computing devices that may be used to implement one or more embodiments of the present invention and does not limit the invention in any way.

[0070] The computing device 1000 may include or interact with a variety of computer-readable media. For example, computer-readable media may include random access memory (RAM); read-only memory (ROM); electrically erasable programmable read-only memory (EEPROM); flash memory or other memory technologies; CD-ROMs, digital versatile disks (DVDs), solid-state drives (SSDs), clouds, or other optical or holographic media; and magnetic cassettes, magnetic tapes, magnetic disk storage, or other magnetic storage devices that may be used to encode information and may be accessed by the computing device 1000.

[0071] Memory 1012 may include computer storage media in the form of volatile and / or non-volatile memory. Memory 1012 may be removable, non-removable, or any combination thereof. Exemplary hardware devices include devices such as hard drives, solid-state memory, optical disc drives, and the like. Computing device 1000 may include one or more processors that read data from components such as memory 1012 and various I / O components 1020. One or more presentation components 1016 present data instructions to a user or other device. Exemplary presentation components include display devices, speakers, printing components, vibration components, and the like.

[0072] I / O port 1018 may allow computing device 1000 to be logically connected to other devices, such as I / O component 1020, using serial, parallel, network, and / or wireless communication protocols. Some of the I / O component 1020 may be built into computing device 1000. Examples of such I / O component 1020 include microphones, joysticks, recording devices, gamepads, satellite antennas, scanners, printers, wireless devices, networking devices, and similar devices. [Example for explanation: Purchase intent]

[0073] In the following example, the assigned quantifiable tendency is purchase intent. [Problem]

[0074] Identifying customer intent from geolocation data enables businesses to target specific areas, tailor promotions, and improve the overall shopping experience. By analyzing patterns and trends in customer behavior, businesses can proactively reach potential customers, offer relevant products or services, and ultimately increase sales and customer engagement. While many businesses rely on traditional marketing and demographic data, integrating geolocation data provides valuable insights into customers' preferences, interests, and behaviors specific to their geographical location. This information can be used to deploy location-based marketing campaigns, recommend nearby stores or services, and optimize inventory management based on demand in different areas. As the retail industry evolves, the shift to understanding customer intent from geolocation data is becoming increasingly important.

[0075] By leveraging predictive analytics and machine learning techniques, businesses can anticipate customer needs, optimize resource allocation, and deliver personalized shopping experiences tailored to individual customers and their geographical locations. In short, predicting customer intent from geolocation data is an emerging field offering numerous benefits to businesses. By effectively utilizing this data, companies can gain a competitive advantage, improve customer satisfaction, and boost sales by delivering targeted marketing campaigns and personalized shopping experiences.

[0076] This invention provides a solution to this problem. This technology enables the use of geolocation data collected from mobile phones in a privacy-protected manner and, utilizing proprietary cloud computing, predicts customer visits to QSR with greater effectiveness than conventional state-of-the-art techniques. [Proposed Solution]

[0077] Embodiments of the present invention relate to proactive computer simulations of likely company visits when user demographic data or data on a user's past behavior, such as purchase expenditure or response to previous promotions, is unavailable. The present invention is an end-to-end process for generating proxy variables of personal consumption factors to make this prediction, which is accomplished by (a) collecting geolocation data from mobile phones; (b) organizing and managing this data; (c) computationally analyzing the data on a private, cloud-based server; and (d) timely communication of the predictions to users and companies through user applications and APIs. [Framework]

[0078] The proposed framework aims to extract proxy variables for user preferences and consumption tastes from location data and simulate user selection for a company. Here, we introduce a notation and state our research topic.

[0079] Definition 1 (Trajectory): The trajectory T of individual i. i This is a set of tuples ordered in time.

number

number

[0080] Definition 2 (Activity Trajectory) Activity trajectory D of individual i i This involves mapping activities that exhibit patterns of behavior, consumption, and leisure. i It is defined as D iThis is a set of tuples ordered in time.

number

[0081] Lifestyle, one of the proxy variables, captures consumer behavior on a daily basis.

[0082] Definition 3 (Lifestyle) Individual i's lifestyle L i is, T i Activities that comprehensively represent an individual's daily temporal activities and their corresponding timestamps.

number

[0083] Next, we look at the consumer trajectory (T i ) Activity trajectory (D i We illustrate the transformation to (location and trajectory). In Section 2.2.1, we illustrate the transformation from consumer location trajectory to lifestyle (L i) will be described in detail. In Section 2.3, we will describe the identification of user consumption data that captures consumer mobility, accessibility, and socio-demographics. i , L i We will examine our methodology for simulating company visits using these and other proxy variables. [Table 1] [From location to activity log]

[0084] Mapping consumer locations to points of interest (e.g., place types, i.e., restaurants, grocery stores, or business types, i.e., Walmart, 5 Guys, etc.) opens up new avenues for understanding and simulating the social and behavioral determinants of consumer consumption from both macro and micro perspectives. For example, macro-level movement and temporal patterns across different competing brands inferred from such mapping have been used to determine the location of new franchises. Furthermore, micro-level daily consumer-specific patterns, such as the number of visits to a place type or the time spent at various business types, can predict the next most likely location for consumers.

[0085] Based on sociologists' definitions of lifestyles—that is, activities exhibiting patterns of behavior, consumption, or leisure—we design our mapping to capture consumer activity. To achieve this, we utilize publicly available resources, mapping locations to place types (see the second column in Table 1) and using the definitions of home and work. Next, we group place types with similar semantics (the first column in Table 1) to form activity groups that represent consumer behaviors in consumption (restaurants, unhealthy activities), leisure (entertainment, personal care, hotels, home, fitness), shopping (necessities shopping, leisure shopping), and commuting (public transport, private transport). These 15 activity groups encompass all activity a i j It forms the entire set of.

[0086] Furthermore, to abstract away the precise time variability in daily activities, a coarser timestamp t i j (Granular timestamp associated with each consumer location), each activity has c i j In other words, 12-2 AM, 3-5 AM, 5-7 AM, 7-9 AM, 9-11 AM, 11-2 PM, 2-5 PM, 5-7 PM, 7-9 PM, and 9-12 PM are associated. The resulting tuple d across the consumer trajectory. i j =( a i j c i j ) is all temporal activity d i j It forms the universal set (W as defined in Def. 2). Location (T i ) from activity trajectory (D i A more detailed explanation of our mapping to ) is presented below. [Proxy variables derived from location and activity trajectories] [Lifestyle as a determinant of consumption]

[0087] Lifestyle, measured as work, daily life, travel, physical activity, diet, leisure, healthcare, and consumption, may demonstrate a strong correlation with an individual's economic preferences and future consumption behavior.

[0088] Automated discovery and simulation of lifestyles from location data is not a simple problem given its scale and high dimensionality. Furthermore, daily variations in individual activities and differences with other individuals add further complexity. We employ an unsupervised topic modeling approach that has shown potential to reveal complex temporal and behavioral patterns to identify work, home, and consumption routines in smaller sets of location data. Specifically, we model consumers' daily activities by leveraging the concept of a probabilistic author topic model (ATM) designed for text documents. A probabilistic graphical model of this model using plate notation is shown in Figure 11. Leveraging our granular location data, we extend the lineage of this research by incorporating a broad set of 15 activity types that represent patterns of behavior, consumption, and leisure.

[0089] The Author-Topic Model (LDA) is a probabilistic unsupervised learning model for a collection of bags and a collection of hidden discrete variables called topics. For text modeling, we may consider each document as a mixture of various topics, where each topic is characterized as a distribution on words. ATM encompasses LDA, where the authors of a document represent a multinomial distribution on topics, where each topic is assumed to be a probability distribution on words. A document with multiple authors has a distribution on topics, which is a mixture of distributions associated with the authors. Figure 11 shows a graphical model of ATM. When generating a document, for each individual word in the document, an author is randomly selected. This author selects a topic from their multinomial distribution on topics, and then samples a word from the multinomial distribution on words associated with that topic. We repeat the process for all words in the document. Formally, the probability w of a word t Assuming K topics, A authors, D documents, and W unique words,

number

Number

Number

[0090] From Activity Trajectory to Lifestyle: To identify lifestyles, we make analogies between text documents and daily activities, authors, and consumers. We consider each activity d i in the mapped activity trajectory D i j as a word w. We represent each day's activities of a consumer (author) as a bag of words, i.e., a document d. We consider multiple days of a consumer i as unique documents of an author a. Based on these, we use Equation 1 to estimate two ATM model parameters

Number

[0091] We represent each lifestyle as a convex combination of the top Y activities ranked by their relevance (Sievert and Shirley 2014), that is, the topic-specific probability of each activity (the first term in Equation 2) and the lift (the second term in Equation 2,

Number

Number

[0092] Next, we assign the topic with the highest probability from the estimated author-topic distribution θ i k as the consumer's major lifestyle. Combining this with the top Y activities ranked by relevance, we can represent the lifestyle of consumer i as L i ={d i 1,d i 2,...,d i Y},d i j =(a i j ,c i j )∈W which can be expressed as. This is the individual's lifestyle L i from T iTo complete the identification, we detail below the selection of various hyperparameters for our ATM-based lifestyle identification. Next, we consider the proposed sequential deep learner to simulate the likelihood of future corporate visits. [Corporate Visit Simulation]

[0093] To assess whether consumers visit a company, we overlay consumers' daily location trajectories onto a publicly available location repository of the company's facilities.

[0094] To enrich our signals for simulation, we augment the identified lifestyles with other healthcare proxy variables extracted from location data. [Proxy variables from location and activity trajectories]

[0095] In Table 2, we describe the different aspects of consumer attributes that we extract from location data and the proxy variables we used to illustrate the outcomes of consumer company visits.

[0096] Lifestyle: We use the ATM-based techniques discussed to identify consumers' weekday and weekend lifestyles from their respective activity trajectories. [Table 2-1] [Table 2-2] [Table 2-3]

[0097] Activity: While lifestyle captures consumers' comprehensive daily routines, the rich behavioral nature of location data also allows us to capture our daily micro-activity behaviors. We can then analyze our transformed activity trajectories D (as defined above). i Using this, 15 activity groups a ij For each of these, the consumer's daily visit frequency and dwell time are calculated as additional time-varying numerical consumer attributes.

[0098] Mobility: Mobility indicators have been studied in the past and have been found to be correlated with health outcomes. This set of attributes is T i We capture daily consumer mobility patterns based on the locations visited, such as the frequency of visits, the time spent there, and the distance traveled. We also calculate other richer mobility features, such as entropy and turning radius. All of these are numerical consumer attributes measured at a daily level and vary over time.

[0099] Accessibility: To incorporate this aspect of consumer characteristics, we have transformed activity trajectories (D i This system utilizes consumer accessibility data to calculate the shortest distance from a consumer's home location to various public facilities such as hospitals, parks, fitness centers, pharmacies, public transportation, and workplaces. These attributes are static (time-invariant) and numerical.

[0100] Sociodemography: Based on consumer home locations from transformed activity trajectories and publicly available census data, we also calculate several block-level sociodemographics. These are static and consist of both categorical (e.g., employment POI, consumer workplace location decoration) and numerical attributes (consumer population in census blocks). [Modeling of Business Visits]

[0101] Table 2 shows that we have multiple types of consumer attributes, namely time-varying categorical (lifestyle, weekday / weekend) and numerical (work_freq, daily), static categorical (census_block_id), and numerical (commute_access). The broad nature of the attributes captures different aspects related to consumers' outcomes of their company visits, but this comes with several modeling challenges. Firstly, these unique types of attributes will need to be represented collectively to model the outcomes. Secondly, this representation will need to consider the potential for interactions between these attributes to lead to better signals for predicting health outcomes. For example, temporal correlations between different daily activity attributes, rather than day-level trends in each attribute, may lead to a better predictive model. Thirdly, given that these attributes capture multiple aspects of consumer attributes, simply concatenating all representations may lead to a suboptimal predictive model. For example, a simple way of representing all attributes as time series is to concatenate static categorical / numerical features with each timestamp, which can result in a potentially overfitted model with a very large number of parameters.

[0102] We address these challenges by separately learning representations of time-varying and static features, taking into account the interactions (temporal and static) between different types of attributes. Then, we combine these and enable interactions between the two representations to learn a final joint representation of all attributes.

[0103] To achieve this, we represent time-varying attributes using a Context-LSTM (CLSTM) cell, which is a modification of the conventional LSTM cell that has been widely used for word transformation and time series modeling. As its name suggests, CLSTM allows for the incorporation of both time-varying and static contextual features into a time series. In its original application, the time-varying contextual feature considered was the latent topic of a word, which was represented co-present with the word (each word in the time series is linked with a topic embedding) to predict the next most likely word in a sentence. Extending this to our setting, we note that lifestyles (lifestyle attributes in Table 2) are latent topics learned from different activities. Therefore, we consider these as contexts for different activity-related time-varying attributes (activities in Table 2). Furthermore, we observe that considering lifestyles as contexts for other time-varying attributes (mobility in Table 2) leads to empirically better predictive performance. Next, we concatenate these representations for a given time period with embeddings of static categorical and numerical attributes (socio-demographic and accessibility) to collaboratively learn representations of all consumer attributes that predict consumer health outcomes. Such concatenation of multiple perspectives of consumer attributes to form a unified representation has been widely studied in multimodal learning.

[0104] An overview of the proposed sequential deep learning model architecture is presented in Figure 12. The upper left box in the diagram illustrates the modeling of consumers' temporal attributes at the day level using CLSTM cells (multiple days as CLSTM layers), where lifestyle is considered the context of activity and mobility attributes. The lower left box shows the representation of static consumer attributes, which will later be concatenated with the temporal representation to predict consumer health outcomes. Next, we formally detail the transformations performed by the various layers of our proposed learner. [Proposed learning device]

[0105] X TN The time-varying numerical attribute tensor (number of users × number of days in the observation period × number of time-varying numerical features) is X TC Let this represent the time-varying categorical attribute tensor (number of users × number of weeks × number of time-varying categorical attributes), and matrix X SN and X SC The (number of users × number of categorical / numerical features) values ​​shall be expressed as static numerical values ​​and categorical consumer attributes, respectively. To simplify notation, in the following discussion, we use x TC , x TN , x SN and x SC This study focuses on the attributes of a single consumer, as expressed by the given method, and their transformation into probabilities of health outcomes (health risks).

[0106] Embedding: The embedding layer is a one-hot encoded categorical attribute (x TC , x SC ) is transformed into a fixed-dimensional continuous vector representation. Formally, it is as follows:

number

[0107] Here, W e TC In other words, the number of time-varying categorical attributes × N e TC , W e SC In other words, the number of static categorical attributes × N e SC N are learnable weight parameters. e TC , N e SC These are tunable model hyperparameters. In our configuration, x TC This is a lifestyle for weekdays and weekends (L i ) and both include the top 10 related activities (d i jPlease remember that it is represented by the total set of activities (W). Therefore, a consumer's weekday and weekend lifestyle can both be represented as a vector of length |W|, that is, we have a number of dimensions |W| × N. e TC Learn the two weight matrices of W, e TC The following procedure is followed for the transformations of other static categorical attributes (employment_poi, census_block_id).

[0108] CLSTM layer: The CLSTM layer shown in Figure 13 is composed of multiple CLSTM cells, each of which is x TN ,e TC It acts on different days. x t TN ,e t TC Assuming that this corresponds to all numerically embedded categorical time-varying consumer attributes on any given day (e t TC The calculation is performed depending on whether the day is a weekday or a weekend, because our lifestyle is derived in terms of weekdays / weekends rather than days, and each CLSTM cell undergoes the following transformation:

number

number

[0109] Concatenation: Concatenated layers do not contain any learnable parameters and are simply used to combine different intermediate representations. We perform two concatenations (pointing to Figure 2). The first one is (as seen in Figure 13) we composite input [x t TN e t TC The above discussion concerns constructing the ]. Next, the CLSTM layer ({h t Hidden temporal representation obtained from}), embedded static (e SC ) and numerical attributes (x SN It is a concatenation of ). Due to the recursive nature of Equation 5, the hidden layer representation of the last day of observation is h T However, if we focus on capturing the temporal relationship across preceding days, we can see that this is e SC , x SN Combined with [h T e SC x SN It forms [ ].

[0110] Shopping Intent: We pass the final representation to a fully connected layer, allowing interaction between temporal and static attributes, and assign a quantified likelihood of shopping intent as follows: r=σ([W T W SC W SN 1][h T e SC x SN b r ])T (6) Here, W T ,W SC ,W SN ,b r is a learnable parameter. For a given binary health outcome, various weights (W in equations 3, 5, and 6) are given. * To learn the model (denoted as ), we minimize the binary cross-entropy loss between the observed shopping outcome (e.g., shopping_visit) and r, which is a vector of shopping intentions from the above formula. The remaining hyperparameters are tuned via cross-validation. These details are discussed below. [Empirical research] [data]

[0111] For the purposes of our analysis, we combined several datasets: individual-level GPS location tracking data, census block-level demographic data from the American Community Survey (2016), and publicly available data sets from HILFD on hospitals, emergency medical services, and emergency care facilities. For location data, we partnered with leading data collectors who aggregate location data across hundreds of commonly used mobile applications ranging from news to weather, map navigation, and fitness. Location data collection was carried out through GDPR and CCPA compliant frameworks. The data covers a quarter of the U.S. population across Android® and iOS® operating systems. Each row of data corresponds to a recorded location for an individual. Each row represents: (Personal ID: An anonymized unique identifier of the individual using the mobile app) (Latitude, longitude, and timestamp of the visited location) (Speed ​​when location is captured) Includes information about this.

[0112] In summary, we obtained individual location data from Boston over a five-month period in 2019 (September to January). We consider only consumers who appeared throughout all four months and were tracked for more than 10 days in each month. Furthermore, we exclude consumers whose work and home locations could not be estimated based on the heuristics discussed in Appendix B. Our final data set consists of over 23,000 individuals. In Tables 3 and 4, and Figure 4, we detail and provide summary statistics for different types of consumer attributes calculated from location trajectories, their activity mappings, census blocks, and public healthcare facility data. We then examine them in detail. [Summary statistics]

[0113] Location Trajectory: In Table 3 (Mobility row), we present summary statistics for the raw location data

[12] . On average, we had approximately 31 locations per consumer per day over a four-month period, of which approximately 18 were unique. The average speed at which these locations were captured was 6.92 km mph. For the remaining measurements, we excluded locations captured at speeds greater than 5 km mph and considered only dwell locations, i.e., locations where consumers spent at least 5 minutes. The average dwell distance between dwell locations was 7.82 km, and the average time spent at these locations was approximately 2.16 hours. Overall, the location data summary statistics provide a detailed consumer observation across both cities.

[0114] Activity Trajectories: Tables 3 (Activity and Accessibility rows) and 4 detail summary statistics for activity trajectories (see Section 2.1 for the transformation from location to activity trajectory). From Table 3, we observe that of the 15 predefined activity groups (referring to Table 1), home, work, and public transport are the top three activity groups in terms of both average daily occurrence rate and time spent. When separated into weekdays and weekends (Table 4), as expected, we observe that work occurs less frequently (0.79) on weekends compared to weekdays (4.27). Home also occurs more frequently on weekends (5.57) compared to weekdays (8.97). To address the differences in the top-performing activities, we learn weekday and weekend lifestyles separately in our empirical analysis. On weekdays, on average, we include 14.12 daily activities per consumer, which translates to an average of 14 words per document, based on our bag-of-word representation of consumer activities for lifestyle identification (considered above). [Table 3-1] [Table 3-2] This number remains similar over the weekend. The average number of unique daily activities is 10.16 for weekends and 9.82 for weekdays, respectively.

[0115] In Figure 14, we plot a heatmap of the frequency of activities in activity trajectories between weekends and weekdays. For a given row (activity group), in Figures 14a and 14b, lighter / darker red cells indicate lower / higher frequency during the corresponding time slot. From Figure 14a, we observe that the highest frequency of workplace activity occurs between 2 and 5 PM, while home activity occurs between 12 and 3 AM; the majority of consumers work between 9 AM and 7 PM on weekdays, meaning they are less likely to stay at home. In contrast, on weekends (Figure 14b), we observe that consumers are more likely to stay at home during the same time period. We also observe that other leisure, shopping, and consumption activities occur earlier on weekdays (between 9 and 11 PM) compared to 12 and 3 AM on weekends. For example, unhealthy activities within activity groups are more likely to occur between 9 and 11 p.m. on weekdays compared to between 12 and 5 a.m. on weekends.

[0116] Block Social Demographics: For the assignment of census blocks, the hubsine distance between the latitude and longitude of individuals' home locations and the interpolated center of census block groups was calculated. Table 3 (Social Demographics) details the summary statistics for the social demographic attributes.

[0117] Where used herein, the terms “includes” and “contains” are intended to be interpreted as comprehensive, not exclusive. Where used herein, the terms “exemplary,” “example,” and “illustrative” are intended to mean “to serve as an example, case, or illustration,” and should not be interpreted as indicating or not indicating a configuration that is preferable or advantageous to other configurations. Where used herein, the terms “about,” “generally,” and “approximately” are intended to cover variations that may exist at the upper and lower limits of a subjective or objective range of values, e.g., variations in characteristics, parameters, size, and dimensions. In one non-limiting example, the terms “about,” “generally,” and “approximately” mean exactly, plus 10 percent or less, or minus 10 percent or less. In one non-limiting example, the terms “about,” “generally,” and “approximately” mean close enough to be considered included by a person skilled in the art. As used herein, the term “substantial” means the degree or extent to which an action, characteristic, feature, state, structure, item, or result is complete or nearly complete, as would be recognized by a person skilled in the art. For example, an object that is “substantial” circular would mean that the object is perfectly circular to the extent that can be mathematically determined, or as close to a circle as would be recognized or understood by a person skilled in the art. The exact degree of permissible deviation from absolute completeness may, in some cases, depend on the specific context. However, generally, closeness to completeness is such that it has the same overall result as absolute and overall completeness would have been achieved or obtained. The use of “substantial” is equally applicable when used in a negative sense to refer to a complete or nearly complete lack of an action, characteristic, feature, state, structure, item, or result as would be recognized by a person skilled in the art.

[0118] Numerous modifications and alternative embodiments of the present invention will be obvious to those skilled in the art in light of the foregoing description. Therefore, this specification should be construed as merely illustrative and aims to teach those skilled in the art the best mode for carrying out the invention. Structural details may be substantially altered without departing from the spirit of the invention, and the exclusive use of all modifications within the appended claims is reserved. Within this specification, embodiments are described in a manner that allows for a clear and concise specification, but it is intended and will be recognized that embodiments may be combined or separated in various ways without departing from the invention. The present invention is intended to be limited to the extent required by the appended claims and applicable law.

[0119] Furthermore, the following claims should be understood to encompass all general and specific features of the invention described herein, and all descriptions of the scope of the invention that may be said to fall between them as a matter of language. (Other possible items) (Item 1) A method for modeling user trends based on location data, I. A stage in which at least one model is trained to determine the user's tendencies toward a specific behavior or experience / state, A) The stage of acquiring location data for multiple users; B) For each of the multiple users, the stage of formatting the location data into one or more trajectories; C) The stage of inferring activity data from location data; D) For each of the multiple users, the step of converting activity data and location data into time-varying and static behavioral attributes; and E) The stage of determining a tendency toward a particular behavior or experience / state, For each user, The stage of modeling time-varying attributes; The stage of combining modeled time-varying attributes with static attributes; The stage of assigning a quantifiable tendency to a specific behavior or experience / state; and The stage where parameters are adjusted based on the results to produce a trained model. The decision-making stage, A training stage having; and II. The stage in which we predict trends for specific behaviors or experiences / states for individual users, A) The step of acquiring the location data of each individual user; B) The step of inputting the location data into the at least one trained model generated by the training step of the at least one model; and C) Receiving assigned quantitative trends for specific behaviors or experiences / states for each individual user from the trained model. Having, the prediction stage A method that includes [a certain feature]. (Item 2) The aforementioned tendency toward a specific behavior or experience / state is a purchase intention, as described in item 1. (Item 3) The aforementioned tendency toward a specific behavior or experience / condition is a risk of hospitalization, as described in item 1. (Item 4) The aforementioned tendency toward a specific behavior or experience / state is employment, as described in item 1. (Item 5) The aforementioned tendency toward a specific behavior or experience / state is job change, as described in item 1. (Item 6) The aforementioned tendency toward a specific behavior or experience / state is a travel intention, as described in item 1. (Item 7) The aforementioned tendency toward a specific behavior or experience / state is an intention to relocate, as described in item 1. (Item 8) The aforementioned tendency toward a specific behavior or experience / state is a healthcare risk, as described in item 1. (Item 9) The aforementioned tendency toward a specific behavior or experience / state is associated with an opportunity window, as described in item 1. (Item 10) The method according to item 1, wherein the location data is geospatial data of the user hardware device over a period of time. (Item 11) The location data of the user device is provided by the location provider / vendor as described in item 1. (Item 12) The method according to item 1, wherein the user's location data is identified without using one or more of personally identifiable information (PII), demographic information, and socioeconomic information relating to the user. (Item 13) The stage of inferring activity data from location data is, i) The step of mapping location data to location types; ii) The step of grouping location types into activity groups based on the function of the location type; and iii) Using the activity group, transform the one or more trajectories of the user into one or more activity trajectories of the user. The method described in item 1, further including the method described in item 1. (Item 14) The method described in item 13, wherein the activity group includes one or more selected from groups including hospitals, health, essential shopping, fitness, public transport, private transport, religion, entertainment, travel, personal care, leisure shopping, unhealthy activities, restaurants, home, and workplace. (Item 15) For each of the aforementioned users, the step of converting activity data and location data into behavioral attributes is: The stage of defining time-varying attributes; and The stage of defining static attributes The method described in item 1, including the method described in item 1. (Item 16) The step of defining time-varying attributes is, The stage of determining lifestyle attributes; The stage of determining activity attributes; and The stage of determining mobility attributes The method described in item 15, including the method described in item 15. (Item 17) The stage of determining lifestyle attributes is, The method described in item 16, which includes a step of using an unsupervised learning model that can identify similarities in activity patterns between users. (Item 18) The aforementioned unsupervised learning model is the method described in item 17, which includes clustering and dimensionality reduction models. (Item 19) The aforementioned unsupervised learning model includes the method described in item 17, which includes a hidden Markov model. (Item 20) The aforementioned unsupervised learning model is the method described in item 17, which includes LDA and topic models. (Item 21) The LDA and topic models are described in item 20, including the Author Topic Model (ATM), in which the user is the author, the activity is a word, the duration of the activity is a document, and the lifestyle is the topic. (Item 22) The step of defining static attributes is, The stage of determining accessibility attributes; and The stage of determining socio-demographic attributes The method described in item 15, including the method described in item 15. (Item 23) The step of defining time-varying attributes is, The step of embedding the lifestyle attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters; The step of linking the embedded lifestyle attributes and the attributes of the activities and mobility; and The step of transforming the concatenated embedded lifestyle attributes and time-varying numerical attributes into a hidden representation having shared learnable weights and bias parameters, and tunable model hyperparameters. The method described in item 16, including the method described in item 16. (Item 24) The method described in item 23, wherein the step of transforming the concatenated embedded lifestyle attributes and time-varying numerical attributes into hidden representations includes the step of using a supervised machine learning model that models both spatial and temporal information from the user trajectory. (Item 25) The supervised machine learning model described above includes a non-deep learning regression model or a classification model, as described in item 24. (Item 26) The non-deep learning regression or classification model is the method described in item 25, including a decision tree-based model, a random forest-based model, or a gradient boosting model. (Item 27) The supervised machine learning model described above includes deep learning models, as described in item 24. (Item 28) The supervised machine learning model described above includes a deep neural network-based model, as described in item 24. (Item 29) The method according to item 28, wherein the deep neural network-based model includes at least one of a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory (LSTM) model, a radial basis function network (RBFN), or a transformer-based attention model. (Item 30) The method according to item 29, wherein the deep neural network-based model includes a convolutional long short-term memory (CLSTM) model that takes the aforementioned time-varying attributes as input. (Item 31) The step of combining modeled time-varying attributes with static attributes involves embedding categorical static attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters; and The step of concatenating the embedded categorical static attribute, the numerical static attribute, and the hidden representation of the modeled time-varying attribute. The method described in item 15, including the method described in item 15. (Item 32) The model described above is tuned via cross-validation, according to the method described in item 1. (Item 33) The method according to item 1, further comprising the step of outputting the assigned quantitative trend for a specific behavior or experience / state for the specific user. (Item 34) The method according to item 33, wherein the outputting step includes providing a graphical representation of the assigned quantifiable trend for a specific behavior or experience / state for the specific user. (Item 35) The method according to item 33, wherein the output step includes a step of providing a web service API that provides access to the assigned quantification trend for a specific behavior or experience / state for the specific user. (Item 36) A system for modeling user trends based on location data, A data collection module configured to acquire location data for multiple users; A model training module configured to train a model for determining user tendencies toward specific behaviors using the aforementioned location data; and A predictive module configured to predict trends for specific behaviors for each individual user, using the trained model and the location data of each individual user. A system equipped with these features. (Item 37) The system described in item 36, wherein the model training module is further configured to infer activity data from location data, convert the activity data and location data into behavioral attributes, and determine tendencies for specific behaviors for each user. (Item 38) The system according to item 36, wherein the prediction module is further configured to input the location data into the trained model and receive from the trained model an assigned quantitative trend for a particular user's specific behavior.

Claims

1. A method for modeling user trends based on location data, I. A stage in which at least one model is trained to determine the user's tendencies toward a specific behavior or experience / state, A) The stage of acquiring location data for multiple users; B) For each of the multiple users, the stage of formatting the location data into one or more trajectories; C) The stage of inferring activity data from location data; D) For each of the multiple users, the step of converting activity data and location data into time-varying and static behavioral attributes; and E) The stage of determining a tendency toward a particular behavior or experience / state, For each user, The stage of modeling time-varying attributes; The stage of combining modeled time-varying attributes with static attributes; The stage of assigning a quantifiable tendency to a specific behavior or experience / state; and The stage where parameters are adjusted based on the results to produce a trained model. The decision-making stage, A training stage having; and II. The stage of predicting trends for specific behaviors or experiences / states for individual users, A) The step of acquiring the location data of each individual user; B) The step of inputting the location data into the at least one trained model generated by the step of training at least one model; and C) Receiving the assigned quantifiable trends for specific behaviors or experiences / states for each individual user from the trained model. Having, the prediction stage Equipped with, Here, for each of the multiple users, converting activity data and location data into behavioral attributes involves a step of defining time-varying attributes and a step of defining static attributes; and Here, the step of combining the modeled time-varying attributes with static attributes is: The step of embedding categorical static attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters; and The step of concatenating the hidden representations of the embedded categorical static attributes, numerical static attributes, and modeled time-varying attributes. Methods that include...

2. The method according to claim 1, wherein the aforementioned tendency toward a specific behavior or experience / state is a purchase intention.

3. The method according to claim 1, wherein the aforementioned tendency toward a specific behavior or experience / state is a risk of hospitalization.

4. The method according to claim 1, wherein the aforementioned tendency toward a specific behavior or experience / state is employment.

5. The method according to claim 1, wherein the aforementioned tendency toward a specific behavior or experience / state is changing jobs.

6. The method according to claim 1, wherein the aforementioned tendency toward a particular behavior or experience / state is a travel intention.

7. The method according to claim 1, wherein the aforementioned tendency toward a specific behavior or experience / state is an intention to move.

8. The method according to claim 1, wherein the aforementioned tendency toward a specific behavior or experience / state is a healthcare risk.

9. The method according to claim 1, wherein the tendency toward a particular behavior or experience / state is associated with an opportunity window.

10. The method according to claim 1, wherein the location data is geospatial data of a user hardware device over a period of time.

11. The method according to claim 1, wherein the location data of the user device is provided by a location provider / vendor.

12. The method according to claim 1, wherein the user's location data is identified without using one or more of personally identifiable information (PII), demographic information, and socioeconomic information relating to the user.

13. The stage of inferring activity data from location data is, i) The step of mapping location data to location type; ii) The step of grouping location types into activity groups based on the function of the location type; and iii) Using the activity group, transform the one or more trajectories of the user into one or more activity trajectories of the user. The method according to claim 1, further comprising:

14. The method according to claim 13, wherein the activity group includes one or more selected from the group including hospitals, health, essential shopping, fitness, public transport, private transport, religion, entertainment, travel, personal care, leisure shopping, unhealthy activities, restaurants, home, and workplace.

15. The step of defining time-varying attributes is, Lifestyle attribute determination stage; The stage of determining activity attributes; and The stage of determining mobility attributes The method according to claim 14, including the method described in claim 14.

16. The stage of determining lifestyle attributes is, The method according to claim 15, further comprising the step of using an unsupervised learning model capable of identifying similarities in activity patterns between users.

17. The method according to claim 16, wherein the unsupervised learning model includes a clustering and dimensionality reduction model.

18. The method according to claim 16, wherein the unsupervised learning model includes a hidden Markov model.

19. The method according to claim 16, wherein the unsupervised learning model includes an LDA and a topic model.

20. The method according to claim 19, wherein the LDA and topic model include an author topic model (ATM) in which the user is the author, the activity is a word, the duration of the activity is a document, and the lifestyle is the topic.

21. The step of defining static attributes is, The stage of determining accessibility attributes; and The stage of determining socio-demographic attributes The method according to claim 14, including the method described in claim 14.

22. The step of defining time-varying attributes is, The step of embedding the lifestyle attributes into a fixed-dimensional continuous vector representation having learnable weight parameters and tunable model hyperparameters; The step of linking the embedded lifestyle attribute, the activity attribute, and the mobility attribute; and The step of transforming the concatenated embedded lifestyle attributes and time-varying numerical attributes into a hidden representation having shared learnable weights and bias parameters, and tunable model hyperparameters. The method according to claim 15, including the method described in claim 15.

23. The method according to claim 22, wherein the step of transforming the linked embedded lifestyle attributes and time-varying numerical attributes into hidden representations includes the step of using a supervised machine learning model that models both spatial and temporal information from the user trajectory.

24. The method according to claim 23, wherein the supervised machine learning model includes a non-deep learning regression model or a classification model.

25. The method according to claim 24, wherein the non-deep learning regression model or classification model includes a decision tree-based model, a random forest-based model, or a gradient boosting model.

26. The method according to claim 23, wherein the supervised machine learning model includes a deep learning model.

27. The method according to claim 23, wherein the supervised machine learning model includes a deep neural network-based model.

28. The method according to claim 27, wherein the deep neural network-based model includes at least one of a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory (LSTM) model, a radial basis function network (RBFN), and a transformer-based attention model.

29. The method according to claim 28, wherein the deep neural network-based model includes a convolutional long short-term memory (CLSTM) model that takes the time-varying attribute as input.

30. The method according to claim 1, wherein the model is tuned through cross-validation.

31. The method according to claim 1, further comprising the step of outputting the assigned quantitative trend for a specific behavior or experience / state for the specific user.

32. The method according to claim 31, wherein the outputting step includes providing a graphical representation of the assigned quantitative trend for a specific behavior or experience / state for the specific user.

33. The method according to claim 31, wherein the output step comprises providing a web service API that provides access to the assigned quantification trend for a specific behavior or experience / state for the specific user.

34. A system for modeling user trends based on location data, A data collection module configured to acquire location data for multiple users; A model training module configured to train a model for determining user tendencies toward specific behaviors using the aforementioned location data; and A predictive module configured to predict trends for specific behaviors for each individual user, using a trained model and the location data of the individual user. A system that includes these features.

35. The system according to claim 34, wherein the model training module is further configured to infer activity data from location data, convert the activity data and location data into behavioral attributes, and determine tendencies for specific behaviors for each user.

36. The system according to claim 34, wherein the prediction module is further configured to input the location data into the trained model and to receive from the trained model an assigned quantifiable trend for a particular user to a particular behavior.

Images