WO2020178829A1 - Système et procédé de génération de réponses d'interaction pour participants de programme de bien-être - Google Patents

Système et procédé de génération de réponses d'interaction pour participants de programme de bien-être Download PDF

Info

Publication number
WO2020178829A1
WO2020178829A1 PCT/IL2020/050252 IL2020050252W WO2020178829A1 WO 2020178829 A1 WO2020178829 A1 WO 2020178829A1 IL 2020050252 W IL2020050252 W IL 2020050252W WO 2020178829 A1 WO2020178829 A1 WO 2020178829A1
Authority
WO
WIPO (PCT)
Prior art keywords
participant
interaction
profile
data points
given
Prior art date
Application number
PCT/IL2020/050252
Other languages
English (en)
Inventor
Lilia Leye SHWARTSMAN
Michael SHPARBER
Yael PORTNOY
Original Assignee
Wis2Biz Ltd.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wis2Biz Ltd. filed Critical Wis2Biz Ltd.
Publication of WO2020178829A1 publication Critical patent/WO2020178829A1/fr

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/70—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to mental therapies, e.g. psychological therapy or autogenous training
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks

Definitions

  • the present invention relates to the fields of customer management and risk analysis.
  • Wellness programs include fitness, exercise and weight loss programs, including programs designed to improve participants' health and/or physical appearance, or to deal with health issues such as obesity, diabetes, and hypertension.
  • Types of wellness programs may include gym programs, fitness club programs, or fitness courses such as cross-fit, yoga, and Pilates, Certain wellness programs may be designed to assist participants in overcoming medical or emotional problems, including problems of addiction and stress, or blood sugar or blood pressure levels that are too high or too low.
  • Embodiments of the present invention provide systems and methods for generating interaction responses for participants of a wellness program, including performing steps of: training a machine learning (ML) classifier, wherein the training comprises: receiving a first set of participant profiles, each participant profile including multiple first set data points; applying an attrition label to each of the first set of participant profiles indicating subsequent attrition of a respective participant from the wellness program; training the ML classifier to predict a profile attrition risk;
  • ML machine learning
  • receiving interaction examples for discouraging participant attrition wherein the interaction examples include message examples intended for participants and counselor recommendation examples intended for counselors; labeling each interaction example with one or more attributes according to respective participant profile data points and an intended participant attrition risk level; receiving a second set of participant profiles, each including multiple second set data points and each associated with a given participant; applying the ML classifier to predict an attrition risk for each of the second set of participant profiles; for each of the second set of participant profiles, responsively to one or more of the predicted attrition risk of the given participant profile, the second set data points of the given participant profile, and a timed interval between interaction responses issued to the associated participant of the given participant profile, triggering a response trigger; and, responsively to the response trigger for the given participant profile, generating an interaction response by selecting an interaction example whose labeled attributes and intended participant attrition risk level are correlated with the second set data points and the attrition risk of the given participant profile, and customizing the selected interaction example according to the second set data points of the given participant
  • the first set data points of the first set of participant profiles may also include a response history of a given interaction response
  • training the ML classifier may include training according to the first set data points including the response history
  • labeling the interaction responses may include determining, according to the ML classifier, an effect of the given interaction response on the predicted profile attrition risk.
  • Further embodiments may include clustering trend data of multiple participant profiles, wherein clustering comprises creating a trend graph of a change over time of an attribute of a participant, clustering trend graphs of multiple participants into multiple trend graph clusters, and designating a trend graph cluster as a data point of a participant profile.
  • Clustering the trend data may also include normalizing, time-based trend data of participant profiles and clustering the normalizing, time -based trend data.
  • Further embodiments may include receiving at least one of: an intervention availability parameter indicating a ratio of counselor intervention recommendations versus automated messages to be generated over a given period of time, a number of clients that a consultant has, and a backlog of intervention recommendations and accordingly generating the given ratio of counselor intervention recommendations versus automated messages.
  • Further embodiments may include: dividing each message example into message fragments, including openings, bodies, and closings; labeling each fragment according to the labels of the message example; and wherein generating the interaction response comprises selecting appropriately labeled, opening, body, and closing message fragments to construct a full automated message.
  • the message fragments may include variables for substituting personalized data on a participant, wherein the personalized data includes at least one of a name, a weight, and a trend.
  • the data points of the first and second set of participant profiles may include Neuro Linguistic Programming (NLP) attributes according to results of an NLP Meta Survey, and the clustered participant profiles and the interaction response labels may include the NLP attributes.
  • NLP Neuro Linguistic Programming
  • the data points of the first and second set of participant profiles may include at least one parameter indicative of a participant gender, profession, pregnancy status, weight, BMI, demographic, medications, dietary restrictions and preferences, hobbies, sleep habits, wellness program subscription type, weight at drop-out from prior program, and number of prior program drop-outs.
  • the data points of the first and second set of participant profiles may include at least one parameter indicative of a long-term trend, wherein the at least one parameter is one of a percent improvement in physical endurance, a percent distance to a target BMI, and a percent change in weight from an initial program weight.
  • the data points of the first and second set of participant profiles may include at least one parameter indicative of a participant's attendance or absence from a session of the wellness program.
  • the data points of the first and second set of participant profiles may include at least one parameter indicative of an upcoming or recent special event, wherein the special event is one of upcoming or past parties or personal or family events, including weddings or health issues, such as an illness, or exams, holidays, or trips.
  • the data points of the first and second set of participant profiles may include at least one parameter indicative of a mood or emotional change of the participant, including at least one of satisfaction, dissatisfaction, hunger, or thirst.
  • the ML classifier may be one of a random-forest, gradient boosting, logistic regression, naive Bayes classifier, or neural network ML model.
  • the data points of the first and second set of participant profiles may include data generated by participants during a program session, an appointment, a training session, an exercise performed on fitness equipment, a weight measurement, a survey, a phone call, or an email, or from a participant phone app, social media posting, smart watch or other computing or IoT device, or from data collected by a social media provider tracking participant interaction with a website.
  • Generating the automated message may also include receiving from a participant a preferred messaging method and transmitting the automated message by the preferred messaging method.
  • Generating the automated message may also include generating by the ML classifier a preferred messaging method and transmitting the automated message by the preferred messaging method.
  • the preferred messaging method may be one of an email, an SMS, or a mobile phone messaging application.
  • the counselor intervention recommendation may be a recommendation to a counselor or other service professional of the wellness program to perform one or more of: personally contacting a participant by phone, arranging a meeting with a participant, suggesting a change in a program activity to a participant, or addressing a potential problem of the participant.
  • the counselor intervention recommendation may also be a recommendation to send a message.
  • Generating the counselor intervention recommendation may include selecting and customizing a message example, whose labeled attributes and intended attrition risk level are correlated with the second set data points and the with the attrition risk of the given participant profile, and including the selected and customized message example with the counselor intervention recommendation.
  • the counselor intervention recommendation may be a notification to an administrator of the wellness program regarding a potential problem regarding a counselor or service professional of the wellness program.
  • the first and second set data points may be physical measurements including at least one of weight, muscle percentage, fat percentage, and waist, bust, or leg circumference.
  • triggering a response trigger may include determining to trigger or not to trigger generation of the automated message or the counselor intervention recommendation.
  • the second set data points may include trend data of a given participant profile
  • receiving the second set of participant profiles may include clustering the trend data of the given participant profile
  • triggering the generation of the interaction response according to the second set data points may include triggering the interaction response responsively to a participant profile being assigned to a new trend cluster group.
  • the second set data points include trend data of a given participant profile
  • receiving the second set of participant profiles includes clustering the trend data of the given participant profile
  • triggering the generation of the interaction response according to the second set data points includes triggering the interaction response responsively to a participant profile being assigned to a the same trend cluster group for more than two consecutive time periods.
  • FIG. 1 is a schematic diagram of modules of a system for generating interaction responses of a wellness program, in accordance with an embodiment of the present invention
  • Figs. 2A-C are tables indicating clustering methods of the system with respect to participant data trends, in accordance with an embodiment of the present invention
  • FIG. 3 is a flow diagram of a process for generating interaction responses of a wellness program, in accordance with an embodiment of the present invention.
  • FIGs. 4A and 4B are flow diagrams of processes for preparing the system for operation, in accordance with an embodiment of the present invention.
  • Embodiments of the present invention provide a system and methods for automatically triggering and generating interaction responses for a wellness program (also referred to herein simply as the "program").
  • interaction responses includes 1) automated messages that may be generated by the system and sent to participants in the wellness program (i.e., messages providing interaction between the system and the participants), and 2) automated recommendations that may be generated by the system and sent to counselors whose role is to monitor and to interact with the participants.
  • Both the automated messages and the automated recommendations are intended ultimately to encourage participants to persevere in the program rather than dropping out, i.e., to encourage participant retention.
  • a machine learning classifier improves the targeting of messages and counselor intervention to reduce participant attrition.
  • the term "participant” refers to people who have joined a wellness program to achieve their specific wellness goals. Depending on the type of program, participants may also be referred to by other terms, such as members, customers, clients, employees, or patients.
  • the term “counselor” herein refers to wellness program staff, including administrators and staff who are involved in monitoring and interacting with participants in the program, and/or in providing other resources to the program, in order to assist participants in reaching their wellness goals.
  • counselors may include, for example, group leaders, coaches, instructors, guides, or trainers, or to any professional in the appropriate field of the wellness program, such as doctors, psychologists, nutritionists, dieticians, or physical therapists.
  • wellness programs typically have more participants than counselors, such that counselor one-to-one time with participants can be viewed as a limited resource.
  • wellness goals refers to the goals of the specific type of wellness program, such as weight loss, weight gain, improved endurance, reduced stress, reduced blood sugar level or blood pressure, improved self-confidence, or addiction cessation.
  • Fig. 1 is a schematic diagram of modules of a system 100 for generating interaction responses of a wellness program, in accordance with an embodiment of the present invention.
  • System 100 typically includes a set of computer modules that perform respective tasks indicated in the blocks of the schematic diagram of Fig. 1.
  • the computer modules may be configured to operate on any appropriate network-connected computing device, such as a web-accessible server or within a cloud-based computing environment.
  • Participants 102 and counselors 104 communicate with the system through respective computer user interfaces, participants 102 communicating through participant user interface 106 and counselors communicating through counselor user interface 108.
  • Such computer interfaces may include client computer applications running on computing devices, such as personal computers or mobile devices.
  • the client computer applications may be web-based applications, and may include screens for entering data as described hereinbelow and for receiving computer-generated results of the system also described hereinbelow.
  • the participant user interface 106 may also include internet-of-things (IoT) devices, such as biometric smart watches, fitness club equipment, or other types of fitness trackers, which may acquire physical data of a participant, such as heart rates during exercise.
  • the participant user interface 106 may also include intemet-of- medical-things (IoMTs), such as blood pressure or blood sugar devices, or mobile device apps.
  • IoMTs intemet-of- medical-things
  • System 100 typically includes a profile repository 110, which receives and stores data about participants. This includes data received from the participants through their participant user interfaces 106.
  • participant user interfaces 106 are configured to include diary-like forms, into which a participant may enter data on a regular basis, such as daily physical measurements or exercise accomplishments, or even several times a day, for example when tracking meals.
  • a participant may also enter significant events into the diary, such as upcoming or past parties or personal or family events, including weddings or health issues, such as an illness, or exams, holidays, or trips (also referred to hereinbelow as "life events.")
  • a participant may also enter his personal moods or mood changes, such as sense of depression or satisfaction, or physiological feelings, such as hunger or thirst, such attributes also typically quantified, for example on scales of 1 to 10).
  • Additional data that may be stored in the profile repository 110 may include results of surveys of psychological or behavioral surveys, such as results of NLP surveys 112.
  • Data may also be acquired through session interfaces 114, i.e., by interfaces operating during sessions (e.g., group meetings, private appointments, or personal or group workouts) related to the wellness program.
  • the session interfaces 114 may acquire data on attendance or absence of the participants at sessions (for example, by time clocks or electric turnstiles at a fitness center) as well as other relevant session data (such as fitness machine data).
  • Additional data may be entered during personal or clinic-based medical checks, including measurements such as blood pressure, blood sugar level, or other blood tests.
  • an additional source of participant data may be external data providers 116, such as social media sites that may provide data on participants' interests and habits (based, for example, on participants' postings or website interactions).
  • External systems providing data may also include a corporation's internal customer-related systems, such as a customer relationship management (CRM) system.
  • CRM customer relationship management
  • Wellness programs may also occasionally call or email participants to give impromptu surveys of the participants' views on aspects of the program, which are also a form of participant data.
  • participant profile repository 110 stores data acquired regarding an individual participant as attributes, i.e., data points, of a participant profile.
  • system 100 correlates participant profiles 1) with participant attrition risk, 2) with appropriate interaction examples, and 3) with triggers that determine when appropriate interaction responses are generated.
  • a table of the appendix hereinbelow includes a listing of some of the different categories of attributes that may be stored with a participant profile. These categories may include, for example: personal details, program details, NLP results, participant diary entries, meeting data, counselor input, and trend data.
  • Personal details may be acquired from a participant at the time of the participant's registration with a program. Examples of such details are date of birth, gender, height, and address. Some attributes may be extracted or calculated from the entered data: for example, a participant's age may be calculated according to the date of birth, a city attribute may be extracted from the address data, a body mass index (BMI) may be calculated from weight and height data. Attributes may be stored in the profile repository 110 as key-value pairs, and some attributes may have multiple entries. For example, for the attribute "current medical issues," two attribute entries may be stored for a participant who has two issues (such as having both a pace-maker and dysphagia).
  • attributes may be binary, numeric, or limited to a set. For example, an attribute "marital status”, may be assigned values from the set of terms, “yes,” “no,” “single,” “married,” “divorced,” or “widowed.” Free text is not typically used, given that the attributes are used as data points of the machine learning classifier described below.
  • other personal detail attributes may include additional demographic information, such as number of children, employment status, education, profession, hobbies, sleep habits, types of dietary restrictions, types of medications taken, as well as details of prior program participation, such as number of prior enrollments, number of previous drop-outs, weights at previous drop-outs, and average time in previous programs.
  • a participant's profile may also include program details related to an individual participant, such as the participant's assignment to a group ID and a counselor ID or name.
  • Programs may also have multiple subscription plan types, such as a plan for weight loss and a separate plan for cardio improvement.
  • a plan type attribute may therefore also be included in the participant profile.
  • a concrete goal of the participant with respect to the program type is also an attribute of the profile, such as a 10% weight loss.
  • participant's may also complete a psychological or behavioral survey, such as a Neuro-Linguistic Programming (NLP) Macro survey, which may provides additional profile attributes, such as a participant's behavioral pattern indicators.
  • NLP Neuro-Linguistic Programming
  • data entered into the profile repository 110 are time -based data, such as the data described above as entered through the participant's "diary," or at sessions.
  • data may include, for example, measurements taken at regular intervals (e.g., daily or weekly) including physical parameters (e.g., weight, muscle percentage, fat percentage, and waist, bust, or leg circumference).
  • Physical activity parameters may be entered for or by participants in programs that include such activity, such as time spent on exercise apparatus, such as a treadmill, or time spent swimming or running.
  • Such data may also be converted by the profile repository 110 into normalized trend data, such as percent progress towards a weight target or BMI from a starting weight, or a percent "endurance improvement" for improved times or durations on exercise apparatus.
  • trend data may be clustered into groups of similar trends, such that an attribute of a given participant for a given trend, such as a trend of weight measurements, may be indicated by an identifier representative of a particular cluster group.
  • Additional data for attributes may be entered by a participant's counselor (i.e., by any professional program staff interacting with the participant). Such data may include both physical measurements and judgmental "impressions," such as a counselor's view of a participant's level of satisfaction or participant's likelihood to drop out (attributes that may be indicated, for example, on scales of 1 to 10).
  • attributes may be added to the system over time. (Conversely, if over time an attribute is shown to not affect the ML classifier, it may be dropped from the participant profiles.)
  • the repository may also store with each participant profile a history of counselor interactions with the participant, as well as a history of automated messages that the participant receives over time, these interactions indicated in Fig. 1 as a response history 118.
  • automated messages refer to messages generated by the system, according to message examples stored in an interaction repository 130.
  • the system may also issue counselor recommendations to counselors, which recommend interactions counselor should initiate with participants (such counselor interactions also referred to hereinbelow as "counselor interventions").
  • Counselor recommendations issued by the system are based on counselor recommendation examples, which may also be stored in the interaction repository 130.
  • both message examples and counselor recommendation examples stored in the interaction repository 130 are referred to as "interaction examples" of the system.
  • a message example stored by the system typically includes various variables that may be replaced when the message example is issued as an automated message to a given participant.
  • a message example might be stored as: "[Participant name] you have [event type] [event time] ! Drink plenty of water and keep up the protein intake. For desert, pick one favorite— don't compromise! You'll have a great time! [Counselor name].”
  • An automated message issued by the system based on the given message example, might read: "John, you have a party tonight! Drink plenty of water and keep up the protein intake. For desert, pick one favorite— don't compromise! You'll have a great time!
  • automated messages are typically customized according to profile attributes, such as an upcoming event type and a participant name.
  • profile attributes such as an upcoming event type and a participant name.
  • a label assigned to a message example determines the attributes of a participant who will get the message (in this case, a participant for whom the message, "Drink plenty of water... " is appropriate.
  • automated messages are typically issued in the name of a participant's assigned counselor (e.g., "Susan").
  • Message examples input into the system may also be divided into message fragments, including opening, body, and closing fragments, such that a wider variety of messages may then be reconstructed before the messages are issued.
  • Types of counselor recommendation examples might be, for example, a recommendation that a counselor personally contact a participant by phone, or arrange a meeting with a participant, or change in a program activity of a participant, or otherwise address a potential problem of the participant.
  • An additional type of counselor recommendation example might be a notification to an administrator of the wellness program regarding a potential problem regarding a counselor or service professional of the wellness program, such as a lack of fit between a participant, as indicated by the participant profile, and a personality of the counselor or service professional.
  • Both types of interaction examples of the system are designed to discourage participant attrition.
  • the interaction examples are created by counselors, who typically have specific, target participants in mind.
  • the interaction examples are stored in the interaction repository 130, with labels that are assigned according to one or more attributes of participant profiles of appropriate target participants.
  • the message example described above (regarding "Drinking water”) might be labeled with attributes corresponding to a participant profile with the following attributes: a high protein diet program type, a trend indicating that a participant is making progress towards a participant's weight goal, and an up-coming event that includes a meal.
  • the interaction examples may also be labeled according to attrition risk levels that are determined for participant profiles. (The attrition risk levels used for labeling interaction examples are typically designated as ranges of attrition risks, such as "greater than 50% " or "between 10% and 20%.")
  • each fragment may be labeled according to attributes of the intended participant profile and the intended level of attrition risk of the full message. Fragments, as well as entire message examples, may also be labeled for multiple alternative attributes. For example, a given message example may be labeled with attributes that correlate the given message example both to a participant profile having a program type attribute of a weight loss program and to a participant profile having a program type attribute of a blood pressure reduction program.
  • Labeling of an interaction example may also subsequently be modified according to the success or failure of an interaction response— issued based on the given interaction example— in reducing attrition risk. Such a determination may be made, for example, by determining, after each interaction response is issued, a revised attrition risk, and a the contribution of the given interaction response to changing the attrition risk, as described below.
  • a machine learning (ML) classifier 140 is trained as a predictor of attrition, that is, it is trained to generate a measure of risk of a participant leaving a program, by correlating attrition of participants to the attributes (i.e., data points) of participant profiles. Attrition of participants may be tracked over time and stored as attrition records 132, each record being associated with a respective participant profile of a corresponding participant.
  • a variety of ML classifier models may be applied, such as models based on methods of random-forest, gradient boosting, logistic regression, naive Bayes classifier, or neural networks.
  • profile repository 110 may store multiple "instances" of participant's profiles over time, such that training may also be performed by training the ML classifier with multiple, time -based "snapshots" of participant profiles, generated prior to the attrition of the respective participant. Training is performed by labeling (or "tagging") each participant profile used for training according to the attrition record of the respective participant.
  • the attrition label of a participant profile may be, for example, a binary flag, indicating whether the participant has or has not subsequently dropped out.
  • the label may be a value indicating the amount of time elapsed between the generation of a given participant profile (the "snapshot date") and the actual determination of drop-out. This labelling facilitates prediction of an expected time interval leading to attrition, for example, a prediction of a 50% likelihood of attrition in 3 weeks.
  • participant profiles for training may be group profiles, generated, for example, by average attribute values of multiple participants.
  • the ML classifier may be trained to predict an expected attrition rate of a group, such as a specific of wellness class, based on attributes of a class, such as meeting time or program type.
  • training of the ML classifier may be an iterative process, being repeated as more participant data is collected, in order to improve prediction reliability.
  • the ML classifier may determine patterns of profile attributes that predict future attrition even when such patterns might be counter-intuitive to human analysis.
  • the ML classifier training is described further hereinbelow with respect to Fig. 4B.
  • participant profiles based on newly acquired data may be analyzed by the ML classifier to determine attrition risks (or risk levels) of the new participant profiles. Analysis may be performed on a frequent basis, such as daily.
  • the ML classify may be deployed to run on an ML platform, such as Azure ML, AWS SageMaker, or Databricks Spark.
  • ML interpretation methods such as LIME or SHAP may be applied to the ML classifier model on a frequent basis to determine which attributes (i.e., "data points") of a profile are more significant to the determination of attrition risk.
  • interpretation methods may be applied to determine the positive or negative contribution of specific interaction examples. That is, such interpretation methods may determine whether a given automated messages delivered to participants (based on a selected message example), or a given recommendation implemented by counselors (based on a selected recommendation example), proved to be helpful in reducing the attrition risk of profiles that generated such responses.
  • a response trigger module 142 is applied to determine whether an interaction response should be triggered, and if so, which interaction response. Whether or not to trigger an interaction response may be controlled by a set of rules, which are typically based on factors such as attrition risk level, key profile attributes, and a timed interval between interaction responses issued to a given participant. For example, a significant change in the attrition risk (for example, a change in expected drop-out from 30% to 60%, or a change from 50% likelihood of drop-out within a month to 50% likelihood of drop-out in one week) may trigger a response, but may be affected by a time interval between responses, for example, a rule to not issue more than one response per day.
  • a significant change in the attrition risk for example, a change in expected drop-out from 30% to 60%, or a change from 50% likelihood of drop-out within a month to 50% likelihood of drop-out in one week
  • triggering rules may also be analyzed by machine learning interpretation of contributing factors. That is, trigger of a certain response to a given participant profile may also be analyzed by methods such as SHAR or LIME to determine the extent to which the given trigger increased or decreased attrition risk levels.
  • the triggering module 142 selects an appropriate interaction example from the interaction repository 130, i.e., the module selects a message example or a counselor recommendation example that is labeled with attributes corresponding to the profile attributes, and which is additionally labeled with an attrition risk level matching the attrition risk determined for the profile by the ML classifier. Determining whether to issue an automated message or a counselor recommendation may be straightforward, if, for example, messages examples are generally labeled for low attrition risk levels and counselor recommendation examples are labeled for higher attrition risk levels.
  • the triggering module 142 may be configured to issue an automated message (corresponding to one of the message examples), rather than a counselor recommendation, as automated messages are essentially "free,” as opposed to counselor interventions, which require the time of the program's counselors.
  • Additional rules may be set with respect to "life events" of a participant, such as a timing of issuing an interaction response related to a given life event, that is whether before, during, or after the event, or some combination of such options.
  • Additional rules may also determine whether an automated message or a counselor recommendation is issued. In some cases, both may be issued, as a rule may be set indicating that, given a certain attrition risk level and certain profile attributes, both an urgent automated message is required, as well as a counselor follow-up.
  • some automated messages may be "open" messages, suggesting that a participant perform some type of action, which in some cases may include initiating contact with a counselor.
  • a counselor recommendation example may also include a recommendation that a counselor first review and then send an automated message, meaning that the counselor recommendation example may include a system instruction to generate an automated message (based on an appropriate message example, which may be stored with the recommendation example), which is then sent to a counselor instead of being sent directly to a participant.
  • a counselor who receives such a "recommended” message would be expected to review and perhaps change the message before sending it on to the participant.
  • Triggering rules may also prioritize profiles that trigger counselor recommendations that are issued within a given period of time.
  • Recommendations that are triggered may then be prioritized, given that counselor time is limited, such that a counselor may not be able to handle all recommended interventions if no limits were imposed. Prioritization may be based either on preset rules, such as those described above, or by machine learning analysis of which interventions, given the associated participant profile and attrition risk, are expected to contribute the most to reducing the attrition risk.
  • the triggering module 142 may also use an intervention availability ratio for a given counselor, which may specify ratio of counselor intervention recommendations versus automated messages to be generated over a given period of time. The triggering module 142 may then, accordingly, generate the given ratio of counselor intervention recommendations versus automated messages, typically by limiting the counselor intervention recommendations to those that have the highest priority.
  • the triggering module 142 typically checks (from the response history) that identical (or similar) automated messages are not sent twice to the same participant.
  • a method of sending an automated message to a participant such as by SMS or email, may be determined by the triggering module based on a predefined participant preference (which may be stored in the profile repository or other participant database), or by other preset rules.
  • a predefined participant preference which may be stored in the profile repository or other participant database
  • An ML analysis of which method has been more effective for participants with similar profiles may also be performed.
  • the timing of message delivery may be set according to a participant preference or an analysis of what times have proven effective (for a given participant or for a given set of profile attributes).
  • the triggering module may a participant may have requested that messages only be delivered [0067]
  • automated messages are sent by means of a messaging or mail system that indicates the sender as the participant's assigned counselor (i.e., a mail delivery subsystem, or SMS subsystem, of the system 100 may be configured to send automated messages from the email addresses of the assigned counselor).
  • a messaging or mail system that indicates the sender as the participant's assigned counselor (i.e., a mail delivery subsystem, or SMS subsystem, of the system 100 may be configured to send automated messages from the email addresses of the assigned counselor).
  • Figs. 2A-C are tables indicating clustering methods of the system with respect to participant data trends, in accordance with an embodiment of the present invention.
  • Fig. 2A shows a list of trends of a given data measurement for participants ("clients") of a given wellness program. The trend may be, for example, daily weight changes of participants over the course of the previous month. If all the data points were applied directly to the ML classifier, it would be difficult to achieve an estimated correlation with the target label (i.e., attrition risk). Consequently, the trends are first clustered, as indicated in Fig. 2B, which shows a set of ten clustered trends, which are representative of clusters of trends achieved by clustering.
  • Fig. 2B shows a set of ten clustered trends, which are representative of clusters of trends achieved by clustering.
  • Clustering algorithms that may be implemented include, for example: hierarchical clustering, DBSCAN, HDBSCAN, k-means, Gaussian mixture, and time-series, dynamic time warping.
  • FIG. 3 is a flow diagram of a process 300 for generating interaction responses of a wellness program as implemented by system 100, in accordance with an embodiment of the present invention.
  • the system receives multiple types of customer data from multiple sources. These data points (data types and their values) are stored as attributes of respective customer profiles in a profile repository. Some trend data is also clustered, such that some attributes indicate the clustered groups.
  • the previously trained ML classifier 140 is applied at regular intervals to current participant profiles to determine participant attrition risks.
  • ML interpretation methods may also be applied to determine significant contributing factors to the attrition risks, which may indicate the success of specific interaction responses and of specific trigger rules in reducing attrition risks.
  • Key profile attributes may include changes in trend clustering of a profile. For example, a trigger may be caused by a trend attribute of a profile changing from one cluster group to another, or by a trend attribute remaining constant over multiple consecutive measurements (indicating, for example, that a participant is not progressing towards a goal).
  • a determination may be made as to whether to issue an automated message and/or a counselor intervention recommendation. The determination is typically made based on finding an optimal correlation between the attributes and attrition risk of the participant profile and the attribute labels of stored message examples and/or counselor recommendation examples. Additional rules may determine the selection of an appropriate interaction example, including a rule with respect to a ratio between the two types of interaction responses that should be issued within a given time frame.
  • an automated message is to be issued, then, at a step 318, appropriately labeled message fragment examples may be selected and customized to generate a full message, which may then be sent to the participant (i.e., the participant corresponding to the given profile).
  • a counselor intervention recommendation is to be issued, then, at a step 320, an appropriately labeled counselor recommendation example is selected, customized and issued as a recommendation (or "intervention") "response" to a counselor assigned to the given participant.
  • step 312 when an interaction response is not triggered at step 312, the system returns to step 310 to get another profile to classify.
  • receipt of participant data is an on-going process that typically takes place concurrently with the other steps of process 300.
  • the other steps following step 310 may be implemented in parallel, in multiple threads or on multiple processors of the system.
  • Figs. 4A and 4B are flow diagrams of respective processes 400 and 410 for preparing the system 100 for operation, in accordance with an embodiment of the present invention.
  • Process 400 indicates steps of preparing the interaction response repository with message examples and counselor recommendation examples that are to be issued by the system, according to attributes and attrition risks determined for participant profiles.
  • the process includes two basic steps, a first step 402 of receiving the interaction examples (examples of messages and counselor intervention recommendations) from program counselors (i.e., program professionals/experts, as described above).
  • the interaction examples are typically created with specific participants, or categories of participants, in mind.
  • the interaction examples are typically created for associated, intended attrition risk levels, as well.
  • attribute labels are assigned to the interaction examples at step 404.
  • Assignment of the attributes and the attrition levels may be performed automatically using the same attributes for labelling interaction examples as those already assigned to the intended participants. Additionally or alternatively, assignment of the attributes and the attrition risk levels as labels for the interaction examples may be done manually by counselors i.e., program professionals/experts). Labels of interaction examples may be subsequently updated during operation. Updating will occur as responses issued on the basis of interaction examples are shown to be either effective or ineffective in reducing attrition risks of participant profiles.
  • Automated messages as well as interactions that a participant has with a counselor may also be recorded as attributes of a participant profile. Such attributes would affect the prediction of attrition risk generated by the ML classify.
  • interaction examples may be similarly labeled with attributes indicating such interactions.
  • a message example may be labeled to correlate with a type of counselor interaction that a participant had, for example, on the same day.
  • the profile attribute (i.e., data point) of same day interaction could trigger an interaction response.
  • the trigger module would then identify the given message example as correlating with the participant profile attribute of a same day interaction. Consequently, an automated message would be sent soon after the participant has the given type of counselor interaction, in order to reinforce the interaction.
  • Process 410 indicates four basic steps of training the ML classifier.
  • participant data points are received from multiple sources, for multiple participants, and participant profiles including these participant data points (i.e., attributes) are stored in a profile repository. Participant profiles are also typically "time- stamped" in order to be associated, according to their date, with subsequent dates of participant attrition. Multiple snapshots of participant profiles may also be saved at different dates (and times).
  • profile clusters are typically generated from participant trend data, to provide cluster groups as attributes for participant profiles.
  • each of the one or more time-stamped profiles, for each of the participants, are labeled by the participant attrition records to generate the ML classifier for classifying (i.e., predicting) attrition risks correlated with participant profiles.
  • the system or application implementing the above described method may be an add-on, or upgrade, or a retrofit to a software application for contact or customer management.
  • Processing elements of the system described herein may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or in combinations thereof. Such elements can be implemented as a computer program product, tangibly embodied in an information carrier, such as a non-transient, machine -readable storage device, for execution by, or to control the operation of, data processing apparatus, such as a programmable processor, computer, or deployed to be executed on multiple computers at one site or one or more across multiple sites.
  • Memory storage for software and data may include multiple one or more memory units, including one or more types of storage media.
  • Examples of storage media include, but are not limited to, magnetic media, optical media, and integrated circuits such as read-only memory devices (ROM) and random access memory (RAM).
  • Network interface modules may control the sending and receiving of data packets over networks. Method steps associated with the system and process can be rearranged and/or one or more such steps can be omitted to achieve the same, or similar, results to those described herein.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • Data Mining & Analysis (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Child & Adolescent Psychology (AREA)
  • Developmental Disabilities (AREA)
  • Hospice & Palliative Care (AREA)
  • Psychiatry (AREA)
  • Psychology (AREA)
  • Social Psychology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

L'invention concerne un système et des procédés destinés à générer des réponses d'interaction pour des participants à un programme de bien-être, faisant intervenir les étapes consistant à: entraîner un classificateur à apprentissage automatique (ML) pour prédire un risque d'attrition de profil d'un profil d'un participant au programme de bien-être; recevoir des exemples d'interaction et étiqueter chaque exemple d'interaction avec un ou plusieurs attributs d'après des points de données du profil du participant considéré et un niveau prévu de risque d'attrition d'un participant prévu; recevoir un ensemble de profils de participants, comprenant chacun de multiples points de données et dont chacun est associé à un participant donné; appliquer le classificateur à ML pour prédire un risque d'attrition pour chaque participant d'un second ensemble de profils de participants; et générer une réponse d'interaction en sélectionnant un exemple d'interaction dont les attributs étiquetés sont corrélés avec les points de données du second ensemble et le risque d'attrition du profil de participant donné, la réponse d'interaction incluant un message automatisé et/ou une recommandation d'intervention d'accompagnateur.
PCT/IL2020/050252 2019-03-04 2020-03-04 Système et procédé de génération de réponses d'interaction pour participants de programme de bien-être WO2020178829A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201962813417P 2019-03-04 2019-03-04
US62/813,417 2019-03-04

Publications (1)

Publication Number Publication Date
WO2020178829A1 true WO2020178829A1 (fr) 2020-09-10

Family

ID=72337603

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/IL2020/050252 WO2020178829A1 (fr) 2019-03-04 2020-03-04 Système et procédé de génération de réponses d'interaction pour participants de programme de bien-être

Country Status (1)

Country Link
WO (1) WO2020178829A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11770307B2 (en) 2021-10-29 2023-09-26 T-Mobile Usa, Inc. Recommendation engine with machine learning for guided service management, such as for use with events related to telecommunications subscribers

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110218407A1 (en) * 2010-03-08 2011-09-08 Seth Haberman Method and apparatus to monitor, analyze and optimize physiological state of nutrition
US20180012242A1 (en) * 2016-07-06 2018-01-11 Samsung Electronics Co., Ltd. Automatically determining and responding to user satisfaction
US10068494B2 (en) * 2016-10-14 2018-09-04 Invoy Holdings, Llc Artificial intelligence based health coaching based on ketone levels of participants

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110218407A1 (en) * 2010-03-08 2011-09-08 Seth Haberman Method and apparatus to monitor, analyze and optimize physiological state of nutrition
US20180012242A1 (en) * 2016-07-06 2018-01-11 Samsung Electronics Co., Ltd. Automatically determining and responding to user satisfaction
US10068494B2 (en) * 2016-10-14 2018-09-04 Invoy Holdings, Llc Artificial intelligence based health coaching based on ketone levels of participants

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11770307B2 (en) 2021-10-29 2023-09-26 T-Mobile Usa, Inc. Recommendation engine with machine learning for guided service management, such as for use with events related to telecommunications subscribers

Similar Documents

Publication Publication Date Title
US20200005928A1 (en) System and method for personalized wellness management using machine learning and artificial intelligence techniques
US11930087B1 (en) Matching patients with decentralized clinical trials to improve engagement and retention
CN109416820B (zh) 对用户满意度进行自动确定和响应的方法和系统
McTavish et al. How patients recovering from alcoholism use a smartphone intervention
US20150356701A1 (en) Monitoring and adapting a patient's medical regimen and facilitating communication with a caregiver
US20180286509A1 (en) Computer-Implemented Multi-Factor Patient Education Method
WO2017019417A1 (fr) Systèmes et procédés d'appariement de patients aux fournisseurs les mieux adaptés de programmes de prévention de maladies chroniques
WO2017165621A1 (fr) Assistance en ligne en temps réel ou juste à temps pour des individus pour les aider à atteindre des objectifs de santé personnalisés
US20170208021A1 (en) Adaptive nudge messages to motivate individuals to achieve certain wellness goals
US11574714B2 (en) Remote health assertion verification and mortality prediction system
US11417418B1 (en) Recruiting for clinical trial cohorts to achieve high participant compliance and retention
US20200104935A1 (en) Systems and methods for wealth and health planning
US12002064B1 (en) Adapting computerized processes for matching patients with clinical trials to increase participant engagement and retention
WO2017176653A1 (fr) Systèmes et procédés de suggestion d'actions bénéfiques
US20190065684A1 (en) Computer program products, methods, and systems for assisting a user to achieve a health-related goal
Van Der Swaluw et al. Commitment lotteries promote physical activity among overweight adults—a cluster randomized trial
US20120221345A1 (en) Helping people with their health
US20230017196A1 (en) System and method for rules engine that dynamically adapts application behavior
US20120278414A1 (en) System and method for electronic event correlated message delivery
US20210110924A1 (en) System and method for monitoring system compliance with measures to improve system health
WO2020178829A1 (fr) Système et procédé de génération de réponses d'interaction pour participants de programme de bien-être
KR102476612B1 (ko) 인공지능에 기반한 심리 맞춤 솔루션 제공 방법 및 시스템
US20230075408A1 (en) Methods and systems for interactively counseling a user with respect to supervised content
Kovalchik et al. Scaled inverse probability weighting: A method to assess potential bias due to event nonreporting in ecological momentary assessment studies
US20120219931A1 (en) System, method, and program product for modifying eating behavior

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1)EPC DATED 24.06.222(EPO FORM 1205A)

122 Ep: pct application non-entry in european phase

Ref document number: 20766449

Country of ref document: EP

Kind code of ref document: A1