Systems and methods for trusted self-checkout at retail stores
Post pandemic, retailers are adopting more contactless services for shopper's checkout such as self-checkout, hybrid checkout and mobile checkout, and these touchpoints have become the potential areas for fraudulent activity during check out process. For detecting fraud carried out by a customer at the time of self-checkout, existing approaches require respective customer identity and his purchase history. Embodiments of the present disclosure do not require customer identity and information about his historical shopping carts and provide a method and system for approving a user shopping cart for self-checkout from the items picked by the customers in real time.
1 . A processor implemented method comprising:
obtaining in real-time, via one or more hardware processors, information pertaining to a user shopping cart associated with a user for a context further comprising one or more real-time factors, wherein the user shopping cart comprises one or more items being added therein;
extracting in the real-time, via the one or more hardware processors, a plurality of historical universal shopping carts containing the one or more items being added in the user shopping cart and which is done in a similar context of the user shopping cart;
identifying, a behavioral item and a noise item when a new item is added in the user shopping cart, wherein based on items being added, variations within a group of historical shopping carts increase or decrease, wherein when the within group variation is reduced then the new item is noted as the behavioral item and a level of reduction is noted as a signal carried by the new item in positive direction towards identifying intention of the user, wherein when the within group variation is increased then the new item is noted as the noise item and a level of increase is noted as a signal carried by the new item in negative direction, wherein behavioral items are noted as first set of items and noise items are noted as second set of items;
measuring in the real-time, via the one or more hardware processors, similarity of the plurality of historical universal shopping carts based on one or more formats of the plurality of historical universal shopping carts;
determining in the real-time via the one or more hardware processors, during addition of the one or more items in the shopping cart, an improvement in the similarity among the extracted historical universal shopping carts containing the one or more items being added in the user shopping cart for the context;
identifying in the real-time, via the one or more hardware processors, a first set of items amongst the one or more items being added in the user shopping cart based on the improvement in the similarity;
determining in the real-time, via the one or more hardware processors, a context specific expected distribution for the user shopping cart based on a distance distribution of the plurality of historical universal shopping carts containing the first set of items pertaining to the context;
categorizing in the real-time via the one or more hardware processors, in an earlier phase of a self-checkout development, the user shopping cart as a first type or a second type based on a position of the user shopping cart in comparison to dynamic cut off value of the context specific expected distribution for an approval of the user shopper cart;
monitoring correctly categorized via the one or more hardware processors, the first type or the second type of one or more shopping carts through one or more monitoring devices and storing information further comprising at least one context and a format of a plurality of correctly categorized shopping carts;
continually improving an accuracy of the categorization via the one or more hardware processors, in the earlier phase of the self-checkout development, through (i) a self-learning mechanism by using the at least one context, the format, and the dynamic cut off value of the context specific distribution of the plurality of correctly categorized shopping carts, and (ii) by finding (a) an ideal cut off value of the context specific distribution, (b) real time specific ideal formats of the plurality of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors in an iterative manner, wherein the plurality of historical useful shopping carts are classified into groups where each group has useful shopping carts with same format of real time factors and same format of items as a useful subset, wherein the useful subset acts as a separate training set and is termed as a training useful set, wherein each training set has format of real time factors and format of items for each useful shopping cart and their class such as (a) first type, or (b) second type, wherein the training useful set, format of real time factors and format of items are used as independent variable and class as (a) suspect of fraud, or (b) no suspect for fraud is used as dependent variable;
developing and training, via the one or more hardware processors, a machine learning model in an advanced phase of the self-checkout development, and predicting a user shopping cart for approval based on the trained machine learning model trained using the plurality of correctly categorized user shopping carts and associated context and format of the plurality of correctly categorized user shopping carts to enable categorization in real-time thereof and without extracting the plurality of historical universal shopping carts;
approving in the real-time via the one or more hardware processors, the user shopping cart for a check out predicted as the first type by the trained machine learning model based on a comparison of the predicted probability of the user shopping cart with dynamic cut off value of the predicted probability by the trained machine learning model;
monitoring outcome via the one or more hardware processors, for the first type and the second type through the one or more monitoring devices and storing information pertaining to the at least one context, the format of the plurality of correctly approved user shopping carts and associated predicted probabilities;
continually improving the accuracy of the approval of the user shopping carts via the one or more hardware processors, in the advanced phase of the self-checkout development through the self-learning mechanism by using the at least one context, the format and the dynamic cut off value of the predicted probability of the plurality of correctly categorized shopping carts and by finding (a) the ideal cut off value of the predicted probability for the approval of the one or more user shopper cart, (b) real time specific ideal formats of the plurality of historical universal shopping carts and (c) real time specific ideal formats of associated real time factors in an iterative manner, wherein the shopping cart is approved based on the items picked by the user in real time by deriving context specific expected behavior of the customer and using the same for approval of a shopping cart for self-checkout; and
performing deployment of the self-checkout in response to obtaining improved accuracy of the approval of the user shopping carts in the advanced phase of the self-checkout development and reducing the frequency of monitoring activity to periodic partial monitoring to validate performance of system randomly.
2 . The processor implemented method of claim 1 , wherein in the earlier phase of the self-checkout development, (a) the ideal cut of value of the context specific distribution for the approval of user shopper cart, (b) real time specific ideal formats of the plurality of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors are determined by maximizing a matching score between the categorized first type and actual first type in an iterative manner, wherein the plurality of historical universal shopping carts having behavioral items by omitting noise items are filtered and is formed as a standard group, wherein central point of the standard group is derived using multivariate distance by considering all the shopping carts within the standard group, wherein multi variate distance of a shopping cart with centroid is found and repeated for all the shopping carts of the standard group, wherein the calculated multivariate distance is used to derive a distribution noted as standard behavior and mean and standard deviation of distribution are calculated as per standard procedures used for multivariate distribution, wherein an ideal shopping cart is derived for a context in real time.
3 . The processor implemented method of claim 1 , wherein in the advanced phase of the self-checkout development, (a) the ideal cut of value of the predicted probability for the approval of user shopper cart, (b) real time specific ideal formats of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors are determined by maximizing a matching score between predicted shopping cart for approval of the trained machine learning model and actual first type in an iterative manner, wherein entities including a shopping cart, a timestamp and a store ID are converted into suitable formats to ensure that the converted formats are mapped with the format owned by the useful subset, wherein the converted format is used to predict the probability for the shopping cart, wherein the process is repeated for each useful subset and the one with lowest error is selected, wherein the dynamic entities (a) the cut off value assigned for the context specific expected behavioral distribution, (b) the cut off value of the probability for the approval of shopping cart, and (c) the format of shopping carts used for finding similarity among the group of shopping carts undergo fine tuning to maintain accuracy in allowing shopping cart for approval for self-checkout, wherein the machine learning model enables categorization of user shopping carts in real time without extracting the plurality of historical universal shopping carts, wherein the cut off value is readjusted depending on a real time system performance.
4 . The processor implemented method of claim 1 , wherein the context comprises one or more real-time factors which include at least one of a time, a location, and weather.
5 . The processor implemented method of claim 1 , wherein the one or more formats of the plurality of historical universal shopping carts comprises at least one of (i) one or more formats of items present, (ii) one or more formats of value spread, and (iii) one or more formats of order of items picked, and wherein the one or more formats are based on an associated importance level.
6 . The processor implemented method of claim 1 , wherein the step of determining the context specific expected distribution of the user shopping cart enables to measure the position of the user shopping cart in real time in comparison to the context specific expected distribution of the user shopping cart without a user identity and without historical shopping carts of the user, wherein the position is determined by comparing a current shopping cart comprising both the behavioral items and the noise items with a standard behavior, so as to measure a level of deviation of the current shopping cart from the standard behavior and to determine a severity of the deviation.
7 . The processor implemented method of claim 1 , wherein the step of identifying the first set of items amongst the one or more items further comprises identifying a second set of items amongst the one or more items.
8 . The processor implemented method of claim 1 , wherein the first set of items and the second set of items are different from each other.
9 . A system, comprising:
a memory storing instructions;
one or more communication interfaces; and
one or more hardware processors coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processors are configured by the instructions to:
obtain in real-time, information pertaining to a user shopping cart associated with a user for a context further comprising one or more real-time factors, wherein the user shopping cart comprises one or more items being added therein;
extract in the real-time, a plurality of historical universal shopping carts containing the one or more items being added in the user shopping cart and which is done in a similar context of the user shopping cart;
identify, a behavioral item and a noise item when a new item is added in the user shopping cart, wherein based on items being added, variations within a group of historical shopping carts increase or decrease, wherein when the within group variation is reduced then the new item is noted as the behavioral item and a level of reduction is noted as a signal carried by the new item in positive direction towards identifying intention of the user, wherein when the within group variation is increased then the new item is noted as the noise item and a level of increase is noted as a signal carried by the new item in negative direction, wherein behavioral items are noted as first set of items and noise items are noted as second set of items;
measure in the real-time, similarity of the plurality of historical universal shopping carts based on one or more formats of the plurality of historical universal shopping carts;
determine in the real-time, during addition of the one or more items in the shopping cart, an improvement in similarity among the extracted historical universal shopping carts containing the one or more items being added in the user shopping cart for the context;
identify in the real-time, a first set of items amongst the one or more items being added in the user shopping cart based on the improvement in similarity;
determine in the real-time, a context specific expected distribution for the user shopping cart based on a distance distribution of the plurality of historical universal shopping carts containing the first set of items pertaining to the context;
categorize in the real-time, in an earlier phase of a self-checkout development, the user shopping cart as a first type or a second type based on a position of the user shopping cart in comparison to dynamic cut off value of the context specific expected distribution for an approval of the user shopper cart;
monitor correctly categorized, the first type or the second type of one or more shopping carts through one or more monitoring devices and storing information further comprising at least one context and a format of a plurality of correctly categorized shopping carts;
continually improve an accuracy of the categorization, in the earlier phase of the self-checkout development, through (i) a self-learning mechanism by using the at least one context, the format, and the dynamic cut off value of the context specific distribution of the plurality of correctly categorized shopping carts, and (ii) by finding (a) an ideal cut off value of the context specific distribution, (b) real time specific ideal formats of the plurality of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors in an iterative manner, wherein the plurality of historical useful shopping carts are classified into groups where each group has useful shopping carts with same format of real time factors and same format of items as a useful subset, wherein the useful subset acts as a separate training set and is termed as a training useful set, wherein each training set has format of real time factors and format of items for each useful shopping cart and their class such as (a) first type, or (b) second type, wherein the training useful set, format of real time factors and format of items are used as independent variable and class as (a) suspect of fraud, or (b) no suspect for fraud is used as dependent variable;
develop and train, a machine learning model in an advanced phase of the self-checkout development, and predicting a user shopping cart for approval based on the trained machine learning model trained using the plurality of correctly categorized user shopping carts and associated context and format of the plurality of correctly categorized user shopping carts to enable categorization in real-time thereof and without extracting the plurality of historical universal shopping carts;
approve in the real-time, the user shopping cart for a check out predicted as the first type by the trained machine learning model based on a comparison of the predicted probability of the user shopping cart with dynamic cut off value of the predicted probability by the trained machine learning model;
monitoring outcome, for the first type and the second type through the one or more monitoring devices and storing information pertaining to the at least one context, the format of the plurality of correctly approved user shopping carts and associated predicted probabilities;
continually improve the accuracy of the approval of the user shopping carts, in the advanced phase of the self-checkout development through the self-learning mechanism by using the at least one context, the format and the dynamic cut off value of the predicted probability of the plurality of correctly categorized shopping carts and by finding (a) the ideal cut off value of the predicted probability for the approval of the one or more user shopper cart, (b) real time specific ideal formats of the plurality of historical universal shopping carts and (c) real time specific ideal formats of associated real time factors in an iterative manner, wherein the shopping cart is approved based on the items picked by the user in real time by deriving context specific expected behavior of the customer and using the same for approval of a shopping cart for self-checkout; and
perform deployment of the self-checkout in response to obtaining improved accuracy of the approval of the user shopping carts in the advanced phase of the self-checkout development and reducing the frequency of monitoring activity to periodic partial monitoring to validate performance of system randomly.
10 . The system of claim 9 , wherein in the earlier phase of the self-checkout development, (a) the ideal cut of value of the context specific distribution for the approval of user shopper cart, (b) real time specific ideal formats of the plurality of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors are determined by maximizing a matching score between the categorized first type and actual first type in an iterative manner, wherein the plurality of historical universal shopping carts having behavioral items by omitting noise items are filtered and is formed as a standard group, wherein central point of the standard group is derived using multivariate distance by considering all the shopping carts within the standard group, wherein multi variate distance of a shopping cart with centroid is found and repeated for all the shopping carts of the standard group, wherein the calculated multivariate distance is used to derive a distribution noted as standard behavior and mean and standard deviation of distribution are calculated as per standard procedures used for multivariate distribution, wherein an ideal shopping cart is derived for a context in real time.
11 . The system of claim 9 , wherein in the advanced phase of the self-checkout development, (a) the ideal cut of value of the predicted probability for the approval of user shopper cart, (b) real time specific ideal formats of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors are determined by maximizing a matching score between predicted shopping cart for approval of the trained machine learning model and actual first type in an iterative manner, wherein entities including a shopping cart, a timestamp and a store ID are converted into suitable formats to ensure that the converted formats are mapped with the format owned by the useful subset, wherein the converted format is used to predict the probability for the shopping cart, wherein the process is repeated for each useful subset and the one with lowest error is selected, wherein the dynamic entities (a) the cut off value assigned for the context specific expected behavioral distribution, (b) the cut off value of the probability for the approval of shopping cart, and (c) the format of shopping carts used for finding similarity among the group of shopping carts undergo fine tuning to maintain accuracy in allowing shopping cart for approval for self-checkout, wherein the machine learning model enables categorization of user shopping carts in real time without extracting the plurality of historical universal shopping carts, wherein the cut off value is readjusted depending on a real time system performance.
12 . The system of claim 9 , wherein the context comprises one or more real-time factors which include at least one of a time, a location, and weather.
13 . The system of claim 9 , wherein the one or more formats of the plurality of historical universal shopping carts comprises at least one of (i) one or more formats of items present, (ii) one or more formats of value spread, and (iii) one or more formats of order of items picked, and wherein the one or more formats are based on an associated importance level.
14 . The system of claim 9 , wherein the step of determining the context specific expected distribution of the user shopping cart enables to measure the position of the user shopping cart in real time in comparison to the context specific expected distribution of the user shopping cart without a user identity and without historical shopping carts of the user, wherein the position is determined by comparing a current shopping cart comprising both the behavioral items and the noise items with a standard behavior, so as to measure a level of deviation of the current shopping cart from the standard behavior and to determine a severity of the deviation.
15 . The system of claim 9 , wherein the step of identifying the first set of items amongst the one or more items further comprises identifying a second set of items amongst the one or more items.
16 . The system of claim 9 , wherein the first set of items and the second set of items are different from each other.
17 . One or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause:
obtaining in real-time, information pertaining to a user shopping cart associated with a user for a context further comprising one or more real-time factors, wherein the user shopping cart comprises one or more items being added therein;
extracting in the real-time, a plurality of historical universal shopping carts containing the one or more items being added in the user shopping cart and which is done in a similar context of the user shopping cart;
identifying, a behavioral item and a noise item when a new item is added in the user shopping cart, wherein based on items being added, variations within a group of historical shopping carts increase or decrease, wherein when the within group variation is reduced then the new item is noted as the behavioral item and a level of reduction is noted as a signal carried by the new item in positive direction towards identifying intention of the user, wherein when the within group variation is increased then the new item is noted as the noise item and a level of increase is noted as a signal carried by the new item in negative direction, wherein behavioral items are noted as first set of items and noise items are noted as second set of items;
measuring in the real-time, similarity of the plurality of historical universal shopping carts based on one or more formats of the plurality of historical universal shopping carts;
determining in the real-time, during addition of the one or more items in the shopping cart, an improvement in the similarity among the extracted historical universal shopping carts containing the one or more items being added in the user shopping cart for the context;
identifying in the real-time, a first set of items amongst the one or more items being added in the user shopping cart based on the improvement in the similarity;
determining in the real-time, a context specific expected distribution for the user shopping cart based on a distance distribution of the plurality of historical universal shopping carts containing the first set of items pertaining to the context;
categorizing in the real-time, in an earlier phase of a self-checkout development, the user shopping cart as a first type or a second type based on a position of the user shopping cart in comparison to dynamic cut off value of the context specific expected distribution for an approval of the user shopper cart;
monitoring correctly categorized the first type or the second type of one or more shopping carts through one or more monitoring devices and storing information further comprising at least one context and a format of a plurality of correctly categorized shopping carts;
continually improving an accuracy of the categorization, in the earlier phase of the self-checkout development, through (i) a self-learning mechanism by using the at least one context, the format, and the dynamic cut off value of the context specific distribution of the plurality of correctly categorized shopping carts, and (ii) by finding (a) an ideal cut off value of the context specific distribution, (b) real time specific ideal formats of the plurality of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors in an iterative manner, wherein the plurality of historical useful shopping carts are classified into groups where each group has useful shopping carts with same format of real time factors and same format of items as a useful subset, wherein the useful subset acts as a separate training set and is termed as a training useful set, wherein each training set has format of real time factors and format of items for each useful shopping cart and their class such as (a) first type, or (b) second type, wherein the training useful set, format of real time factors and format of items are used as independent variable and class as (a) suspect of fraud, or (b) no suspect for fraud is used as dependent variable;
developing and training a machine learning model in an advanced phase of the self-checkout development, and predicting a user shopping cart for approval based on the trained machine learning model trained using the plurality of correctly categorized user shopping carts and associated context and format of the plurality of correctly categorized user shopping carts to enable categorization in real-time thereof and without extracting the plurality of historical universal shopping carts;
approving in the real-time, the user shopping cart for a check out predicted as the first type by the trained machine learning model based on a comparison of the predicted probability of the user shopping cart with dynamic cut off value of the predicted probability by the trained machine learning model;
monitoring outcome, for the first type and the second type through the one or more monitoring devices and storing information pertaining to the at least one context, the format of the plurality of correctly approved user shopping carts and associated predicted probabilities;
continually improving the accuracy of the approval of the user shopping carts, in the advanced phase of the self-checkout development through the self-learning mechanism by using the at least one context, the format and the dynamic cut off value of the predicted probability of the plurality of correctly categorized shopping carts and by finding (a) the ideal cut off value of the predicted probability for the approval of the one or more user shopper cart, (b) real time specific ideal formats of the plurality of historical universal shopping carts and (c) real time specific ideal formats of associated real time factors in an iterative manner, wherein the shopping cart is approved based on the items picked by the user in real time by deriving context specific expected behavior of the customer and using the same for approval of a shopping cart for self-checkout; and
performing deployment of the self-checkout in response to obtaining improved accuracy of the approval of the user shopping carts in the advanced phase of the self-checkout development and reducing the frequency of monitoring activity to periodic partial monitoring to validate performance of system randomly.
18 . The one or more non-transitory machine-readable information storage mediums of claim 17 , wherein in the earlier phase of the self-checkout development, (a) the ideal cut of value of the context specific distribution for the approval of user shopper cart, (b) real time specific ideal formats of the plurality of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors are determined by maximizing a matching score between the categorized first type and actual first type in an iterative manner, wherein the plurality of historical universal shopping carts having behavioral items by omitting noise items are filtered and is formed as a standard group, wherein central point of the standard group is derived using multivariate distance by considering all the shopping carts within the standard group, wherein multi variate distance of a shopping cart with centroid is found and repeated for all the shopping carts of the standard group, wherein the calculated multivariate distance is used to derive a distribution noted as standard behavior and mean and standard deviation of distribution are calculated as per standard procedures used for multivariate distribution, wherein an ideal shopping cart is derived for a context in real time.
19 . The one or more non-transitory machine-readable information storage mediums of claim 17 , wherein in the advanced phase of the self-checkout development, (a) the ideal cut of value of the predicted probability for the approval of user shopper cart, (b) real time specific ideal formats of historical universal shopping carts, and (c) real time specific ideal formats of associated real time factors are determined by maximizing a matching score between predicted shopping cart for approval of the trained machine learning model and actual first type in an iterative manner, wherein entities including a shopping cart, a timestamp and a store ID are converted into suitable formats to ensure that the converted formats are mapped with the format owned by the useful subset, wherein the converted format is used to predict the probability for the shopping cart, wherein the process is repeated for each useful subset and the one with lowest error is selected, wherein the dynamic entities (a) the cut off value assigned for the context specific expected behavioral distribution, (b) the cut off value of the probability for the approval of shopping cart, and (c) the format of shopping carts used for finding similarity among the group of shopping carts undergo fine tuning to maintain accuracy in allowing shopping cart for approval for self-checkout, wherein the machine learning model enables categorization of user shopping carts in real time without extracting the plurality of historical universal shopping carts, wherein the cut off value is readjusted depending on a real time system performance.
20 . The one or more non-transitory machine-readable information storage mediums of claim 17 , wherein the context comprises one or more real-time factors which include at least one of a time, a location, and weather, wherein the one or more formats of the plurality of historical universal shopping carts comprises at least one of (i) one or more formats of items present, (ii) one or more formats of value spread, and (iii) one or more formats of order of items picked, and wherein the one or more formats are based on an associated importance level, wherein the step of determining the context specific expected distribution of the user shopping cart enables to measure the position of the user shopping cart in real time in comparison to the context specific expected distribution of the user shopping cart without a user identity and without historical shopping carts of the user, wherein the step of identifying the first set of items amongst the one or more items further comprises identifying a second set of items amongst the one or more items, wherein the first set of items and the second set of items are different from each other.