WO2011005072A2

WO2011005072A2 - Personalized shopping list recommendation based on shopping behavior

Info

Publication number: WO2011005072A2
Application number: PCT/MY2010/000110
Authority: WO
Inventors: Mohamad Zaini Norliza; Akmar Omar Hasmila; Mohd. Yassin Norlidza
Original assignee: Mimos Bhd.
Priority date: 2009-07-09
Filing date: 2010-06-30
Publication date: 2011-01-13
Also published as: WO2011005072A3

Abstract

A method of providing purchasing recommendations to a user by generating a purchasing behavior model of the user is disclosed. The model is generated by capturing user's purchase preference information during registration and/or by tracking and analyzing purchases of the user at physical and virtual stores. The products purchased by the user are identified to be in one or more of the following categories: next interval due; seasonal; store dependent; promotion dependent; commodity-brand dependent; commodity dependent; brand dependent; and price-dependent. When a request is made for shopping recommendations, a server generates a list of personalized purchase-recommendations by performing a reasoning process on the user's purchasing behavior model and based on the user's current context, e.g. visited store, current time and current date. The generated list of personalized purchase-recommendations is then forwarded to the user.

Description

Personalized Shopping List Recommendation Based On Shopping

Behavior

FIELD OF INVENTION

The present invention relates to a method of providing a personalized shopping list recommendation to a user.

BACKGROUND OF INVENTION

There are currently already a number of methods and systems that provide shopping list recommendations to shoppers. Jacobi et al (US 7113917) claims a computer-implemented service which recommends items to a user based on items previously selected by the user. A shopping recommendation is made based on the user's shopping history. This as well as many other shopping list recommendation systems are online based - users have to be browsing online or connected in some other way to the internet to receive the recommendations generated.

There is a rising need for having a system that enables recommendations on demand; i.e. both online and offline, due to the emerging trend of frequent mobility of users. A system that is capable of integrating features for both online and offline modes will be able to provide seamless recommendations between the different modes.

Furthermore, many of the existing recommendation methods generate the purchase history of the user from the user's online product selection history. There is no data captured from purchases made at physical stores and in real-time. A method that collects purchasing information from purchase receipts at the actual stores will be inherently more accurate and robust. Many known methods of generating shopping lists also rely on collaborative filtering to choose the products to be recommended. This relies on the ratings given by other users. The problem with this approach is there are cases where a suitable item is not recommended merely because no other users have purchased the item and thus no ratings have been given. Other approaches that recommend items purchased by shoppers who are in a particular group of which the user is also a member suffers from the same disadvantage. As such, a filtering method which does not rely exclusively on purchase history of other users is desirable. Some other methods of capturing purchase data for the purpose of generating a user profile with shopping preferences do not associate purchased items with the stores where they were purchased. The items are also not associated with the intervals at which they are usually purchased, nor are they associated with special seasons or occasions when particular items may be more likely to be sought. The items are also not associated with any current promotions being offered. There is a need for a method that takes these aspects into account when generating a shopping behavior model.

SUMMARY OF INVENTION

This invention relates to a method of providing purchasing recommendations to a user by generating a purchasing behavior model of the user. The model is generated by capturing purchase information of the user during a registration of the user and/or by tracking and analyzing purchases of the user at physical and virtual stores. The products purchased by the user are identified to be in one or more of the following categories: next interval due; seasonal variations; store purchased; promotions; commodity-brand; commodity; brand; and price-dependent. When a request is made for shopping recommendations, a server generates a list of purchase recommendations by performing a reasoning process on the purchasing behavior model and based on the current time, date and store location of the user. The generated list of purchase recommendations is then forwarded to the user.

This invention also relates to a method of providing purchase recommendations to a user, comprising the steps of:

- generating a purchasing behavior model for the user;

- receiving a request to provide purchase recommendations;

- generating purchase recommendations by performing a reasoning process on said model; and

- providing to the user said purchase recommendations

wherein the said model of the user's purchasing behavior is generated based on preferences of the user. The preferences are manually captured either during a registration of the user or by tracking and analyzing the user's purchases at physical and virtual stores. The tracking and analyzing of the user's purchases comprises identifying whether each of the products purchased by the user fall into any one of the following categories: next interval due; seasonal variations; store purchased; promotions; commodity- brand; commodity; brand; and price-dependent. In a preferred embodiment, the purchase recommendations are dependent on the current context of the user, e.g. visited store, current time and current date. The purchase recommendations can also be dependent on any current promotions associated with the products in the user's purchasing behavior model.

This invention further relates to a process of recommending shopping-lists to users that requires tracking and analyzing users' shopping activities in order to understand their buying behavior. This learnt knowledge is required to provide good recommendations to users in the future. The process starts by keeping track of the user's purchasing history. A new update to the database is performed every time a user makes a purchase. Each purchase record is tagged with a user identifier, product / item identifier, time and location of purchase before it is sent to the server for tracking and analysis. The resulting user's shopping-behavior model is regarded as closely representative of the user's habits or behavior of buying items. Such a model is important to predict the likelihood for the user to purchase a particular item in the future. Predicting the most likely items desired by the user for purchase is one source of recommending a shopping list since such recommendations are made on the basis of reminding the user of what items need to be purchased at a particular store and time. Upon entering (offline/physical) or login (online) into a pre-visited store, the system is able to identify that this is the right time to request for a shopping-list recommendation. Upon receiving such request, a personalized recommended shopping list will be constructed by performing inference on the user's shopping model. The inference will be performed based on the user's current context; i.e. visited store and current time. Once the recommended shopping list is constructed, it will then be returned to the requesting application as to enable it to present the recommendation to the user. Other objects and advantages will be more fully apparent from the following disclosure and appended claims.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 shows an overall process flow in an embodiment of this invention.

Figure 2 shows a process flow of a model generation process in an embodiment of this invention.

Figure 3 shows a system architecture in an embodiment of this invention.

Figure 4 shows a process flow of sub-processes in an embodiment of this invention.

Figure 5 shows a process flow of sub-processes in an embodiment of this invention. Figure 6 shows a process flow of sub-processes in an embodiment of this invention.

Figure 7 shows a system architecture in an embodiment of this invention.

DETAILED DESCRIPTION OF INVENTION

It should be noted that the following detailed description is directed to a system and method for personalized shopping list recommendation and is not limited to any particular size or configuration of the system but in fact a multitude of sizes and configurations within the general scope of the following description. Referring to Figure 1 , there is shown a flow chart of a process for recommending shopping-lists to users that requires tracking and analyzing users' shopping activities in order to understand their buying behavior. This learnt knowledge is required to provide good recommendations to users in the future. The process starts by keeping track of the user's purchasing history (1). A new update to the database is performed every time a user makes a purchase. Each purchase record is tagged with a user identifier, item identifier, time and location of purchase before it is sent to the server for tracking and analysis. Based on such recorded purchase-data (historical), a process to generate the user's shopping behavior model is activated. The resulting user's shopping-behavior model (2) is regarded as closely representative of the user's habits or behavior of buying items. Such a model (2) is important to predict the likelihood for the user to purchase a particular item in the future. Predicting the most likely items desired by the user for purchase is one source of recommending a shopping list since such recommendations are made on the basis of reminding the user of what items need to be purchased at a particular store and time. Upon entering (physical/offline) or login (online) into a pre-visited store, the system is able to identify that this is the right time to request for a shopping-list recommendation. Upon receiving such request, a personalized recommended shopping list (3) will be constructed by performing inference on the user's shopping model. The inference will be performed based on the user's current context; e.g. visited store and current time. Once the recommended shopping list is constructed, it will then be returned to the requesting application as to enable it to present the recommendation to the user (4).

Referring to Figure 2, there is shown a generation process of a user's shopping behavior model of this invention. The system captures and stores features of each purchased item. The system attempts to identify whether a purchased item is store dependent, interval dependent, a seasonal buy, promotion dependent, commodity dependent, brand dependent, commodity- brand dependent, or price dependent. Such features are described in detail below based on an item called X.

Store dependent: X is classified to be store-dependent if the user only buys X at one particular store even though the user also goes to other stores. Next interval due: X is classified to be an interval dependent item if the system learns that the user always buys X between fixed time intervals, e.g. every two weeks, every 10 days, etc.

Seasonal buy: X is classified to be a seasonal buy if it is bought only for special occasions, e.g. festive seasons, school openings, etc.

Promotion dependent: X is classified to be a promotion dependent item if the system learns that it is purchased only when there is a promotion. The promotion or discount information is normally included in the purchase receipts.

Commodity dependent: Commodity here refers to a specific product class, e.g. chocolate drinks, writing instruments, clothing apparel. X is classified to be a commodity dependent item if the system learns that the user likes any items within a specific product class.

Brand dependent: X is classified to be a brand dependent item if the system learns that the user has a preference for a particular brand of items, regardless of which product class they fall under.

Commodity brand dependent: X is classified as a commodity-brand dependent item if the user only purchases X (never other brands) for a particular product class.

Price dependent: X is classified as a price dependent item only if the user purchases items that fall within a price range, whether it is the lowest price, or, for more affluent users, the highest price.

To learn all these features from purchased items, a minimal table structure as in Table 1 is required to track all purchasing records of a user. Table 1 shows the purchase records for a sample item, i.e. MiIo (a type of chocolate drinks) and for a user named Jack. The "itemlD" refers to the product universal identifier; e.g. Universal Product Code (UPC) that is printed as the product's barcode. The "storelD" and "userlD" should be unique as to identify a particular store or user. The recorded date/time is when the item is purchased. The "on_promotion" field expects a binary value of yes/no which reflects whether the purchased item was on promotion or not. The lowest_price field shows whether this particular item was sold with the lowest price in comparison to other items of the same commodity.

Table 1 : A sample table structure for storing historical purchase records

Each purchase feature, e.g. store-dependent; may have some other data associated with it. For example, if an item is classified as a store-dependent item, the associated data would be the name or identifier of the dependent store. The associated data for each purchase feature is described in Table 2. Note that the associated data is only meaningful if the item is classified to possess the respective purchase feature.

Table 2: Purchase features and their associated data

The "Promotion dependent" feature is tied to one embodiment where the system has access to promotional information and advertisements being offered at the visited store. The said external server may be a content-server where the promotional information and advertisements are tagged according to different stores. In such a case, our Recommender server requests for the promotional information or advertisement (if there are any) of the said item, specific to those being offered at the current store being visited by the user. Once the promotional value is captured, it will then be assigned as the current offering of the low-price or discount value for the said item.

The calculation to classify items based on their purchase features can be done in two ways. One solution is based on a strict 0 or 1 value assignment, where an item can either be e.g. store-dependent or not. Another approach would be to give a value between 0-1 to denote the fuzzy value of a particular purchase feature. For example, the store-dependent value of "0.8" may mean that 80% of the time, the item was bought at Store A, while the rest of the time it was bought at other stores. For such approach, the fuzzy value can easily be captured by calculating the probabilistic value of the respective historical purchasing data.

Having presented the method of modeling user's shopping behavior model, Figure 3 shows how the method for recommending a shopping list is implemented (in the form of a system architecture). The Shopping Behavior Modeler (10) performs the user's shopping-behavior modeling based on the data extracted from the i) User purchasing records database (11), ii) User profiles database (12) and iii) Products and Promotions database (13). The resulting models are stored in the User shopping behavior models database (14). The Products and Promotions database (13) can be external databases that can be accessed from this system. Such databases contain all products and promotional information offered at different stores / locations. A recommended shopping list is created based on past purchasing behavior of a user. The past purchasing behavior of a user is represented as the user's shopping-behavior model. The method of this invention for constructing a personalized shopping list is by performing reasoning on the generated user's shopping-behavior model (from 14) once requested.

The recommended shopping-list construction process is initiated once a request is received from a client application; e.g. a context-aware application, a browser application, etc. The request consists of the user and visited-store identifiers. Upon reception of such request, the server performs a reasoning process on the user's shopping-behavior model based on the given store identifier and current time/date. Once a recommended shopping list is constructed, it will be returned to the requesting application that will present the recommendation to the user.

The reasoning process requires the following data prior to its execution: i) User's shopping-behavior model (stored on the server)

ii) Store identifier (from the request)

iii) Current time/date (from the request)

iv) Products & Promotions database

The user's shopping-behavior model is retrieved from the server's datastorage based on the given user identifier; i.e. included in the request- message from the user's mobile terminal. The store identifier is also captured from the same request-message while the current time/date is based on the server's current time/date. The reasoning process alone is further divided into sub-processes as shown in Figure 4. The sub-processes include i) extracting purchase-due or in-season items (20), ii) Extracting best-value items (21) and iii) sort the recommended items based on priority (22).

Extracting purchase-due or in-season items.

The process flow of extracting purchase-due or in-season items (20) is illustrated in Figure 5. It is dependent of the following three purchase features:

i) Next-interval-due

ii) Seasonal buy

iii) Store-dependent

Based on these features, we can see that there are two ways of determining that an item (i.e. purchase-due item) needs to be purchased based on time, i.e. firstly if the item has reached or is almost reaching its purchase-due (related to the next-interval-due dependent feature), or secondly if the current date matches a particular season or special occasion date (related to the Seasonal buy feature).

The purchase-due value of an item is calculated as follow: purchaseDue = (currentDate - lastBuvDate) ^* 100

purchaselnterval

The variables are defined as follow:

• currentDate: today's date

• lastBuyDate: the date on which the item was last bought

• purchaselnterval: the normal (average) purchase time-interval between two purchases, e.g. 7 days. The higher the purchaseDue value of an item denotes the closer that item is to its purchase-due date. This calculation may be performed automatically at every 12:00 am by the system as to update each item's purchaseDue value in daily manner.

As for seasonal buy items, the in-season items are determined firstly by identifying the current seasons or special occasions. This is done by determining if any specified events fall on the current date, in which case said events are entered into the user's calendar. The user's calendar is a part of the user's static profile, which is stored in the user profiles database. Once the respective season/event (seasonX) is identified, the system will extract all seasonal items that have seasonX as their related-season.

Once the system has captured all purchase-due and in-season items, the filtering process based on the currently visited store is then performed. If item X is store-dependent, then the dependent-store [where the item should be purchased (storeX)] is compared with the user's current location

(visitedStore). If both values matched, then item X is then identified as commodity or brand dependent. If item X is commodity dependent, then item X is added to List-A together with other items-on-sale of the same commodity. If item X is identified to be a brand-dependent item, then it is added to List-C together with other items-on-sale of the same brand. If item

X is identified neither as a commodity or brand dependent item, then only item X is added to List-A. In the case where item X is not store-dependent, then it is processed just as if item X is store-dependent and the user's current location matches with item X's purchase store. Extracting best-value items.

A best-value item is defined as an item that is on discount or promotion. In constructing shopping-list recommendation based on such an attribute, the scope is extended not only to cover a specific item but broader product categories, e.g. commodity and brand. The related item's purchase features are as follows:

i) Next-interval-due;

ii) Store-dependent;

iii) Promotion dependent;

iv) Commodity-brand dependent;

v) Commodity dependent;

vi) Brand dependent; and

vii) Priced-ranged dependent. We are still referring to next-interval-due dependent and store-dependent purchase features since these are the main filtering measures of the recommended items since we want to ensure that:

• If an item is a next-interval-due dependent item, then it will only be recommended when the item has reached or almost reaching (e.g. > 80%) its purchase-due, but in the case where there is a promotional offer for the item, then the purchase-due measure is made more flexible where all promotional items that have purchase-due value e.g. > 50% are eligible for recommendations.

• If an item is a next-interval-due dependent and store-dependent item, then it will be recommended only when the user is visiting the right store.

• If the item is either (next-interval-due dependent, store-dependent and promotion-dependent) or (next-interval-due dependent and promotion-dependent item), then the next-interval-due dependent measure is not taken into account since the decision of buying the item is mainly reliant on the availability of a promotion.

The general flow diagrams adopted in extracting the best-value items are illustrated in Figure 6, which shows a filtering and matching process that affects in general all next-interval-due dependent items that have purchase-due values e.g.≥ 50%. Once retrieved, these next-interval-due dependent items will be filtered to store-dependent and non-store- dependent items. All store-dependent items will be processed in such a way that only items that have a match between the dependent store (storeX) with the one being visited by the user (visitedStore) are allowed to proceed. While all non-store-dependent items skip such filtering process. The remaining processes are basically filtering items according to their classes and extracting related promotional information, in the following sequence:

i) For a (next-interval-due dependent, store-dependent and commodity- brand-dependent / promotional-dependent) OR (next-interval-due dependent and commodity-brand-dependent / promotional- dependent item), get all promotional information specific to itemX applicable to visitedStore; ii) For a (next-interval-due dependent, store-dependent and commodity- dependent) OR (next-interval-due dependent and commodity- dependent item), get all promotional information for any items categorized in the same commodity-category as itemX, which are applicable to visitedStore ; iii) For each (next-interval-due dependent, store-dependent and brand- dependent) OR (next-interval-due dependent and brand- dependent), get all promotional information for any items categorized in the same brand-category as itemX, which are applicable to visitedStore; iv) For each (next-interval-due dependent, store-dependent and low- cost-dependent) OR (next-interval-due dependent and low-cost- dependent items), get all promotional information for any items categorized in the same commodity-category as itemX, which are applicable to visitedStore; and then compare the promotional price of the item with the lowest-price recorded. If the promotional price offered is the same or lower than the lowest-price recorded, then add the promoted item to the recommendation. The current lowest-price is then updated according to the current offered lowest price.

For these processes (i-iv), if there is related promotional information returned by the advertisement server, then this information is added to the recommendation list. All processes being explained in this section refer to the sub processes being performed in the process of "extracting best-value items" (21) (see Figure 4). Following this process is the "sort items based on priority" (22) process, which simply sorts the recommended shopping list based on priority. We opt to have List-A (from Figure 5) to appear first in the final recommendation list followed by List-B and List-C (from Figure 5-6). The rationale of prioritizing these lists in this order (List-A, List-B, List-C) is due to their significance as described below:

■ List-A: This list contains all items that are known to be of interest to user since the user has always buy this item in the past and these items are identified to be urgent as they have reached their purchase- due or are currently in-season.

^■ List-B: This list contains all promotional items that are categorized under commodities that are known to be of interest to user and these items are less urgent.

^■ List-C: This list contains all promotional items that are categorized under the brands that are known to be of interest to user and these items are less urgent. One embodiment in which the method of this invention can be applied is illustrated in Figure 7. The userlD (30), storelD (31) and date/time (32) are the inputs required to be supplied by the client application when requesting for a recommendation. These inputs reflect the user's current context. The User profiles database (12), User shopping behavior models database (14), and Products and promotions database (13) are required for the reasoning process in constructing the recommendation. Upon request, the Personalized Shopping List Recommender (40) will construct the personalized shopping list (50), which will be returned to the requesting application to be presented to user.

While several particularly preferred embodiments of the present invention have been described and illustrated, it should now be apparent to those skilled in the art that various changes and modifications can be made without departing from the spirit and scope of the invention. Accordingly, the following claims are intended to embrace such changes, modifications, and areas of application that are within the spirit and scope of this invention.

Claims

1. A method of providing purchase recommendations to a user, comprising the steps of:

a. generating a purchasing behavior model for the user; b. receiving a request to provide purchase recommendations; c. generating purchase recommendations (50) by performing a reasoning process on said model; and

d. providing to the user said purchase recommendations (50).

2. A method of providing purchase recommendations according to claim 1 wherein the said model of the user's purchasing behavior is generated based on preferences of the user.

3. A method of providing purchase recommendations according to claim 2 wherein the said preferences are manually captured during a registration of the user.

4. A method of providing purchase recommendations according to claims 2 or 3 wherein the said preferences are automatically captured by tracking and analyzing the user's purchases at physical and virtual stores.

5. A method of providing purchase recommendations according to claim 4 wherein the said tracking and analyzing the user's purchases comprises identifying whether the products purchased by the user fall into any one of the following categories:

next interval due;

seasonal; store dependent;

promotion dependent;

commodity-brand dependent;

commodity dependent;

brand dependent; and

price-dependent.

6. A method of providing purchase recommendations according to any of the preceding claims wherein the said purchase recommendations are dependent on the user's current context, e.g. visited store, current time and current date.

7. A method of providing purchase recommendations according to any of the preceding claims wherein the said purchase recommendations are dependent on any promotions associated with the products captured in the user's purchasing behavior model.

8. A system of providing purchase recommendations to a user, comprising:

generating a purchasing behavior model for the user;

receiving a request to provide purchase recommendations;

generating purchase recommendations (50) by performing a reasoning process on said model; and

providing to the user said purchase recommendations (50).

9. A system of providing purchase recommendations according to claim 8 wherein the said model of the user's purchasing behavior is generated based on preferences of the user.

10. A system of providing purchase recommendations according to claim 9 wherein the said preferences are manually captured during a registration of the user.

11. A system of providing purchase recommendations according to claims 9 or 10 wherein the said preferences are automatically captured by tracking and analyzing the user's purchases at physical and virtual stores.

12. A system of providing purchase recommendations according to claim 11 wherein the said tracking and analyzing the user's purchases comprises identifying whether the products purchased by the user fall into any one of the following categories:

next interval due;

seasonal;

store dependent;

promotion dependent;

commodity-brand dependent;

commodity dependent;

brand dependent; and

price-dependent.

13. A system of providing purchase recommendations according to any of claims 8 through 12 wherein the said purchase recommendations are dependent on the user's current context, e.g. visited store, current time and current date.

14. A system of providing purchase recommendations according to any of claims 8 through 13 wherein the said purchase recommendations are dependent on any promotions associated with the products captured in the user's purchasing behavior model.