An effective mobile solution for grocery list-creation process
INTRODUCTION
Software solutions based on mobile platforms have become very popular in today’s technology world in a massive scale [1] and are used in vivid range of world domains. Not only mobile applications have become a lifestyle [2] but also the technology related is evolving by the day and demonstrating the promising potential to undertake most of the human technological necessities in a comprehensive manner. Similarly, data mining is also a broad concept which is employed in almost every field in the world [3], heavily in data sciences. Whilst applying the practicality of data mining on sales and consumer data is prevalent, in identifying purchase and interest patterns, association and association rules, is widely taken into consideration. Although a substantial amount of applications have been produced in various mobile platforms for online grocery shopping [4][5] and comprehensively data mining techniques have been applied on sales data for interest mining, the requirement has always been present for a solution to enhance static grocery shopping experience of people. https://codeshoppy.com/android-projects-titles-ieee.html Up to the present time people use traditional ways of carrying out these activities such as writing on a piece of paper [6] or memorizing items to be bought [7] which are not reliable and eventually leads to time and money loss. Coming up with a solution which has a capacity of properly addressing the above issues would be much beneficial for the users in terms of both financially and lifestyle improvement. This research presents our work on a mobile-assisted software application namely ‘The Smart Shopping List’ for enabling the users to enrich their grocery shopping experience in to a new level. It allows the users to create shopping lists, find the ideal shop using geolocation services, receive item recommendations filtered using pattern matching recognition with higher level of accuracy using association algorithms, and share shopping lists via text-to-app service. The experimental runs and the comparisons with Weka software have shown promising results in recommending items to the user.
RELATED WORK
There have been several technological developments in mobile-assisted shopping in the past few decades. Today’s reach of mobile smartphones and the availability of the mobile development technology have certainly contributed these innovative advancements. Below are some notable studies and improvements carried out on mobile-assisted shopping-lists creation. A.The Hybrid Shopping List Heinrichs F., Schreiber D. and Schöning J. under the patronage of Prof. Dr. Antonio Krüger, worked on a project to create a prototype for a hybrid mobile application combining the advantages of paper and electronic shopping lists [11]. An initial study was carried out to understand the creation and usage of paper based shopping lists. Then a secondary investigation was made to identify the properties of electronic shopping lists. Findings which were extracted from the initial study put together with the secondary analysis, a functional matrix was created to design the prototype. Prototype was a system which uses Client-server architecture with a Mobile GUI. Users would write the grocery list using a digital pen and the activity would be captured and be available at the mobile GUI. The project laid paved the way for mobile assisted grocery shopping creation process, however further studies were much required to investigate how many other modalities than pen and paper can be applied and facilitated in the shopping list creation process. B.Intelligent Shopping List Marcus Liwicki and the three team members (Sandra Thieme, Gerrit Kahl, and Andreas Dengel) developed a system which automatically extracts the intended items for purchasing from a handwritten shopping list [12]. This intelligent shopping list relies on a categorization of the products which is provided by the supermarket. The system identifies handwritten items in a shopping list by the use of a digital paper. Using the data transmitted to computer the handwritten data is understood by matching the data against an ontology. Promising results were shown however since the ontology is provided by the store and very specific and narrow, it has to be enlarged so that the users are able to any items they prefer. C.Multimodal shopping lists Another interesting development would be the work concluded by Jain and the team regarding a developing a prototype for creating a shopping list from multiple source devices like desktop, smart phones, landline or cell phone in different formats, essentially structured text, audio, still images, video, unstructured text and annotated media [13]. An evaluation was done with 10 participants in two week. Their goal was to further analyze the shopping list creation and management process. Based on their findings they give the recommendations to develop interactive features for the systems made for managing shopping lists. D.Grocery Retrieval System & Mobile services Similar to Marcus Liwicki’s study, Nurmi and the group introduce a product retrieval system that maps the content of shopping lists written in natural language into the relevant real world products in a supermarket [14]. The system was developed having shopping basket data as the base which they had gathered from a large local Finnish supermarket. Furthermore the new architecture designed by Wu H. and Natchetoi Y. enabling efficient integration between mobile phone applications and Web Services with the help of XML compression features [15]. Using this architecture, they have implemented a mobile shopping assistant which has multiple input modes such as camera, voice and Bluetooth. While there are still more promising work to be done to further improve the framework, they conclude their study with their plan of releasing the work to the public as a generic library. Interestingly, most of the applications focused on improving the input method to effectively create a shopping list, while none of them have taken a look at the broader view on the shopping list creation process and the consumer patterns to further enrich the shopping list creation experience, such as applying data mining techniques for interest mining.
DATA MINING ON PURCHASE INTERESTS
Statistical studies often find that there could be a significant relationship between the customer behavior and the items that he purchases. Raicu and the team has done a study to understand the customer preferences on products physical characteristics using data mining [16]. Their data mining was done using 3 approaches; Clustering with K-means, Association via Fuzzy and LSA, and Visualization with MDS. Findings show that the data mining does offer the industry world with some certainty to leverage sales risks and to focus on targeted marketing to maximize profits. Additionally, the two Guptas, namely A.K Gupta and Chakit Gupta have conducted another study to identify the importance of the role played by data mining in the field of financial and sales domain [17]. Their learnings converse about the tactics to be used as data mining techniques to analyze customer data. Classification, Regression, Link Analysis and segmentation were taken in to account. Followed by a case study which differentiate the traditional way of marketing and data mining approach they conclude their study. A.Association Rule Mining (ARM) Raorane A. has done a very comprehensive study with Kulkarni R.V. to identify the consumer behavior, his psychological state at the time of purchase and how data mining method can be applied to improve ordinary method [18]. A similar study on evaluating data mining techniques to find purchase items association towards the customer had previously been made by Watada and Yamashiro [19]. Above studies clearly emphasize that association rule mining would be the most preferred way of data mining techniques to be applied to uncover the consumer behavior and their correlations to their purchases. B.Apriori Algorithm While there are several association algorithms that are available, selecting an appropriate one for the proposed solution is at the utmost importance. Two Czech researchers namely Turþínek and Turþínková had explored the use of association rules in determining the consumer behavior [20]. They carried out a survey with 1127 individuals which would be a close representation to Czech population to identify problems of shopping for meat products. During the data analysis, they use 2 methods for generating ARMs; Apriori algorithm and FP-grow algorithm which both of them run in Weka software. Results were more favorable towards the Apriori algorithm as it had provided finer data and, the data had to be reduced to extreme values prior to be fed to the FP-Growth since the algorithm works only with binary input data. Similar comparison has been made by B. YÕldÕz and B. Ergenç in comparing the two association rule mining techniques [21]. They used FP algorithm and Matrix Apriori another enhanced version of basic Apriori algorithm since the native implementations suffers from bottlenecks in its candidate generation phase as it requires multiple passes over the source data. Two case studies were carried out using two datasets per algorithm to see their performance over datasets with different characteristics and the causes for the performance differences. Overall Apriori outperformed FP-Growth in total performance and finding item sets were faster as well Adding further value, Prof. Venkatachari has done an in-depth case study of a Mumbai retail store to do a market basket analysis using FP growth and Apriori algorithm [22]. The main objective of the research was is to see how different products in a grocery store inter-relate. The algorithms were used to find frequent items using rapid miner and R programming. Proving the above studies, FP- Growth showed poor performance results in both the tools which bring us to the conclusion that native Apriori algorithm is ideal for identifying customer purchase patterns using sales data in a small retail business. Another advantage of Apriori algorithm is that it calculates more sets of frequent items which is very advantageous especially when the database is smaller and generating more frequent items is always better if more suggestions are to be made.
EVALUATION
A.Algorithmic Analysis A test criteria of 50 items were considered with 20 grocery lists with 140 item instances. The test cases were designed accordingly by changing values of the algorithm input parameters (Min. sup and Min. Conf.) and observing the outputs. Out of 5 test cases, min. support = 10% (0.1) and min. confidence = 70% (0.7) produced sufficient amount of results were favorable when compared with WEKA software outputs.
Eggs,Butter=>0.2,
confidence = 1
Milk,Rice=>0.15,
confidence = 1
Rice,Coconut oil=>0.15,
confidence = 1
Milk, Tea=>0.25,
confidence = 1
Milk,Biscuits=>0.15,
confidence = 1
Tea,Biscuits=>0.15,
confidence = 1
Biscuits,Bread=>0.15,
confidence = 1
Butter, Bread=>0.25,
confidence = 0.83333333333333
Soap,
Toothpaste=>0.15,
confidence = 1
Soap,Toilet paper=>0.15,
confidence = 1
Soap, Toilet paper=>0.15,
confidence = 1
Toothpaste,
Toilet paper=>0.15,
confidence = 1
Coconut oil,
Papadam=>0.15,
confidence = 1
Chips, Pepsi=>0.15,
confidence = 1 B.
Performance and Security Analysis Performance and security aspects were also considered from the design and throughout all stages of the development of the solution. Application weighs roughly around 25MB. This allows the application to be installed on devices without needing large storage requirements. Data usage for a session of 30 days with moderate app usage was only 3.55MB, which 128
is a good news for the users who have limited data plans. The system design decisions taken to make use of server-side query processing and the use of JSON data has undisputedly helped in achieving these numbers. Average memory usage was also at a healthy level of 75MB. The Smart Shopping List requires a few device permissions; Location for geolocation services by the mapper module and send SMSs by BringMe. The application requires no further permissions and this makes the app desirable for the users who are concerned about their privacy. C.Usability Analysis A usability analysis for the Smart Shopping List has also been conducted to ensure that the software application is built in a way that the usability of the application has been properly maintained. This has been achieved by a feedback taken after a 3 weeks of application usage by 17 individuals. Selected 9 were from 22-40 years that ranged from an equal distribution of genders, and 7 picked from 40-50 years range with a distribution of 5 females and 2 males. One female was taken from the age group of 50 to 65. All the participants except the last had an average prior knowledge in mobile applications. Participants who were 40+ were novice to average smartphone users. According to the feedbacks given, over 76% were happy with the design of the application which let us make the direct judgement that the design allowed the users to get their intended activities done smoothly. One person who had difficulty understanding the application content was a novice smartphone user and was not very confident with mobile applications, although she found the application concept is useful, also the mapper component was the most popular module. This suggests that participants would be more benefited from the geolocation services than others however majority voted for this module were millennials. Item suggestions and BringMe! were popular among other age groups which proved to be more extensive grocery shoppers. The algorithm is proved to be functioning well and the application provides a great use to majority of the participants to carry out their grocery shopping and demonstrating an overall approval net score of 76% with more than 82% positive ratings for a population of 17 participants.
CONCLUSION
In this paper we have described the mobile solution, The Smart Shopping List for enhancing people’s grocery list creation process. As future work the native ARM algorithm can be improved to a more subtle version such as Matrix Apriori. Furthermore, stronger filtrations can be made to strengthen the algorithm by introducing interestingness measures such as Kappa or J-measure. It is further suggested to implement user models of preferred items and used abbreviations so that in a scenario, the application might be able to understand which specific kind of milk (brand, quality) is intended to be bought when the user just writes “milk”. Lastly, enabling the Smart Shopping List on other mobile platforms such as growing IOS base and windows phone would also be beneficial for non-android users.