Portal de Periódicos da CAPES

Handling consumer vulnerability in e-commerce product images using machine learning

2022; Elsevier BV; Volume: 8; Issue: 9 Linguagem: Inglês

10.1016/j.heliyon.2022.e10743

2405-8440

Sarvjeet Kaur Chatrath, Gautam Batra, Yogesh Chaba,

Cybercrime and Law Enforcement Studies

NeedIn recent years, secondhand products have received widespread attention, which has raised interest in them. The susceptibility issues that consumers encounter while buying online products in reference to the display images of the products are also not well researched.MotivationRetailers employ clever tactics such as ratings, product reviews, etc., to establish a strong position thereby boosting their sales and profits which may have an indirect impact on the consumer purchase that was not aware of that retailer's behavior. This has led to the novel method that has been suggested in this work to address these issues.Proposed methodologyIn this study, a handling method for reused product images based on user vulnerability in e-commerce websites has been developed. This method is called product image-based vulnerability detection (PIVD). The convolutional neural network is employed in three steps to identify the fraudulent dealer, enabling buyers to purchase goods with greater assurance and fewer damages.SummaryThis work is suggested to boost consumers' confidence in order to address the issues they encounter when buying secondhand goods. Both image processing and machine learning approaches are utilized to find vulnerabilities. On evaluation, the proposed method attains an F1 score of 2.3% higher than CNN for different filter sizes, 4% higher than CNN-LSTM when the learning rate is set to 0.008, and 6% higher than CNN when dropout is 0.5. In recent years, secondhand products have received widespread attention, which has raised interest in them. The susceptibility issues that consumers encounter while buying online products in reference to the display images of the products are also not well researched. Retailers employ clever tactics such as ratings, product reviews, etc., to establish a strong position thereby boosting their sales and profits which may have an indirect impact on the consumer purchase that was not aware of that retailer's behavior. This has led to the novel method that has been suggested in this work to address these issues. In this study, a handling method for reused product images based on user vulnerability in e-commerce websites has been developed. This method is called product image-based vulnerability detection (PIVD). The convolutional neural network is employed in three steps to identify the fraudulent dealer, enabling buyers to purchase goods with greater assurance and fewer damages. This work is suggested to boost consumers' confidence in order to address the issues they encounter when buying secondhand goods. Both image processing and machine learning approaches are utilized to find vulnerabilities. On evaluation, the proposed method attains an F1 score of 2.3% higher than CNN for different filter sizes, 4% higher than CNN-LSTM when the learning rate is set to 0.008, and 6% higher than CNN when dropout is 0.5. 1. IntroductionBe it irregular, occasional, or irreversible, vulnerability is a problem influencing everyone at any stage as it will become erratic as the time evolves. A consumer who is vulnerable will experience damage as a result of their unique situation, especially when a business is not acting with reasonable standards of care. Vulnerable personal circumstances can also include, but are not limited to, long-term or short-term enduring health problems (physical or mental), emotional trauma or neglect, having a physical impairment, having trouble understanding English or having weak language skills, struggling with addiction, or taking care of someone in multiple situations.The most common choice for businesses is an e-commerce website created by a technology company because it offers a wide range of online purchasing and selling options. In comparison to a standard website, which is typically used to verify and gather information, e-commerce website platforms enable the consumer to make purchases without visiting a physical store. Shopping on e-commerce websites has several benefits for customers because they offer a big selection of things at fantastic discounts. The consumers' need for the best website production firm for e-commerce will be a crucial step. In e-commerce, the marketing firm acts as the primary customer representative for the business.Vendors and the sellers on E-commerce websites had to be very careful and pay attention to attract customers for their products displayed online thereby preventing the loss as the disruptive attitudes and the imbalances may impact the business greatly. The new customers will get know about the quality and the reliability of the products by exploring the feedbacks and ratings given by the customers for the products which will, in turn, increase the business sales. The volume and nature of online payment network protection attacks have increased in parallel with the massive expansion in electronic payments.Common third-party modules used by websites, including shopping cart software, have been identified as vulnerable to these kinds of attacks. These growing web program vulnerabilities, SQL link insertion, and cross-site scripting are used by many attacks. Price manipulation, SQL injection, information disclosure, path disclosure, cross-site scripting, and buffer overflow are some of the various kinds of vulnerabilities. Secondhand product consumption has expanded as a result of online auction and trade platforms. Secondhand product consumption has been performed for decades and is defined as the reusability of old things without losing their principal function.The possibility of secondhand online trading mainly lies in the possibility of extending the product's life span thereby ignoring extra environmental stresses due to the purchase of newer products. Some European countries had a long history of using second-hand products. The practice of purchasing secondhand or used products from the United Kingdom, for instance, is deeply rooted in the society. Focusing on the current financial crisis, individuals from several nations, including Spain and France, have successfully entered the second-hand market due to various economic causes.There are conventional indications, such as socioeconomic class, that may discourage some groups from engaging in the purchase of used second-hand goods. However, the current state of the internet and its associated technologies has led to the creation of numerous new electronic gadgets that offer consumers the best options for shopping and making purchases. The market for used goods has been popularized by the use of social networks and smartphones by all socioeconomic groups in society, and as a result, demand for used second-hand products is growing daily. Thus, it is crucial to focus on the variables affecting the purchasing of used goods.According to a survey (Carufel, 2017Carufel Richard Consumers Say Visuals Are Most Important Factor in Online Shopping.2017https://www.agilitypr.com/pr-news/public-relations/consumers-say-visuals-important-factor-online-shopping/Google Scholar), which asked U.S. consumers about their preferences when making purchases online, many respondents preferred the consistency of a company's online promotional images over other factors like social networking, branding, or product specification. Innovative imaging technologies have come into the spotlight as a result of the expanding market need for high-quality images and a variety of product shots. 47 percent of American internet users said that high-quality product photos are important when deciding whether or not to buy a particular brand.The standards and image of a company are now represented by digital marketing, thus firms are required to provide the highest quality online images in order to move clients through the purchase process. This is because brands will give advertising imagery precedence over social network posts. One third, or 37.9%, of online shoppers in the United States said that outside factors have an influence on their decision. Another third, or 33.7 percent, however, thinks that it might have an impact. As a result, retailers and marketer will concentrate on that as well as enhance their online marketing branding, which has been shown to increase the additional sales.Additionally, 50.5 percent of online shoppers in the United States like to check the product's photographic images (taken from all angles) before making a purchase. Consumers were used to measure the quantity of online promotional images as well as the accuracy of the images, according to the research. Even the decision of when to tap "purchase" by the customer may be significantly influenced by adding the product or services to a model in the product picture. The study's findings show that brands shouldn't rely solely on real-world photos, and website photography tends to increase the likelihood that customers will shop online.However, there are no studies that specifically address the risks that customers take when making purchases based on online product photos. Hence based on the Machine learning method, the proposed system is developed to prevent high false-positive rates thereby improving the reliability of the consumers. The suggested system encourages consumers to buy goods or services with greater assurance and without feeling vulnerable.The objectives of the proposed method are given below:•While e-commerce is still expanding, retailer competition for sales and revenue has also been growing significantly. As a result, most retailers today have adopted innovative tactics to take advantage of e-commerce websites and to strengthen their positions. sn addition to code modification, the risky technique also involves click farms, and false customer reviews and ratings. Retailers employ these smart tactics, along with others, to boost their earnings and capture more market share.•Consumers on e-commerce websites are extremely vulnerable. A significant portion of consumers are impacted by the character, quality, or risk profile of a product, which has grown increasingly complex. A method based on machine learning and image processing has been presented for accurately identifying product damage in order to get over these drawbacks.The contributions of this work are:•To identify risk factors for consumers when buying things online by looking at the provided product photos.•A machine learning-based methodology called the Product image-based vulnerability detection (PIVD) approach is suggested to address reused product image-based consumer vulnerability. In this proposed method, Convolutional Neural Network (CNN) is also utilized.•Four alternative methods were applied to the displayed product pictures to determine the efficacy of the proposed (PIVD) method. The dealer's products and their vulnerability are also determined using this way.2. Literature surveyThe previous research on used goods, image analysis, machine learning, and emotion detection approaches is covered in this part.2.1 Secondhand products(Kim et al., 2017Kim Dae-Kyung Lim Jae-Hak Park Dong Ho Optimization of post-warranty sequential inspection for secondhand products.J. Syst. Eng. Electron. 2017; 28: 793-800Crossref Scopus (6) Google Scholar) discovered the ideal performance level to improve the second-hand product for every inspection and to address the anticipated maintenance costs throughout the product's life cycle from the perspective of the user. They derived an explicit method for calculating the routine maintenance costs incurred over the product's life cycle from the given cost structures. Additionally, they examined the method for determining the ideal stage of development when failure rates followed the Weibull distribution.On the other hand, a cost model was presented by (N.Darghouth et al., 2015Darghouth M.N. Chelbi Anis Ait-Kadi Daoud On Reliability Improvement of Second-Hand Products". Elsevier, 2015: 2158-2163Google Scholar) to identify the reliability enhancement level for second-hand products that are acquired with a free warranty repair (FRW) form. For the purpose of enhancing the dealer's cost reliability, they examined the periodic PM. This model aids in determining whether or not the PM behavior used during the warranty cycle is advantageous from the perspective of the dealers. They consider choosing the best update standard while forgoing routine preventive maintenance (PM) during the warranty period in order to reduce the likelihood of early defects.(Krishna et al., 2014Krishna Dr.V. Mohan P.l. Padmaja A study on customer attitude towards reuse goods shop: a case study with respect to hard off corporation, Japan.Eur. J. Bus. Manag. 2014; 6: 34Google Scholar) evaluated customer perception with a study of 50 people from various 5 local stores in and near Tochigi Prefecture, Japan. From the study, it was found that the buyers were interested in purchasing the used products, and also, pricing has been identified as fair, as the customers will not experience blotting or social guilt while buying those used products. This paper also explored the cause for consumer loyalty in those shops and noticed that this store shouldn't be wrong for antique merchandise and artifact shops.Also (Valerie and Thomas, 2003Valerie M. Thomas Demand and dematerialization impacts of second-hand markets.J. Ind. Ecol. 2003; 7Google Scholar), addressed secondhand business development and the competition for new items has been decreasing as there are surplus products available that can be sold on the market. As there is no steady availability of existing used goods, there will be rising competition amongst second-hand buyers to buy fresh items which will also boost the materials cost. Moreover, if second-hand sales reduce the supply of new goods, this will not be generally one-for-one. When the price of recycled items exceeds the purchasing of fresh merchandise, this will be a simple beneficial feature given by recycled merchandise vs. new ones (Hristova, 2019Hristova Y. The second-hand goods market: trends and challenges. Izvestia journal of the union of scientists - varna.Economic Sciences Series. 2019; 8: 62-71Google Scholar). investigated some of the major trends in the second-hand goods market, along with their causes and impacts over retail in the digital society.In the presented project, a conventional fuzzy reasoning tool was employed (Ghosh et al., 2021Ghosh S. Thang D.V. Satapathy S.C. Mohanty S.N. Fuzzy rule based cluster analysis to segment consumers' preferences to eco and non-eco friendly products.Int. J. Knowl. Base. Intell. Eng. Syst. 2021; 24: 343-351Google Scholar). developed a fuzzy rule-based system for decision-making based on Perceived Environmental Knowledge, Perceived Environmental Attitude, and Green Purchase Behavior among consumers who are related to Eco-friendly products. Green purchase behaviour (GPB) perceived environmental knowledge (PEK), and perceived environmental attitude (PEA) were chosen as fuzzy input variables, each of which has a set of five language factors. Consumers who are "Eco friendly" or "Non-ecofriendly" were divided into two sets in the output.2.2 Identification of fake reviewsNave Bayes Classifier, Logistic Regression, and Support Vector Machines were the classifiers utilised in this research by (Kolli et al., 2015Kolli Shivagangadhar Sagar H. Sathyan Sohan Vanipriya C.H. Fraud detection in online reviews using machine learning techniques.Int. J. Comput. Eng. Res. 2015; 5: 52-56Google Scholar) to investigate and determine whether the review is truthful or untrue. To build a model to identify false reviews, they have used linguistic variables including the existence of unigrams, the frequency of unigrams, the presence of bigrams, the frequency of bigrams, and the length of the review. The Yelp Challenge Dataset has been used by this research to locate the phone reviews since it provides both linguistic and behavioural data that can be used to identify fraudulent reviews.Also (Rodrigo et al., 2019Rodrigo Barbado Oscar Araque Iglesias Carlos A. A framework for fake review detection in online consumer electronics retailers.Elsevier. 2019; 56: 1234-1244Google Scholar), suggested a system for identifying the false feedback measured in the consumer electronics industry and for classifying the fake reviews in the consumer electronics domain, they have built a dataset for the classification of fake reviews in four various cities centered on scraping methods and defined a feature structure for the detection of fakes. For classifying reviews, they have also established a fake system depending on the suggested structure. They have obtained an F-Score of 82% on the classification function and as per the Friedman test; the Ada Boost classifier was seemed to be effective through statistical means.Ranking systems that are frequently harmed by profile injection attacks or anomalous scores caused by collective suggestion processes were addressed by (Yang et al., 2018Yang Zhihai Sun Qindong Zhang Beibei Evaluating prediction error for anomaly detection by exploiting matrix factorization in rating systems.IEEE Access. 2018; 6: 50014-50029Crossref Scopus (8) Google Scholar). The primary issues attackers constantly running into are introducing malicious profiles that strongly score a particular item or injecting malicious profiles that tend to lower an item's popularity. Due to this, the vulnerable client encounters numerous issues when they believe fake ratings on e-commerce websites. In order to get around the challenging challenges of calculating similarity and extracting features, a method for spotting anomalous ratings or attacks has been created.In addition (Liu et al., 2017Liu Yuhong Zhou Wenqi Chen Hong Efficiently promoting product online outcome: an iterative rating attack utilizing product and market property.IEEE Trans. Inf. Forensics Secur. 2017; 12: 1444-1457Crossref Scopus (6) Google Scholar), incorporated a quintile regression model for analyzing the important variables in online consumer preferences and exposed the promotional impact on the sales outcomes of the goods. Such effects are measured not only by the ability of the intruder to exploit but also by the unique properties of the desired commodity and the self-exciting force of the sector. Motivated by these findings, a novel iterative rating attack was formulated, and its efficiency is also validated through the experiments. They also researched and listed the economic effect of various influencing factors on product sales/download.2.3 Image analysis and emotion detection techniquesNew neural network models that combine the conventional bag-of-words, word context, and user emotions have been proposed by (Hajek et al., 2020Hajek Petr Barushka Aliaksandr Munk Michal Fake Consumer Review Detection Using Deep Neural Networks Integrating Word Embeddings and Emotion Mining". Springer, 2020Crossref Scopus (49) Google Scholar). These models specifically use three sets of features to attempt to understand at the text level: (1) n-grams, (2) phrase embedding, and (3) various lexicon-based emotion signals. The misleading feedback has been divided into four categories using a high-dimensional classification. The proposed approach has been contrasted with the other methods to show the effectiveness of classification.(Budhi et al., 2021Budhi G.S. Chiong R. Pranata I. Hu Z. Using machine learning to predict the sentiment of online reviews: a new framework for comparative analysis.Arch. Comput. Methods Eng. 2021; 28: 2543-2566Crossref Scopus (17) Google Scholar) suggested a novel framework for measuring the ratings of online reviews using machine learning techniques. They found that a number of texts preprocessing methods, including negation word identification, word elongation correction, and part of speech lemmatization paired with Terms Frequency and N-gram words, can improve accuracy. Additionally, they showed how well the general emotion of lexicons like SenticNet 4 and SentiWordNet 3.0 can be leveraged to generate the features.A novel method for discovering de-blocking that would dynamically acquire object demonstrations from a deep learning system was presented out by (Liu et al., 2019Liu Xianjin Lu Wei Liu Wanteng Luo Shangjun Liang Yaohua Image deblocking detection based on a convolutional neural network.IEEE Access. 2019; 7: 26432-26439Crossref Scopus (12) Google Scholar). Prior to deblocking, hierarchical characteristics were analysed with a convolutional neural network (CNN) that was administered, and the best features were extracted by CNN by utilising the sliding window.The main requirement for the development of e-commerce is the achievement of online automatic product categorisation. Jia et al., 2010Jia S. Kong X. Jin G. Automatic fast classification of product-images with class-specific descriptor.J. Electron. 2010; 27: 808-814Google Scholar developed a quick supervised image classifier based on the class-specific Pyramid Histogram of Words (PHOW) descriptor and Image-to-Class distance (PHOW/I2C) by examining the characteristics of product photos. During the training phase, the local features are heavily sampled and represented as soft-voting PHOW descriptors. Following this, the means and variances of the distribution of each visual word from each labelled class are used to construct the class-specific descriptors.The growth of used goods decreases the desire for new ones, as shown by the aforementioned works, yet the warranty period is discovered to have an impact on sales and how consumers were behaving while purchasing secondhand items. The market for second-hand products is very strong. Also, it examines the difficulties customers encounter when making purchases based on falsified reviews and how machine learning models were used to identify these reviews.However, most attempts result in defects and untraceable outcomes (Karode and Werapun, 2021Karode T. Werapun W. Robustness against fraudulent activities of a blockchain-based online review system.Peer-to-Peer Networking and Applications. 2021; Google Scholar). The aforementioned papers also covered the application of machine learning techniques to sentiment and emotion analysis. In conclusion, none of the aforementioned research projects have concentrated on both image analysis of product photos and emotion recognition in product reviews. Since these works do not compare the same goods with other dealers and don't take into consideration factors like product quality, appearance, wear and tear, or quality, all of these factors were taken into account in the proposed model. These works have not accounted for the factors like quality, appearance, wear and tear of the product, and also, this work doesn't compare the same product with other dealers, and hence, all these factors were accounted for in the proposed model.3. Methodology3.1 Factors contributing to consumer vulnerabilityThe consumers are getting cheated and disappointed just by trusting the product images that are displayed on the E-commerce website and ordering online. Following are the consequences that a customer will face if they order a product just by trusting the image displayed alone.1)Quality and appearance of ProductThe quality of the product is of utmost important while purchasing a product from online Ecommerce websites as it is the only thing that helps in retaining the satisfaction and loyalty of the customers. This will greatly reduce the risks associated thereby eliminating the cost of replacing the damaged products. Today, if a consumer is not satisfied with the cost and quality of the product anytime, they are buying from an Ecommerce website, they will definitely move on to some other competitors to buy the same product.Hence, quality is what that matters everything and it must be the sole commitment made by the sellers to the consumers and those kind of quality products are said to commonly called as the premium products. On the other hand, there is a chance that the product will malfunction after being purchased because the original quality of the product cannot be determined from the product pictures that are posted on e-commerce websites. The product might not fulfill the standard as per the description viewed online. The Product size, colour, and material type are all examples of the Quality indicated here.It is not possible for the vulnerable consumers to identify the size of the property with the displayed images alone as the appearances and sizes of the image may get varied depending upon the position or the angle of the image that is taken. Also, the product color may also get varied from the original displayed image due to reasons like the lighting conditions at which they are being captured with high resolution. As a result, people rely on the product's external look to judge its quality.2)Physical damageAdditionally, e-commerce companies are being accused of selling defective and damaged goods, which accounts for the fact that more than 20% of delivered goods are returned because of physical defects. Simultaneously, few buyers will purchase things online that they cannot return because they would have done so had the sellers of e-commerce websites offered them a discount or free shipping instead (Saleh, 2018Saleh K. E-Commerce product return rate – statistics and trends [Infographic].Invesp. 2018; (available at:)https://www.invespcro.com/blog/ecommerce-product-return-rate-statistics/Google Scholar). This is taking place because customers were purchasing the product only on the basis of the product's displayed photos, despite the fact that any physical defects in the product may be hidden or not apparent. There is a possibility of a financial risk involved when the service does not meet the acceptable quality based on the vendor's online records.3)Reimbursed productsIt will be challenging to fight the urge to buy a premium quality item at an expensive "throwaway" price because some retailers or dealers give large discounts for reimbursement products. Although e-commerce businesses may have lower administrative costs than conventional retailers, this market difference wouldn't be significant when it comes to marketed items.For branded or premium goods, there will typically be a standard discount accessible. If the customers are receiving greater discounts than these, the product needs to be taken into account as suspicious product. Reimbursed products cannot be identified solely by looking at the product photos that are provided, and as there are few opportunities to test or physically examine the product before making an online purchase, customers must rely on the website's minimal information of the item.4)Increasing natural quality of the imageIn general, the quality of the product picture matters more than the actual quality of the goods because a nice image will influence customers' decisions about whether or not to purchase the product from an e-commerce website. Customers may leave the website and stop buying things if the product's image is of poor quality. As a result, vendors must be careful to provide high-quality resolution pictures that enhance the natural quality of the product image. However, in practice, sellers weren't paying attention to the quality of the product as they advertised or presented.5)Mismatched product from the displayThe product should be exactly as agreed upon or ordered on the product image presented, however occasionally customers have reported receiving a different product than what they had seen when completing the order. Customers can get the faulty goods or the wrong color instead of what they requested, which makes them unhappy with the sellers or vendors on e-commerce websites. They begin to doubt online shopping as a result of this.6)Wear of ProductThe damage that occurs naturally and eventually as a result of frequent use or ageing has been known as wear and tear of the material. It is impossible to determine how much a product has been used from the images that are provided, making it uncertain as to how long the goods will remain in good condition.7)Hiding sensitive information of the productHowever, the dealers were hiding the essential information of the displayed product images from customers. Every image provides crucial information about the goods, and high-quality photographs can enhance customer interest in and trust in the product as well as conversion rates. Consumers may be misled into purchasing a fake product when they purchase goods online because they neglect to verify such details while making purchases or placing orders.8)Additional or Extra things with the productThe goods and other products could be included in the displayed product image. This might trick customers into purchasing things. Therefore, the consumers expecting the product to look exactly as it does on the display may lead to disappointment.3.2 Handling product images-based vulnerability using 4 waysAn important part of deciding whether the products will get success or not during the test is to classify the defects. There are 4 different ways available for analyzing and finding the factors that pay way to consumer vulnerability and these ways are very useful for the consumers with the vulnerability as it helps them buying the products b checking and comparing it with the others right before ordering that product online. This would ensure the consumer with trust and confidence in buying the products online and the four ways for evaluating the product images are:•Identifying incorrectly uploaded dealer/retailer images of damaged goods.•Determining how long a product has been in use by comparing the displayed product images with those from other retailers or the related brand company.•Reviewing consumer feedback comments after making a purchase.•Analyzing past purchases or transactions with the same dealers.The aforementioned methods can be carried out manually, however manual inspection may frequent