Editorial Board

A Refereed Monthly International Journal of Management

 Examining Customer Satisfaction with Recommender Systems: Impact of Personalization and Diversification

Amrita Todarwal

Research Scholar

Faculty of Management PAHER University, Udaipur

E-Mail - amritatodarwal@gmail.com

Dr. Narendra Singh Chawda

Assistant Professor,

Faculty of Management PAHER University, Udaipur

E-Mail - chawdans@gmail.com

Abstract - The efficiency of e-commerce recommendation systems from the customer perspective is a critical area of study, particularly concerning the impact of personalized and diversified recommendations. This research investigates the implications of personalized recommendations on e-commerce platforms, focusing on customer satisfaction and efficiency. Additionally, it explores the effects of diversified recommendations on the effectiveness of recommendation systems in enhancing customer satisfaction. The exponential growth of the e-commerce market has intensified the need for streamlined shopping experiences. Personalized recommender systems have emerged as vital tools for global companies, offering tailored product suggestions based on individual preferences. These systems aim to enhance customer satisfaction, drive sales, and increase conversion rates by presenting relevant products to users. However, there is a growing concern regarding the repetitive nature of recommendations, potentially limiting customer engagement. Furthermore, while accuracy has traditionally been a key metric for evaluating recommender systems, diversity has gained significance in recent studies. Balancing accuracy and diversity poses a challenge, as maximizing one metric may compromise the other. Achieving an optimal balance is crucial for ensuring customer satisfaction and overall system efficiency. By examining the impact of these recommendation strategies on customer satisfaction and efficiency, this research aims to contribute to the ongoing discussion on improving e-commerce recommendation systems.

Keywords: E-commerce recommendation system, Personalized recommendations, diversified recommendations, customer satisfaction

Introduction 

The rapid growth of the e-commerce market, coupled with the constant influx of new products, has led customers to spend considerable time and effort selecting items. To address this, recommender systems have emerged as crucial tools for global companies like Amazon, Netflix, and Google, aiming to maintain a competitive edge.

In the realm of e-commerce recommendation systems, Personalized Recommendations, which include suggestions tailored to individual preferences such as "Recommend for you," "Top picks for you," and "Because you watched” provide significant benefits for businesses and customers, including enhanced customer satisfaction, increased conversion rates, improved customer retention, and opportunities for cross-selling and upselling.

However, a prevalent challenge emerges with personalized recommendations as they frequently propose items similar to those previously bought, underscoring the necessity for diversification (Liang Hu, 2017). For instance, if a person recently bought a smartphone charger, they're less likely to need another one immediately. Instead, they might find recommendations for a phone case or screen protector more useful and appealing.

Thus, diversified recommendations are essential in overcoming these challenges. Diversified recommendations encompass a broader range of suggestions, including "Customers who bought this also bought," "Frequently bought together," "New items for you," and "People with similar interests bought” are important in enhancing user satisfaction within e-commerce recommender systems.

While previous studies have primarily focused on improving recommender algorithm performance based on customer purchasing history or preferences, accuracy metrics have been key indicators of success. Accuracy gauges how well-predicted preferences align with actual ones. While diversity measures the variety of recommended items that customers have not previously purchased. Therefore, the accuracy-diversity matrix has become indispensable for assessing the effectiveness of e-commerce recommendation systems.

Review of Literature -

Conventional recommendation systems prioritizing personalized recommendations drive higher conversion rates by presenting customers with relevant products, increasing the likelihood of purchase. Personalized recommendations foster customer loyalty and retention by building trust and delivering tailored experiences. They also enable effective cross-selling and upselling by recommending complementary products or upgrades. A notable case study is Netflix's recommendation engine, which has significantly increased user engagement and retention by suggesting highly relevant items (FasterCapitol, 2024).

By tailoring recommendations to individual interests, these systems streamline the shopping experience, ultimately enhancing customer satisfaction and driving sales (Jaekyeong Kim, 2021). By tailoring product suggestions based on individual preferences, businesses can create a more enjoyable shopping experience, leading to higher satisfaction levels (FasterCapitol, 2024).

While personalized recommendations can make people more interested, they can also make people look at fewer different kinds of items. This shows that we need to find a way to balance making things personal with seeing lots of different items(Shin, 2024). Also, this traditional recommendation system often encounters issues such as cold start, data sparsity, and malicious attacks due to reliance on user feedback. To mitigate these challenges multiple new approaches were introduced.  HRPCS (Hybrid Recommendation Approach for Personalized Cloud Services) which utilizes a combination of user and service clustering techniques. Experimental findings validate the efficacy of HRPCS in tackling these obstacles and delivering both personalized and diversified recommendations for cloud services (Hajer Nabli, 2023).TRIER (reTrospective and pRospective Transformers for dIversified Equential Recommendation), is a novel approach for enhancing recommendation diversity in short interaction sequences. TRIER addresses the challenge of limited information by retrospectively predicting users' historical interactions and prospectively capturing their potential intents from augmented sequences. Experimental results on benchmark datasets show significant improvements in diversity and accuracy compared to state-of-the-art methods, particularly for short interaction sequences (Chaoyu Shi, 2024).

RGRec tackles the issue of filter bubbles in recommendation systems by using Generative Artificial Intelligence (GAI). It stands out by providing recommendations that are rich in context and actively promote exploring diverse content based on user preferences. This approach follows principles of libertarian paternalism and transparency, setting it apart from traditional recommendation systems. RGRec represents a significant advancement in responsible AI and recommendation systems, as it works to break filter bubbles, encourage diverse information, and respect user autonomy (Mengyan Wang, 2024).

The Demand for diversity varies not only among different users but also across different shopping scenarios for the same user. Moreover, users' behaviors are influenced by the diversity of impressions, a factor often disregarded in traditional session-based recommendation models. (Bin Hao, 2021). Therefore, there is a growing recognition of diversity as a crucial factor for enhancing user satisfaction and improving business outcomes (Qiong Wu, 2019).

By developing and testing several recommender systems it is clear that accuracy and diversity positively impact customer satisfaction, Interestingly, while accuracy alone drives satisfaction in traditional systems, both accuracy and diversity are crucial in modern ones. These findings underscore the significance of identifying factors that enhance customer satisfaction in recommender systems, promoting sustainable e-commerce development (Jaekyeong Kim, 2021).

DivMF (Diversely Regularized Matrix Factorization), a novel approach was introduced to diversifying recommendations while maintaining accuracy. DivMF significantly improves both diversity and accuracy compared to existing approaches, as demonstrated through extensive experiments on real-world datasets. Specifically, DivMF achieves up to 34.7% improvement in aggregate-level diversity without sacrificing accuracy, and up to 27.6% improvement in accuracy while maintaining diversity levels similar to the best competitors (Jongjin Kim, 2023).

Offering uniform diversity degrees for all users, and neglecting personalized diversity preferences is again a prevalent challenge in the recommendation systems. To overcome this limitation, PDPP (Personalized Determinantal Point Process), is a recommendation algorithm that mines a user's diversity preference from their service invocation history. PDPP generates personalized service recommendation lists with preferred diversity levels for each user. Experimental results demonstrate that PDPP can effectively deliver personalized and diversified web services while maintaining recommendation accuracy (Guosheng Kang, 2023) An accuracy-diversity dilemma exists for recommender systems, where maximizing one metric may come at the expense of the other. While accurate recommendations are crucial, repeatedly suggesting the same item can lead to customer dissatisfaction. Hence, balancing accuracy and diversity is essential for optimal performance (Jaekyeong Kim, 2021). Therefore, it is imperative to comprehend the influence of personalized and diversified recommendations on customers' purchasing decisions and satisfaction.

Research Methodology

Objectives of the study Many research studies have delved into the influence of personalized recommendations on customers' purchasing choices and satisfaction levels. However, there has been a noticeable lack of research dedicated to diversified recommendations and the comparative evaluation of both recommendation types. To bridge this research gap, this study examines the effects of personalized and diversified recommendations on the overall shopping experience of customers. The main objectives of the study would be –

• To investigate the impact of personalized recommendations on the efficiency of e-commerce recommendation systems from the customer perspective.
• To investigate the impact of diversified recommendations on the efficiency of e-commerce recommendation systems from the customer perspective.
• To compare the effectiveness of personalized and diversified recommendations in enhancing customer satisfaction within e-commerce platforms.

The study will focus on Examining the effectiveness of personalized and diversified recommendations in improving customer engagement and satisfaction within e-commerce environments. For businesses, the scope of this study could involve Comparing the strengths and weaknesses of personalized and diversified recommendation approaches in meeting customer needs and preferences and identifying best practices and strategies for optimizing the effectiveness of personalized and diversified recommendations to enhance overall customer satisfaction and business outcomes in e-commerce settings.

• Research Design - The present research is descriptive by nature
• Sample Description- 111 Online shoppers have been contacted through the Judgmental sampling method.
• Data Collection Method - Primary data was collected by survey technique using a structured questionnaire.
• Data Analysis Tools -The Chi-Square test and Regression analysis were used for the analysiswith the help of IBM-SPSS 21.0 software.

Demographic Analysis

                                                       Table 1: Demographic Details

(Source: Primary Data)

Table 1 represents female respondents are slightly higher as compare to male respondents. 51.4% (57) of the respondents are female, while 48.6% (54) are male. Many respondents lie in the age group of 20-30 years, with 72.1% (80) with minimum age respondents around 40-50 years of age with 2.7% (3). 48.6% (54) of respondents hold postgraduate degrees, whereas only 3.6% (4) have a background of higher secondary education. 66.7% (74) of the respondents are students, while only 7.2% (8) are involved in business.

Results and Discussion

Ho1: Personalized recommendations do not significantly impact the efficiency of e-commerce recommendation systems from the customer's perspective.

Table 2: Tests of Normality

1. Lilliefors Significance Correction (Source: SPSS Output)

The Kolmogorov-Smirnov test was applied to test the normality of the data. The table number 1 shows P-values is lesser than 0.05, so it can be said that data is normally distributed.

Table 3: ANOVA^a (Ho1)

1. Dependent Variable: Purchase Decision (Source: SPSS Output)
2. Predictors: (Constant), Personalized Recommendations

Table 4: Coefficients^a (Ho1)

1. Dependent Variable: Purchase Decision (Source: SPSS Output)

Table 5: Residuals Statistics^a (Ho1)

1. Dependent Variable: Purchase Decision (Source: SPSS Output)

The Linear Regression test was applied to explore the impact of personalized recommendations on the efficiency of e-commerce recommendation systems, with the purchase decision as the focal variable representing the impact from the customer's viewpoint. In this analysis, personalized recommendations serve as the independent variable, while the purchase decision acts as the dependent variable. It can be observed thatP-values is lesser than 0.05 which interpretsa significant impact of personalized recommendations on the efficiency of e-commerce recommendations (Table 4). These findings entail the rejection of the null hypothesis, affirming the influence of personalized recommendations on enhancing the system's efficiency.

Ho2: Diversified recommendations do not significantly impact the efficiency of e-commerce recommendation systems from the customer's perspective.

Table 6: ANOVA^a(Ho2)

1. Dependent Variable: Purchase Decision (Source: SPSS Output)
2. Predictors: (Constant), Diversified Recommendations

                                                    Table 7: Coefficients^a(Ho2)

1. Dependent Variable: Purchase Decision (Source: SPSS Output)

 Table 8:  Residuals Statistics^a (Ho2)

1. Dependent Variable: Purchase Decision (Source: SPSS Output)

The Linear Regression test was applied to explore the impact of diversified recommendations on the efficiency of e-commerce recommendation systems, with the purchase decision as the focal variable representing the impact from the customer's viewpoint. In this analysis, diversified recommendations serve as the independent variable, while the purchase decision acts as the dependent variable. It can be observed that P-values is lesser than 0.05 which interprets a significant impact of diversifies recommendations on the efficiency of e-commerce recommendations (Table 7). These findings entail the rejection of the null hypothesis, affirming the influence of diversified recommendations on enhancing the system's efficiency.

H03: There is no significant difference in the effectiveness of personalized and diversified recommendations in enhancing customer satisfaction and engagement within e-commerce platforms.

                                                 Table 9: Chi-Square Test (Ho3)

(Source: SPSS Output)

The Chi-square test was applied to compare the effectiveness of personalized and diversified recommendations in improving customer satisfaction and engagement on e-commerce platforms. It can be observed that P-values is lesser than 0.05 which interprets that there is a statistically significant difference between the effectiveness of personalized and diversified recommendations in enhancing customer satisfaction and engagement within e-commerce platforms (Table 9). In other words, the results indicate that one type of recommendation method is more effective than the other in positively impacting customer satisfaction and engagement within e-commerce environments. These findings entail the rejection of the null hypothesis.

Inferences

1. The application of personalized recommendations has been found to significantly enhance the efficiency of e-commerce recommendation systems.
2. By leveraging customers' purchase history, preferences, and buying behavior, personalized recommendations offer heightened relevance, aiding users in making informed purchasing decisions.
3. The application of diversified recommendations has been found to significantly enhance the efficiency of e-commerce recommendation systems.
4. Diversified recommendations are crucial in addressing user fatigue stemming from repetitive personalized recommendations.
5. By offering a variety of product suggestions, diversified recommendations play a pivotal role in enabling users to explore new and diverse items within the e-commerce ecosystem. This not only enhances user engagement but also fosters a more dynamic and satisfying shopping experience.

In summary, the findings underscore the importance of recommendation strategies in e-commerce environments, highlighting the significant impact they have on system efficiency and customer satisfaction and engagement. Moreover, the results emphasize the need for tailored recommendation approaches to optimize user experience and drive desired outcomes in e-commerce platforms.

Recommendations

1. Leverage Personalized Recommendations: E-commerce platforms should continue to prioritize the implementation of personalized recommendation systems due to their demonstrated efficiency enhancement. Utilizing customers' purchase history, preferences, and buying behaviour can significantly increase relevance and aid users in making informed purchasing decisions.
2. Integrate Diversified Recommendations: In addition to personalized recommendations, it is crucial for e-commerce platforms to incorporate diversified recommendation strategies. Diversified recommendations address user fatigue associated with repetitive suggestions and enable users to explore new and diverse items. This variety enhances user engagement and fosters a more dynamic and satisfying shopping experience.
3. Optimize Recommendation Algorithms: E-commerce businesses should invest in optimizing their recommendation algorithms to strike a balance between personalized and diversified recommendations. This ensures that users receive tailored suggestions while also being exposed to a range of product options, maximizing satisfaction and engagement.

4. Continuous Monitoring and Adaptation: Regular monitoring of user feedback and behaviour is essential for refining recommendation strategies. E-commerce platforms should continuously adapt their recommendation approaches based on user preferences, market trends, and evolving customer needs to maintain effectiveness and relevance over time.

References

1. Bin Hao, M. Z. (2021). Diversity or not: Dynamic Diversification for personalized recommendation. Springer, 461-472.
2. Chaoyu Shi, P. R. (2024). Diversifying sequential recommendation with retrospective and prospective transdormers. ACM digital library.
3. FasterCapitol. (2024). The Benefits of Personalized Recommendations For Customers .
4. Guosheng Kang, B. L. (2023). Qos- centric diversified web service recommendation based on personalized determinantal point process. MDPI.
5. Hajer Nabli, R. B. (2023). Improved clustering - based hybrid recommendation system to offer personalized cloud services. Springer.
6. Jaekyeong Kim, l. C. (2021). Customer Satisfaction of Recommender System: Examining Accuracy and Diversity in Several Types of Recommender approaches. MDPI, 11-13.
7. Jongjin Kim, H. J. (2023). Diversely regularized matrix factorization for accurate and aggregately diversified recommendation. Springer, 361-373.
8. Liang Hu, L. C. (2017). Diversifying personalized Recommendation with User-session Context. IJCAI. IJCAI.
9. Mengyan Wang, Y. H. (2024). Nudging towards responsible recommendations: a graph-based approach to mitigate belief filter bubbles. IEEE Xplore, 1-15.
10. Qiong Wu, Y. L. (2019). Recent Advances in Diversified Recommendation. Arxiv.
11. Shin, D. (2024). Misinformation and Diversity: Nudging Away from Misinformation Nudging Toward Diversity. Springer, 139-169.