Academia.eduAcademia.edu

Figure 6 – uploaded by Nasser Al Musalhi

See full PDFdownloadDownload figure
The following recommended list can be used as a suggestion that is effective since all these factors are subject to vary depending on the amount of by a single user or the often used by various users. So, we can recommend a list based on the number of sales as shown in Figure6. Also, we can recommend an average number of sales as shown in Figure 7 below. Figure6: Recommended items based on the number of sales

Figure6 6 The following recommended list can be used as a suggestion that is effective since all these factors are subject to vary depending on the amount of by a single user or the often used by various users. So, we can recommend a list based on the number of sales as shown in Figure6. Also, we can recommend an average number of sales as shown in Figure 7 below. Figure6: Recommended items based on the number of sales

Related Figures (13)

Se ae ae eee eee. ge eee eee es ee ee wor ee Content-based filtering, also known as cognitive filtering, recommends products by comparing the product’ properties with the buyer’s characteristics. Tags and labels describe the product’s features, and use interactions create a customer profile. This profile outlines the user’s interests, such as music preferences o: cosmetic choices, which the system uses to search for relevant recommendations within its database Content-based systems focus on a user’s explicit and implicit input to build their profile and suggest item: based on their past purchases or preferences. Unlike collaborative filtering, content-based systems do no consider other users’ activities. Both user and item qualities play a crucial role in content reeommendatior systems, contributing to accurate predictions[3]—[5]. Figure 1 shows how the content-based approach works. Content-based filtering in recommender systems offers several advantages, including enabling users to create personalized profiles based on their ratings, educating users about the system’s functioning, and recommending items not yet rated by any user, which is particularly beneficial for new users. However, there are limitations to content-based filtering, such as the challenge of generating certain item qualities, the potential for overspecialization in recommending similar items, and the reliance on user ratings, which may be sparse and make evaluation challenging[5]. 1.2 Collaborative Filtering 1.1 Content-Based Filtering
Se ae ae eee eee. ge eee eee es ee ee wor ee Content-based filtering, also known as cognitive filtering, recommends products by comparing the product’ properties with the buyer’s characteristics. Tags and labels describe the product’s features, and use interactions create a customer profile. This profile outlines the user’s interests, such as music preferences o: cosmetic choices, which the system uses to search for relevant recommendations within its database Content-based systems focus on a user’s explicit and implicit input to build their profile and suggest item: based on their past purchases or preferences. Unlike collaborative filtering, content-based systems do no consider other users’ activities. Both user and item qualities play a crucial role in content reeommendatior systems, contributing to accurate predictions[3]—[5]. Figure 1 shows how the content-based approach works. Content-based filtering in recommender systems offers several advantages, including enabling users to create personalized profiles based on their ratings, educating users about the system’s functioning, and recommending items not yet rated by any user, which is particularly beneficial for new users. However, there are limitations to content-based filtering, such as the challenge of generating certain item qualities, the potential for overspecialization in recommending similar items, and the reliance on user ratings, which may be sparse and make evaluation challenging[5]. 1.2 Collaborative Filtering 1.1 Content-Based Filtering
eee 2D ee ee Hybrid filtering strategies combine multiple recommendation algorithms to enhance system performance and overcome the limitations of individual methods. This approach acknowledges that a mix of algorithms can provide more effective suggestions by leveraging the strengths of each. Various recommendation approaches are integrated to mitigate the shortcomings of standalone strategies. Integration methods include implementing techniques separately and combining results, blending content-based and collaborative filtering, or creating a unified recommendation system that combines both approaches, offering a versatile solution to recommendation system optimization[11]. Figure 2 shows the hybrid filtering approach. Hybrid filtering offers advantages including overcoming the limitations of individual methods, leading to improved recommendation outcomes, and the ability to work effectively with sparse data. However, it also presents drawbacks, such as increased expenses, heightened complexity in system design and management, and the occasional need for external data that may not always be readily accessible. These considerations underline the trade-offs associated with the use of hybrid recommendation systems.
eee 2D ee ee Hybrid filtering strategies combine multiple recommendation algorithms to enhance system performance and overcome the limitations of individual methods. This approach acknowledges that a mix of algorithms can provide more effective suggestions by leveraging the strengths of each. Various recommendation approaches are integrated to mitigate the shortcomings of standalone strategies. Integration methods include implementing techniques separately and combining results, blending content-based and collaborative filtering, or creating a unified recommendation system that combines both approaches, offering a versatile solution to recommendation system optimization[11]. Figure 2 shows the hybrid filtering approach. Hybrid filtering offers advantages including overcoming the limitations of individual methods, leading to improved recommendation outcomes, and the ability to work effectively with sparse data. However, it also presents drawbacks, such as increased expenses, heightened complexity in system design and management, and the occasional need for external data that may not always be readily accessible. These considerations underline the trade-offs associated with the use of hybrid recommendation systems.
In this research, we have selected the dataset and examined the three types of RS to specify the efficiency of each model of RS. Figure 3 shows the methodology sequences for this research. 3.1 Data Preparation aS Sees ee ee To achieve the methodology approaches, we used in this research the Turkish market sales dataset with +9000 items[25]. This Turkish-language dataset encompasses supermarket data, featuring over 600,000 sales entries, 52,000+ distinct Turkish customer profiles with attributes like names, gender, age, and birthdates, 9,000+ categorized and subcategorized items grouped into categories 1 through 4, information on 81 store locations, a wor. and records spanning a t kforce of +1,200 sales personnel, geographical coordinates (latitude and longitude), hree-month sales period. It's important to note that all the data within thisdataset is synthetic; the sales records were randomly generated based on city population distributions, and the customer names are en tirely artificial and unrelated to real individuals.To prepare the dataset we have normalizedit and chose to work with a single branch based on thehighest total number of transactions that have been recorded in the branchas appears in Figure 4:
In this research, we have selected the dataset and examined the three types of RS to specify the efficiency of each model of RS. Figure 3 shows the methodology sequences for this research. 3.1 Data Preparation aS Sees ee ee To achieve the methodology approaches, we used in this research the Turkish market sales dataset with +9000 items[25]. This Turkish-language dataset encompasses supermarket data, featuring over 600,000 sales entries, 52,000+ distinct Turkish customer profiles with attributes like names, gender, age, and birthdates, 9,000+ categorized and subcategorized items grouped into categories 1 through 4, information on 81 store locations, a wor. and records spanning a t kforce of +1,200 sales personnel, geographical coordinates (latitude and longitude), hree-month sales period. It's important to note that all the data within thisdataset is synthetic; the sales records were randomly generated based on city population distributions, and the customer names are en tirely artificial and unrelated to real individuals.To prepare the dataset we have normalizedit and chose to work with a single branch based on thehighest total number of transactions that have been recorded in the branchas appears in Figure 4:
3.2 Implementation
3.2 Implementation
Tablea: Correlation between a selected user and recommended users
Tablea: Correlation between a selected user and recommended users
3.3 Content-Based Filtering Implementation
3.3 Content-Based Filtering Implementation
Figure 5: User profile vectors graph Py ee i eS a ee See a ee, Cee eee CE age eee ae ae Le eve ee Te re ee ae The content-based approach generates recommendations for particular products by analyzing a user's profile. In doing so, we merge item categories with item names to extract the distinctive features of each item, resulting in a column named "combined category." This amalgamation enables us to provide personalized recommendations. We used the cosine similarity function in Eq.2 to calculate the vector distance between the user and the item.Figure 5 shows the vector for user history items. Table3 shows the top 10 user recommendation list using the cosine similarity approach score.
Figure 5: User profile vectors graph Py ee i eS a ee See a ee, Cee eee CE age eee ae ae Le eve ee Te re ee ae The content-based approach generates recommendations for particular products by analyzing a user's profile. In doing so, we merge item categories with item names to extract the distinctive features of each item, resulting in a column named "combined category." This amalgamation enables us to provide personalized recommendations. We used the cosine similarity function in Eq.2 to calculate the vector distance between the user and the item.Figure 5 shows the vector for user history items. Table3 shows the top 10 user recommendation list using the cosine similarity approach score.
Occasionally, operations w a specific item may be purchased in large quanti ties but with a relatively low number of sales hich may fall under minimum sales required to be counted. This situation suggests that the item could be a preferred choice, but it might be overlooked when using the weighted average formula, which places more emphasis on the number of sales rather than the quantity sold. To enhance the accuracy of our recommenda tions, we adopt a balanced approach, allocating 50 percent weight to the weighted average and 50 percent weight to the number of sales for each item Eq.5. To achieve this, we employ a scaling technique to standardize t he quantity units, thereby streamlining the dataset and addressing the minimum sales issue. In Figure 7, we illustrate the final recommended system, which effectively combines the weighted average and the number of sales for each item as a combined weigh recommenda tion approach. ted average (CWA), resulting in a hybrid
Occasionally, operations w a specific item may be purchased in large quanti ties but with a relatively low number of sales hich may fall under minimum sales required to be counted. This situation suggests that the item could be a preferred choice, but it might be overlooked when using the weighted average formula, which places more emphasis on the number of sales rather than the quantity sold. To enhance the accuracy of our recommenda tions, we adopt a balanced approach, allocating 50 percent weight to the weighted average and 50 percent weight to the number of sales for each item Eq.5. To achieve this, we employ a scaling technique to standardize t he quantity units, thereby streamlining the dataset and addressing the minimum sales issue. In Figure 7, we illustrate the final recommended system, which effectively combines the weighted average and the number of sales for each item as a combined weigh recommenda tion approach. ted average (CWA), resulting in a hybrid
Table 5:Recommended items based on a combined weighted average
Table 5:Recommended items based on a combined weighted average
Figure8: Graph of Recommended list based on the weighted average and number of sales for each item.
Figure8: Graph of Recommended list based on the weighted average and number of sales for each item.
Related topics:
Recommendation SystemsHybrid Recommender Systems
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved