Problem Description

The internet contributes a major role in speeding and modifying the manner in which daily tasks like online shopping, communication, paying bills and watching movies are conducted. Markets have to offer different products and services to various customers with diverse needs. Producers and retailers have been forced by the shift to online shopping to customize for the needs of customers while providing more options than ever before (Sivapalan et al., 2014). As a result, the customer is facing the problem of analyzing every offering with the aim of determining what they need and will be of benefit to them. The paper addresses how the issues can be solved using the common recommender systems techniques and their associated trade-offs.

Proposed Solutions

Recommender systems have been developed with the aim of overcoming this problem. Recommender systems are used for rapid and automatic customization and personalization of e-commerce sites. They assist sites in generating more sales by showing products and services that are tailored to meet the needs of visitors, thereby turning them into consumers, up-selling additional products by bundling together closely-related products, and increasing the loyalty of products (Ricci, Rokach & Shapira, 2015).  The recommendation in e-commerce entails the provision of products and services that users are interested in. Recommendation systems in e-commerce have emerged to be extremely popular in recent times. Different techniques are used by e-commerce website in providing users with better experience in online shopping. There are different types of information that are used in making recommendation decisions. They include demographics of users, item attributes, and the preferences of users (Konstan & Riedl, 2012).

Demographics refers to the attribute of users which the outcome of a recommendation made. Demographic information includes age, likes and dislikes, gender, occupation, age group, income level, and hobbies. Item attributes can either be extrinsic and intrinsic. The identification of extrinsic features cannot be possible by automatic analysis of content. On the contrary, intrinsic can easily obtain from the contents. Features of items can be obtained from the item description (Konstan & Riedl, 2012). The preferences of users can either be categorized as presence scores such as likes or dislike, or numerical score that indicates the degree to which a user likes or dislikes a product. The implicit indicator of users rating is considered to be the amount of time that is spent by a user on a specific page reading and analyzing its content (Ricci, Rokach & Shapira, 2015). Despite the difficulty of gathering implicit indicators, they are effecting in providing more information more user information, that the user may not be capable of providing.

Data Mining

Knowledge discovery database (KDD) is employed in describing extraction of useful information from a dataset. It is also known as data mining. The information extracted can either be implicit or explicit. KDD is used in finding patterns in the buying behavior of users such as the time of the year that particular products are more likely to be bound and make recommendations that lead to the generation of revenue. Association rules are considered to be one of the essential algorithms used in KDD. The regulations provide an association between different sets of product in a manner that the presence of one from a specific set implies there is a high possibility of another product being in the same set. The common association rules include Apriori, Direct Hashing and Pruning, FP-tree algorithms, and Tree projection algorithms (Maimon & Rokach, 2009).

Types of Recommendations

Recommendation systems are categorized as follows: attribute-based, personalized, non-personalized, people-to-people correlation, and item-to-item correlation. Based implementation, a recommendation can either be described as short-lived or long-lived. When the system requires minimal or no input from user, then it is considered to be automatic. A system that needs some is considered to be manual. Personalized recommendations are automatic and depend on the preference of users. Non-personalized recommendations are based on the rating of products from other users of the system (Sivapalan et al., 2014). These two recommendations are considered to be straight forward since they are automatic require no input from the user. They are not short-lived since other users can apply them.

The

attribute-based recommendation is whereby item description is done using various features and attributes used in the generation of recommendation. The technique is considered manual since the user is required to explicitly search for a specific type of product on which recommendation will be based on. These types of recommendation are short-lived and do not rely on how long the system Item-to-item correlation recommenders remember the preferences of the user are based on other items that are of interest to users. They tend to be prevalent in e-commerce sites where the recommendation of products are based on what is on the shopping cart of users (Bobadilla et al., 2013). The recommendations are considered to be manual since the shopping cart must have something, and are short-lived since the user since it ceases to exist when the shopping cart is empty, or purchase has been actualized. These systems often use association rule. People-to-people correlation system entails finding similarity between the active and other users in the system and makes recommendations based on the products that have been purchased and rated by other customers. The system mostly uses collaborative filtering. The method is manual since users are required to have purchased or rated products in the past. The design of the system determines whether the system can last or not (Sivapalan et al., 2014).

Techniques for Recommendation Systems

Recommender systems use different algorithms and methods to generate recommendations. The most common ones include content-based filtering, association rules, hybrid filtering and collaborative filtering (Lu et al., 2015).

Association rules are employed in recommending products based on their presence alongside other products. The presence of one product in a transaction can be used in determining the second product when the purchase of two products is done simultaneously. The approach is essential in recommending for new users who wish to make purchases. The main limitation of the technique is that it supports a lot of rules, and are slower and not very useful when a lot of mining rules are utilizing in making recommendations (Sivapalan et al., 2014).  A site that has a lot of products and transaction would have trouble scaling with multiple rules since they need to it on the entire database and still find recommendation for users within a time frame that is considered acceptable.

Collaborative filtering utilizes the rating and details of the customer, as well as the aggregated reviews from all the customers with the aim of building recommendations. The approach is considered effective in analyzing existing active customers with similar preferences and the features of current customers to make recommendation. The method of filtering is achieved through a model-based method, a heuristic-based or a hybrid model (Lu et al., 2015). The limitation of the method is that it depends on ratings and reviews of customers.

Content-based filtering entails analyzing products and finding similarity with active users with the aim of making recommendation. Contrary to association rules and collaborative filtering, it does not require an active database of purchase history. It depends on information retrieval, analysis, and filtering. It is mainly employed in a situation where content can be read or analyzed. As a result, it is applicable in contents like news, movies, news or anything with metadata attached. It is also capable of providing recommendation from items viewed by the user in the past. The content description can be done using labels, and the tags are given weight based on how well they describe the article (Sivapalan et al., 2014). Nearest neighbors or clustering algorithms can use these labels in recommending other articles to active users. New users with limited information and labels cannot benefit from the technique. Common algorithms used in content filtering include artificial neural network, K-nearest neighbors, Bayesian, and clustering. The drawbacks of the system include: content must be structured and easy to parse; the system is not capable of differentiating between a bad and good item based on retrieved information; lack of user information; and overspecialization (Bobadilla et al., 2013).

Hybrid filtering was developed with the aim of avoiding the problems that exist in both content-based and collaborative filtering systems. Proposed hybrid filtering systems include the implementation of both filtering separately and combining results, incorporation of a characteristic of content-based and collaborative filtering, and new algorithms incorporating the techniques of both system. The approach of combining different recommender systems entails building two different recommender systems from content-based and collaborative-based approaches (Ricci, Rokach & Shapira, 2015).

Conclusion

Recommender systems make it possible for e-commerce sites to highly customized products and services as per the unique needs and preferences of users and buyers. As a result, companies are capable of having a better understanding of their users, provide stores that are highly personalized, thereby increasing the loyalty and satisfaction of customers. The implementation of recommendation is done using various data mining tools which are adapted to current needs. Renowned techniques entail the use of association rules, content-based filtering, collaborative filtering, and hybrid filtering. These techniques have their fair share of limitations and strengths and are treated as virtual salesmen since they are not capable of actively marketing new products as they only five suggestions. Recommendations range from personalized to community-driven outcomes that allow a broad range of possibilities.

References

Bobadilla, J., Ortega, F., Hernando, A., & Gutiérrez, A. (2013). Recommender systems survey. Knowledge-based systems, 46, 109-132.

Konstan, J. A., & Riedl, J. (2012). Recommender systems: from algorithms to user experience. User modeling and user-adapted interaction, 22(1-2), 101-123.

Lu, J., Wu, D., Mao, M., Wang, W., & Zhang, G. (2015). Recommender system application developments: a survey. Decision Support Systems, 74, 12-32.

Maimon, O., & Rokach, L. (2009). Introduction to knowledge discovery and data mining. In Data Mining and Knowledge Discovery Handbook (pp. 1-15). Springer, Boston, MA.

Ricci, F., Rokach, L., & Shapira, B. (2015). Recommender systems: introduction and challenges. In Recommender systems handbook (pp. 1-34). Springer, Boston, MA.

Sivapalan, S., Sadeghian, A., Rahnama, H., & Madni, A. M. (2014, August). Recommender systems in e-commerce. In World Automation Congress (WAC), 2014 (pp. 179-184). IEEE.

Do you need high quality Custom Essay Writing Services?