3.5.3 Personalization
If we take a look into our daily life, then we will understand that we also take recommendations to buy a product. These recommendations are mostly given by our family members and friends. We followed their recommendations because we trust them. This is the sole element which is implementing the MCRS to build among system and the user. The more suitable product system will predict for the user, the more the user will trust the MCRS and will buy the recommended product. That is how by personalization a customer will come to that platform again and again.
3.5.4 Discovery
Every recommender system can be used as a tool. The leading companies discover new ways to apply the recommender system as a tool. For example, iTunes uses “Genius Recommendations” which is a very efficient multi-criteria recommender tool, and Amazon uses “Frequently Bought together” which can also be customized by the user also. These are some innovative ways of using the multi-criteria recommender systems.
3.5.5 Provide Reports
This is an integral part of personalization system. It is also used to make trust with the clients. It helps to make solid decisions about the site or the direction of a campaign. In order to create a drive-in, sales client generates offers on slow moving product based on their report.
In the present time, online sales are generally more satisfying. So, every company wants to do something extra. An e-commerce company can use different types of filtering technique like collaborative, content-based, hybrid filtering as we have studied before to make an effective recommendation engine. For example, Amazon has super successful multi-criteria recommender engine which recommends several things as add on very efficiently and that attract a client or a customer to buy more and more things from the same platform. It follows the sole of every recommender system that the only way to truly engage with customer is to communicate with each as an individual.
3.6 Challenges of Multi-Criteria Recommender System
There are some common challenges which every MCRS faces. Some of them are as follows.
3.6.1 Cold Start Problem
This problem faces recommender system when a client came for the first time in that platform or when a new item is added on that platform. When someone comes for the first time in any platform, recommender system does not have any information about that client. Because the client have not rated any product or have not surfed anything. If a new item is added in platform, then it also does not have any rating of its quality, durability, etc. This is known as cold start problem. A huge number of research works has done to overcome this this problem [34].
3.6.2 Sparsity Problem
This is also a very important problem that all recommender systems are facing nowadays. It happens when user has large history list. The list contains everything like list of items he bought or list of movies he watched or listed of music he listened and even his previous surfing list. Sparsity happens when the user does not give rating after buying items or after watching movies. Rating is the most fundamental element for a MCRS. Because of lack of rating, a recommender system does not able to understand that whether the client liked that particular thing or not [34].
3.6.3 Scalability
Scalability is related to the multi-criteria recommender systems own performance. Generally, recommender system does not consume many resources. It is designed such a way that it gives best accuracy by using minimum resource. Recommender system needs to recommend a list of items to the user. But with time, the number of users increases, the number of items also increases, and the recommendation list is also increasing [34].
3.6.4 Over Specialization Problem
When a MCRS is able to know about the choices of a particular client, then it creates a boundary and shows according to their choice without discovering new items and other options with time. This situation is known as over specialization problem [34].
3.6.5 Diversity
If a user spends a long time in a platform, then the recommender system has a lot of information about him. If he orders almost same kind of things, then the recommender system generates the recommendation list that is based on the same category only [34].
3.6.6 Serendipity
Every recommender system should have an objective that to surprise people by recommending new product and make interest in the user [34].
3.6.7 Privacy
For recommender system privacy is very important. As the recommender system knows some information about the user, it should not be going to the outside world. The users need to understand which data is required to approve more ideally items to them and how it used [34].
3.6.8 Shilling Attacks
It happens when a user became malicious or unethical. Many times, it is observed that user starts giving false rating in some items. If it happens, then either the rating of that good item will drop down and that item will no longer recommended or a bad quality item will get more good ratings and will be recommended more by MCRS. This is known as shilling attack [34].
3.6.9 Gray Sheep
Gray sheep takes place normally in CF systems. Here, the belief of a client does not identify with any group and consequently. It is no able to acquire the advantage of recommendations [34].
3.7 Conclusion
In this chapter of MCRS, we understood the definition of a recommender system, its growing importance, and applications in various sectors. Further, we discussed about the different phases of a recommender system. Moreover, we deeply analyzed the three different filtering techniques associated with the recommender system that includes collaborative, content-based, and knowledge-based filtering techniques. In addition, we also discussed about the hybrid filtering techniques. To have a better understanding on the concept of the recommender system, we considered five research activities conducted by various researchers across the globe. Lastly, we also discussed about the various advantages and the challenges prevailing in the recommender system.
Nowadays, recommender systems are widely used in various aspects, with the passage of time, the recommender systems are being modified and are showing better performance. New techniques are implemented in the recommender system to derive a better prediction [1].
References
1. Al-Ghuribi, S.M. and Noah, S.A.M., Multi-criteria review-based recommender system–the state of the art. IEEE Access, 7, 169446–169468, 2019.
2. Wasid, M. and Ali, R., An improved recommender system based on multi-criteria clustering approach. Proc. Comput. Sci., 131, 99–101, Jan. 2018.
3. Zheng, Y., Utility-based multi-criteria recommender systems, in: Proc. 34th ACM/SIGAPP Symp. Appl. Comput., pp. 2529–2531, 2019.
4. Tallapally, D., Sreepada, R.S., Patra, B.K., Babu, K.S., User preference learning in multi-criteria recommendations using stacked auto encoders. Presented at the Proc. 12th ACM Conf. Recommender Syst., Vancouver, BC, Canada, 2018.
5. Musto, C., de Gemmis, M., Semeraro, G., Lops, P., A multi-criteria recommender system