Dr. Madhukar Dayal & Dr. Sanjeev Tripathi
Information Systems Area & Marketing Area, IIM Indore
Email: madhukar@iimidr.ac.in & sanjeevt@iimidr.ac.in; Phone: 0731-2439531/489
Recommender systems (RS) are used in multiple applications and are actively used by people to find interesting places to visit or to eat, movies to watch, books to read, and in a variety of other ways. With the growth in applications in the industry, the study of such systems has emerged as specialized courses in technical institutes, universities and more importantly, in business schools. As such, there is a need for a simple model explaining the concept and working of recommender systems and their applications to real world situations for business school students. While there has been an extensive research to improve recommender systems, little work has been done to develop pedagogical tools for teaching recommender systems. In this article we present an easy to understand model which explains the core concepts behind recommender systems. This model has been tested in class, and has interactive ability not found in existing recommender systems. This allows a user to alter preference parameters and obtain improved recommendations.
Application of existing RS algorithms is plagued with real time computational requirements of an immense magnitude making implementation of interactive RS even more difficult, due to which the development of interactive RS has trailed behind. Unfortunately, while the technology and the application has gathered pace, the understanding of RS among management students and teachers has remained limited.
Libraries are an essential part of every management school and almost everyone has had an experience of reading in a library and getting books issued from the library. Hence, explanation of RS in the library environment makes understanding of the concepts simpler. Given the data of other library users for the books that they have read, the problem is to find the recommended books for the new target user. Mathematical statement of the problem is presented. A two-dimensional array comprises the available data (please refer to the full paper on the journal’s website). Microsoft Excel formulas are used to derive cold start (generalized) recommendations. A drawback of these recommendations is that it does not take into account whether the target user has read the book or not. In the advanced model described later, a book read by a target user is eliminated from recommendation.
In the paper, we describe an enhanced solution to the problem using Microsoft Excel to overcome the limitations mentioned above (please refer to the full paper on the journal’s website). The complexity of computations in recommender systems is assessable from this model. For frequent users, for whom a reading (borrowing) history is available, the system should have the capability to create a group of users, based on similarity of reading preferences from history, and map the target user to this group to determine the recommendations.
The personalized recommendations are an improvement over the generalized recommendations, as they, (i) do not recommend a book which the target user has read, (ii) compute recommendations based on only such users which have a matching taste with that of the target user, and (iii) more the users based on whom a book becomes recommendable for the target user, the higher is the recommendation score for that book. The incorporation of feedback (star ratings of the books) below increases the computational complexity even further. Data pertaining to feedback (star ratings) for the books read, by all users, if available, can help us overcome this limitation, thus, providing improved recommendations.
In the advanced model, the system should be able to utilize the feedback (star ratings from one star to five stars, five stars representing excellent) given to the read books by the library users, to suggest highly rated books for the target user. Libraries have a mechanism of electronically capturing user ratings for the books the users have read. These are usually star (*) ratings ranging from 1* to 5* for each book, where 5* represents highly recommended and 1*, otherwise. Further, a user may or may not like all the books s/he issues and reads from the library. Hence, the target user is better served by providing an input of minimum star rating for the books s/he is interested in, thus, enabling the system to filter out disliked books. Both these features are incorporated in the model that follows (please refer to the full paper on the journal’s website). A detailed discussion on the class usage of the model and teaching experience is presented in the paper.
There is no doubt that as compared to theoretical learning, experiential learning leads to better understanding of a subject, and the lessons learnt are retained longer. The lack of appropriate pedagogical tools for teaching the concepts of RS poses a challenge for faculty. We hope that the model presented in this paper proves to be useful to management students and faculty. User feedback is an important part in recommendation systems for improved recommendations. The failure to leave online feedback, a common trait in the online community, with a bias towards negative feedback (Zhou, Dresner & Windle, 2008) is a potential detriment for generating good positive recommendations in the long run. Instructors should make the students aware of these traits.
Similar to the mechanism of screening and sequential screening for multi-criteria decision making (MCDM) presented by Chen, Kilgour & Hipel, 2008, this spreadsheet model allows interactive step by step screening for the decision making. In a way, it simulates a complex interactive RS in a very simple manner without compromising any of the features of the RS. Jiang, Shang & Liu (2010) have pointed out that “a truly successful recommendation system should be one that maximizes the customer’s after-use gratification”. Interactive RS have the potential to achieve a higher degree of user involvement, as well as, improved user selection, and hence, achieve user delight. We hope that this tool will excite management teachers to teach RS in class and those who are already teaching to use this tool in providing in class experiential learning.
ORIGINAL ARTICLE: Dayal, M. & Tripathi, S. (2018). Teaching Interactive Recommender Systems – Recommending an Excel Based Approach. Journal of International Business Education, 13, link: http://www.neilsonjournals.com/JIBE/abstractjibe13dayaltripathi.html.