Enhancing Prediction Accuracy of a Multi-criteria Recommender System Using Adaptive Genetic Algorithm
CHAPTER ONE
The aim of this project is to use an adaptive genetic algorithm to model a multi-criteria recommendation problem using an aggregation function-based approach to achieve a more accurate and efficient prediction.
The specific objectives were:
- To formulate an adaptive genetic algorithm
- To use an adaptive genetic algorithm to model multi-criteria recommendation
- To develop a system that will be proficient enough to recommend the most appropriate item to a
- To compare the predictive performance of the multi-criteria recommender technique using an adaptive genetic algorithm with the traditional recommender
CHAPTER TWO
LITERATURE REVIEW
Introduction
The Recommender Systems (RSs) and its application, as well as the overview of collaborative filtering, will be discussed. We will also present the genetic algorithm and its component multi- criteria RS and provide a review of existing relevant literature. The literature review was carried out in order to identify what has been done before on the subject matter as regards this thesis, highlight weaknesses and suggest how the proposed system to be developed intends to solve these identified weaknesses.
Overview of Recommender Systems and their applications
RSs are technology-based systems that primarily use relevant ratings data given by a user to make personalized recommendations to active users. RSs make use of opinions and actions of other users with similar preferences to build recommendations, thereby helping users avoid information overload. These systems contain ubiquitous features in most e-commerce sites such as Netflix.com, Ebay.com, Amazon.com, Instagram, Facebook and Last.fm. RSs are information-processing systems that gather numerous kinds of data (the items to suggest and the users who will receive the recommendations) and knowledge sources in order to generate recommendations for users. These data used by the RSs refer to three kinds of objects: items, users and transactions (Ricci et al., 2015).
A RS is a technique by which an algorithm is used to detect or predict what a user is buying. One would always prefer that recommendations made are similar to the user’s taste or items already bought in the past. Recommendation engines help to do just that. The first factor to consider in developing any recommendation system is its application domain, which has a major effect on the algorithmic approach taken (Ricci et al., 2015). RSs help users to find interesting items in a given domain (such as movies, books, applications, music, hotels and restaurants) by recommending items that match their preferences. Important application domains have been mentioned, but below is a centralized list of the application of RSs:
E-commerce or product recommendation: Ebay.com, Ali-express and Amazon, which are big online stores that sell a large variety of products, use recommendation techniques to display a list of recommended items a user might be interested in, by using the information from users’ past actions, or decisions made by similar users. In addition, users can rate back their purchases using a five-star rating scale, thereby improving the accuracy of the recommendations.
CHAPTER THREE
RESEARCH METHODOLOGY
Introduction
This chapter presents the outline of the problem we intend to solve and our approach to tackling it. The tools, packages and data set used will be discussed, and we will also introduce analysis of the aggregation function-based technique and adaptive genetic algorithm. This chapter, finally, describes the design of the aggregation function-based adaptive genetic algorithm.
CHAPTER FOUR
IMPLEMENTATION
Introduction
This chapter presents the detailed implementation of the proposed system using the adaptive genetic algorithm. In order to prove our methodology works, we implemented it in a feasible manner, ran, generated results and compared them with traditional collaborative filtering. In this chapter, methods that are well accepted were used to evaluate the effectiveness of the algorithms and the result was discussed.
CHAPTER FIVE
SUMMARY, CONCLUSION AND RECOMMENDATION
Introduction
This chapter summarizes the work conducted within this thesis, and the contributions of this work are highlighted. It also concludes the work done with discussion on future work and provides recommendations.
Summary and contributions
The most critical problems of a multi-criteria recommender system are prediction accuracy and multi-criteria optimization. In a multi-criteria rating system a user rates an item giving priority to some specific dimensions over others, and this varies from user to user according to each user’s personal interest. Thus, it is important to find a suitable relationship between the individual criteria ratings and overall ratings, because each user has different priorities on various dimensions of an item.
This thesis was aimed at determining the impact of an adaptive genetic algorithm in enhancing the prediction accuracy of a multi-criteria recommender system. In our approach we treated the multi-criteria problem as an optimization problem and applied a weighted sum average which was achieved by combining values derived from an adaptive genetic algorithm. To achieve this, a model-based approach focusing on the use of the aggregation function-based technique was used. The aggregation function-based technique consists of mainly three steps after the data acquisition.
Conclusion
The proposed approach had a number of aims and objectives as discussed in Chapter One of the thesis. Among the objectives set out, and from the analysis of the implementation methodology, we were able to develop a novel adaptive genetic algorithm in which the crossover rate and mutation rate were based on the fitness function, which provided significant performance improvements over standard implementation to generate better potential solutions.
The developed adaptive genetic algorithm was then used to solve a multi-criteria recommendation problem by generating optimum weights whereby a user rated an item based on the priority of item criteria. The weighting was aggregated with the predicted ratings generated using one of the traditional recommendation techniques to achieve the overall ratings. This approach bridged the gap between each individual criterion rating and the overall rating.
Recommendation and future work
Recommender systems are now widely being employed in various types of application and domains (such as e-commerce, e-business, social media, entertainment etc.) and there are many possibilities to further continue to improve the algorithm to increase prediction accuracy and efficiency. A potential researcher who might be interested in this topic in the near future should use a different stochastic or evolutionary algorithm to improve the prediction accuracy of the multi-criteria recommender system.
In future we plan to extend the current approach by employing a hybrid of an adaptive genetic algorithm and fuzzy logic techniques on more than one real-world data set to learn appropriate aggregation functions for combining individual criterion ratings.
REFERENCE
- Abdullah, S., & Turabieh, H. (2008). Generating university course timetable using Genetic Algorithms and local search. In Proceedings – 3rd International Conference on Convergence and Hybrid Information Technology, ICCIT 2008 (Vol. 1, pp. 254–260). https://doi.org/10.1109/ICCIT.2008.379
- Adomavicius, G., & Kwon, Y. (2007). New recommendation techniques for multicriteria rating systems. IEEE Intelligent Systems, 22(3), 48–55. https://doi.org/10.1109/MIS.2007.58
- Adomavicius, G., & Tuzhilin, a. (2005). Toward the Next Generation of recommender systems: a Survey of the State of the Art and Possible Extensions. IEEE Transactions on Knowledge and Data Engineering, 17(6), 734–749. https://doi.org/10.1109/TKDE.2005.99
- Adomaviciuszan, G., & Tuzhilin, H. (2014). INFORMS Tutorials in Operations Research Personalization and recommender systems. https://doi.org/10.1287/educ.1080.0044
- Alhijawi, B. (2016). Using genetic algorithms for measuring the similarity values between users in collaborative filtering recommender … 8VLQJ * HQHWLF $ OJRULWKPV IRU 0HDVXULQJ WKH 6LPLODULW \ 9DOXHV EHWZHHQ 8VHUV LQ & ROODERUDWLYH, (June). https://doi.org/10.1109/ICIS.2016.7550751
