A Fuzzy-based Approach for Modelling Preferences of Users in Multi-criteria Recommender Systems
Chapter One
Aim and Objectives
The thesis aims to implement a fuzzy-based algorithm that would model the preferences of users in multi-criteria recommender system. The thesis would be tested to ensure that it improves the predictive performance of the fuzzy-based multi-criteria technique and compare it with some of the existing methods (traditional RS). The thesis is aimed at achieving the following objectives:
- Decrease in prediction errors, increase in ranking accuracy,
- To obtain high correlation between the predicted and actual values.
CHAPTER TWO
LITERATURE REVIEW
In this chapter, we presented a review of the existing related works done to determine what has been done before and current thinking in the interest of this research area. Therefore, we looked at the various works done in solving the problems of Collaborative filtering RS and the various approaches used. Issues of prediction accuracy were discussed as well as the technique applied in improving them, some benefits of applying multi-criteria in RS were also discussed.
Problems of Collaborative Filtering Recommender System
RS constitutes a problem-rich research area whose abundance of practical applications has helped users to deal with information that has continuously improved businesses. The major problems of CF RS would be discussed in this section together with the various approaches used by different scholars in solving these problems.
Data Sparsity
Recommender Systems became an important research area since the appearance of the first paper on collaborative filtering since the mid-1990s (Gediminas, 2005). Over the past decades, information technology and the internet have resulted in e-commerce flourishing and emerging as an important gateway to business papers on collaborative filtering (Nilashi, M., & Ibrahim, O. Bin. 2014). Recommender System has been a vast growing tool that has enhanced the massive growth of e-commerce websites (Adomavicius & Tuzhilin, 2003). Yet, interest in this area remains high because it constitutes a problem rich research area and its abundance of practical applications has helped users to deal with information overload.
However, although there is vast growth of RS, Multi-Criteria rating problems have recently emerged in the recommender systems research literature as an important next-generation issue (Adomavicius & A. Tuzhilin 2005). Multi aspects in the CF recommender systems presents new challenges such as sparsity problem in criteria and overall ratings, scalability problem with increasing new dimensions, representation and rating (Nilashi, 2014). Therefore, there is a need to address these problems of RS. Researchers have proposed that the problem of inflexibility has been addressed using Recommendation Query Language (RQL) (Adomavicius & Tuzhilin, 2007). Using user profile information when calculating user similarity may overcome the problem of rating sparsity (Adomavicius & Tuzhilin, 2007). Sparsity problem has been circumvented due to the application of matrix factorization (Chen, G., Wang, F., & Zhang, C. 2009).
Clustering-based privacy preserving collaborative filtering is a technique proposed to solve the problem of scalability and sparsity in order to enable users to supply their information without fear of insecurity trusting that their privacy is well protected. This is achieved through masking the user’s confidentiality before submitting them to the data holder and disguising the rated items and the ratings (Bilge, A., & Polat, H. 2013). The use of self- organizing map (SOM) clustering to CF schemes is designed to preserve users’ confidentiality (Alper & Polat, 2012). SOM is a type of artificial neural network that reduces dimensions by producing a map of usually one or two dimensions that plots the similarities of the data by grouping similar objects together. Fuzzy methods can predict the users’ preference more accurately and even better alleviate the sparsity problem in overall rating by considering user perception about items’ features (Nilashi & Ibrahim, 2014).
Orthogonal Non-negative Matrix Tri-Factorization (ONMTF) is a novel framework for collaborative filtering that alleviates the sparsity problem via matrix factorization and also solves the scalability problem by simultaneously clustering rows and columns of the user- item matrix (Chen, Wang, & Zhang, 2009). Integrating both subjective and objective information to generate recommendations for an active consumer is a novel collaborative filtering framework proposed to solve the problem of sparsity and the cold-start (Li-Chen Cheng & Hua-An Wang 2014). Fuzzy linguistic model, which is a more natural way for the consumer to present their preferences, is adopted within their proposed framework, based on their concepts, two algorithms, a simple aggregated (SA) algorithm and aggregated subjective was proposed.
Cold Start
Cold start problem occurs when a new user or item has just entered the system and it is difficult to find similar ones because there is not enough information about the user nor the item (Su & Khoshgoftaar, 2009). Multi-criteria CF recommender systems suffer more from this problem on two sides, missing values in overall and criteria, with the system having to predict these missing ratings with new approaches. Recommendation problem is reduced to the problem of estimating ratings for the items that have not been seen by a user (Bilge & Polat, 2013). Many recommender systems are intrusive in the sense that they require explicit feedback from the user and often at a significant level of user involvement, leading to new user problems (Breese, et al. 1998). The unrated items of an individual user in a group can be predicted by using the rating information from a group of closely related users (Gediminas, 2005). However, prediction of an unrated item can be derived by the composition of matrix operation (Cheng & Wang, 2014) and Hybrid Recommender System to solve the New items problem of CF (Adomavicius & Tuzhilin, 2001).
CHAPTER THREE
RESEARCH FRAMEWORK AND METHODOLOGY
This Chapter presents the different approaches used during implementation of the proposed system. This research adopted collaborative filtering techniques for the traditional (single) rating RS using Asymmetric Singular Value Decomposition, it further discussed the model based CF, aggregation function that was used to find the relationship f between the r0 (overall rating) and r1 (criteria rating). It described how the model was trained to determine a user’s preference. It presented the fuzzy logic system, the flow diagram and the programming language used.
Fuzzy logic system
Fuzzy logic is used to represent things in the real world that are crisp or put simply so unclear. The concept of Fuzzy Logic (FL) is derived from fuzzy set theory and was conceived by Lotfi Zadeh in 1975, a professor at the University of California at Berkley. It was not presented as a control methodology, but as a way of processing data by allowing partial set membership rather than crisp set membership or non-membership. Fuzzy Logic (FL) is a method of reasoning that resembles human reasoning. The approach of FL imitates the way of decision making in humans that involves all intermediate possibilities between digital values YES and NO. The conventional logic block that a computer can understand takes precise input and produces a definite output as TRUE or FALSE, which is equivalent to a human’s YES or NO, whereas a fuzzy set is determined by a membership function with a range of values between 0 and 1.
Definition 1.1: A fuzzy set A defined as a function [0,1] where X is the universe of discourse is represented by membership function that enhances its characteristic function of a set.
In a fuzzy set A, a membership function (MF) expressed as is defined as [0,1]. The value at element determines the degree of its membership of the element in the fuzzy set A. consequently,
CHAPTER FOUR
IMPLEMENTATION
This Chapter presents how the proposed system was modelled and simulated, it discussed the datasets used, the programming language used as well as the different evaluation metrics used to measure the prediction accuracy of the system, the experiment carried out and the results of the experiments.
CHAPTER FIVE
CONCLUSION
This chapter presents the summary of all the work conducted during this research, the contributions made as well as suggestions for future work in this subject matter.
Conclusion
Recommender System (RS) is well known for its great contribution in e-commerce and decision making. They are software tools and techniques that provides users with suggestions for items that are most likely to be of interest to them. RS makes these suggestions based on the implicit or explicit user feedback or rating submitted by the user in the form of transaction log that describes the amount of satisfaction they derived from an item. User ratings in RS might be in single rating, or multi-criteria rating, the single rating RS also known as traditional RS is an RS that determines the right item to direct to users using single rating techniques, whereas Multi-Criteria RS uses several features of an item to predict an item that a user might like. The main goal of this thesis was to improve the prediction accuracy of the system; a goal which was achieved by adopting the model based approach. This Collaborative Filtering technique uses prior activities or history of the users to learn the predictive model, and uses the aggregation function to determine the relationship between the overall rating and the criteria rating. To obtain the aggregation function f, we applied fuzzy logic features as the machine learning technique. Fuzzy logic is a model of reasoning that resembles human reasoning. With all these put together we built two RS, a single rating Recommender System using Asymmetric Singular Value Decomposition and a multi-criteria RS using fuzzy logic integrated with Asymmetric SVD. The dataset obtained from the Yahoo movie site is defined as ratings provided by users on four different rating aspects; story, direction, action and visual.
The test data was selected on a 10-fold cross validation and offline experiments were conducted to simulate the interactions of real users with the real-world system, while different evaluation metrics were used to measure the prediction of the proposed system and the traditional RS to determine the RS that can provide better prediction accuracy under the same circumstance. The results from these experiments showed that the proposed Fuzzy based MCRS provided the highest accuracy compared to the single rating Asymmetric SVD technique used. From indication, this study proposes that using Fuzzy logic together with single-rating technique built with Asymmetric SVD proved to be the best way to model an MCRS.
Challenges
During this research, many a challenge was encountered. However, we strategized different means to surmount these underlying challenges. Among some of these challenges are:
Timee constraint: Research is usually time consuming, this is because of the optimal need for the thoroughness at all stages of the research process, hence, a lot of hard work was applied to manage the limited timeframe and meet the deadline without compromising the quality of the research.
Insufficient material: Sourcing for research materials for this thesis was challenging because of the practical nature of our work, very limited amount of resources was available both in the library and on the internet, but we made maximum use of the resources we had which finally lead to the success of this work.
Future work
Although the fuzzy based approach used improved the prediction accuracy of the system, there is still a need to incorporate other features and techniques that can enhance the efficiency and precision of MCRS. In our work, we focused more on using the linguistic terms to model the user preferences, there is however still a need to consider incorporating the user mood, climatic and seasonal conditions as part of the criteria to which an item would be recommended. We also suggest applying other machine learning techniques such as combining Adaptive Genetic algorithm and Fuzzy logic, artificial neural network and fuzzy logic, Bayesian deep learning and Convolutional Neural Network on different datasets to enhance the prediction accuracy in modelling a MCRS.
REFERENCES
- Adomavicius, G., & Kwon, Y. (2007). New Recommendation Techniques for Multi-Criteria. Adomavicius, G., & Tuzhilin, A. (n.d.). Towards the Next Generation of Recommender
- Systems : A Survey of the State-of-the-Art and Possible Extensions, 1–43.
- Adomavicius, G. & A. Tuzhilin. Multidimensional recommender systems: a data warehousing approach. In Proc. of the 2nd Intl. Workshop on Electronic Commerce (WELCOM’01). Lecture Notes in Computer Science, 2232, Springer, 2001b.
- Zadeh, A. (1994). Fuzzy Logic, Neural Networks, and Soft Computing. Comm. ACM, 37
- (3), 77-84.
- Zadeh, A. (1975). The concept of a linguistic variable and its applications to approximate reasoning. Part I. Information Sciences 8, 199–249.
- Balabanovic, M., Shoham, Y. (1997). Content-based, collaborative recommendation.
- Communication of ACM 40(3), 66–72.
- Breese, J. S., D. Heckerman, and C. Kadie. Empirical analysis of predictive algorithms for collaborative filtering. In Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence, Madison, WI, July 1998.
