An Efficient and Intelligent Recommender System for Mobile Platform

Recommender Systems are valuable tools to deal with the problem of overloaded information faced by most of the users in case of making purchase decision to buy any item. Recommender systems are used to provide recommendations in many domains such as movies, books, digital equipment’s, etc. The massive collection of available books online presents a great challenge for users to select the relevant books that meet their preferences. Users usually read few pages or contents to decide whether to buy a certain book or not. Recommender systems provide different value addition factors such as similar user ratings, users past history, user profiles, etc. to facilitate the users in terms of providing relevant recommendations according to their preferences. Recommender systems are broadly categorized into content based approach and collaborative filtering approach. Content based or collaborative filtering approaches alone are not sufficient to provide most accurate and relevant recommendations under diverse scenarios. Therefore, hybrid approaches are also designed by combining the features of both the content based and collaborative filtering approaches to provide more relevant recommendations. This paper proposes an efficient hybrid recommendation scheme for mobile platform that includes the traits of content based and collaborative filtering approaches in addition of the context based approach that is included to provide the latest books recommendations to user.Objective and subjective evaluation measures are used to compute the performance of the proposed system. Experimental results are promising and signify the effectiveness of our proposed hybrid scheme in terms of most relevant and latest books recommendations.


INTRODUCTION
a laptop then it can be very hectic experience for him/ her to analyze all available reviews about the laptop of different brands and models before deciding to Authors E-Mail: (m.jabbar@uog.edu.pk, qaisar@iiu.edu.pk, arifmuhammad36@hotmail.com, asim@iiu.edu.pk, ali.javed@uettaxila.edu.pk) combine the useful features of both domains to present more effective recommendations to the users.
Existing mobile based books recommender systems still need many improvements in terms of providing accurate and most relevant recommendation among the books recommendations provided to the user. Moreover, existing systems that are either using collaborative filtering or content based approaches provide limited precision and recall. In most cases the recommendation results of these systems deviate from the potential interests or preferences of the users. Existing research work on books recommendation have applied different techniques to provide the most suitable systems that can provide more relevancy in recommendations. Zhu et. al. [4] designed abook recommendation system using association rule mining. There are some limitations in mining association rule using apriori algorithm. However, efficiency of apriori algorithm is improved through filtering the useful records while ignoring the irrelevant data. This improved algorithm is then used to build a book recommender system using data set from university circulation library records. A service model has also been introduced in [4] to offer the information about the recommendations to users that can also be utilized in other domains i.e. information retrieval, book store service, etc [4]. Rana et. al. [5] pr oposed a r ecommen der system to pr ovide recommendations about relevant books to the users by finding other users having same preferences. This system [5] introduces a concept of temporal dimension which computes the frequency of user liking of any item in a specific time period. This system provides diverse recommendations in case an item has fewer ratings for specific user.

An Efficient and Intelligent Recommender System for Mobile Platform
There exist many book recommender systems in the form of web applications that are used to provide recommendations about different books to the user based on his/her input query. However, all recommendations provided by these systems are not relevant or most relatable to the users' interests or preference and contains much irrelevant information. In addition, we have seen a drastic technology shift in the last decade where we experienced the smart phones revolution. In modern era, every user prefers to use applications on the mobile as compared to desktop/web applications a decade back.
Mobile applications are more convenient for users as he/ she can have easy access of application anywhere/ anytime without the need of a big size computer/laptop as mobiles are easy to carry. In addition, mobile phones have constraints of memory and processing which must be considered before development. Therefore, researchers are focusing more towards developing efficient mobile based recommender systems nowadays to meet the needs of the mobile users.

LITERATURE REVIEW
This section provides a critical analysis of existing state-  [7]. In recent years, online recommender systems are providing the functionality of a social recommendation system. They predict whether a particular user will like a particular product or not. They also identify a set S of items that will be of interest to a particular user [8]. the books to e-user community [9]. Fig. 1 shows the pictorial representation of existing techniques of recommender systems.
The first term coined in the context of recommender ystem is "collaborative filtering". It was developed to generate the list of recommended documents from newsgroups for group of users [10]. Collaborative filtering approach identifies the users that have similar interest with the active user and find the ratings provided by these users to different products, then recommends the list of products to active user that are highly rated by these similar users.
Collaborative filtering are further divided into two categories i.e. model based and neighborhood based approach. Neighborhood based approach is commonly

FIG. 1. TECHNIQUES FOR RECOMMENDER SYSTEMS
known as memory based approach [11]. In this approach, a small number of users are selected based on similarity with the active user, then predictions are performed based a weighted mixture of ratings of these users. This approach assigns the weights to all users based on their similarity with the active user. In the next step m number of users are selected having highest similarity to the active user.
Finally, prediction is achieved using the weighted ratings of selected users. For similarity computation mechanism it is assumed that unrated products are not the ones with negative or zero ratings [12]. al. [15] have used the data such as title synopses, author, subject items and ratings to train a classifier for books recommendation. The scale of rating is from 1 to k where mapping is performed among k different classes. However, different other classification algorithms have also been used purely for content based filtering such as decision tree, K-nearest neighbor and neural network [16].
They presented an argument based hybrid approach for books recommendation. Argumentation facility has been provided to convince the user regarding the utility of the

PROPOSED SYSTEM
The

Proposed Methodology
This section provides the discussion of the proposed methodology of our hybrid recommendation approach. The proposed application then search the database to find the books similar to the ones user liked in step-3. The filtered books are assigned the weights in the next stage as Equation (1): Here r ji is the rating of ith book from list J.
In step-4, the proposed system identifies other users from the user dataset who liked to read the similar category of books which the active user wants to read.
After finding the similar users, the system checks the rating of these users from rating table to find their highly Where P is the total number users and Pci is the number of users whom interest is similar to the active user.
In step-7, the system combine the difference of rating calculated in step-5 with active user given rating to get the predicted rating. In step-8, the system calculates a weighted rating list from the predicted rating computed in step-6 and rating frequency computed in step-5. In step-9, the system finds the books of user's interest from the books dataset. This list of books is then assigned a weight W 2 in step-10 (Equation (4)).
Here r ki is the rating of ith book from the list K.
In step-11, three lists of books are combined using an aggregated function. The aggregation function finally creates a single consolidated list of books using the weights of these three lists of books. The aggregation function is computed as Equation (5): The proposed algorithm of our hybrid recommendation approach is provided in Table 1.

PERFORMANCE EVALUATION
Performance of the proposed recommender system is evaluated using two methods i.e. objective and subjective

Dataset
A standard dataset "Book Crossing" is used to measure the performance of the proposed method. Ziegler et. al.

Experimental Results
Experimental results of the proposed method are

Objective Evaluation
For objective evaluation of the proposed method, we have measured the performance on precision, recall, accuracy,  (6): (iv) Calculate average dissimilarity in ratings that users gave in regard to user 'U'.
(v) Calculate predicted rating by adding the average variance in rating and user 'U' provided rating.
(vi) Get weighted list 'J2' from predicted rating and rating's frequency. Where P avg (I,U) represents the average precision value.
As shown in Table 1, the proposed approach is very effective in terms of providing relevant recommendation books among the total books provided to the user.
We also computed the F-Measure as well against the recommendations offered by the proposed method.
From Table 1 it can be clearly observed that the proposed

Subjective Evaluation
The proposed hybrid recommendation approach is also evaluated in a subjective way. Subjective evaluation is based on user feedback or ratings on a subjective scale provided by multiple users based on their experience. We

Performance Comparison
We designed an experiment to compare the performance of the proposed method against existing state-of-the-art recommender systems [21] including collaborative and content based approaches. We have implemented both

CONCLUSION
In this paper we proposed an efficient hybrid books recommender system for mobile platform. The