IJCST Vol. 6, ISSue 3, July - SepT 2015

w w w . i j c s t . c o m InternatIonal Journal of Computer SCIenCe and teChnology 113

ISSN : 0976-8491 (Online) | ISSN : 2229-4333 (Print)

A Hybrid Online Genre-based Recommender System 1Dr. Amit Verma, 2Harpreet Kaur Virk

1Dept. of Computer Science, Chandigarh University, Gharuan, India

Abstract Recommendation is a sort of web intelligence technique used for filtering information to provide relevant data to the people on daily basis. There are various approaches used for recommendation. In content-based filtering, the systems examine items previously chosen by the actual user, whereas in collaborative filtering, recommendations are based on the information of similar users or items. The recommendations are also influenced by the factors such as age, gender and some other user profile information. In this paper, both content and collaborative techniques along with some demographic information are combined into a hybrid approach, where additional content features are used to improve the accuracy of collaborative filtering. And the genetic algorithm along with kNN approach and correlation is used to provide recommendations to the user. The system is evaluated using precision and recall parameters.

KeywordsClusters, Classification, kNN, Demographic Information etc

I. IntroductionEarly in the history of computing capability of computers to provide recommendations was fairly recognized. An early step towards the automatic recommendation systems was Grundy [34]; a computer based librarian. It provides recommendations by grouping users into stereotypes based upon some short interviews. A solution for online information overload was provided by collaborative filtering in the early 1990s. A manual collaborative filtering recommender system - Tapestry[35] allow users to query for items based on other users’ actions or opinions. To determine the relevance of recommendations it allows the users to exploit the reactions of prior readers. Soon automatic collaborative-based systems followed to automatically locate related opinions and combining them to make recommendations. Amazon.com is the most popular application using this technology in which they recommend items for the users based on their purchase and browsing history. While adoption of recommender technology by Amazon frequently based on collaborative filtering, has been incorporated in various e-commerce and online systems. A considerable inspiration for doing this is to boost sales degree as the customers may pay for an item if it is recommended to them but might not look for it otherwise. Beyond collaborative filtering, the toolbox of recommendation techniques has also incorporate content-based techniques based on information retrieval, Bayesian inference, and case-based interpretation methods [8-9]. However, collaborative filtering has remained a valuable technique, both unaided and hybridized with content-based approaches. To provide an overview of the different types of recommender systems, we want to cite a taxonomy provided by [12].

In content-based recommendation approach, the system analyze a set of documents or descriptions of items that are previously rated by a user, and create a model or profile of user preferences based on the features of the items rated by that user[4]. The profile is an organized representation of user preferences which is adopted to recommend new interested items. Basically, the recommendation process comprise of matching up the features of the user profile

against the features of a content item. The result is a pertinence judgment that represents the level of user interest in that item.

Content based

Collaborative based

Demographic based

Knowledge based

Hybrid

Recommender System

Fig. 1: Classification of Recommender System

Most of all collaborative filtering is the most familiar and broadly accepted technique. It works on gathering and analyzing the huge amount of users’ data [13] by tracking user actions and preferences to predict recommendations based on the relationship with other users [14]. It provides recommendations based on simple principle that like-minded people have alike interest [15]. Knowledge-based recommendation systems provide suggestions based on specific domain knowledge with reference to user preferences. Constraint-based and case-based [16-17] are types of knowledge-based systems. Knowledge-based systems have a tendency to work better systems than others only if they are equipped with learning mechanism.

Hybrid methods combine any of two or more techniques. Mostly collaborative and content filtering approaches are combined to provide hybrid approaches.

In this paper, a hybrid movie recommender system is proposed in which both content and collaborative techniques along with some demographic information are combined, where additional content features are used to improve the accuracy of collaborative filtering. And the genetic algorithm along with correlation is used to provide recommendations to the user.This paper consists of five sections. Section II describes related work. Section III describes the methodology to be followed to recommend movies to the user. Section IV is the experimental analysis. Section V wraps up this paper with its conclusion and future scope.

II. Related WorkMaria S. Pera et.al.[21] presented a system named as GroupReM which provide recommendations to the members of a group considering individual members’ interest and build a profile that reflects the needs of the group on movies, find the similar content using word correlation and also consider the popularity of the movies. GroupReM mainly employs tags for capturing the information of movies for recommendation task. This system can easily be extended to generate group recommendations for items other than movies. Experimental studies have been conducted using MovieLens dataset which show that it is an effective and highly efficient system for group recommendations.Punam Bedi et.al.[22] introduced a Trust based Ant Recommender System (TARS) which generates relevant recommendations by including the concept of dynamic trust among users and

IJCST Vol. 6, ISSue 3, July - SepT 2015 ISSN : 0976-8491 (Online) | ISSN : 2229-4333 (Print)

w w w . i j c s t . c o m 114 InternatIonal Journal of Computer SCIenCe and teChnology

selecting the best neighborhood based on genetic image of ant colonies. It provides additional information for explanation of recommendations about the connectedness in trust graph from where recommendations are produced. Two datasets Jester dataset and MovieLens dataset are used for performance evaluation of TARS. Sang Hyun Choi et.al.[23] proposed HYRED, a hybrid recommendation algorithm which combines collaborative filtering by means of modified Pearson’s binary correlation coefficients with content-based filtering by means of generalized distance to boundary based rating. It utilizes the nearest and farthest neighbors of a target user to yield useful information from large dataset avoiding sparsity and scalability problem. This increases the performance of the system and reduces the effort to perform computations.

Beel, J., Langer et.al.[24] analyzed several recommendation techniques with both online and offline evaluations and they found contradiction among the results of both online and offline evaluations. Offline evaluations may be unsuitable for evaluation of research paper recommender systems.Jonathan L. Herlocker et.al.[28] reviewed the collaborative filtering RSs based on user tasks, types of datasets, how the prediction quality is calculated and assessment of prediction attributes. Presented experimental outcome from the analysis of several precision or accuracy metrics on one content field where every tested metric distorted approximately into three equivalence classes. Each equivalence class contains strongly correlated metrics, whereas metrics from diverse classes were not correlated.

J. Bobadilla et.al.[29] a metric is presented to calculate similarity among users and formulated through simple linear combination of values and weights that can be applied in collaborative filtering based recommender systems. Weights are computed only once while the values are computed for every pair of users with similar preferences. The results present valuable improvements in prediction quality and performance.

H. Chang et.al.[30] proposed a context model to offer users appropriate services and using context to manage resources efficiently in Mobile Cloud environment. The proposed context model optimizes the personalized services. And an aggregated information based recommendation approach is presented which classifies the aggregated information into time, location and resources and create a profile. This information is analyzed by recommender system to generate context. Mainly it is not a filtering approach. Its role is to improve the legacy filtering technique and escalating recommendation approach into mobile cloud computing.

From the literature survey we conclude that the sometime recommendations are not according to the need and preferences of the users. Collaborative filtering approach suffers from several shortcomings. Scalability factor is attempted to improve the performance of the system in this paper. Recommendations are also influenced by the factors such as age, gender and some other user profile information. Considering this, age and gender information of the user along with collaborative filtering to provide recommendations is used. Further the genetic algorithm along with genre correlation to provide recommendations is used in this paper.

III. Proposed WorkRecommendation process for our work mainly consists following steps. First, collect the data of different users and secondly, extraction of valuable user information from offline database SQLite in android and movie information from RottenTomatoes and TMdb online. Then analyze this information to provide relevant recommendations. The explanation of steps of proposed methodology:

A. Collect User InformationFirst of all, we have to collect user data to provide recommendations. The new users will register themselves and their information will be stored in the user database.

B. Extract User Profile FeaturesIn next step, we will extract the features from the user profile such as its search history, age, gender etc. in order to provide recommendations.

C. Similarity MeasureFor finding the neighbor set of the active user, similarity measure is used. First we gather the users who have related taste with the active user then the distance measure d and feature weight w between the active user and his neighbors can be found by the Euclidean Distance function:

Where a is the active user, j is the neighbor, and a≠s.i is the common liking between a and j.k is the total number of common items.f is the feature number.n is the total number of features.wf is the weight of feature f for user a.ua.i.f is the value of feature f on item I for active user a.Then the matrix will be generated.

1. InitializationIn this step the population will be initialized as

and on the basis of initial population, new individuals are produced which are evaluated by fitness function in each iteration.

2. Fitness FunctionFitness is an estimated function to evaluate the features weight of individual and judge its prediction accuracy.

3. SelectionThe choosing of selection operators is an essential part in genetic algorithms. This part is independent of other parts in genetic algorithms and has no direct relation with the problem and with the fitness function, crossover operator and mutation operator used in genetic algorithms. In this research, a probabilistic selection is performed based upon the individual’s fitness such that the better individuals have an increased chance of being selected.

4. CrossoverThen we perform crossover for the selected neighbors. The probability for crossover is 0.7.

IJCST Vol. 6, ISSue 3, July - SepT 2015

w w w . i j c s t . c o m InternatIonal Journal of Computer SCIenCe and teChnology 115

ISSN : 0976-8491 (Online) | ISSN : 2229-4333 (Print)

5. MutationThen mutation will be performed to select the best recommendation. The probability for mutation is 0.05.

User Database

Online Movie

Database

New User

Store user information

Apply correlation on genre attribute

Generate n*n correlation matrix

Initialize population as

Convert xi to binary

For all combinations perform crossover

Perform mutation

Is correlation = best

correlation???

Save for recommendation

Fig. 2: Flowchart of Proposed Work

IV. Experimental AnalysisThe proposed system generates recommendations based on the genre preferred by users considering the two factors age and gender. The figures below show the output for two different users (6 and 7). The first user has preferred the genre 6 and 8 whereas the second user respectively preferred genre 7 and 8.

Fig. 6: System Output 1

Fig. 7: Recommendations for user6

Fig. 8: Recommendations for user7

V. ConclusionTo find out related content is very difficult task in current scenario where there are huge amount of data is stored in the databases. Recommender systems are solution to this problem and attracting researchers to explore this area in past few years. This paper tries to solve the problem for recommending movies. In this paper, an application is proposed which recommends relevant data to the user according to their preferences and history in movies. Wrong recommendations assigned to the user tend to decrease the efficiency of the system but this problem is reduced by using content and collaborative approach and optimal recommendations are provided to the user. . In our work, both content and collaborative techniques and some demographic information are combined into a hybrid approach, where additional content features are used to improve the accuracy of collaborative filtering. Also the genetic algorithm is used to provide recommendations to the user.Further research can be performed in the field of non-technical aspects like privacy resulting from user based tagging.

VI. AcknowledgementsWe would like to thank Dr. (Prof.) Amit Verma the head of department at Chandigarh University) for their ideas that helped guide our research.

References[1] Simon Dooms, Toon De Pessemier , Luc Martens,“Online

optimization for user-specific hybrid recommender systems,” In: Springer , Multimed Tools Appl, 2014.

[2] D. Goldberg, D. Nichols, B. M. Oki, D. Terry,“Using collaborative filtering to weave an information tapestry,” In: Communications of the ACM, Vol. 35, No. 12, pp. 61–70, 1992.

[3] Linas Baltrunas, Francesco Ricci,“Experimental evaluation of context-dependent collaborative filtering using item splitting,” In: Springer, User Model User-Adap Inter, pp.

IJCST Vol. 6, ISSue 3, July - SepT 2015 ISSN : 0976-8491 (Online) | ISSN : 2229-4333 (Print)

w w w . i j c s t . c o m 116 InternatIonal Journal of Computer SCIenCe and teChnology

7–34, 2014. [4] W. Hill, L. Stead, M. Rosenstein, G. Furnas, “Recommending

and evaluating choices in a virtual community of use.” In: ACM CHI ’95, pp. 194–201, ACM Press/Addison-Wesley Publishing Co., 1995.

[5] U. Shardanand, P. Maes,“Social information filtering: Algorithms for automating “word of mouth,” In: ACM CHI ’95, pp. 210–217, ACM Press/Addison-Wesley Publishing Co., 1995.

[6] A. Ansari, S. Essegaier, R. Kohli,“Internet recommendation systems,” In: Journal of Marketing Research, Vol. 37, No. 3, pp. 363–375, August 2000.

[7] P. M.West, D. Ariely, S. Bellman, E. Bradlow, J. Huber, E. Johnson, B. Kahn, J. Little, D. Schkade,“Agents to the Rescue?,” In: Marketing Letters, Vol. 10, No. 3, pp. 285–300, 1999.

[8] J. B. Schafer, J. A. Konstan, J. Riedl,“E-Commerce recommendation applications,” In: Data Mining and Knowledge Discovery, Vol. 5, No. 1, pp. 115–153, Janaury 2001.

[9] B. Smyth,“Case-based recommendation,” In: The Adaptive Web, vol. 4321 of Lecture Notes in Computer Science, (P. Brusilovsky, A. Kobsa, and W. Nejdl, eds.), pp. 342–376, Springer, 2007.

[10] R. Burke,“Hybrid recommender systems: Survey and experiments,” In: User Modeling and User-Adapted Interaction, Vol. 12, No. 4, pp. 331–370, November 2002.

[11] Bamshad Mobasher,“Recommender Systems,” In: Kunstliche Intelligenz, Special Issue on Web Mining, No. 3, pp. 41-43, 2007.

[12] K. Choi, D. Yoo, G. Kim, Y. Suh,“A hybrid online-product recommendation system: Combining implicit rating-based collaborative filtering and sequential pattern analysis,” In: Electron. Commer. Res. Appl., Vol. 11, No. 4, pp. 309–317, Jul. 2012.

[13] Burke, R., “Hybrid web recommender systems. In: The Adaptive Web,” In: Springer Berlin / Heidelberg, pp. 377–408, 2007.

[14] J. Lee, M. Sun, G. Lebanon,“A comparative study of collaborative filtering algorithms,” In: arXiv Prepr. arXiv1205.3193, pp. 1–27, 2012

[15] V Formoso, F Cacheda, V Carneiro,“Algorithms for Efficient Collaborative Filtering,” In: Efficiency Issues in Information Retrieval Workshop, p. 17. 2008.

[16] Bridge, D., G¨oker, M., McGinty, L., Smyth, B.,“Case-based recommender systems,” In: Knowledge Engineering review 20(3), pp. 315–320, 2006.

[17] Ricci, F., Cavada, D., Mirzadeh, N., Venturini, A., “Case-based travel recommendations,” In: D.R. Fesenmaier, K.Woeber, H.Werthner (eds.) Destination Recommendation Systems: Behavioural Foundations and Applications, pp. 67–93. CABI,2006.

[18] P Han, B Xie, F Yang, R Shen, “A scalable P2P recommender system based on distributed collaborative filtering.” In: Expert systems with applications 27, pp: 203-210, 2004.

[19] Janusz Sobecki, “Implementations of Web-based Recommender Systems Using Hybrid Methods”, In: International Journal of Computer Science & Applications, Vol. 3. Issue 3, pp. 52 – 64, 2006.

[20] Claypool, M., Gokhale, A., Miranda, T., Murnikov, P., Netes, D., Sartin, M. “Combining Content-Based and Collaborative Filters in an Online Newspaper,” In: Proceedings of ACM

SIGIR Workshop on Recommender Systems, 1999.[21] Maria S. Pera, Yiu-Kai Ng, “A group recommender for movies

based on content similarity and popularity,” In: Information Processing and Management 49, pp. 673–687, 2013.

[22] Punam Bedi, Ravish Sharma, “Trust based recommender system using ant colony for trust computation,” In: Expert Systems with Applications 39, pp. 1183–1190, 2012.

[23] Sang Hyun Choi, Young-Seon Jeong, Myong K. Jeong, “A Hybrid Recommendation Method with Reduced Data for Large-Scale Application”, In: IEEE Transactions on Systems, Man, And Cybernetics—Part C: Applications And Reviews, Vol. 40, No. 5, September 2010.

[24] Beel, J., Langer, S., Genzmehr, M., Gipp, B. and Nurnberger, A., “A Comparative Analysis of Offline and Online Evaluations and Discussion of Research Paper Recommender System Evaluation,” In: Proceedings of the Workshop on Reproducibility and Replication in Recommender Systems Evaluation (RepSys) at the ACM Recommender System Conference (RecSys), 2013.

[25] Nan Zheng, Qiudan Li, “A recommender system based on tag and time information for social tagging system,” In: Expert Systems with Applications 38, pp. 4575–4587, 2011.

[26] Qing Li, Byeong Man Kim, “Clustering Approach for Hybrid Recommender System,” In: Proceedings of the IEEE/WIC International Conference on Web Intelligence, 2003.

[27] Bamshad Mobasher, Robin Burke, Runa Bhaumik, Chad Williams,“Towards Trustworthy Recommender Systems: An Analysis of Attack Models and Algorithm Robustness,” In: ACM Transactions on Internet Technology, Vol. 7, No. 2, May 2007.

[28] Jonathan L. Herlocker, Joseph A. Konstan, Loren G. Terveen, And John T. Riedl, “Evaluating Collaborative Filtering Recommender Systems,” In: ACM Transactions on Information Systems, Vol. 22, No. 1, pp. 5–53, January 2004.

[29] Jesus Bobadilla, Fernando Ortega, Antonio Hernando, Javier Alcala, “Improving collaborative filtering recommender system results and performance using genetic algorithms,” In: Knowledge-Based Systems 24, pp. 1310–1316, 2011.

[30] Changbok, Hyokyung Chang, Yongho Kang, Yumi Bae, Hyosik Ahn, Euiin Choi, “Filtering Technique on Mobile Cloud Computing,” In: Energy Procedia 16, pp. 1305 – 1311, 2011.

[31] Linas Baltrunas, Xavier Amatriain, “Towards Time-Dependant Recommendation based on Implicit Feedback.” In: ACM CARS, New York, USA, 2009.

[32] Mojtaba Salehi, Mohammad Pourzaferani, Seyed Amir Razavi, “Hybrid attribute-based recommender system for learning material using genetic algorithm and a multidimensional information model,” In: Egyptian Informatics Journal 14, pp. 67–78, 2013.

[33] Kate Starbird, Grace Muzny, Leysia Palen, “Learning from the Crowd: Collaborative Filtering Techniques for Identifying On-the-Ground Twitterers during Mass Disruptions,” In: Proceedings of the 9th International ISCRAM Conference – Vancouver, Canada, April 2012.

[34] E. Rich, “User modeling via stereotypes,” In: Cognitive Science, Vol. 3, No. 4, pp. 329–354, October 1979.

[35] P. Resnick, N. Iacovou, M. Suchak, P. Bergstrom, J. Riedl, “GroupLens: an open architecture for collaborative filtering of net news.” In: ACM CSCW ’94, pp. 175–186, ACM, 1994.