Literature Survey on Various Recommendation Techniques in Collaborative Filtering

Transcription

1 Literature Survey on Various Recommendation Techniques in Collaborative Filtering Mr. T. Sunil Reddy #, Mr. M. Dileep Kumar *, Mr.N. Vijaya sunder sagar # # M.Tech., Dept. of CSE, Ashoka Institute of Engineering & Technology, JNTU Hyderabad *Asst.Professor, Dept. of CSE, Ashoka Institute of Engineering & Technology, JNTU Hyderabad # Head, CSE Department, Ashoka Institute of Engineering & Technology, JNTU Hyderabad 1 dhyasaniss@gmail.com 2 dileepcool@gmail.com 3 vijaysundern@gmail.com Abstract: Recommendation system turn into a most important in web applications that provide many services and suggest some services automatically as per user s interest. In the achievement of current marketing field, different individual services in business play important roles. In websites, one of the precious tools of personal service is personalize recommendation technique. This technique makes great worth in internet marketing activities of e-commerce. To construct the recommendation system, the collaborative filtering is very useful technologies in the field of recommender systems.now a days the utilization of internet is expanding quickly, people approaches to internet to retrieve and share the information through numerous online tools. The achievement of the recommendation system relies on upon the value of the system. The usefulness can be measured in terms of accuracy, reliability, flexibility and diversity. To emphasis on improving recommendation accuracy, but diversity is neglected. There are many problems in today s web related to the accuracy of recommendation engine. So there are various techniques are used to improve the diversity and accuracy of recommendation system. Collaborative filtering (CF) is an important and popular technology for recommender systems. However, current CF methods suffer from such problems as data sparsity, recommendation inaccuracy, and big-error in predictions

2 I. INTRODUCTION There is tremendous measure of information available and the recommendation system is proven to be useful to prune the redundant information and making useful recommendation to the users. In E-Commerce, recommendation system plays an important role. COLLABORATIVE filtering (CF) is an important and popular technology for recommender systems. There has been a lot of work done both in industry and academia. These methods are classified into user-based CF and item-based CF. The basic idea of user-based CF approach is to find out a set of users who have similar favor patterns to a given user (i.e., neighbors of the user) and recommend to the user those items that other users in the same set like, while the item-based CF approach aims to provide a user with the recommendation on an item based on the other items with high correlations (i.e., neighbors of the item). In all collaborative filtering methods, it is a significant step to find users (or items ) neighbors, that is, a set of similar users (or items). Currently, almost all CF methods measure users similarity (or items similarity) based on corated items of users (or common users of items). Although these recommendation methods are widely used in E-Commerce. Most of recommendation systems focused on the preference given by user. User s preference for an item is 1. rating prediction: From the repository of information available, the ratings of unrated items are predicted. 2. Ranking: To rank the users rated items and then recommended to users to maximize the user s utility. There are different recommendation techniques like collaborative filtering, content based, demographic, utility based, knowledge based, typicality- based and hybrid. II. Recommendation Techniques: 2.1. Content based recommendation: The meaning of content based recommender system is that to get features from the content of item, in many cases they are words which describe the item. This recommendation system also has a ability to introduce new items to the user. For recommending items it also provides explanation to users. The main task of this method is analyze the set of documents of items, which is rated by

3 user and construct a model of user interest based on features of the items. Origins of content based filtering approach is in information retrieval and information filtering. It gives only textual information. Such as news webs and documents. 2.2 Knowledge based recommendation This recommendation technique use knowledge about product and users needs for making recommendations. Recommendation is made by matching the similarity between user s preference and product description. It does not suffer from ramp up problem since it does not depend on the ratings given by user. This system needs a database and needs to be updated for making useful recommendations. 2.3 Graph Based Recommendation Most of the recommender system uses two dimensional information like user and item. Homogeneous and heterogeneous graphs can be used which provides the capability to deal with multidimensional information like user s intensions. A graph-theoretic approach [9] based on maximum flow or maximum bipartite computations represent user and item as vertices and the flow is calculated. This technique improves aggregate diversity of recommendations. One of the major drawback with the graph-based approach is when the input data becomes large it becomes tedious to rebuild graph. 2.4 Hybrid recommendation It is combination of content based filtering and collaborative filtering approach. The content filtering is based on relationship between content or descriptions. While collaborative filtering is based on user s preference and the similarity with active user. It adopts nearest neighbour algorithm, so calculate distance between the users according to users preference and find out target user s nearest neighbour. The recommendation system recommends resources to user in a personalized way and this system helps users find out the needed resources. This technique improves the recommendation accuracy but diversity is not considered. 2.5 collaborative based recommendation

4 The main important method in recommendation system is collaborative filtering methods that find a group of people who share the same interest with you. These people could be determined by the similar ranking on items. These people are called the neighbourhood of the current user. If an item is liked by many people in this neighbourhood, then it is very possible to be liked by current user. Netflix is using collaborative filtering method. If a recommendation system involves rating on items then it is probably using collaborative filtering method. Another kind of collaborative filtering method is called item-to-item collaborative filtering method. Collaborative filtering recommendation system is further classified into two algorithms. They are depicted below. 1. Memory based CF: This collaborative filtering method recommend the items based on the past activities of users. 2. Model based CF: This collaborative filtering method has an ability to learn a predictive model based on the past user activities using statistical or machine learning model. In order to have the good recommender system it is clear that a good mechanism to find neighbors of users is very important. A better way to select neighbors of users or items for collaborative filtering can facilitate better handling of the challenges. We note that using rated items to represent a user, as in conventional collaborative filtering, only captures the user s preference at a low level (i.e., item level). Measuring users similarity based on such a low-level representation of users (i.e., corated items of users) can lead to inaccurate results in some cases. In reality, people may like to group items into categories, and for each category there is a corresponding group of people who like items in the category [3]. Cognitive psychologists find that objects (items) have different typicality degrees in categories in real life [4], [5], [6]. For instance, people may consider that a sparrow is more typical than a penguin in the concept of bird, and Titanic is a very typical romance movie, and so on. Similarly, different people may have different degrees of typicality in different user groups (i.e., sets of persons who like items of particular item groups). For instance, Raymond is a very typical member of the concept users who like war movies while not so typical in the concept users who like romance movies. The typicality of users in different user groups can indicate the user s favor or preference on different kinds of items. The typicality degree of a user in a particular user group can reflect the user s preference at a higher abstraction level than the rated items by the user

5 IV. CONCLUSION: This paper discusses the current ranking techniques to improve the accuracy of recommendation system and diversity is neglected. First it introduces recommender system and its importance in today s world and in real application. Than it summarizes some of recommendation technique like collaborative filtering, content-based filtering, graph-based technique, knowledge-based and hybrid approach. Even in collaborative filtering it is explained about typicality based CF where, a user is represented by a user typicality vector that can indicate the user s preference on each kind of items. A distinct feature of TyCo is that it selects neighbors of users by measuring users similarity based on their typicality degrees instead of co rated items by users. Such a feature can overcome several limitations of traditional collaborative filtering methods. VI. REFERENCE [1] Zhonghuo Wu,Jun Zheng,Su wang Hongfeng Feng Combined predictor for item based collaborative filtering th International Conference on Intelligent Networking and Collaborative Systems. [2] Chandima Hewa, Nadungodage, Yuni Xia GPU Accelerated item based collaborative filtering for big data applications 2013 IEEE International Conference on Big Data [3] Jinbo Zhang, Zhiqing Lin, Bo Xiao, Chuang Zhang An optimized item based collaborative filtering recommendation algorithm Proceedings of IC-NIDC2009 [4] Heung-Nam Kim 1, Ae-Ttie Ji 1, and Geun-Sik Jo 2 Enhanced prediction algorithm for item- based collaborative filtering recommendation Springer-Verlag Berlin Heidelberg 2006 [5] Badrul Sarwar, George Karypis,Joseph Konstan, and John Riedl item-based collaborative filtering recommendation algorithms ACM /01/0005. [6] Suyun Wei, Ning Ye, Shuo Zhang, Xia Huang, Jian Item-based collaborative filtering recommendation algorithm combing item category with interestingness measure 2012 International Conference on Computer Science and Service System [7] Yi Cai, Ho Fung Leung, Qing Li, Hauqing Min, Jie Tang and Juanzi Li Typicality based collaborative filtering algorithm IEEE transactions knowledge and data engineering. [8] Mohhammad Hamidi Isfahani and Farid Khosh Alhan New hybrid recommendation system based on C-means clustering method th conference on information and knowledge technology

6 [9] Cheng Qaio, Huang Jian, Gong Jian- Xing, Hao Jian-Guo Simulation resource recommendation system based on collaborative filtering IEEE Computer Society [10] Greg Linden, Brent Smith, Jeremy York and AMAZON.COM item- to item collaborative filtering method IEEE Computer Society, january february 2003 [11] Gediminas Adomavicius, Member, IEEE, and Alexander Tuzhilin, Member, IEEE Toward the Next Generation of Recommender Systems: A Survey of the State-of-the-Art and Possible Extensions IEEE june [12] Kangning Wei 1, Jinghua Huang 2, Shaohong Fu 3 A Survey of E-Commerce Recommender Systems IEEE [13] Sarwar B karypis g,konstan.j,riedl.j, Item based collaborative filtering recommendation algorithm ACM, New York, [14] Breiman.L,Random forests, in journal. Machine learning. Vol 45, pp. 5-32, Kluwer Academic Publishers Massachusetts(2001). [15] Resnick P, Iacovou N, et al, Grouplens: an open architecture for collaborative filtering of netnews, Proceedings of the ACM Conference on Computer Supported Cooperative Work, pp175-86, [16] Deng Ai-lin, Zhu Yang-yong, Shi Bo-le. A collaborative filtering recommendation algorithm based on item rating prediction. Journal of Software, 2003 [17] M. Deshpande and G. Karypis, Item based top-n recommendation algorithms, ACM Transactions on Information Systems, 22: ,