Collaborative Filtering Based Recommendation

Published Date: 02 Nov 2017

3.1 Introductions

3.2 Recommendation System

With the increase in amount of information across the world, it is necessary to process data more quickly in the exigent environment. For processing the data in the real world scenario, two terms i.e. Data Mining and Recommendation System plays a key role in social network. Data mining is defined as the process of mining unnecessary information and summarize them into useful form, while recommendation system is suggesting that information to user and incorporate the data mining techniques like clustering, association rules etc. which helps in decision making. So it can be said that recommendation system [1] is one which processes enormous data and suggests useful and interesting data, to a set of users on the basis of user interest. Recommendation System takes item ranking given by user on the basis of item feature and recommend item to a set of users as shown in Figure 1.

C:\Users\hp\Desktop\nai.png

Fig 1: Recommendation System

One of the major applications of recommendation system is in online shopping where user finds an item to purchase. An item has many features and it is responsibility of recommendation system to suggest the useful item to a user on the basis of their feature requirements. A recommendation system is prominent part of networking environment and applies in many different areas like tag recommendation, music recommendation, item recommendation, movie recommendation etc. Movie recommendation system is one of the thrust areas in recommendation systems.

3.3 Recommendation Method Classification

Currently Recommendation methods are mainly classified into three broad categories.

1. Collaborative filtering based recommendation.

2. Content based recommendation.

3. Hybrid recommendation system.

3.3.1 Collaborative Filtering based Recommendation

Collaborative filtering method is one of the most popular researched techniques of Recommendation system which predict the user’s likeness based on their similarity with others users. Collaborative filtering is capable of recommending more accurately even for complex items because this method does not rely on machine analyzable content. The idea of collaborative filtering is to finding users in a community that share appreciations. Collaborative filtering builds a group (also called neighborhood) of similar taste users or who have same rated items or almost same rated items in common. If an item is not rated earlier by a user, but neighbourhood of user rated already them, than recommendation of that item given to the user. For example a neighborhood is build between the similar tastes of three users who rated the movie with similar marks. User A hasn’t rated that movie which he hasn’t watched yet but that movie was positively rated by other user, than movie will recommend to user A. Two types of similarity, either user based or item based is find out in collaborative filtering method. In user based collaborative filtering method, similarity is find out between users and item is recommend to those users who have similar interest. In items based user is provided to items that has high correlation from other items. Collaborative filtering is widely used in e-commerce and often suffer from the problems of cold start, scalability, and sparsity.

3.3.2 Content based Recommendation

Based on the subjective evaluation of some items in the past, the idea of this kind of recommendation will helps in evaluating the other similar items in the future. Content-based recommender systems work with profiles of users that contain the information about user and taste and created at the beginning of time based on a weighted vector of item features. For creating a profile, a survey is made by recommendation system for collecting the initial information about a user and that will avoid the new user problem. In the recommendation similarity is find out between the item positively rated by user and unrated item and the item similar to positively rated items will recommend to users. Profile of other users is not essential for recommendation and does not affect the recommendation process because recommendation is based on individual information. Collaborative filtering method is generally based on the information about and characteristic of recommended items or item that are going to be recommended. For efficient and better filtering, content based recommendation system mostly uses tags or keywords.

3.3.3 Hybrid Recommendation System

For better results and overcome the common problem in recommendation system like cold start, scalability and sparsity problem, some recommender systems combine different techniques of collaborative and content based approaches, which produced Hybrid recommendation system. This method can be implemented in several ways; one of them is making content based and collaborative based prediction separately and then combine them, another one is by adding the content based capabilities to collaborative-based approach(or vice versa). NetFlix and CinemaScreen is an example of a hybrid recommendation. Result of several experiment show that hybrid method is better than the pure content based and collaborative filtering method.

3.4 Modern Recommendation Approaches

Context-aware approaches

Semantic based approaches

Cross-domain based approaches

Peer-to-Peer approaches

Cross-lingual approaches

3.4.1 Context Aware Approaches

Context contain the detailed of situation he/she is in at the time of rated the items along with the information about the user. Detail of recommendation may play an important role in recommendation rather than rating of items because rating alone does not specify under which circumstances they were given by users. For some users, recommendation system can be more suitable in evening and does not his preferences in morning and for some of user it is easy to do one thing when it’s cold and completely another when it’s hot outside. These types of recommendation system which focus and utilize such information are called context aware system. Unlike content information that is saved in profiles, context information changes dynamically, saved permanently and after a certain period of time it’s loses his currency. Context aware is very important for periodically refreshing the information, increased the quality of recommendation, and in some areas this approaches become more specific. Contextual information and user items relationship can be represented in many ways, one of them is by contextual graph. In contextual graph node represent the items, users, and contextual object while edge represents the ratings of items and similarity indicator between users. Another way apart from representation is use of graph algorithm which is used to improve the prediction. There having some problem in working with contextual information as it is difficult to obtain the contextual information. This contextual information is obtained from making a survey by directly interact with user and asking him to fill out a form. Another way to obtaining the contextual information is using the sources like GPS, to get location or a time stamp on transaction. Lastly by analysing users, their behaviour or data mining techniques contextual information can be obtained.

3.4.2 Semantic based Approaches

Textual form is the way of representing the users in recommendation system, description of items and web. Use of tags and keyword does not improve the accuracy of recommendation if they haven’t any semantic meaning because some keyword may be homonyms. This is the reason understanding and structuring of text is very significant part of recommendation. Based on lexical and syntactical analysis, a traditional recommendation system show description that users can understand but which not understood by a recommendation system or computer. To overcome this new text mining techniques is required which is based on the semantic analysis and system is known as semantic based recommendation system. The performance of such type of recommendation system is knowledge based. TripFromTV is an example of recommendation agent which is based on semantic analysis and work with digital television. Agent create a profile of user by creating the classification form collecting the information about what kind of channels, programs and films user like and based on that profile and history agent can give a recommendation to user.

3.4.3 Cross Domain based Approaches

Important part of recommending process in collaborative recommender systems is finding similar users and building an accurate neighbourhood. Based on their appreciations of items similarities of two users is determined. But there is not sureties of similar appreciation in one domain are similar with another domain valuation. Standard recommendation is based on collaborative filtering where users are compared without splitting in different domain. In cross domain system similarity of users is determined which are domain dependent in which for each user according to domain an engine creates local neighbourhoods. These computed similarity and set of nearest neighbour are sent for overall similarities computation. Recommendation system determines these overall similarities; create overall neighbourhoods and makes prediction and recommendation.

3.4.4 Peer-to-Peer Approaches

The recommender systems with P2P(Peer to peer) approaches are decentralized. Each peer can relate itself to a group of other peers with same interests and get recommendations from the users of that group. Recommendations can also be given based on the history of a peer. Decentralization of recommender system can solve the scalability problem.

3.4.5 Cross Lingual Approaches

The recommender system based on cross-lingual approach is occured when users receive recommendations to the items that have descriptions in languages they don’t speak and understand. The main idea is to map both text and keywords in different languages into a single feature space, that is to say a probability distribution over latent topics. Using dictionaries the system parses keywords from the descriptions of items than translates them in one defined language. After that, using collaborative or other filtering, the system gives recommendations to users. With the help of semantic analysis it’s possible to make a language-independent representation of text. Cross-lingual recommender systems break the language barrier and gives opportunities to look for items, information, papers or books in other languages.

3.5 Challenges and Issue in Recommendation System

There are several issue and challenges related to recommendation.

3.5.1. Scalability: Scalability is the problem occurred with the increase in large amount of users and items. System needs more number of resources for processing the large amount of users and items that finds users with similar taste and makes recommendation. Combination of various type of filters and physical improvement of system can solved the scalability problem in recommendation system.

2. Sparsity- Sparsity problem is the problem of lack of information or problem of having too few ratings. Sparsity problem is occurred when available data are insufficient for identifying similar users and items; hence it is difficult to find out correlation between users and items. For example in online shopping where a huge amount of users and items are present, user rate item according to his interest but the sparsity problem is occurred when user rated just a few items. In that case it is difficult to determine the user true taste and there is possibility that user could be related to wrong neighbourhood.

3. Cold start: Cold Start problem is occurred when recommendation is take place to new user. At starting, profile of new user is empty because he hasn’t rated any item yet so taste of user is unknown to the system. This problem is solved by making a survey in which profile is created. Cold start problem can also occur in items when they are new in system and haven’t rated before. Cold start problem for either user or item can be solved by hybrid approaches.

4. Privacy: Now a day’s most important problem in recommendation is privacy. For receiving most accurate and correct recommendation, the information regarding user such as the location of particular user should be validated. By using the specialized algorithm and programs, effective protection of privacy is provided by many online shops.

5. Recommendation accuracy: Main goal of recommendation system is to predict user preferences and rating as accurately as possible. Increase in error of prediction may, reduce the trust of users on recommendation.

3.6 Applications of Recommendation Systems

Several important application of recommendation system is discussed below.

Product Recommendations: Most important use of recommendation system is at on-line retailers. Some of the on-line vendors like Amazon take the suggestion on items which they want to buy. These suggestions are based on the purchasing decision made by the customer who they are similar.

Movie Recommendations: It is also one of the most important applications of recommendation system. Netflix provide movie recommendation facility to its customers, whose rate the movie depend on their likeness. Based on rating provided by users to movie, recommendation is performed.

News Articles: application of recommendation system in News article is that it finds the interested article to readers based on the article they have read in the past. For recommending the article, similarity of important word in the document is determined with the article to be recommended or on the articles that are read by people having similar reading taste.

1.7 Limitation of Recommendation System

Recommendation system is used for suggesting the items, movie, video etc to a set of similar users, having some limitation. New user and new item are two distinct but related limitation of recommendation system. It is difficult to recognize a new user having few rating. There are different limitations for using recommender systems. The most two distinct but related well problems are new user and new item problems. A new user with few ratings becomes hard to recognize in recommender systems. Similarly a new item with few ratings cannot be easily recognized by the recommendation system, so there is a need to encourage users to rate items in such systems.

3.8 Summary and Discussion

http://www.snet.tu-berlin.de/fileadmin/fg220/courses/WS1011/snet-project/recommender-systems_asanov.pdf

http://en.wikipedia.org/wiki/Recommender_system

Yi Cai, Ho-fung Leung, Qing Li, Huaqing Min,Jie tang and Juanzi Li, "Typicality-based Collaborative Filtering Recommendation",IEEE TRANSACTION ON KNOWLEDGE AND DATA ENGINEERING,Jan 2013.