Algorithmic Approach To Address Item Recommendation And Cold Start Issues In Social Networks

Social networks are a digital representation of real world relationships like friends, relatives, acquaintances, colleagues, etc. Just like the way humans interact with each other in the real world by talking about different issues, giving opinions and sharing things with each other; social networks provide the humans with same capabilities in the digital world. With the advancements in Web2.0applications, social networks have enabled people to keep in touch even with those people whom a person will not be interacting on a regular basis. This results into a rapid and enormous influx of information and finding and prioritizing the information becomes a tedious task.Social networks generally use collaborative filtering, content filtering or hybrid systems to recommend items to the users that are consistent with their choices and previous browsing patterns. Earlier approaches have been adopted to gauge the importance of the content by calculating relevancy score for a particular user. The users are served with the item with maximum score in the recommendation system eco-system. The scoring system encompasses multiple attributes for calculating the scores which are different for different contexts. For example, social networks would use users’ closeness and previous browsing patterns as vital attributes in calculating score where as other general recommender system may use similar users browsing patterns as the vital attribute in generating recommendations. The study aims at deducing scores from the attributes that are common across social networks like privacy settings of the content, creation time, degree of closeness with the other user, etc.Another important aspect pertaining to the recommender systems is the cold start problem where the lack or sparseness of data becomes a bottleneck for the recommender system. The study aims at addressing this problem by finding community leaders and serving content to the cold start users that originated from the community leaders.The study also proposes a solution to store the data and an algorithm to compute the scores in such a way that it can be used in live applications. The study was conducted on a live social network Wospace (www.wospace.net). Experimental results were gauged by the user response to changes of recommendation mechanisms. The user response based on clicks, views and activities increased significantly showing the effectiveness of the approach.