SURVEY ON RECOMMENDER SYSTEMS FOR SOLVING COLD START PROBLEM

Naveena M, Shivakumar N
2019 International Journal of Recent Trends in Engineering and Research  
Recommender system is an information filtering system which is used to extract the exact content from massive amount of data set based on user preference or behavior. It will become popular in various areas include news, research article, artificial intelligence, data mining, big data analytics, products, etc. The task of recommender system is to predict the user's ratings for each item and ranking the items. In research area RS plays a major role. There are different approaches in recommender
more » ... ystem such as time-aware, event-aware, content-aware and location-aware. This paper mainly focuses on the aspects of RS, issues and challenges of RS. It also includes the survey on cold start problem. The major concern of the cold start problem is non-accessibility of data needed to achieve recommendations. Thus non-accessible data is gathered using the various proposed methods. The gathering of information is done by asking the user explicitly or making use of existing data implicitly. Hence the solutions are divided into two groups depending on the method of gathering this data. There are various methods for gathering the missing data based on the type of information gathering. A survey on the existing solutions of how to gather the missing information is described as follows: Accuracy of Related Recommendation A user is attracted if the recommendation is related to them. Accuracy of related recommendation is defined as the ratio of the amount of related recommendations to the amount of total recommendations. The various metrics for evaluating the accuracy is described in [1] . The efficiency and helpfulness of the scheme is measured using this metric. If small related recommendations are done as gathering the user's profile, the scheme's rate of accuracy might drop. Hence, the solutions proposed must protect the total accuracy. This can be done by choosing the fewer amounts of query pieces which have great information. Decreasing Bias The communication among the users and items are assumed to be obtained by ratings. The communication among them is independent for some ratings. For example, the ratings are very high for the most popular items and some items are rated without considering the knowledge in utilizing it. These unfair ratings hindrances cause special recommendations. The solution includes the usage of baseline predictors to decrease the bias, although it needs the history of ratings to achieve maximum accuracy.
doi:10.23883/ijrter.conf.20190322.055.4z5km fatcat:l7man4gpvndwhmavchvj2mouse