Challenges of Recommendation Systems - Assignment Example

 Challenges of recommendation systems Introduction Recommender systems are the inventions of new age technologies. They are nothing but basic kinds of software that help users to find the information that they require to clear their ambiguity while taking a decision. It basically performs an advisory role where it takes inputs from the user and based on his/her input, it provides recommendations in that specific region that the user asked for. The user could ask for anything right from books, movies to things that people find difficult to ask for in a realistic situation. It basically uses the power of web services in providing a solution to users to discover things that they would never had an opportunity to find earlier. This technology has turned the corner stone for resource search in its way. It has become the precious piece of software that was searched forever to complete the information search puzzle. Example of such systems includes sites that are used by Yahoo, Google, Amazon, etc. Standardization of these systems still has a long way to go. Following are the different challenges that are posed to developing these systems with their possible solutions. ( ACM 2009) Challenges posed by data The most integral part of the recommender system is data. The data needs to be consistent and should suffice the requirements of the user. In order to make effective recommendations to the user, the system needs huge amounts of data. But, to get lot of data, it depends on the number of users who are using the system. The availability of good recommendations to the user depends on the number of inputs given by the user and hence, a systematic manipulation of data behavior, leads to satisfying recommendations. Therefore collection of data provides a good recommendation system and vice versa. (MacManus 2009) Next challenge posed by data is its accuracy. It is the determining factor of data quality. If it goes wrong, then the entire system is in trouble. There are several possibilities for data to be inaccurate. Hence appropriate measures are required to identify the mistakes in the entry stage itself. It can either be due to typographical errors, ineffective preliminary training etc. Another cause of concern is the wrong information entered by users. Though these are unavoidable, proper steps can be followed to ensure some kind of accuracy in this area. Other problems for data accuracy are that most of the data are constantly changing and so up-to-date information becomes all the more important. Other causes include transfer of data from one place to other and often irrelevant algorithms applied to find recommndations. (Olson 2002) User’s trust is one of the constant challenges of developing a recommender system. The trust of a user varies on both technical and sociological perspectives. Trust in electronic commerce can be evaluated to two models. One is the interpersonal way, where a user can trust other users of his choice but, not necessarily all. Second is the impersonal way where a user keeps his full trust on the entire system rather than on individuals. Since the benefits of trust are distinct to everyone, several models of trust framework have been proposed for the recommendation systems to manage quality in data and in the recommendations provided. (MacManus 2009) Challenges posed by Models There are basically three broad categories of models that are being currently followed for developing recommendation systems. They are: Content based; Collaborative and Hybrid models. Choosing what model to follow while developing this system is a difficult question to answer. The choice of the system depends on various factors such as domain knowledge, product database, demographics, user categories, basic interest of majority of users and the data being explored. Selection of exact model for interaction depends on whether the user just wants to be encapsulated of the happenings in the system or not. This is explored as the data or process explorative models. Hence the main challenge remains to be extracting articulate answers from these models instantly. (Burke 2005) Shielding and manipulation When there is a proposal for sharing of data between users to provide recommendations, there comes a scenario of security and integrity. The challenges posed by them are too large. But, with the effective advancements in technologies, the attainment of security in any kind of open systems has been well structured and well implemented. (Herlocker 2009) Cold Start The problem of cold start arises in a recommender system when the system is not able to support the user’s inputs by providing appropriate recommendations. It may be due to lack of enough information or not able to identify the behavioral inferences from user’s input. An example of cold start in a content based recommendation model is when the system is not able to provide apt recommendation until it gets input about the user’s preferences either manually or through the recent activities of user. One apt solution that could be considered to solve this scenario is through the methodology of cloud computing, where agents from different areas assist users and get input from them to provide a combined set of data useful to the system. All these collected data’s are placed in common repository, where different businesses share their part of information and vice versa. The proposed hybrid model is developed as a solution to the cold start problem were content based collaborative filtering models are run together to gather maximum possible data that could reiterate to every user actions. (Yezdi, Max & Pattie M) Similarity of recommendation systems Identifying similarities in a recommendation system is an advanced concept in providing recommendations. Although implementing it seems to be a tough ask for the developers, the results yielded are far better than other systems without similarity based approach. For example, two people purchasing same set of items would indicate a similarity in their tastes. This reading could prove vital for the development of the system. These results could be used as a valuable input in providing the recommendations. But the challenge here remains to be the security of the user details. It could happen that the users may not want their identity to be revealed in public. Hence the system should be developed in such a way that only the data is being used as an input to the system rather than user identities. Measurements of similarity through techniques like Euclidean distance help in a long way in achieving the proposed benefits of models based on similarity metrics. (Renals 2009) Delivery of recommendations Recommendations delivered by the system should be dynamic and should be on demand for the user. The recommendations have to be smart enough to satisfy the user requirements and should be more personalized to the user. The delivery should be prompt with accurate and up to date information. The ease of usability should be prevalent to the developers and the system should be developed on the lines of customer’s perspectives. The usage of cloud computing increases the delivery recommendations to be more accurate since the data has been approved after going through several site verifications. Though the availability of resources is unpredictable, the level of service offered through this is magnanimous and the quality remains to be unmatchable.( Tongchuay & Praneetpolgrang 2008) The different aspects of a recommender system were discussed. The challenges posed by them and the possible solutions to counter affect the challenges were detailed. The adoption of cloud computing in determining recommendations as forwarded the development of recommender systems to an altogether new level. References MacManus, R 2009, 5 problems of recommender systems, Read write web, viewed 29 October 2009, http://www.readwriteweb.com/archives/5_problems_of_recommender_systems.php. ACM 2009, 3rd ACM Conference on Recommender Systems 2009, viewed 29 October 2009, http://recsys.acm.org/. Tongchuay , C., Praneetpolgrang, P 2008, ’ Knowledge quality in knowledge management systems using trust –recommendations‘, IEEE Conference 2008, viewed 29 October 2009, http://www.ieee.th.org/IEEEConference2008/Proceedings2008/papers/IEEE_Full_Paper_Chawanrat.doc_Paper_1.pdf Olson J, 2002, Data accuracy: the challenge, Infomanagement direct, viewed 29 October 2009, http://www.information-management.com/infodirect/20021108/6019-1.html Burke R, 2005, Recommender Systems and user modeling, viewed 29 October 2009, http://www.cri.haifa.ac.il/events/2005/recommender/airs05-1.pdf Renals S 2009, Similarity and recommender systems, viewed 29 October 2009, http://www.inf.ed.ac.uk/teaching/courses/inf2b/learnnotes09/inf2b09-learn02-lec.pdf Herlocker, J. L., 2009 Position Statement | Explanations in Recommender Systems, viewed 29 October 2009, http://www.patrickbaudisch.com/interactingwithrecommendersystems/WorkingNotes/JonHerlockerExplanationsInRecommenderSystems.pdf Yezdi L., Max M. & Pattie M., 1994. ‘Collaborative Interface Agents’, Proceedings of the Twelfth National Conference on Artificial Intelligence. AAAI Press. Seattle, Washington pp. 444–449. Read More