Collaborative Filtering and Recommender System in Knowledge Management
During live interactions, techniques enabling discovery of knowledge are used in recommender systems. Their use is normally in solving the issue of recommending information, services or products. This is the concept referred to as collaborative filtering. The systems, particularly the k-nearest neighbor based on collaborative filtering have achieved increasing effectiveness on the Web. The techniques have a critical role of filling in the user-item association matrix’s missing entries. Model-based collaborative filtering is currently supported by MLlib, where the defining of products and users is done by a miniature set of concealed factors that can apply to predicting the entries that are missing. The alternating least squares (ALS) algorithm is applied by MLlib in learning the concealed factors (Adomavicius and Tuzhilin, 2005).
From a knowledge management point of view, collaborative filtering is basically information filtering process or models utilizing techniques that involve collaboration amongst some viewpoints, agents, data sources, etc. use of these techniques typically entails data sets that are extremely large. According to Francesco, Lior and Bracha (2011), the methods of collaborative filtering are applicable to diverse types of data such as sensing and monitoring data, financial data, environmental sensing, or in e-commerce and web-based applications. From a simpler perspective, the concept applies to the means of rendering automatic predictions regarding a user’s interests by collecting information relating to an individual’s taste or preferences from multiple users (collaborative processes).
In collaborative filtering, future predictions are made from the past user behavior. The information that is used is the interests and preferences that a large number of users express. Such preferences could be openly expressed, in the form of ratings that show the satisfaction level or could be obtained from implicit sources, like purchase histories or views expressed on user’s pages. With the Internet, such information is readily accessible explaining the increasing popularity of collaborative filtering-based solutions (Beck and Schenker-Wicki, 2014). Also, collaborative filtering is a highly adaptable approach, given the fact that it is not dependent on the recommended content.
While collaborative filtering systems come in various forms, the most commonly applied one uses a process reduced in two steps. The first step is looking for the users using the common patterns of rating with the user identified as active or the user that the prediction is being made for. The second step involves using the ratings of the users that were established in the first step in calculating the active user’s prediction. This is the form that is referred to as the user-based collaborative filtering. The Nearest Neighbor algorithm that is user-based is the particular application of this form (Francesco, Lior and Bracha, 2011).
Otherwise, one can use the item-based collaborative filtering, which is based on the assumption that the users who purchased product X also purchased product Y. The algorithm proceeds in an item-centric way. The first step is building a matrix based on an item to determine the association between items’ pairs. The second step is inferring the tastes of the present user through examination of the matrix and matching the data of the user (Francesco, Lior and Bracha, 2011).
Depending on a rating or scoring system is argued by Beck and Schenker-Wicki (2014) to provide even more reliable predictions. While it has challenges given the fact that it assumes the user’s particular demands, it remains widely used in association with data clustering and web search. Following is a problem formulation that enables an alternative definition of collaborative filtering. This is the most commonly used form of the concept and is the algorithm for estimating the matrix’s missing entries.
Let U denote sets of users and I items.
|U| = nU and |I| = nI .
r: U ×I ! S, a rating function is a user-item pairs’ mapping into a rating scale S. in most cases; S is represented as a series of natural numbers, normally of 5 or 10 in length. For specified sets of users and items, the values of S can form a matrix through tabularization.
R = [rui]U×I , in which rui is a rating that user u gives to the item I (Adomavicius and Tuzhilin, 2005).
There is a challenge posed to recommender systems by the increasing amount of information available on the Web as well as the increase in the number of those visiting different sites. Some of the challenges include the performance of recommendations per second for the users and items coming in millions, production of recommendations that are of high quality and achievement of high-level coverage within the reality of data sparsity. While the challenges are facing the traditional systems of collaborative filtering, new technologies have brought in new possibilities that can be implemented by organizations to achieve greater level of efficiency and effectiveness for people, products’ and services (Adomavicius and Tuzhilin, 2005).
Therefore, organizations need to seek new technologies and innovative algorithms to increase operational efficiency. New systems that generate recommendations of high quality even in large-scale problems are a must for any organization that seeks to achieve efficiency and effectiveness (Francesco, Lior and Bracha, 2011).Based on the needs of the organization for information, any of the three approaches can be utilized, but for more accurate and reliable recommendations, it would be better to implement the system based on rating. This is because the productions based on users or items can fail to provide the desired outcomes.
Adomavicius, G. and Tuzhilin, A. (2005). “Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions”, IEEE Trans. Knowl. Data Eng., 17(6): 734–749
Beck, M. and Schenker-Wicki, A. (2014). Cooperating with external partners: The importance of diversity for innovation performance. Eur. J. Int. Manag. 8, 548–569.
Francesco, R. Lior R. and Bracha S. (2011). Introduction to Recommender Systems Handbook, Recommender Systems Handbook, Springer, 2011, pp. 1-35