Imputed Neighborhood Based Collaborative Filtering

Abstract

Collaborative filtering (CF)

A Novel Imputation-Boosted Technique to Overcome the Unrated Items Issue and Improving the Performance of Collaborative Filtering

“…The missing data can be managed by training models for various combinations of modalities and by selecting an convenient model for each combination [9,10] or by applying generic methods to combine all modalities in the presence of missing data, such as imputation of the missing data or modification of the fusion algorithm [11]. [12,13] proposed an imputation boosted collaborative filtering technique (IBCF). They imputes the user-item rating matrix with predicted ratings to alleviate the missing data using different machine learning models.…”

Scholarly paper recommendation via user's recent research interests

“…However, it suffers from "coldstart problem," in which it cannot generate accurate recommendations without enough initial ratings from users. Recent works alleviate this problem by introducing pseudo users that rate items [23] and imputing estimated rating data using some imputation technique [39].…”

Proposing a New Metric for Collaborative Filtering

“…A widespread approach in memory-based collaborative filtering is the k-nearest neighbor algorithm. In [1] two imputed neighborhood based collaborative filtering algorithms are proposed that improved the performance of recommender system in very sparse rating data. Another popular technique is the top-N recommendations which recommends a set of N top-sorted items which probably will be interesting items by a target user [2,3].…”

“…For example in Figure 1, 1 , u D P =4 In order to determine the distance order (interaction sequence) of u 1 from u 2 and u 3 in Figure 1, (10) is applied. As Figure 1 shows, 7 items are in common among all users.…”

“…Users, who are fan of these kinds of movies, would watch and rate it as soon as possible, but those who are not very interested in mentioned type of movie maybe watch it later or when there is no one left in the list which is fascinating from their point of view. Hence we can infer, this group of users is different from pervious users and one of the most important factors which help us to recognize this issue is measuring the 1 Collaborative Filtering level of time distances in interactions of users. An intelligent recommender system must be able to adapt itself with current requirements of users.…”