Improving the test time of M-distance based recommendation system
By: Hasanzadeh, Narges
.
Contributor(s): Forghani, Yahya.
Publisher: New York Springer 2022Edition: Vol.103(1), Feb.Description: 119-129p.Subject(s): Humanities and Applied SciencesOnline resources: Click here
In:
Journal of the institution of engineers (India): Series BSummary: The M-distance-based recommendation system (MBR) is one of the most successful types of recommendation systems. In the training phase of MBR, the average of ratings given to each item is used to determine similar items. To estimate the active user’s rating on an item in the test phase of MBR, only rated items of the active user are necessary, whereas in MBR, all nearest neighbors of that item are examined and then, unrated neighbors of the active user are ignored. In most datasets, the number of unrated items is very high and, therefore, the most of nearest neighbors are unrated, examining all nearest neighbors in the test phase of MBR is unnecessary and time-wasting. In this paper, a new data structure is proposed to improve the test time of MBR. In the training phase, the rated items of each user are stored in this data structure. By employing this data structure in the test phase, it will be then unnecessary to examine unrated items. Strictly speaking, the nearest neighbor of each unrated item of the active user is determined in the test phase by examining the small set of rated items of the active user stored previously in the data structure. According to the experiments conducted on five real datasets, the runtime of our proposed method is 2.05, 10.68, 132.19 and 21.77 times less than that of MBR for the datasets ML-1 m, ML-10 m, Douban and EachMovie, respectively.
| Item type | Current location | Call number | Status | Date due | Barcode | Item holds |
|---|---|---|---|---|---|---|
Articles Abstract Database
|
School of Engineering & Technology Archieval Section | Not for loan | 2022-1578 |
The M-distance-based recommendation system (MBR) is one of the most successful types of recommendation systems. In the training phase of MBR, the average of ratings given to each item is used to determine similar items. To estimate the active user’s rating on an item in the test phase of MBR, only rated items of the active user are necessary, whereas in MBR, all nearest neighbors of that item are examined and then, unrated neighbors of the active user are ignored. In most datasets, the number of unrated items is very high and, therefore, the most of nearest neighbors are unrated, examining all nearest neighbors in the test phase of MBR is unnecessary and time-wasting. In this paper, a new data structure is proposed to improve the test time of MBR. In the training phase, the rated items of each user are stored in this data structure. By employing this data structure in the test phase, it will be then unnecessary to examine unrated items. Strictly speaking, the nearest neighbor of each unrated item of the active user is determined in the test phase by examining the small set of rated items of the active user stored previously in the data structure. According to the experiments conducted on five real datasets, the runtime of our proposed method is 2.05, 10.68, 132.19 and 21.77 times less than that of MBR for the datasets ML-1 m, ML-10 m, Douban and EachMovie, respectively.
Click on an image to view it in the image viewer
There are no comments for this item.