Situation recognition for recommendation using merge-split approach
In one embodiment, data relating to usage patterns of the user is stored, wherein the data includes information as to items which were used and the context in which they were used. The data is then clustered into input clusters of data points. It is determined if there are any input clusters that are similar to each other. Similar clusters are merged if there are any input clusters similar to each other. Any non-merged input clusters are divided into split clusters if the split clusters would not be similar to each other. The determining, merging, and dividing are then repeated using the merged, divided, and remaining unmerged and undivided clusters as input clusters.
1. A method for making recommendations to a user, the method comprising:
storing data relating to usage patterns of the user, wherein the data includes information as to items which were used and the context in which they were used, wherein context includes a situation in which the user and device were operating when the items were used;
randomly initializing a set number of cluster centroids;
performing an initial clustering algorithm to cluster the data into input clusters of data points based upon the cluster centroids;
performing a merge-split clustering algorithm on the input clusters, wherein the merge-split clustering algorithm comprises:
determining if there are any input clusters that are similar to each other;
merging the similar clusters if there are any input clusters similar to each other:
dividing any non-merged input clusters into split clusters if the split clusters would not be similar to each other; and
repeating the determining, merging, and dividing using the merged, divided, and remaining unmerged and undivided clusters as input clusters.
2. The method of claim 1 , wherein the merge-split clustering algorithm further comprises not repeating the determining, merging, and dividing if there are no input clusters that are similar to each other.
3. The method of claim 1 , wherein the dividing is performed before the determining and the merging.
4. The method of claim 1 , further comprising:
determining a centroid for each of the input clusters;
selecting clusters similar to a current context of the user by comparing a data point representing the current context of the user to one or more of the centroids;
computing, for each of one or more items, a probability that the user wishes to use the corresponding item, based on the selected similar clusters; and
recommending one or more of the items having based on the probability that the user wishes to use them.
5. The method of claim 1 , wherein the performing an initial clustering algorithm is performed using a K-means technique.
6. The method of claim 1 , wherein the performing an initial clustering algorithm is performed using a co-clustering technique.
7. The method of claim 1 , wherein the performing an initial clustering algorithm is performed using a hierarchical technique.
8. The method of claim 1 , wherein the context includes the user's location.
9. The method of claim 1 , wherein the user is operating a mobile device.
10. The method of claim 1 , wherein the context includes a time of day in which the items were used by the user and device.
11. The method of claim 1 , wherein the context includes a location in which the items were used by the user and device.
12. An apparatus comprising:
an interface; and
one or more processors configured to:
store data relating to usage patterns of the user, wherein the data includes information as to items which were used and the context in which they were used;
randomly initialize a set number of cluster centroids;
perform an initial clustering algorithm to cluster the data into input clusters of data points based upon the cluster centroids;
perform a merge-split clustering algorithm on the input clusters, wherein the merge-split clustering algorithm comprises:
determining if there are any input clusters that are similar to each other;
merging the similar clusters if there are any input clusters similar to each other:
dividing any non-merged input clusters into split clusters if the split clusters would not be similar to each other; and
repeating the determining, merging, and dividing using the merged, divided, and remaining unmerged and undivided clusters as input clusters.
13. The apparatus of claim 12 , wherein the apparatus is a mobile device.
14. A system for making recommendations to a user, the system comprising:
means for storing data relating to usage patterns of the user, wherein the data includes information as to items which were used and the context in which they were used, wherein context includes a situation in which the user and device were operating when the items were used;
means for randomly initializing a set number of cluster centroids;
means for performing an initial clustering algorithm to cluster the data into input clusters of data points based upon the cluster centroids;
means for performing a merge-split clustering algorithm on the input clusters, wherein the merge-split clustering algorithm comprises:
determining if there are any input clusters that are similar to each other;
merging the similar clusters if there are any input clusters similar to each other:
dividing any non-merged input clusters into split clusters if the split clusters would not be similar to each other; and
repeating the determining, merging, and dividing using the merged, divided, and remaining unmerged and undivided clusters as input clusters;
a computer processor configured to interact with the means for storing, means for performing an initial clustering algorithm, and means for performing a merge-split algorithm and to perform calculations based on input therefrom, providing results.
15. The system of claim 14 , wherein the context includes a time of day in which the items were used by the user and device.
16. The system of claim 14 , wherein the context includes a location in which the items were used by the user and device.
17. A non-transitory storage device readable by a machine tangibly embodying a program of instructions executable by the machine to perform a method for making recommendations to a user, the method comprising:
storing data relating to usage patterns of the user, wherein the data includes information as to items which were used and the context in which they were used, wherein context includes a situation in which the user and device were operating when the items were used;
randomly initializing a set number of cluster centroids;
performing an initial clustering algorithm to cluster clustering the data into input clusters of data points based upon the cluster centroids;
performing a merge-split clustering algorithm on the input clusters, wherein the merge-split clustering algorithm comprises:
determining if there are any input clusters that are similar to each other; merging the similar clusters if there are any input clusters similar to each other:
dividing any non-merged input clusters into split clusters if the split clusters would not be similar to each other; and
repeating the determining, merging, and dividing using the merged, divided, and remaining unmerged and undivided clusters as input clusters.
18. The non-transitory storage device of claim 17 , wherein the context includes a time of day in which the items were used by the user and device.
19. The non-transitory storage device of claim 17 , wherein the context includes a location in which the items were used by the user and device.