Cross-listed property matching using image descriptor features
Two sets of data, each containing property listings, are obtained from two discrete merchant platforms. Each property listing in a set of data of a first merchant is sequentially paired with each of the property listings in a set of data of a second merchant. For each pair, each image of the property listing of the first merchant is compared to each image of the property listing of the second merchant, and images of statistically sufficient similarity are identified. The similarity of images, and in particular, of similar images likely to be rooms of the property, are considered in a determination of whether the product listings of the first and second merchant are for the same cross-listed product.
1. A method for identifying correlation between property listings, the method comprising:
identifying, within a geographic area of a predetermined size, a source property listing and a target property listing;
storing, in a first database, at least one image corresponding to the source property listing;
storing, in a second database, at least one image corresponding to the target property listing;
for each image corresponding to the source property listing and each image corresponding to the target property listing, identifying a group of keypoints in the image, each keypoint corresponding to a keypoint descriptor;
for each image corresponding to the source property listing, determining an image category for the image;
for each image corresponding to the source property listing,
(i) comparing the keypoint descriptors of the group of keypoints in the image to the respective keypoint descriptors of the group of keypoints in each of the at least one images corresponding to the target property listing;
(ii) determining, based on the comparing, a number of shared keypoints between the image corresponding to the source property listing and each respective image corresponding to the target property listing;
(iii) determining, for each image corresponding to the target property listing, based on the determination of the number of shared keypoints, whether to pair the image corresponding to the source property listing with the image corresponding to the target property listing;
determining a likelihood of correlation between the source property listing and the target property listing based on one or more of: (1) a number of shared keypoints between a first image corresponding to the source property listing and a second image corresponding to the target property listing, (2) whether the first image is paired with one or more images corresponding to the target property listing, and (3) the image category for the first image; and
removing, from the first database, at least one image corresponding to the source property listing having an image category of view.
2. The method of claim 1 , wherein the image category for the first image is one of: kitchen, bathroom, living room, bedroom, pool, and view.
3. The method of claim 2 , wherein the likelihood of correlation between the source property listing and the target property listing is higher where (i) the first image is paired with one or more images corresponding to the target property listing and (ii) the image category for the first image is one of: kitchen, bathroom, or living room.
4. The method of claim 1 , further comprising:
identifying source description data corresponding to the source property listing; and
identifying target description data corresponding to the target property listing,
wherein the determining of the likelihood of correlation between the source property listing and the target property listing is further based on a comparison of the source description data and the target description data.
5. The method of claim 1 , wherein the identifying of the source property listing and the target property listing comprises:
identifying a first geographic area;
dividing the first geographic area into one or more virtual areas of the predetermined size;
selecting, from among the one or more virtual areas, a second geographic area, the second geographic area being smaller than the first geographic area;
identifying, within the second geographic area, (i) the source property listing and (ii) a location of the source property listing;
identifying a set of one or more target property listings within a first distance from the source property listing; and
selecting a target property listing from the set of one or more property listings.
6. The method of claim 1 , wherein the geographic area is one of: a squared area with a predetermined size of 1.3 km in length, or a circular area with a predetermined size of 1.3 km in diameter.
7. A method comprising:
storing, in a first database, at least one image corresponding to a source property listing;
storing, in a second database, at least one image corresponding to a target property listing;
for each image corresponding to the source property listing, determining an image category for the image;
for each image corresponding to the source property listing, (i) comparing the image to each of the at least one image corresponding to the target property listing and (ii) determining a level of similarity between the image corresponding to the source property listing and each image corresponding to the target property listing;
determining a likelihood of correlation between the source property listing and the target property listing based on: (1) whether the level of similarity between a first image corresponding to the source property listing and a second image corresponding to the target property listing meets a predetermined threshold of similarity and (2) the image category for the first image; and
removing, from the first database, at least one image corresponding to the source property listing having an image category of view.
8. The method of claim 7 , further comprising:
for each image corresponding to the source property listing, determining, in a case that the level of similarity between the image and an image corresponding to the target property listing exceeds a predetermined threshold, that the image corresponding to the source property listing and the image corresponding to the target property listing match,
wherein the determining of the likelihood of correlation between the source property listing and the target property listing is further based on the number of the one or more images corresponding to the source property listing that match with at least one of the one or more images corresponding to the target property listing.
9. The method of claim 7 , wherein the level of similarity between a first image corresponding to the source property listing and a second image corresponding to the target property listing is determined by comparing a plurality of keypoints of the first image with a plurality of keypoints of the second image.
10. The method of claim 7 , wherein the level of similarity between a first image corresponding to the source property listing and a second image corresponding to the target property listing is determined by performing an image hash comparison.
11. The method of claim 7 , further comprising:
identifying source description data corresponding to the source property listing; and
identifying target description data corresponding to the target property listing,
wherein the determining of the likelihood of correlation between the source property listing and the target property listing is further based on a comparison of the source description data and the target description data.
12. A system comprising:
a memory configured to store at least one image corresponding to a source property listing and at least one image corresponding to a target property listing; and
at least one processor configured to:
for each image corresponding to the source property listing and each image corresponding to the target property listing, identify a group of keypoints in the image;
for each image corresponding to the source property listing, determine an image category for the image;
for each image corresponding to the source property listing, (i) compare the image to each of the at least one image corresponding to the target property listing and (ii) determine a level of similarity between the image corresponding to the source property listing and each image corresponding to the target property listing;
determine a likelihood of correlation between the source property listing and the target property listing based on: (1) whether the level of similarity between a first image corresponding to the source property listing and a second image corresponding to the target property listing meets a predetermined threshold of similarity and (2) the image category for the first image; and
remove, from the memory, at least one image corresponding to the source property listing having an image category of view.
13. The system of claim 12 , wherein the image category for the first image is one of: kitchen, bathroom, living room, bedroom, pool, and view.
14. The system of claim 13 , wherein the likelihood of correlation between the source property listing and the target property listing is higher where (i) the first image is paired with one or more images corresponding to the target property listing and (ii) the image category for the first image is one of: kitchen, bathroom, or living room.
15. The system of claim 12 , wherein the level of similarity between a first image corresponding to the source property listing and a second image corresponding to the target property listing is determined by comparing a plurality of keypoints of the first image with a plurality of keypoints of the second image.
16. The system of claim 12 , wherein the memory is further configured to store (a) source description data corresponding to the source property listing and (b) target description data corresponding to the target property listing, and
wherein the determining of the likelihood of correlation between the source property listing and the target property listing is further based on a comparison of the source description data and the target description data.
17. The system of claim 12 , wherein the identifying of the source property listing and the target property listing comprises:
identifying a first geographic area;
dividing the first geographic area into one or more virtual areas;
selecting, from among the one or more virtual areas, a second geographic area, the second geographic area being smaller than the first geographic area;
identifying, within the second geographic area, (i) the source property listing and (ii) a location of the source property listing;
identifying, a set of one or more target property listings within a first distance from the source property listing; and
selecting a target property listing from the set of one or more property listings.
18. The system of claim 12 , wherein the geographic area is one of: a squared area with a predetermined size of 1.3 km in length, or a circular area with a predetermined size of 1.3 km in diameter.