Suppression and Deduplication of Place-Entities on Online Social Networks

Abstract

In one embodiment, a method includes receiving, from a client system, a search query, identifying a plurality of place-entity nodes matching the search query, determining that one or more of the identified place-entity nodes are low-quality place-entity nodes based on the place names of the place-entity nodes having anomalous language characteristics, filtering the plurality of identified place-entity nodes to exclude the determined low-quality place-entity nodes, and generating one or more search results corresponding to the plurality of filtered place-entity nodes.

What is claimed is: 
     
         1 . A method comprising, by one or more computing systems:
 receiving, from a client system, a search query;   identifying a plurality of place-entity nodes matching the search query;   determining that one or more of the identified place-entity nodes are low-quality place-entity nodes based on the place names of the place-entity nodes having anomalous language characteristics;   filtering the plurality of identified place-entity nodes to exclude the determined low-quality place-entity nodes; and   generating one or more search results corresponding to the plurality of filtered place-entity nodes.   
     
     
         2 . The method of  claim 1 , further comprising:
 sending, to the client system, instructions for presenting one or more of the search results responsive to the query.   
     
     
         3 . The method of  claim 1 , further comprising identifying each of the post-filtered place-entity nodes as a valid place-entity node. 
     
     
         4 . The method of  claim 1 , further comprising identifying each of the determined low-quality place-entity nodes as a junk place-entity node. 
     
     
         5 . The method of  claim 1 , further comprising deleting each of the determined low-quality place-entity nodes. 
     
     
         6 . The method of  claim 1 , wherein identifying the plurality of place-entity nodes matching the search query comprises:
 identifying the plurality of place-entity nodes in a heterogeneous graph, wherein the heterogenous graph comprises a plurality of place-entity nodes, user nodes, and n-gram nodes, and wherein each place-entity node corresponds to a place-entity associated with a particular geographic location.   
     
     
         7 . The method of  claim 6 , wherein determining that one or more of the identified place-entity nodes are low-quality place-entity nodes comprises:
 assigning, for each identified place-entity node, an initial quality-score for the place-entity node; and   calculating, for each identified place-entity node, a final quality-score for the identified place-entity node, wherein each final quality-score is calculated by iteratively propagating the initial quality-scores corresponding to the identified place-entity nodes, respectively, through the place-entity nodes, user nodes, and n-gram nodes of the heterogeneous graph until the quality-scores associated with the place-entity nodes, user nodes, and n-gram nodes reach convergence, wherein the final quality-score for one or more of the identified place-entity nodes is below a threshold quality-score based on the place names of the place-entity nodes having anomalous language characteristics with respect to other place-entity nodes in the heterogeneous graph.   
     
     
         8 . The method of  claim 7 , filtering the plurality of identified place-entity nodes to exclude the determined low-quality place-entity nodes comprises:
 excluding each place-entity node having a final quality-score below the threshold quality-score.   
     
     
         9 . The method of  claim 7 , wherein the initial quality-score for each identified place-entity node represents a measure of quality of the place-entity node. 
     
     
         10 . The method of  claim 7 , wherein the initial quality-score for each identified place-entity node is based at least in part on social-networking interactions represented by one or more edges connected to the place-entity node. 
     
     
         11 . The method of  claim 10 , wherein the social-networking interactions comprise one or more of check-ins, likes, comments, views, or reviews of a place-entity corresponding to the identified place-entity node. 
     
     
         12 . The method of  claim 7 , wherein iteratively propagating the initial quality-scores corresponding to the identified place-entity nodes, respectively, through the place-entity nodes, n-gram nodes, and user nodes of the heterogeneous graph comprises performing a label-propagation process on the heterogeneous graph. 
     
     
         13 . The method of  claim 7 , wherein iteratively propagating the initial quality-scores corresponding to the identified place-entity nodes, respectively, through the place-entity nodes, n-gram nodes, and user nodes of the heterogeneous graph comprises an iterative propagation process, each iteration of the propagation process comprising:
 calculating, for each n-gram node connected to one or more identified place-entity nodes having an associated quality-score, an n-gram-node score based on the associated quality-scores of the place-entity nodes connected to the n-gram node;   calculating, for each user node connected to one or more identified place-entity nodes, a user-node score based on the associated quality-scores of the place-entity nodes connected to the user node;   calculating, for each identified place-entity node, an updated quality-score based on:
 the n-gram-node scores associated with the n-gram nodes connected to the identified place-entity node; and 
 the user-node scores associated with the user nodes connected to the identified place-entity node; and 
   based on the updated quality-scores:
 if the updated quality-scores associated with the identified place-entity nodes have converged, then exiting the iterative propagation process and using the updated quality-scores as the final quality-scores, 
 else performing another iteration of the propagation process. 
   
     
     
         14 . The method of  claim 13 , wherein calculating the n-gram-node score comprises averaging the quality-scores associated with the place-entity nodes connected to the n-gram node. 
     
     
         15 . The method of  claim 13 , wherein calculating the updated quality-score comprises determining a weighted average value of:
 the n-gram-node scores associated with the n-gram nodes connected to the identified place-entity node; and   the user-node scores associated with the user nodes connected to the identified place-entity node.   
     
     
         16 . The method of  claim 15 , wherein calculating the updated quality-score further comprises multiplying each of the n-gram-node scores and user-node scores by a respective weighting factor. 
     
     
         17 . The method of  claim 13 , wherein the updated quality-scores associated with the identified place-entity nodes have converged when the sum of updated quality-scores, n-gram-node scores, and user-nodes for the heterogeneous graph changes between two iterations by a value less than a threshold value. 
     
     
         18 . The method of  claim 1 , wherein:
 the search query is associated with a further particular identified place-entity node, the particular identified place-entity node being associated with a particular canonical place-entity node, and   the one or more search results comprise:
 if the particular identified place-entity node corresponds to one of the determined low-quality place-entity nodes, then the response comprises a reference to the particular canonical place-entity node; 
 else the response comprises a reference to the particular identified place-entity node. 
   
     
     
         19 . One or more computer-readable non-transitory storage media embodying software that is operable when executed to:
 receive, from a client system, a search query;   identify a plurality of place-entity nodes matching the search query;   determine that one or more of the identified place-entity nodes are low-quality place-entity nodes based on the place names of the place-entity nodes having anomalous language characteristics;   filter the plurality of identified place-entity nodes to exclude the determined low-quality place-entity nodes; and   generate one or more search results corresponding to the plurality of filtered place-entity nodes.   
     
     
         20 . A system comprising: one or more processors; and a memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:
 receive, from a client system, a search query;   identify a plurality of place-entity nodes matching the search query;   determine that one or more of the identified place-entity nodes are low-quality place-entity nodes based on the place names of the place-entity nodes having anomalous language characteristics;   filter the plurality of identified place-entity nodes to exclude the determined low-quality place-entity nodes; and   generate one or more search results corresponding to the plurality of filtered place-entity nodes.