Method and system for securely storing private data in a semantic analysis system
Disclosed is an approach for allowing an entity to perform semantic analysis in a SaaS semantic analysis platform upon private data possessed by one or more entities. In one or more embodiments, separate processing pipelines may be provided to the plurality of entities thereby keeping private data secure within the semantic analysis platform. In one or more embodiments, a common processing pipeline is provide with data associated a first entity being assigned a first identification code, and data associated with a second entity being assigned a second identification code.
1. A method comprising:
receiving an uncategorized message from a first tenant of a plurality of tenants of a multi-tenant cloud service;
responsive to receiving the uncategorized message:
identifying, by the multi-tenant cloud service, a semantic space associated with the first tenant based at least in part on a first attribute of the first tenant, wherein the semantic space is unique and dedicated to the first tenant relative to other tenants of the plurality of tenants, wherein the semantic space includes a collection of vector values that are unique to the first tenant relative to the other tenants of the plurality of tenants and represent one or more semantic meanings of one or more terms within the semantic space; and
identifying, by the multi-tenant cloud service, a set of one or more topic vectors associated with the first tenant;
generating, by the multi-tenant cloud service, a message vector as a function of at least a subset of vector values from the collection of vector values that are unique to the first tenant relative to the other tenants of the plurality of tenants;
performing, by the multi-tenant cloud service, a semantic analysis on the uncategorized message, the semantic analysis comparing the message vector to the set of one or more topic vectors associated with the first tenant to categorize the uncategorized message for the first tenant.
2. The method of claim 1 , wherein the semantic space is identified based at least in part on a unique identifier associated with the first tenant.
3. The method of claim 1 , wherein the semantic space is further identified based at least in part on a language associated with the uncategorized message.
4. The method of claim 1 , wherein the semantic space associated with the first tenant defines a meaning of a particular term for the first tenant; wherein a second semantic space associated with a second tenant defines a different meaning of the particular term for the second tenant; wherein the first semantic space is used to generate message vectors for the first tenant; wherein the second semantic space is used to generate message vectors for the second tenant.
5. The method of claim 1 , wherein comparing the message vector to the set of one or more topic vectors associated with the first tenant to categorize the uncategorized message for the first tenant comprising computing a similarity measure for a particular topic vector based on said comparing the message vector to the set of one or more topic vectors; and determining whether the similarity measure satisfies a threshold; and responsive to determining that the similarity measure satisfies a threshold, categorizing the uncategorized message for a topic associated with the particular topic vector.
6. The method of claim 1 , wherein generating the message vector comprises: generating a snippet of the uncategorized message; wherein the snippet includes a threshold number of words on either side of an anchor term in the uncategorized message; and generating a semantic vector that includes values representing a semantic significance for terms included in the snippet of the uncategorized message, wherein the semantic significance for the terms is determined based at least in part on the semantic space associated with the first tenant.
7. The method of claim 1 , further comprising: performing semantic analysis of uncategorized messaged for the plurality of tenants using separate private pipelines; wherein different private pipelines perform semantic analysis only for a tenant to which the private pipeline is dedicated.
8. The method of claim 1 , further comprising: assigning the uncategorized message a unique identifier associated with the first tenant; wherein the multi-tenant cloud service identifies the one or more topic vectors and the semantic space based at least in part on the unique identifier associated with the first tenant.
9. The method of claim 1 , wherein the uncategorized message is received through a semantic application programming interface (API) that exposes the semantic analysis of the multi-tenant cloud service to the first tenant.
10. One or more non-transitory computer-readable media storing instructions which, when executed by one or more hardware processors, cause:
receiving an uncategorized message from a first tenant of a plurality of tenants of a multi-tenant cloud service;
responsive to receiving the uncategorized message:
identifying, by the multi-tenant cloud service, a semantic space associated with the first tenant based at least in part on a first attribute of the first tenant, wherein the semantic space is unique and dedicated to the first tenant relative to other tenants of the plurality of tenants, wherein the semantic space includes a collection of vector values that are unique to the first tenant relative to the other tenants of the plurality of tenants and represent one or more semantic meanings of one or more terms within the semantic space; and
identifying, by the multi-tenant cloud service, a set of one or more topic vectors associated with the first tenant;
generating, by the multi-tenant cloud service, a message vector as a function of at least a subset of vector values from the collection of vector values that are unique to the first tenant relative to the other tenants of the plurality of tenants;
performing, by the multi-tenant cloud service, a semantic analysis on the uncategorized message, the semantic analysis comparing the message vector to the set of one or more topic vectors associated with the first tenant to categorize the uncategorized message for the first tenant.
11. The media of claim 10 , wherein the semantic space is identified based at least in part on a unique identifier associated with the first tenant.
12. The media of claim 10 , wherein the semantic space is further identified based at least in part on a language associated with the uncategorized message.
13. The media of claim 10 , wherein the semantic space associated with the first tenant defines a meaning of a particular term for the first tenant; wherein a second semantic space associated with a second tenant defines a different meaning of the particular term for the second tenant; wherein the first semantic space is used to generate message vectors for the first tenant; wherein the second semantic space is used to generate message vectors for the second tenant.
14. The media of claim 10 , wherein comparing the message vector to the set of one or more topic vectors associated with the first tenant to categorize the uncategorized message for the first tenant comprises: computing a similarity measure for a particular topic vector based on said comparing the message vector to the set of one or more topic vectors; and determining whether the similarity measure satisfies a threshold; and responsive to determining that the similarity measure satisfies a threshold, categorizing the uncategorized message for a topic associated with the particular topic vector.
15. The media of claim 10 , wherein generating the message vector comprising generating a snippet of the uncategorized message; wherein the snippet includes a threshold number of words on either side of an anchor term in the uncategorized message; and generating a semantic vector that includes values representing a semantic significance for terms included in the snippet of the uncategorized message, wherein the semantic significance for the terms is determined based at least in part on the semantic space associated with the first tenant.
16. The media of claim 10 , wherein the instructions further cause: performing semantic analysis of uncategorized messaged for the plurality of tenants using separate private pipelines; wherein different private pipelines perform semantic analysis only for a tenant to which the private pipeline is dedicated.
17. The media of claim 10 , wherein the instructions further cause: assigning the uncategorized message a unique identifier associated with the first tenant; wherein the multi-tenant cloud service identifies the one or more topic vectors and the semantic space based at least in part on the unique identifier associated with the first tenant.
18. The media of claim 10 , wherein the uncategorized message is received through a semantic application programming interface (API) that exposes the semantic analysis of the multi-tenant cloud service to the first tenant.
19. A system comprising:
one or more hardware processors;
one or more non-transitory computer readable media storing instructions which, when executed by the one or more hardware processors, cause:
receiving an uncategorized message from a first tenant of a plurality of tenants of a multi-tenant cloud service;
responsive to receiving the uncategorized message:
identifying, by the multi-tenant cloud service, a semantic space associated with the first tenant based at least in part on a first attribute of the first tenant, wherein the semantic space is unique and dedicated to the first tenant relative to other tenants of the plurality of tenants, wherein the semantic space includes a collection of vector values that are unique to the first tenant relative to the other tenants of the plurality of tenants and represent one or more semantic meanings of one or more terms within the semantic space; and
identifying, by the multi-tenant cloud service, a set of one or more topic vectors associated with the first tenant;
generating, by the multi-tenant cloud service, a message vector as a function of at least a subset of vector values from the collection of vector values that are unique to the first tenant relative to the other tenants of the plurality of tenants;
performing, by the multi-tenant cloud service, a semantic analysis on the uncategorized message, the semantic analysis comparing the message vector to the set of one or more topic vectors associated with the first tenant to categorize the uncategorized message for the first tenant.
20. The system of claim 19 , wherein the subset of vector values from the collection of vector values include a respective vector value selected from the collection of vector values unique to the first tenant for each respective word in a snippet within the uncategorized message; wherein generating the message vector comprises averaging the subset of vector values.