Automatically cross-linking application programming interfaces

• US 10,628,527 B2
• Filed: 04/26/2018
• Issued: 04/21/2020
• Est. Priority Date: 04/26/2018
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Williams, Kyle Mark, Zitouni, Imed
• Primary Examiner(s): Patel, Shreyans A
• Application Number:
  US15/963,871
• Publication Number:
  US 20190332667A1
• Time in Patent Office: 726 Days
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/9024 Graphs; Linked lists G06F16/9027 takes precedence
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06N 3/088 Non-supervised learning, e.g. competitive learning
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition
• CPC Class Codes: G06N 5/046 Forward inferencing; Production systems

Abstract

A method for automatically cross-linking a plurality of APIs in an artificial intelligence (AI) graph structure comprises maintaining an AI graph structure defining a plurality of API-agnostic semantic entities, a plurality of function nodes, a plurality of input-adapter edges, and a plurality of output adapter edges. The method further comprises cross-linking a new function from a new API by computer-analyzing documentation of the new API with a natural language processing (NLP) machine in order to recognize the new function, and updating the AI graph data structure to include a new function node based on the new function.

Claims

1. A method for automatically cross-linking a plurality of application programming interfaces (APIs) in an artificial-intelligence graph data structure, the method comprising:

maintaining an artificial intelligence (AI) graph data structure, wherein the AI graph data structure defines;

a plurality of API-agnostic semantic entities, each API-agnostic semantic entity associated with an API-agnostic semantic data type and including a semantic data socket configured to hold a semantic data value having the API-agnostic semantic data type;

a plurality of function nodes, each function node representing a function made available by an API and including an input data socket and an output data socket;

a plurality of input adapter edges, each input adapter edge connecting a semantic data socket of an API-agnostic semantic entity to an input data socket of a function node; and

a plurality of output adapter edges, each output adapter edge connecting an output data socket of a function node to a semantic data socket of an API-agnostic semantic entity; and

for a target API of a plurality of APIs;

computer-analyzing documentation describing the target API with a natural language processing machine previously trained to identify;

a new function made available by the target API,

an upstream API-agnostic semantic entity corresponding to an API-specific input parameter of the function, and

a downstream API-agnostic semantic entity corresponding to an API-specific output parameter of the function; and

updating the AI graph data structure to include;

a new function node representing the new function;

an input adapter edge connecting a semantic data socket of the upstream API-agnostic semantic entity to an input data socket of the new function node, wherein the input data socket corresponds to the API-specific input parameter; and

an output adapter edge connecting an output data socket of the function node to a semantic data socket of the downstream API-agnostic semantic entity, wherein the output data socket corresponds to the API-specific output parameter.

2. The method of claim 1, wherein the natural language processing machine is previously trained via supervised learning on a plurality of labelled APIs, wherein a labelled API includes documentation describing an API annotated by one or more trainers to include one or more semantic labels, wherein a semantic label indicates an API-specific parameter of the API and an API-agnostic semantic entity corresponding to the API-specific parameter.

3. The method of claim 1, wherein computer-analyzing the documentation describing the target API with the natural language processing machine includes:

computer-identifying a natural language feature indicating the new function made available by the target API;

computer-identifying a natural language feature indicating an API-specific input parameter of the new function;

computer-identifying the upstream API-agnostic semantic entity based on the natural language feature;

computer-identifying a natural language feature indicating an API-specific output parameter of the new function; and

computer-identifying the downstream API-agnostic semantic entity based on the natural language feature.

4. The method of claim 1, further comprising:

computer-recognizing an API-agnostic semantic entity including a semantic data socket without an incoming connection via an output adapter edge;

automatically updating the AI graph data structure to include a botlet node, wherein the botlet node includes;

a botlet output data socket configured to transmit a semantic data value having an API-agnostic semantic data type associated with the API-agnostic semantic entity, and

a botlet output edge connecting the botlet output data socket to the semantic data socket of the API-agnostic semantic entity.

5. The method of claim 4, further comprising:

posing a disambiguation question via a natural language user interface;

receiving a response to the disambiguation question via the natural language user interface; and

operating a natural language response processing machine to output a computer-readable description of the response to the question;

wherein the botlet output data socket is configured to transmit the semantic data value based on the computer-readable description of the response.

6. The method of claim 1, wherein:

the input data socket of the new function node is configured to receive an input data value for parametrizing the new function, the input data value having an API-specific input data type of an API-specific input parameter of the new function; and

the input adapter edge between the semantic data socket of the upstream API-agnostic semantic entity and the input data socket of the new function node is configured to ensure compatibility between an API-agnostic semantic data type associated with the upstream API-agnostic semantic entity and the API-specific input data type of the API-specific input parameter.

7. The method of claim 1, wherein:

the output data socket of the new function node is configured to transmit an output data value of the new function, the output data value having an API-specific output data type of an API-specific output parameter of the new function; and

the output adapter edge between the output data socket of the new function node and the semantic data socket of the downstream API-agnostic semantic entity is configured to ensure compatibility between the API-specific output data type of the API-specific output parameter and an API-agnostic semantic data type associated with the downstream API-agnostic semantic entity.

8. The method of claim 1, wherein documentation describing each different target API of the plurality of APIs is received from an API discovery service configured to track, on a computer network, a plurality of available APIs accessible via the computer network and to provide, for each available API, documentation describing the available API in a computer-readable format.

9. The method of claim 1, further comprising:

receiving a query;

based on the query, selecting a target function node of the AI graph data structure;

executing a target function represented by the target function node.

10. The method of claim 9, further comprising:

computer-analyzing the query with a natural language query processing machine to computer-recognize a computer-readable description of an intent associated with the query, and selecting the target function node previously associated with the intent.

11. The method of claim 9, wherein executing the target function includes:

selecting a path through the AI graph data structure ending at the target function node; and

executing a plurality of other functions represented by other function nodes in the path through the AI graph data structure.

12. The method of claim 11, wherein the path through the AI graph data structure is a highest ranked candidate path based on a previously learned path selection policy, and wherein the previously learned path selection policy ranks a plurality of candidate paths ending at the target function node with a machine learning system previously trained to output, for each candidate path, a score indicating a quality of the path.

13. The method of claim 11, further comprising:

receiving feedback indicating a quality of path selection; and

adjusting a previously learned path selection policy based on the received feedback.

14. The method of claim 1, further comprising pruning a low-quality path from the graph data structure by removing, from the graph data structure, one or more API-agnostic semantic entities, function nodes, input adapter edges, and/or output adapter edges.

15. A method for automatically cross-linking a plurality of application programming interfaces APIs, the method comprising:

for a first API of the plurality of APIs;

computer-analyzing documentation describing the first API with a natural language processing (NLP) machine previously trained to identify natural language features associated with APIs, the NLP identifying a first function made available by the first API, and identifying a first natural language feature associated with an API-specific input parameter of the first function; and

computer-recognizing, with an artificial intelligence (AI) machine previously trained to associate natural language features with API-agnostic semantic entities, an API-agnostic semantic entity associated with the natural language feature;

for a second, different API of the plurality of APIs;

computer-analyzing documentation describing the second API with the NLP machine, the NLP machine identifying a second function made available by the second API, and identifying a second natural language feature associated with an API-specific output parameter of the second function; and

computer-recognizing, with the AI machine, an association between the second natural language feature and the API-agnostic semantic entity identified for the first API.

16. The method of claim 15, further comprising providing an artificial-intelligence graph data structure representing the first function and the second function as cross-linked by the API-agnostic semantic entity, wherein the artificial-intelligence graph data structure is useable to:

invoke the second API to execute the second function, to determine an output data value of the second function, the output data value having an API-specific output data type;

update a semantic data socket of the API-agnostic semantic entity to hold a semantic data value having an API-agnostic semantic data type, wherein the semantic data value is based on the output data value; and

invoke the first API to execute the first function, wherein the API-specific input parameter of the first function is parameterized with an input data value having an API-specific input data type, and wherein the input data value is based on the semantic data value.

17. The method of claim 15, wherein the API-specific input data type of the first function is different than the API-specific output data type of the second function.

18. The method of claim 15, wherein the natural language processing machine is previously trained via supervised learning on a plurality of labelled APIs, wherein a labelled API includes documentation describing an API annotated by one or more trainers to include one or more semantic labels, wherein a semantic label indicates an API-specific parameter of the API and an API-agnostic semantic entity corresponding to the API-specific parameter.

19. A computerized personal assistant, comprising:

a graph machine configured to access an AI graph data structure defining;

a plurality of API-agnostic semantic entities, each API-agnostic semantic entity associated with an API-agnostic semantic data type and including a semantic data socket configured to hold a semantic data value having the API-agnostic semantic data type;

a plurality of function nodes, each function node representing a function made available by an API and including an input data socket and an output data socket;

a plurality of input adapter edges, each input adapter edge connecting a semantic data socket of an API-agnostic semantic entity to an input data socket of a function node; and

a plurality of output adapter edges, each output adapter edge connecting an output data socket of a function node to a semantic data socket of an API-agnostic semantic entity; and

a query machine configured to;

receive a query;

operate a natural language query processing machine to computer-recognize a computer-readable description of an intent associated with the query;

pass the computer-readable description of the intent to the graph machine;

wherein the graph machine is further configured to;

select a target function node of the AI graph data structure based on the computer-readable description of the intent associated with the query;

select a path through the AI graph data structure ending at the target function node; and

invoke one or more APIs to execute a plurality of other functions represented by function nodes in the path to the target function and to execute the target function.

20. The computerized personal assistant of claim 19, wherein the natural language processing machine is previously trained via supervised learning on a plurality of labelled APIs, wherein a labelled API includes documentation describing an API annotated by one or more trainers to include one or more semantic labels, wherein a semantic label indicates an API-specific parameter of the API and an API-agnostic semantic entity corresponding to the API-specific parameter.