Systems and methods for mapping a term to a vector representation in a semantic space

1 . An apparatus for mapping a sentence to a vector representation in a semantic space, wherein the apparatus comprises:

at least a computing device, wherein the computing device comprises:

a memory; and

at least a processor communicatively connected to the memory, wherein the memory contains instructions configuring the at least a processor to:

access, using the at least a processor, a document corpus;

predict, using the at least a processor, a plurality of query sentences appearing in the document corpus prior to a semantic search of the document corpus;

identify, using the at least a processor, a first set of sentences that includes the plurality of predicted query sentences appearing in the document corpus;

identify, using the at least a processor, a second set of sentences, the second set being a subset of the first set of sentences;

store, in a first data structure, correlation data generated from the first set of sentences;

store, in a second data structure, a set of vector representations computed from the second set of sentences;

provide, using the at least a processor, the first data structure to a run-time embedding engine configured to, during run-time operation, determine a vector representation of a first query sentence among the first set of sentences;

rank, using the at least a processor, a plurality of documents of the document corpus based on the vector representation of the first query sentence and vector representations of sentences of the plurality of documents using a neural network model, wherein the neural network is configured as a dependency parser that is trained to generate a dependency parse of each sentence in the document corpus; and

display, using a graphical user interface of a downstream device, the ranked plurality of documents, wherein displaying the ranked plurality of documents further comprises:

receiving, using the graphical user interface, user feedback; and

modifying, using the at least a processor, the ranked plurality of documents based on the user feedback.

2 . The apparatus of claim 1 , wherein determining the vector representation of the first query sentence among the first set of sentences comprises:

receiving the first query sentence;

determining that the first query sentence is included in the first set of sentences; and

retrieving correlation data corresponding to the first query sentence from the data structure;

wherein the vector representation of the first query sentence is determined based on a correlation between the first query sentence and one or more sentences in the second set of sentences, the correlation being determined based on the correlation data.

3 . The apparatus of claim 2 , wherein the vector representation of the first query sentence is determined by mapping the first query sentence to its corresponding vector representation based on the retrieved correlation data.

4 . The apparatus of claim 1 , wherein identifying the first set of sentences comprises accessing a database of sentences corresponding to a log of sentences used in historical searches.

5 . The apparatus of claim 1 , wherein the computing device is further configured to generate the vector representations of the sentences of the plurality of documents.

6 . The apparatus of claim 1 , wherein the computing device is further configured to determine, at run-time, a vector representation of a second query sentence not in the first set of sentences.

7 . The apparatus of claim 1 , wherein the computing device is further configured to look up, at run-time, a vector representation of a second query sentence in the second set of sentences.

8 . The apparatus of claim 1 , wherein the correlation data corresponds to an aggregation of intermediate results that are generated during the computing of the set of vector representations.

9 . The apparatus of claim 1 , wherein identifying the first set of sentences comprises identifying sentences whose frequency in the document corpus relative to the frequency of its constituent terms exceeds a predetermined threshold.

10 . The apparatus of claim 1 , wherein identifying the second set of sentences comprises identifying sentences whose frequency in the document corpus relative to the frequency of its constituent terms exceeds a predetermined threshold.

11 . A method comprising:

accessing, using at least a processor, a document corpus;

predicting, using the at least a processor, a plurality of query sentences appearing in the document corpus prior to a semantic search of the document corpus;

identifying, using the at least a processor, a first set of sentences that includes the plurality of predicted query sentences appearing in the document corpus;

identifying, using the at least a processor, a second set of sentences, the second set being a subset of the first set of sentences;

storing, in a first data structure, correlation data generated from the first set of sentences;

storing, in a second data structure, a set of vector representations computed from the second set of sentences;

providing, using the at least a processor, the first data structure to a run-time embedding engine configured to, during run-time operation, determine a vector representation of a first query sentence among the first set of sentences; and

ranking, using the at least a processor, a plurality of documents of the document corpus based on the vector representation of the first query sentence and vector representations of sentences of the plurality of documents using a neural network model, wherein the neural network is configured as a dependency parser that is trained to generate a dependency parse of each sentence in the document corpus; and

display, using a graphical user interface of a downstream device, the ranked plurality of documents, wherein displaying the ranked plurality of documents further comprises:

receiving, using the graphical user interface, user feedback; and

modifying, using the at least a processor, the ranked plurality of documents based on the user feedback.

12 . The method of claim 11 , wherein determining the vector representation of the first query sentence among the first set of sentences comprises:

receiving the first query sentence;

determining that the first query sentence is included in the first set of sentences; and

retrieving correlation data corresponding to the first query sentence from the data structure;

wherein the vector representation of the first query sentence is determined based on a correlation between the first query sentence and one or more sentences in the second set of sentences, the correlation being determined based on the correlation data.

13 . The method of claim 12 , wherein the vector representation of the first query sentence is determined by mapping the first query sentence to its corresponding vector representation based on the retrieved correlation data.

14 . The method of claim 11 , wherein identifying the first set of sentences comprises accessing a database of sentences corresponding to a log of sentences used in historical searches.

15 . The method of claim 11 , wherein the method further comprises generating the vector representations of the sentences of the plurality of documents.

16 . The method of claim 11 , wherein the method further comprises determining, at run-time, a vector representation of a second query sentence not in the first set of sentences.

17 . The method of claim 11 , wherein the method further comprises looking up, at run-time, a vector representation of a second query sentence in the second set of sentences.

18 . The method of claim 11 , wherein the correlation data corresponds to an aggregation of intermediate results that are generated during the computing of the set of vector representations.

19 . The method of claim 11 , wherein identifying the first set of sentences comprises identifying sentences whose frequency in the document corpus relative to the frequency of its constituent terms exceeds a predetermined threshold.

20 . The method of claim 11 , wherein identifying the second set of sentences comprises identifying sentences whose frequency in the document corpus relative to the frequency of its constituent terms exceeds a predetermined threshold.