Generation of imposition type lists for input text

1 . A computing system for generating a tax imposition type prediction list from input text, comprising:

a processor and memory of a computing device, the processor being configured to execute a program using portions of memory to:

receive the input text;

divide the input text into sentences;

generate and output sentence embeddings using a sentence embeddings encoder based on the sentences;

identify matches in the input text with geographical jurisdictions listed in a table comprising imposition types and their corresponding geographical jurisdictions;

generate and output an imposition types list based on the matches identified in the input text;

train a candidate embeddings encoder on imposition names corresponding to imposition types contained in the imposition types list;

generate and output candidate embeddings using the candidate embeddings encoder trained on the imposition names corresponding to imposition types contained in the imposition types list;

perform a vectorized cosine similarity search between the candidate embeddings and the sentence embeddings to generate and output a scored sentence list of sentences comprising cosine similarity scores corresponding to respective top scoring imposition types for each of the sentences in the input text;

aggregate the cosine similarity scores in the scored sentence list to determine the top scoring imposition types in the input text; and

generate and output the tax imposition type prediction list containing the top scoring imposition types in the input text, thereby processing the input text to identify the top scoring imposition types as relevant imposition types based on jurisdictional information in the table comprising imposition types and their corresponding geographical jurisdictions, wherein

the sentence embeddings and the candidate embeddings are one-dimensional tensor vectors configured to be processed as query embeddings and corpus embeddings in a semantic search function which performs the vectorized cosine similarity search.

2 . The computing system of claim 1 , wherein the sentence embeddings encoder is a transformer.

3 . The computing system of claim 2 , wherein the sentence embeddings encoder is a Sentence-BERT (Bidirectional Encoder Representations from Transformers) encoder.

4 . The computing system of claim 1 , wherein the candidate embeddings encoder is a transformer.

5 . The computing system of claim 4 , wherein the candidate embeddings encoder is a Sentence-BERT encoder.

6 . The computing system of claim 1 , wherein a predetermined cosine similarity score threshold is used to determine the top scoring imposition types.

7 . The computing system of claim 1 , wherein the scored sentence list comprises a list with at least one entry for each sentence, each entry being a list of dictionaries with an imposition type key and a score value, sorted by decreasing cosine similarity scores indicated by the score values.

8 . The computing system of claim 1 , wherein the table comprises a currently supported imposition type table comprising currently supported imposition types and their corresponding geographical jurisdictions and a currently unsupported imposition type table comprising currently unsupported imposition types and their corresponding geographical jurisdictions.

9 . A method for generating a tax imposition type prediction list from input text, the method comprising steps to:

receive the input text;

divide the input text into sentences;

generate and output sentence embeddings using a sentence embeddings encoder based on the sentences;

identify matches in the input text with geographical jurisdictions listed in a table comprising imposition types and their corresponding geographical jurisdictions;

generate and output an imposition types list based on the matches identified in the input text;

train a candidate embeddings encoder on imposition names corresponding to imposition types contained in the imposition types list;

generate and output candidate embeddings using the candidate embeddings encoder trained on the imposition names corresponding to imposition types contained in the imposition types list;

perform a vectorized cosine similarity search between the candidate embeddings and the sentence embeddings to generate and output a scored sentence list of sentences comprising cosine similarity scores corresponding to respective top scoring imposition types for each of the sentences in the input text;

aggregate the cosine similarity scores in the scored sentence list to determine the top scoring imposition types in the input text; and

generate and output the tax imposition type prediction list containing the top scoring imposition types in the input text, thereby processing the input text to identify the top scoring imposition types as relevant imposition types based on jurisdictional information in the table comprising imposition types and their corresponding geographical jurisdictions, wherein

the sentence embeddings and the candidate embeddings are one-dimensional tensor vectors configured to be processed as query embeddings and corpus embeddings in a semantic search function which performs the vectorized cosine similarity search.

10 . The method of claim 9 , wherein the sentence embeddings encoder is a transformer.

11 . The method of claim 10 , wherein the sentence embeddings encoder is a Sentence-BERT (Bidirectional Encoder Representations from Transformers) encoder.

12 . The method of claim 9 , wherein the candidate embeddings encoder is a transformer.

13 . The method of claim 12 , wherein the candidate embeddings encoder is a Sentence-BERT encoder.

14 . The method of claim 9 , wherein a predetermined cosine similarity score threshold is used to determine the top scoring imposition types.

15 . The method of claim 9 , wherein the scored sentence list comprises a list with at least one entry for each sentence, each entry being a list of dictionaries with an imposition type key and a score value, sorted by decreasing cosine similarity scores indicated by the score values.

16 . The method of claim 9 , wherein the table comprises a currently supported imposition type table comprising currently supported imposition types and their corresponding geographical jurisdictions and a currently unsupported imposition type table comprising currently unsupported imposition types and their corresponding geographical jurisdictions.

17 . A computing system for generating a tax imposition type prediction list from a tax law article, the computing system comprising:

a processor and memory of a computing device, the processor being configured to execute a program using portions of memory to:

receive input text of the tax law article;

generate and output sentence embeddings using a sentence embeddings encoder based on sentences in the tax law article;

generate and output a tax imposition types list based on the tax law article;

train a candidate embeddings encoder on imposition names corresponding to imposition types contained in the tax imposition types list;

generate and output candidate embeddings using the candidate embeddings encoder trained on the imposition names corresponding to tax imposition types contained in the tax imposition types list;

perform a vectorized cosine similarity search between the candidate embeddings and the sentence embeddings to generate and output a scored sentence list comprising cosine similarity scores corresponding to respective top scoring tax imposition types for each of the sentences in the input text;

aggregate the cosine similarity scores in the scored sentence list comprising cosine similarity scores corresponding to respective top scoring tax imposition types for each of the sentences in the input text; and

generate and output the tax imposition type prediction list containing the top scoring tax imposition types in the input text, the highest scoring tax imposition type being the predicted tax imposition type, thereby processing the input text to identify the top scoring tax imposition types as relevant imposition types based on jurisdictional information contained in the tax imposition types list, wherein

the sentence embeddings and the candidate embeddings are one-dimensional tensor vectors configured to be processed as query embeddings and corpus embeddings in a semantic search function which performs the vectorized cosine similarity search.

18 . The computing system of claim 17 , wherein the tax imposition types list is generated and outputted based on a table comprising a currently supported tax imposition type table comprising currently supported tax imposition types and their corresponding geographical jurisdictions and a currently unsupported tax imposition type table comprising currently unsupported tax imposition types and their corresponding geographical jurisdictions.