Method implemented in computer system for analyzing document versions to identify shared document elements using machine learning, and non-transitory computer-readable storage medium
A present invention embodiment analyzes documents. A first document is received comprising a plurality of sentences that each include one or more words. A matrix is populated with the plurality of sentences, wherein each of the one or more words of each sentence in the matrix is encoded as a numerical value. The matrix is processed using a machine learning model to generate a first feature map. The first feature map is compared to a second feature map of a corresponding second document to identify a shared document element between the first document and the second document based on a common feature in the first feature map and the second feature map. The shared document element is indicated via a user interface.
1 . A computer-implemented method of analyzing documents comprising:
receiving a first document comprising a plurality of sentences that each include one or more words;
populating a matrix with the plurality of sentences, wherein each of the one or more words of each sentence in the matrix is encoded as a numerical value;
processing the matrix using a machine learning model to generate a first feature map;
comparing the first feature map to a second feature map of a second document to identify a shared document element between the first document and the second document based on a common feature in the first feature map and the second feature map, wherein comparing the first feature map to the second feature map comprises comparing a first vector of the first feature map and a second vector of the second feature map to a codebook comprising an embedding space of codebook vectors to determine that the first vector and the second vector are both closest to a same codebook vector; and
indicating the shared document element via a user interface.
2 . The computer-implemented method of claim 1 , wherein the machine learning model comprises an encoder, a decoder, and the codebook.
3 . The computer-implemented method of claim 2 , wherein the machine learning model is trained using a plurality of training matrices by:
processing each training matrix using the encoder, to generate a first training feature map;
processing the first training feature map by accessing embeddings in the codebook to fetch a codebook embedding vector for each vector of the first training feature map that is most similar to each vector of the first training feature map;
generating a second training feature map that includes codebook embedding vectors fetched from the codebook; and
processing the second training feature map, using the decoder, to generate a reconstructed training matrix, wherein the reconstructed training matrix is compared to the training matrix for each iteration of training to learn embeddings for the encoder, the decoder, and the codebook.
4 . The computer-implemented method of claim 1 , wherein the machine learning model is trained using a plurality of training matrices obtained from a plurality of training documents that include different versions of documentation for same products.
5 . The computer-implemented method of claim 1 , wherein each of the one or more words of each sentence in the matrix is encoded using one-hot encoding.
6 . The computer-implemented method of claim 1 , wherein indicating the shared document element comprises applying a same visual indicator to the shared document element in the first document and the second document.
7 . The computer-implemented method of claim 1 , wherein the comparing the first feature map to the second feature map comprises using a nearest-neighbor mapping to compare the first vector of the first feature map and the second vector of the second feature map to the embedding space of codebook vectors.
8 . A computer system for analyzing documents comprising:
one or more memories; and
at least one processor coupled to the one or more memories, wherein the at least one processor is configured to:
receive a first document comprising a plurality of sentences that each include one or more words;
populate a matrix with the plurality of sentences, wherein each of the one or more words of each sentence in the matrix is encoded as a numerical value;
process the matrix using a machine learning model to generate a first feature map;
compare the first feature map to a second feature map of a second document to identify a shared document element between the first document and the second document based on a common feature in the first feature map and the second feature map, wherein to compare the first feature map to the second feature map, the at least one processor is configured to compare a first vector of the first feature map and a second vector of the second feature map to a codebook comprising an embedding space of codebook vectors to determine that the first vector and the second vector are both closest to a same codebook vector; and
indicate the shared document element via a user interface.
9 . The computer system of claim 8 , wherein the machine learning model comprises an encoder, a decoder, and the codebook.
10 . The computer system of claim 9 , wherein to train the machine learning model using a plurality of training matrices, the at least one processor is configured to:
process each training matrix, using the encoder, to generate a first training feature map;
process the first training feature map by accessing embeddings in the codebook to fetch a codebook embedding vector for each vector of the first training feature map that is most similar to each vector of the first training feature map;
generate a second training feature map that includes codebook embedding vectors fetched from the codebook; and
process the second training feature map, using the decoder, to generate a reconstructed training matrix, wherein the reconstructed training matrix is compared to the training matrix for each iteration of training to learn embeddings for the encoder, the decoder, and the codebook.
11 . The computer system of claim 8 , wherein the machine learning model is trained using a plurality of training matrices obtained from a plurality of training documents that include different versions of documentation for same products.
12 . The computer system of claim 8 , wherein each of the one or more words of each sentence in the matrix is encoded using one-hot encoding.
13 . The computer system of claim 8 , wherein indicating the shared document element comprises applying a same visual indicator to the shared document element in the first document and the second document.
14 . The computer system of claim 8 , wherein to compare the first feature map to the second feature map, the at least one processor is configured to use a nearest-neighbor mapping to compare the first vector of the first feature map and the second vector of the second feature map to the embedding space of codebook vectors.
15 . A non-transitory computer-readable medium storing a set of instructions for analyzing documents, the set of instructions comprising:
one or more instructions that, when executed by one or more processors of a device, cause the device to:
receive a first document comprising a plurality of sentences that each include one or more words;
populate a matrix with the plurality of sentences, wherein each of the one or more words of each sentence in the matrix is encoded as a numerical value;
process the matrix using a machine learning model to generate a first feature map;
compare the first feature map to a second feature map of a second document to identify a shared document element between the first document and the second document based on a common feature in the first feature map and the second feature map, wherein the one or more instructions, to cause the device to compare the first feature map to the second feature map, cause the device to compare a first vector of the first feature map and a second vector of the second feature map to a codebook comprising an embedding space of codebook vectors to determine that the first vector and the second vector are both closest to a same codebook vector; and
indicate the shared document element via a user interface.
16 . The non-transitory computer-readable medium of claim 15 , wherein the machine learning model comprises an encoder, a decoder, and the codebook.
17 . The non-transitory computer-readable medium of claim 16 , wherein the one or more instructions further cause the device to train the machine learning model using a plurality of training matrices, and wherein the one or more instructions, to cause the device to train the machine learning model, cause the device to:
process each training matrix, using the encoder, to generate a first training feature map;
process the first training feature map by accessing embeddings in the codebook to fetch a codebook embedding vector for each vector of the first training feature map that is most similar to each vector of the first training feature map;
generate a second training feature map that includes codebook embedding vectors fetched from the codebook; and
process the second training feature map, using the decoder, to generate a reconstructed training matrix, wherein the reconstructed training matrix is compared to the training matrix for each iteration of training to learn embeddings for the encoder, the decoder, and the codebook.
18 . The non-transitory computer-readable medium of claim 15 , wherein the machine learning model is trained using a plurality of training matrices obtained from a plurality of training documents that include different versions of documentation for same products.
19 . The non-transitory computer-readable medium of claim 15 , wherein each of the one or more words of each sentence in the matrix is encoded using one-hot encoding.
20 . The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions, to cause the device to indicate the shared document element, cause the device to apply a same visual indicator to the shared document element in the first document and the second document.