Computing page relevance for tabular contents from a document

A system comprising:

• at least one processor;
• and
• an application that, when executing on the at least one processor, performs operations comprising:
• receiving a table type data associated with a portion of a plurality of documents, wherein the table type data comprises one or more attributes related to one or more columns of a table type;
• executing a term model using frequency features and positioning features extracted from a page of a document to generate a term score of the page, wherein executing the term model comprises:
• determining a first frequency of the frequency features based on a term of the one or more attributes appearing in the page of the document;
• determining a second frequency of the frequency features, wherein the second frequency is associated with an inverse document frequency determined based on a presence of the term within the plurality of documents;
• determining the positioning features of the term in the page based on coordinates of the term within the page;
• and
• determining the term score of the page for the term based on the first frequency, the second frequency, and the positioning features;
• executing an embeddings model using text embeddings, image embeddings, and layout embeddings to generate an embeddings score of the page, wherein executing the embeddings model comprises:
• generating attribute embeddings based on the one or more attributes;
• and
• generating the embeddings score based on a comparison between the attribute embeddings to the text embeddings, the image embeddings, and the layout embeddings;
• executing a summary model using a page summary generated with a language model to generate a summary score of the page;
• executing a neighborhood model using a set of pages from the document to generate a neighborhood score of the page;
• executing a combination model using the term score, the embeddings score, the summary score, and the neighborhood score to generate a combined score of the page, wherein the combined score corresponds to a likelihood the page includes a table of the table type;
• and
• presenting the combined score to indicate the page includes the table of the table type, wherein the combined score is presented with the page via a display of a computing device.

A non-transitory computer readable medium comprising instructions executable by at least one processor to perform operations comprising:

• receiving a table type data associated with a portion of a plurality of documents, wherein the table type data comprises one or more attributes related to one or more columns of a table type;
• executing a term model using frequency features and positioning features extracted from a page of a document to generate a term score of the page, wherein executing the term model comprises:
• determining a first frequency of the frequency features based on a term of the one or more attributes appearing in the page of the document;
• determining a second frequency of the frequency features, wherein the second frequency is associated with an inverse document frequency determined based on a presence of the term within the plurality of documents;
• determining the positioning features of the term in the page based on coordinates of the term within the page;
• and
• determining the term score of the page for the term based on the first frequency, the second frequency, and the positioning features;
• executing an embeddings model using text embeddings, image embeddings, and layout embeddings to generate an embeddings score of the page, wherein executing the embeddings model comprises:
• generating attribute embeddings based on the one or more attributes;
• and
• generating the embeddings score based on a comparison between the attribute embeddings to the text embeddings, the image embeddings, and the layout embeddings;
• executing a summary model using a page summary generated with a language model to generate a summary score of the page;
• executing a neighborhood model using a set of pages from the document to generate a neighborhood score of the page;
• executing a combination model using the term score, the embeddings score, the summary score, and the neighborhood score to generate a combined score of the page, wherein the combined score corresponds to a likelihood the page includes a table of the table type;
• and
• presenting the combined score to indicate the page includes the table of the table type, wherein the combined score is presented with the page via a display of a computing device.