System for document ranking by phrase importance

Patent metadata

Field	Value
Publication number	US-11328022-B2
Application number	US-202016821311-A
Country	US
Kind code	B2
Filing date	Mar 17, 2020
Priority date	Feb 1, 2016
Publication date	May 10, 2022
Grant date	May 10, 2022

Abstract

A method, apparatus, system, and computer program product for searching for documents. A query comprising search terms is received. A search is performed on a collection of documents for each search term in the search terms. Search results are received for the search terms. Each search result includes documents from the collection of documents containing a search term in the search terms and scores associated with the documents. Aggregate scores for the search terms in the search results are associated with each document in the documents returned to form combined results in which a combined result in the combined results comprises the document and an aggregate score associated with the document. The documents in the combined results are ranked based on the aggregate scores to form a ranked order for the documents. The combined results with the aggregate scores in the ranked order are displayed on a display system.

First claim

What is claimed is:

1. A document search system comprising:
a computer system; and
a search manager in the computer system, wherein the search manager is configured to:
receive a query comprising search terms;
perform a search on a collection of documents for each search term in the search terms;
receive search results for each search term, wherein each search result includes documents from the collection of documents containing the search term in question and scores associated with the documents, wherein a score in the scores indicates an importance of the search term to each document in the collection of documents searched using the search term;
associate aggregate scores for the search terms in the search results with the documents returned in the searches to form combined results in which a combined result in the combined results comprises the document in the documents and an aggregate score associated with the document;
rank the documents in the combined results in the search based on the aggregate scores to form a ranked order for the documents; and
display, on a display system, the combined results with the aggregate scores in the ranked order and a graphical display of individual scores for the search terms associated with the document in each combined result in the combined results, wherein the graphical display comprises a heat map displaying a rank of each search result in the search results.

2. The document search system of claim 1 , wherein in displaying, on the display system, the combined results with the aggregate scores in the ranked order, the search manager is configured to:
display, on the display system, a number of highest ranked combined results in the combined results with the aggregate scores for the number of highest ranked combined results in the ranked order.

3. The document search system of claim 1 , wherein in displaying, on the display system, the combined results in the ranked order with the aggregate scores and a graphical display of individual scores for the search terms associated with the document in each combined result in the combined results, the search manager is configured to:
identify a score for each search term contributing to an aggregate score in each combined result in the combined results;
select a graphical indicator for each score identified for each search term contributing to the aggregate score in each combined result in the combined results to form graphical indicators that indicate how often the search term is present in a document relative to other search terms in the document, wherein the graphical indicator comprises one of a color, an icon, a pictogram, an ideogram, an image, text, animation, bolding, a line, or an arrow;
display the documents in a ranked order based on the combined results; and
display the graphical indicators for each of the documents in graphical association with the documents.

4. The document search system of claim 3 , wherein in displaying the graphical indicators for each of the documents in a graphical association with the documents, the search manager is configured to:
display the graphical indicators for each of the documents in the graphical association with the documents as a heat map.

5. The document search system of claim 1 , wherein the search manager is configured to:
compute a commonality score for each of the search terms in the document, wherein the commonality score for the search term based on a number of times the search term is present in the document and the number of times the search term is present in all documents subject to the search.

6. The document search system of claim 1 , wherein the scores are term frequency-inverse document frequency (tf-idf) scores.

7. The document search system of claim 1 , wherein the search manager is configured to:
identify a set of key terms in a selected document;
determine features for the key terms in the selected document in which the document, the set of key terms, and the features form a training dataset; and
train a machine learning model using the training dataset, wherein the machine learning model is trained to determine an importance of candidate search terms from an input document recommend a set of the candidate search terms.

8. The document search system of claim 7 , wherein the search manager is configured to:
add negative samples to the training dataset in which the negative samples include terms that are not key terms.

9. The document search system of claim 7 , wherein the search manager is configured to:
generate, by the machine learning model, the candidate search terms in response to receiving the selected document; and
display the set of the candidate search terms in a graphical user interface on the display system.

10. The document search system of claim 1 , wherein the search manager is configured to:
identify a prior query comprising prior search terms;
determine cooccurrences of the prior search terms in which the cooccurrences of each prior search term with a number of other prior search terms in the prior query comprises how many times a prior search term was used in prior searches in a search term repository and how many times each of the number of other prior search terms was used in the prior searches with the prior search term in the search term repository;

store the prior query and the cooccurrences of each prior search term with a number of other prior search terms in the prior query in a training data set; and
train a machine learning model to generate candidate search terms from an input search term.

11. The document search system of claim 10 , wherein the search manager is configured to:
generate, by the machine learning model, candidate search terms in response to receiving the input search term; and
display the candidate search terms in a graphical user interface on the display system.

12. The document search system of claim 1 , wherein in performing the search on a collection of documents for each search term in the search terms, the search manager is configured to:
perform the search on a collection of documents for each search term in the search terms using a fuzzy searching process.

13. A method for searching for documents, the method comprising:
receiving, by a computer system, a query comprising search terms;
performing, by the computer system, a search on a collection of documents for each search term in the search terms;
receive search results for each search term, wherein each search result includes documents from the collection of documents containing the search term in question and scores associated with the documents, wherein a score in the scores indicates an importance of the search term to each document in the collection of documents searched using the search term;
associating, by the computer system, aggregate scores for the search terms in the search results with each document in the documents returned in the searches to form combined results in which a combined result in the combined results comprises the document and an aggregate score associated with the document;

ranking, by the computer system, the documents in the combined results in the search based on the aggregate scores to form a ranked order for the documents; and
displaying, by the computer system on a display system, the combined results with the aggregate scores in the ranked order and a graphical display of individual scores for the search terms associated with the document in each combined result in the combined results, wherein the graphical display comprises a heat map displaying a rank of each search result in the search results.

Assignees

• S&P Global Inc

Inventors

• Watson William
• Yong Lawrence
• Di Mingyang
• Nourbakhsh Armineh

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06F16/90344 Primary

  by using string matching techniques · CPC title

• A61K31/573

  substituted in position 21, e.g. cortisone, dexamethasone, prednisone or aldosterone · CPC title

• A61K9/1075 Primary

  Microemulsions or submicron emulsions; Preconcentrates or solids thereof; Micelles, e.g. made of phospholipids or block copolymers (A61K9/0026 takes precedence) · CPC title