Generating and presenting a text-based graph object

Patent metadata

Field	Value
Publication number	US-12169526-B2
Application number	US-202117484670-A
Country	US
Kind code	B2
Filing date	Sep 24, 2021
Priority date	Sep 24, 2021
Publication date	Dec 17, 2024
Grant date	Dec 17, 2024

Abstract

The present disclosure relates to extracting key concepts from digital content items and determining associations between the key concepts and candidate terms for use in generating and presenting a correlation graph object based on the determined associations. For example, systems described herein involve determining frequency of co-occurrence between various key concepts and applying a classification model (e.g., a zero-shot classification model) to the key concepts and candidate terms to determine associations between the key concepts and candidate terms for a given domain of interest. The systems further involve generating a graph object and processing graph queries in a way that enables fast and efficient presentation of slices of the graph object that provide a visual depiction of key concepts and edges representing associations between pairs of the key concepts.

First claim

What is claimed is:

1. A method, comprising:
extracting a plurality of key concepts from a collection of digital content items, wherein extracting the plurality of key concepts includes:
applying a first model to text content of the collection of digital content items to identify a first set of terms from the text content, the first model comprising a rule-based model including rules for identifying certain types of terms within the text content of the collection of digital content items; and
applying a second model to the text content to identify a set of candidate terms from the first set of terms, the second model comprising a machine learning model trained to identify one or more key topics within a given text based on the given text and one or more terms within the given text indicated as one or more certain types of terms;

receiving the set of candidate terms associated with a domain of interest by way of a graph query application over a network;
applying a zero-shot classification model to the plurality of key concepts and the set of candidate terms to determine, for each key concept from the plurality of key concepts, a candidate term from the set of candidate terms associated with a respective key concept; and
generating a correlation graph object for the collection of digital content items, the correlation graph object including:
a plurality of nodes associated with respective key concepts from the plurality of key concepts, each node including an indication of a candidate term from the set of candidate terms associated with a corresponding key concept; and
a plurality of edges connecting the plurality of nodes, the plurality of edges being associated with pairs of key concepts corresponding to nodes connected by the respective edges, each edge of the plurality of edges including a correlation value based on a plurality of pre-calculated segment correlation values for associated segments of time, the plurality of pre-calculated segment correlation values based on frequency of co-occurrence of a respective pair of key concepts within subsets of the collection of digital content items associated with the respective segments of time.

2. The method of claim 1 , further comprising:
applying a sentiment model to the collection of digital content items to determine sentiment scores for co-occurring concepts from the plurality of key concepts, the sentiment model being trained to determine a sentiment score for a given digital content item,
wherein the correlation value is further based on sentiment scores for digital content items within which the respective pair of key concepts co-occurs.

3. The method of claim 1 , wherein the zero-shot classification model comprises a zero-shot classification model having been trained based on training data independent from the set of candidate terms associated with the domain of interest.

4. The method of claim 1 , wherein the set of candidate terms includes a first plurality of terms related to a domain of interest and a non-classification term not related to the domain of interest, wherein the zero-shot classification model associates a subset of key concepts from the plurality of key concepts with the non-classification term.

5. The method of claim 4 , wherein the subset of key concepts are excluded from the correlation graph object based on association with the non-classification term by the zero-shot classification model.

6. The method of claim 1 , further comprising:
receiving a graph query including one or more key concepts and a candidate term; and
providing a presentation of a portion of the correlation graph object including a first subset of nodes from the plurality of nodes corresponding to the one or more key concepts and a second subset of nodes associated with other key concepts, the second subset of nodes being determined based on correlation values for respective edges that connect the second subset of nodes to the first subset of nodes within the correlation graph object.

7. The method of claim 1 , further comprising:
receiving a graph query including one or more candidate terms; and
providing a presentation of a portion of the correlation graph object including a set of nodes from the plurality of nodes with key concepts associated with the one or more candidate terms.

8. The method of claim 1 , further comprising:
receiving a query including an indicated range of time; and
providing a presentation of the correlation graph object including nodes and associated edges based on correlation values determined for the indicated range of time.

9. The method of claim 1 , wherein the collection of digital content items comprise text content and visual content.

10. The method of claim 8 , wherein the segments of time include predetermined durations of time, and wherein the indicated range of time includes a selection of one or more segments of time within a duration of time inclusive of the collection of digital content items.

11. The method of claim 1 , wherein generating the correlation graph object includes:
excluding edges for a first set of pairs of key concepts from the correlation graph object based on co-occurrence of the first set of key concepts co-occurring less than a minimum threshold value within the collection of digital content items; and
excluding edges for a second set of pairs of key concepts from the correlation graph object based on co-occurrence of the second set of key concepts co-occurring greater than a maximum threshold value within the collection of digital content items.

12. The method of claim 11 , wherein the minimum threshold value is a first threshold percentile, and wherein the maximum threshold value is a second threshold percentile.

13. The method of claim 1 , wherein extracting the plurality of key concepts from the collection of digital content items includes mining text content from a collection of social networking posts publicly available from one or more social networking platforms.

14. A system, comprising:
one or more processors;
memory in electronic communication with the one or more processors; and instructions stored in the memory, the instructions being executable by the one or more processors to:
extract a plurality of key concepts from a collection of digital content items, wherein extracting the plurality of key concepts includes:
applying a first model to text content of the collection of digital content items to identify a first set of terms from the text content, the first model comprising a rule-based model including rules for identifying certain types of terms within the text content of the collection of digital content items; and
applying a second model to the text content to identify a set of candidate terms from the first set of terms, the second model comprising a machine learning model trained to identify one or more key topics within a given text based on the given text and one or more terms within the given text indicated as one or more certain types of terms;

receive the set of candidate terms associated with a domain of interest by way of a graph query application over a network;
apply a zero-shot classification model to the plurality of key concepts and the set of candidate terms to determine, for each key concept from the plurality of key concepts, a candidate term from the set of candidate terms associated with a respective key concept; and
generate a correlation graph object for the collection of digital content items, the correlation graph object including:
a plurality of nodes associated with respective key concepts from the plurality of key concepts, each node including an indication of a c

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Del Villar Paul Pangilinan
• Zeng Xiaofei
• Xu Mingyang

Classifications

• G06F16/9024

  Graphs; Linked lists (G06F16/9027 takes precedence) · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N5/025

  Extracting rules from data · CPC title

• G06F16/906 Primary

  Clustering; Classification · CPC title

• G06F16/358 Primary

  Browsing; Visualisation therefor · CPC title