Natural language processing and machine-learning for event impact analysis

Patent metadata

Field	Value
Publication number	US-12346346-B2
Application number	US-202217738742-A
Country	US
Kind code	B2
Filing date	May 6, 2022
Priority date	May 7, 2021
Publication date	Jul 1, 2025
Grant date	Jul 1, 2025

Abstract

Systems and methods may use natural language processing (NLP) and machine-learning techniques to detect an impact that an event will have on a domain-specific topic. For example, the system may use multi-stage cleaning using a rules-based and an artificial intelligence (AI)-based filter to filter large quantities of event items that may not be relevant to a domain of interest. The AI-based filter may be trained using labeled event items that were previously known to be impactful. The system may cluster the cleaned event items to group similar event items and eliminate redundancy. The system may then predict and quantify the impact that events described by clustered event items will have on the domain-specific topic. Such prediction may be based on a classifier trained using various model features that correlate with impactful events, including prior similar events.

First claim

What is claimed is:

1. A method of assessing an impact of events in a domain of interest over time, comprising:
accessing, by a processor, a plurality of event items, each event item, from among the plurality of event items, comprising content that describes a corresponding event;
identifying, by the processor, for at least a first event item from among the plurality of event items, an event source of the first event item and a type of domain content that the event source provides;
filtering, by the processor, the plurality of event items via a rules-based filter that uses one or more domain-specific rules, the one or more domain-specific rules comprising at least a first domain-specific rule that filters in or out content based on whether or not the domain of interest matches the type of domain content that the event source provides;
filtering, by the processor, an output of the rules-based filter via an artificial intelligence (AI)-based filter trained via machine-learning on a training corpus comprising event items that are labelled as relating to the domain;
determining, by the processor, that a subset of event items, from among the plurality of event items, relate to the domain based on the rules-based filter and the AI-based filter;
determining, by the processor, that two or more event items, from among the subset of event items, both commonly describe an event-based on content similarity among the two or more event items; and
predicting, by the processor, that the event commonly described by the two or more event items will have an impact on a domain-specific topic.

2. The method of claim 1 , wherein the one or more domain-specific rules include a domain-specific rule that specifies a tag value for filtering, and wherein filtering the plurality of event items via the rules-based filter comprises:
extracting a metadata tag included in a tagged event item, from among the plurality of event items, wherein the tag indicates a corresponding domain to which the tagged event item relates; and
filtering in or out the tagged event item based on the tag and the tag value specified by the domain-specific rule.

3. The method of claim 1 , the method further comprising:
generating a word embedding for each event item in the output of rules-based filter, the word embedding being based on natural language text of each event item in the output of rules-based filter;
generating a word embedding for each item in the training corpus; and
filtering the output of the rules-based filter based on a comparison of the word embedding for each event item in the output of rules-based filter and the word embedding for each item in the training corpus.

4. The method of claim 1 , wherein determining that two or more event items, from among the subset of event items, describe the same event-based on content similarity among the two or more event items comprises:
generating a word embedding for each event item from among the two or more event items, each word embedding being based on natural language text of each event item from among the two or more event items; and
comparing the word embeddings for the two or more event items.

5. The method of claim 1 , wherein predicting that the event commonly described by the two or more event items will have an impact on the domain-specific topic comprises:
providing a plurality of feature values as an input to an AI-based impact classifier trained to correlate the plurality of feature values with various impacts on domain-specific topic.

6. The method of claim 5 , wherein the plurality of feature values comprise:
a geographic origin of the event and/or the quantitative value of the domain-specific topic.

7. The method of claim 1 , wherein predicting that the event commonly described by the two or more event items will have an impact on the domain-specific topic comprises:
generating an impact score that measures an impact of the event on the domain-specific topic, wherein the impact score is a number of standard deviations away from prior average standard deviations; and
comparing the impact score to a threshold value or range of values.

8. The method of claim 1 , wherein predicting that the event commonly described by the two or more event items will have an impact on the domain-specific topic comprises:
performing a check on one or more real-time data of the event;
comparing the real-time data to one or more real-time data threshold checks; and
adjusting the impact score based on the comparison.

9. The method of claim 8 , wherein the one or more real-time data threshold checks comprises: a cluster size threshold check that specifies a minimum number of event items that describe the same event.

10. The method of claim 8 , wherein the one or more real-time data threshold checks comprises: a quantitative value threshold check that requires a minimum movement in the quantitative value.

11. The method of claim 1 , further comprising:
accessing event items from a prior impactful events datastore, wherein each of the event items from the prior impactful events datastore is known to have impacted the domain-specific topic;
determining that content of the two or more event items is similar to content of at least one of the event items from the prior impactful events datastore; and
quantifying the impact that the event will have on the domain-specific topic based on the determination that the content of the two or more event items is similar to the content of the at least one of the event items from the prior impactful events datastore.

12. The method of claim 11 , wherein quantifying the impact comprising:
generating a time series of quantitative values after the event-based on the historic impact of the at least one of the event items from the prior impactful events datastore.

13. A system, comprising:
a processor programmed to:
access a plurality of event items, each event item, from among the plurality of event items, comprising content that describes a corresponding event;
identify, for at least a first event item from among the plurality of event items, an event source of the first event item and a type of domain content that the event source provides;
filter the plurality of event items via a rules-based filter that uses one or more domain-specific rules, the one or more domain-specific rules comprising at least a first domain-specific rule that filters in or out content based on whether or not the domain of interest matches the type of domain content that the event source provides;
filter an output of the rules-based filter via an artificial intelligence (AI)-based filter trained via machine-learning on a training corpus comprising event items that are labelled as relating to the domain;
determine that a subset of event items, from among the plurality of event items, relate to the domain based on the rules-based filter and the AI-based filter;
determine that two or more event items, from among the subset of event items, both commonly describe an event-based on content similarity among the two or more event items; and
predict that the event commonly described by the two or more event items will have an impact on a domain-specific topic.

14. The system of claim 13 , wherein the one or more domain-specific rules include a domain-specific rule that specifies a tag value for filtering, and wherein to filter the plurality of event items via the rules-based filter, the processor is further programmed to:
extract a metadata tag included in a tagged event item, from among the plurality of event items, wherein the tag indicates a corresponding domain to which the tagged event item relate

Assignees

• Refinitiv Us Organization Llc

Inventors

• Ramkumar Nitish
• Whyte Joseph
• Chu Ben
• Chan Jason
• Sircar Swarnima
• Rocha Kelvin
• Tuck Regan
• Melnikova Alena

Classifications

• G06F40/49

  using very large corpora, e.g. the web · CPC title

• G06F16/24564

  Applying rules; Deductive queries · CPC title

• G06Q40/04

  Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange · CPC title

• G06Q30/0202

  Market predictions or forecasting for commercial activities · CPC title

• G06Q10/0637

  Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals · CPC title