Event detection through text analysis using dynamic self evolving/learning module

Patent metadata

Field	Value
Publication number	US-9544361-B2
Application number	US-201414558342-A
Country	US
Kind code	B2
Filing date	Dec 2, 2014
Priority date	Dec 2, 2013
Publication date	Jan 10, 2017
Grant date	Jan 10, 2017

Abstract

A system and method for detecting events based on input data from a plurality of sources. The system may receive input from a plurality of sources containing information about possible events. A method for event detection involves pre-processing and normalizing a data input from a plurality of sources, extracting and disambiguating events and entities, associate event and entities, correlate events and entities associated from a data input to results from a different data source to determine if an event has occurred, and store the detected events in a data storage.

First claim

What is claimed is:

1. A computer-implemented method comprising:
identifying, by a computer, a plurality of features in a data stream associated with a data source;
assigning, by the computer, an initial confidence score to each respective feature of the plurality of features;
determining, by the computer, a candidate score of one or more features of the plurality of features using the initial confidence score of each respective feature in the one or more features, based upon a number of occurrences of each respective feature identified in the one or more features;
identifying, by the computer, an event candidate when the candidate score of the one or more features satisfies a predetermined threshold, wherein the event candidate is defined by the one or more features;
automatically determining, by the computer, whether the one or more features identified in the data stream as the event candidate satisfy one or more event models in a categorization table, based upon the computer comparing the one or more features of the data stream against the one or more event models, wherein an event concept store comprises a non-transitory machine-readable memory storing the one or more event models; and
responsive to the computer determining that the one or more features from the data stream fail to satisfy at least one event model in at least one categorization table stored in the event concept store:
comparing, by the computer, the one or more features against one or more uncategorized event models in an uncategorized event table stored in the event concept store wherein the uncategorized event table store records associated with new unknown event models;
storing, by the computer, the one or more features as a new uncategorized event model in the uncategorized event table, in response to determining the one or more features fail to satisfy at least one uncategorized event model;
generating, by the computer, an increased confidence score of an uncategorized event model of the one or more uncategorized event models in the uncategorized event table when the one or more features satisfy the uncategorized event model of the one or more uncategorized event models in the uncategorized event table;
calculating, by the computer, a probability score for the event candidate based on a likelihood that the one or more features represent a new event model;
comparing, by the computer, the increased confidence score of the event candidate and the probability score of the event candidate with a pre-determined threshold score of the uncategorized event model of the one or more uncategorized event models in the uncategorized event table; and
storing, by the computer, the uncategorized event model in the categorization table when the increased confidence score and the probability score is higher than or matches the pre-determined threshold score.

2. The method according to claim 1 , further comprising receiving, by the computer, one or more data streams from one or more data sources.

3. The method according to claim 1 , wherein determining whether the one or more features identified in the data stream satisfy at least one of the one or more event models further comprises:
automatically comparing, by the computer, the one or more features against the one or more event models of one or more categorization tables stored in the event concept store.

4. The method according to claim 3 , further comprising:
automatically identifying, by the computer, a subset of one or more features from the one or more features, according to an event model of the one or more event models; and
determining, by the computer, a score for the subset of the one or more features, based on a weight assigned to each respective feature in the subset according to the event model.

5. The method according to claim 4 , wherein each respective feature of the subset of the one or more features satisfies a threshold value of the event model.

6. The method according to claim 1 , further comprising:
determining, by the computer, whether the one or more features identified in the data stream satisfy an event model in the categorization table; and
automatically identifying, by the computer, in one or more categorization tables an event model associated with the event candidate, in response to determining that the one or more features satisfy the event model of the one or more event categorization tables.

7. An event extraction system comprising:
an event concept store that comprises a non-transitory machine-readable memory storing one or more event models, one or more event categorization tables, and an uncategorized event table,
wherein each event model of the one or more event models is associated with an event candidate and further comprises a threshold event score and a set of one or more features,
wherein each event categorization table comprises one or more known event models, and
wherein the at least one uncategorized event table comprises a set of one or more features, a set of one or more entities, and a set of one or more topics, each associated with one or more uncategorized event models; and

an event category validation processor communicatively coupled to a network and executes an event extractor module configured to:
receive a set of extracted features, a set of extracted entities, and a set of extracted topics from a normalized pre-processed data stream associated with a data source, wherein at least one feature of the set of extracted features, the set of extracted entities, and the set of extracted topics is an event candidate;
assign an initial confidence score to the set of extracted features, the set of extracted entities, and the set of extracted topics;
calculate a score using the initial confidence score of the set of extracted features, the set of extracted entities, and the set of extracted topics that repeat themselves or overlap in different data sources;
determine that an event candidate occurred when the calculated score is greater than a predetermined threshold;

compare each of the sets in which the at least one feature is the event candidate with the one or more event categorization tables to determine whether the extracted features, entities, and topics, correspond to a known event model; and
compare each of the sets with the uncategorized event table to determine whether the extracted features, entities, and topics, correspond with at least one uncategorized event model, wherein the event extractor module is further configured to increase a confidence score corresponding to the at least one uncategorized event model when the set of extracted features are similar to the at least one uncategorized event model in the uncategorized event table, wherein the uncategorized event table store records associated with new unknown event models and a probability score indicating features representing new event models, and wherein the event extractor module is further configured to calculate a probability score indicating features, wherein the probability score is determined based on a likelihood that the features represents new event models; and store the at least one uncategorized event model as a new event model in the one or more event categorization tables responsive to determining the confidence score and the probability score corresponding to the at least one uncategorized event model satisfies a threshold score.

8. The system according to claim 7 , wherein the event extractor module is further configured to:
automatically identify a subset of one or more features from the set of extracted features, according to an event model of the one or more event models; and
determine a score for the subset of the one or more features, based on a weight assigned to each respective feat

Assignees

• Qbase Llc

Inventors

• Boddhu Sanjay
• Dave Rakesh

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G06F40/279

  Recognition of textual entities · CPC title

• H04L67/10 Primary

  in which an application is distributed across nodes in the network (software deployment G06F8/60; multiprogramming arrangements G06F9/46) · CPC title

• G06F16/951

  Indexing; Web crawling techniques · CPC title

• G06F9/542 Primary

  Event management; Broadcasting; Multicasting; Notifications · CPC title